Learning Orientation Information for Robotic Soccer using Neural Nets

Size: px
Start display at page:

Download "Learning Orientation Information for Robotic Soccer using Neural Nets"

Transcription

1 Learning Orientation Information for Robotic Soccer using Neural Nets Jacky Baltes John Anderson Department of Computer Science University of Manitoba Winnipeg, Canada August 6, 2003 Abstract Robotic soccer teams using both local and global vision traditionally rely on a set of pre-determined markers (e.g., a group of small colored circles mounted on the top surface of the robot) to provide easy targets for visual analysis in order to determine the team membership, identity, and orientation of robots in the visual field. This approach requires calibration before any competition, as well as agreement in advance on color codes different enough between teams to avoid recognition errors at run-time. Even after extensive calibration, small lighting variations can cause extensive misidentification. In this paper, we examine an alternative approach: training a neural network to recognize the orientation of the robots on a team so that visual tracking can occur in real time without special markers of any kind. This paper describes the design and implementation of such an approach, and shows the results of an empirical evaluation of this approach. 1 Introduction In robotic soccer leagues such as the F180 league, visual tracking of robots relies on the use of colored markings identifying teams and individuals, with the placement of those markers in such a fashion that individual orientation can easily be determined. While the F180 league uses an overhead camera for global vision, similar requirements exist for local vision teams as well. There is a significant set of difficulties that accompany this approach, both prior to play and during play itself. For this approach to work in practice, the vision system must be finely calibrated to recognize the colors used for these markers across the range of lighting over the playing field. In advance of play Colors must also be chosen far enough apart in the spectrum from one another to avoid misidentification. During play itself, lighting must vary as little as possible during play, since any variation would change the apparent colors of the markers and misidentification would ensue. Even if lighting variance can be controlled to an unrealistic degree, shadows and occlusion also cause similar problems in visual recognition. A promising alternative to this approach is to attempt to learn to recognize the identity and orientation of robots from their image without using additional markers. There are three major points that would make this an improvement over existing approaches: 1

2 1. Recognition without markers would scale much better to more complex domains. Marker-based approaches do not scale well in situations where larger playing fields or larger teams of robots are employed, for example. 2. Recognition without markers would allow the number of colors used on the robot to be reduced. This would in turn directly support scaling, since more objects could be tracked over a greater visual field if color information could be simplified. 3. The ability to infer robots orientation without prior knowledge allows a team to infer the orientation and identity of the opponent s robots. This allows for more sophisticated tactical decision making. For example, robots that have a strong kicker can be extremely dangerous. If the kicker is facing away from the current locus of activity, the situation is not as dangerous. What would ultimately be desirable is a system that could determine identity and orientation of robots from a visual image without an involved initial calibration process: simply allow it to view the field with the robots moving on it for a reasonably limited time, and the system would self-calibrate. Such a system would also ideally adjust itself as play ensued, allowing it compensate for small changes in lighting, shadow, etc. There are two sub-tasks to accomplishing this: first the ability to isolate the image of a moving robot from the overall visual field, and then the ability to recognize identity and orientation. Video servers already exist that can accomplish the former of these. The Doraemon video server [6, 1], for example, allows capture of moving objects and can supply both location information and a small bit-mapped graphic to any number of recipient machines connected to the video server over a network. This paper presents an approach that attempts to deal with the latter sub-task: take this video server data and extract a robot s orientation from each image frame, given that the robot has no additional features for identification or calibration. An artificial neural network is trained on time-sequenced image data where orientation and identity is provided. We then empirically test this training using real data. The remainder of this paper describes the approach employed and the evaluation of that approach. Section 2 gives an overview of related work in the area. The design and implementation of the neural network learner is explained in section 3. The results of an empirical evaluation of the system is described in section 4. The paper concludes in section 5. 2 Related Work In leagues such as the F180 league, team descriptions illustrate the heavy reliance on pre-defined markers for the recognition of object location and orientation (e.g.[3, 14]). While this is true of teams using global vision, local vision relies heavily upon this as well (e.g. [10]). Colored patches mounted on the robot are the most commonly-employed approach; however bar codes or geometrical features are also used [2]. Using geometric features or bar codes does not change the nature of the recognition problem itself, nor the disadvantages of requiring pre-defined markers that have already been discussed in section 1. Work continues on better ways of performing color calibration and real-time detection of colored blobs and other markers within this approach [13, 4]. Removing the ability to build a recognition and tracking system around pre-defined markers results in a much more difficult object recognition problem, since features for recognition must be discovered as part of the recognition process itself. Some of these features may at times be shadowed or otherwise occluded, but the fact that more subtle features than the obvious colored markers currently used can be employed should result in a more robust recognition and tracking process if these features can be learned. The fact that we are working with subtle features across a noisy image points to the use of a more decentralized approach that can consider recognition across the image as a whole. The natural choice for such a task is a neural-net based approach. Neural nets have been used extensively in image processing tasks, from face recognition [12] and recognition of the human form [7] to oil-spill detection from 2

3 satellite images and other forms of remote sensing [8, 9]. Their robustness in the face of noisy and uncertain data is well-known [11]. Within this range of tasks, it is important to distinguish between the problem of recognition or classification in single images as opposed to dealing with the much larger problem of ongoing recognition in real time over a sequence of temporal data. Detecting an oil spill, classifying vegetation, or or classifying clouds in a single satellite image, for example, are all simpler recognition tasks than performing the ongoing recognition over time that is required to track a cloud formation, ocean current, or other entity over time (e.g. [5]). These previous efforts at recognition over time show that neural networks appear to be a viable choice for performing real-time object recognition and tracking in a robotic soccer domain. The question we attempt to answer in this paper is whether a neural network can be trained to robustly recognize the orientation of a robot solely from a sequence of images, thus completely eliminating the need for special markers on the robot. The next section describes the design of a neural network for this task. Figure 1: A sequence of sample images taken and annotated by the videoserver Figure 2: A second sequence of sample images taken and annotated by the videoserver 3 Design In order to gather data suitable for use with a neural network, we use the output of the Doraemon video server [6, 1]. This server provides a small bitmapped image of moving object(s) on a playing field by analyzing frames from a camera looking for differences. Those differences allow the server to focus on portions of the image containing moving objects, and direction of change also allows orientation information to be derived. The system can therefore be used to record training data by assuming the robot is only being driven in a forward direction so that the orientation information provided will be accurate for reinforcement during training. For a test data set, we place one robot to serve as a test subject for recognition and drive it around at random (including across the lines of the field in order to expose the system to interference from the lines). Other stationary robots are also placed on the field so that the system is exposed to more varied examples of robot images and to avoid over-fitting. Two sets of sample images are shown in Figures 1 and 2. Figure 1 is a set using a robot platform based on a toy car, while Figure reffig:sample-input1 uses a robot based on the Lego Mindstorms platform. Both of these were used in evaluation trials described in section 4. These images are sent in the form of 64x32 bit pixmaps (in 24-bit color) with annotation information from the video server (stored in PPM format). Figure 3 shows the pixmap structure for the upper left image in Figure 1. The orientation shown is in radians, where 0 is approximately straight up on the playing field. We use a 3-layered feed-forward backpropagation network to perform recognition on these datasets. The network architecture is shown in Fig.4. The size of each image supplied by the video server is 6,144 bytes. We tested a neural net with 6,144 3

4 Figure 3: Sample pixmap of the image as transmitted by the videoserver. The pixmap is the one for the first image in Fig. 1. Figure 4: The Neural Network Architecture P6 # Object name=robot1,found=yes # Position x= ,y= ,z=50 #Theta= ,dx= ,dy= (64*32*3 Bytes of pixel data ) input nodes, but found quickly that the net would perform very poorly and training was extremely slow. Therefore, sub-sampling of the data is necessitated. All pixels in a 4x4 neighborhood are averaged to create a 16x8 pixmap and then fed to an input layer consisting of a corresponding 384 nodes. The hidden layer consists of 32 nodes, roughly 10% of the input layer. The output layer is a 1-hot encoding of the orientation angle discretized into 5-degree steps. Therefore, there are 72 output nodes, each corresponding to a 5-degree interval. For a correctly-trained neural net, exactly one of these nodes should have a high activation, with the others low. 3.1 Pre-processing of Input Images In addition to studying the efficacy of this approach with raw image data, we desired to investigate the effects of other representations of the input data on the performance of the system. Because edge information may be more important than pixel values alone, we applied 2x2 Sobel edge detection on the sub-sampled pixmaps, to produce sample images such as those shown in Figure 6. Since the resulting edge map is a gray scale image, the size of the input layer can be reduced. 4 Evaluation This section describes the results of a series of empirical evaluations of the neural network. Firstly, we determined the number of hidden units needed for the network. The results of this investigation are described in subsection 4.1. Secondly, we investigated the influence of the input representation (4x4 sub-sampled color image, 2x2 gray scale, and 2x2 sub-sampled edge map) on the accuracy and the speed of learning. Subsection 4.2. Thirdly, subsection 4.3 describes the investigation into the generalization ability of the neural network. Fourthly, the computational cost of classifying using a neural network is described in subsection Accuracy of Learning versus Number of Hidden Units The first experiment was used to determine a suitable number of hidden units for the network. Figure 7 shows the development of the mean squared error (MSE) for neural nets with 8, 16, 24, and 32 4

5 Figure 5: The output of edge detection applied to the sample images shown in Fig. 1 Figure 7: Evolution of the MSE for Varying Number of Hidden Nodes in the Toycar Image Sequence Figure 6: The output of edge detection applied to the sample images shown in Fig. 1 hidden units. There are two common error measures when using neural nets. The sum of the squared error (SSE) is the sum of the squared errors over the output units over the entire set of training examples (a so-called epoch). The MSE is the SSE divided by the number of training patterns in the epoch. In other words, it is the mean error for a given pattern. Unless otherwise specified, the network was trained using a learning rate of 0.3 and a momentum term of 0.2. The learning rate was slightly increased from the more popular 0.2 value to speed up learning. The training data was randomly reshuffled after each epoch to avoid over-fitting the neural network to the specific sequence of training images. The tests were run on a dual MP Athalon PC with 1 GB of RAM. On this system, a training run took about 30 minutes. We applied 4x4 sub-sampling to reduce the size of the image to 16x8 pixels and the number of nodes in the input layer to 384. Sub-sampling is necessary since otherwise the training of the neural net takes too long to be practical. As can be seen, the neural net with 8, 16, and 24 hidden nodes was not able to learn the correct orientation information from the image sequence. After 5000 epochs, the networks with 8, 16, and 24 hidden units had an accuracy of 23%, 59.5%, and 88%. The network with 32 hidden nodes The network with 32 nodes is able to learn the correct orientation from the training data and achieves 100% accuracy after 2000 epochs. Similar results were obtained when training the network with a 2x2 sub-sampled gray scale images. This results in a network with 512 input nodes. The network with 32 hidden units learned to classify the input images with 100% accuracy after 10,000 epochs over all 200 input images. 4.2 Input Representation versus Learning Accuracy After determining a suitable number for the number of hidden units, we evaluated the learning ability and speed of learning for three different representations of the input data. We compared the following three representations: 1. a 4x4 sub-sampled color image 5

6 Figure 8: Evolution of the MSE for different input representations for the Toycar Image Sequence Figure 9: Evolution of the MSE for different input representations for the Lego Image Sequence 2. a 2x2 sub-sampled gray scale image 3. a 2x2 sub-sampled edge map of the original image Figure 8 shows the MSE for different input representations. This figure shows that the representation makes little difference in the overall performance of the neural network. The accuracy for all three representations is similar (with an MSE of 0.10). Although the edge detected image improves quicker in the first few hundred epochs, the performance after longer training is worse than the color or gray scale representation. However, the differences in performance after training for more than 4000 epochs are too small and not statistically significant. We tested the neural networks and computed not only the accuracy based on the MSE, but also when applied to the task of returning the orientation (as one would use the neural network in the vision server). The results were also very similar. The neural network trained using color images was able to classify 99% of all images within 5 degrees. The neural network trained using gray scale or edge maps was able to classify 97% of all images correctly. There was no correlation between the missed images in our tests. That is, the three representations did mis-classify different images in our test. We are currently investigating more closely why these images posed problems for the neural network. The results of training the neural network on 200 sample images taken from the Lego image sequence are shown in Figure 9. The Lego robots have more features to work with. In fact, both the color and gray-scale representation terminated training after 4500 and 2500 epochs respectively with 100% accuracy. The small residual error shown in the graph is negligible. The edge representation again performed more poorly. After 5000 epochs, its classification accuracy was only 93%. The evaluation gives some indication that preprocessing and selecting features for the neural network is unnecessary and may decrease the performance of the neural network. This result is at first counter intuitive. However, a more careful analysis of the results show that the orientation of a robot depends more on gathering small pieces of evidence and combining it into a consistent view than extracting features such as edges. Each pixel only provides a small amount of information by itself and only its relationship to other pixels makes this information important. The edge information in the images is hard to extract and skews the results. For example, after checking some of the misclassified images in the Lego image sequence we found that the front edge of the robot was occluded by shadows in some of those images. Artificial neural networks, on the 6

7 other hand, are especially suited to combining large numbers of inputs into a consistent view. There are therefore useful and promising tools in this research. 4.3 Generalization Ability We then investigated the generalization ability of the neural net. We extracted 20 images (10%) at random from the set of 200 training images. After training the network on the training set, its performance on the 20 unseen instances was tested. The results of these tests were not as encouraging as the results of the previous experiments. Instead of classifying at least some of the previously unseen images correctly, the network was unable to generalize to unseen patterns well. Since it is infeasible to show all possible input patterns to the system, the generalization ability of the neural network is extremely important. Further investigation showed that the generalization ability of the network was limited since there were few examples for each 5 degree bucket. Simply increasing the number of training examples is not a workable solution since the training time increases. The generalization ability of the network can be improved by selecting a more representative training set. Images can be selected to generate a training set with equal proportions of all 5 degree step. 4.4 Computational Cost of Classification Another concern are the computational requirements of the Neural Net after training. We therefore run the classification program on a modern PC (1.6 GHz Athalon Processor) repeatedly. The speed of the classification using a fully connected 16x8 feed-forward network (i.e., 384 input units, 32 hidden units, 72 output units, weights) was around 0.07 msecs. per classification. In practice, this means that the neural net is fast enough to classify all 10 robots in a typical RoboCup F-180 game. 5 Conclusion This is a first step - we are currently working on adding to the process of matching visual images the ability to look at commands that were sent to the robots, giving a greater basis for reasoning about uncertainty during the tracking process. For example, if we are uncertain of a robot s location and orientation at the current time, we can start with the robot s last known location/orientation at previous time t 1. Knowing that the robot was sent a move-forward command at time t 1 allows an assumption of where the robot should be now, and this assumption can be used to improve the recognition process. This will assist in increasing the recognition capabilities in cases of occlusion, for example. The biggest obstacle to the use of neural nets to determine the orientation of a robot is the inadequate generalization ability of the neural net. As discussed previously, this is partly due to the fact that even in a set of 200 training examples, there are only few samples for each orientation. We are currently addressing this problem by preprocessing the training data to create a training set with similar number of examples for each degree. We are also currently exploring options to increase the speed of the classification after the Neural Net has been trained. A simple example is that once one of the output neurons has fired, the computation can terminate since the there should only be one activated output neuron. Another idea is to extract a small set of rules that can be tested faster than calculating the output of the entire network. References [1] John Anderson and Jacky Baltes. Doraemon user s manual [2] Jacky Baltes. Yuefei: Object orientation and id without additional markers. In Andreas Birk, Silvia Coradeschi, and Satoshi Tadokoro, editors, RoboCup-2001: Robot Soccer World Cup V, pages Springer Verlag,

8 [3] Andreas Birk, Silvia Coradeschi, and Satoshi Tadokoro, editors. RoboCup-2001: Robot Soccer World Cup V. Springer Verlag, Berlin, [4] Paulo Costa, Paulo Marques, Antonio Moriera, Armando Sousa, and Pedro Costa. Tracking and identifying in real time the robots of an f-180 team. In Proceedings of the Third International Robocup Workshop, IJCAI-99, Stockholm, IJCAI. [13] Mirk Simon, Sven Behnke, and Raul Rojas. Robust real time color tracking. In Peter Stone, Tucker Balch, and Gerhard Kraetszchmar, editors, RoboCup-2000: Robot Soccer World Cup IV, pages Springer Verlag, Berlin, [14] Peter Stone, Tucker Balch, and Gerhard Kraetszchmar, editors. RoboCup-2000: Robot Soccer World Cup IV. Springer Verlag, Berlin, [5] S. Cote and A. R. L. Tatnall. The hopfield neural network as a tool for feature tracking and recognition from satellite sensor images. International Journal of Remote Sensing, 18(4): , [6] Jacky Baltes et al. The doraemon video server [7] Tony Jan, Massimo Piccardi, and Thomas Hintz. Detection of suspicious pedestrian behavior using modified probabilistic neural networks. In Proceedings of Image and Vision Computing-02, pages , Auckland, New Zealand, November [8] Miroslav Kubat, Robert C. Holte, and Stan Matwin. Machine learning for the detection of oil spills in satellite radar images. Machine Learning, 30(2-3): , [9] Schaale M. and Furrer R. Land surface classification by neural networks. International Journal of Remote Sensing, 16(16): , [10] Emanuele Menegatti, Francesco Nori, Enrico Pagello, Carlo Pellizzari,, and David Spagnoli. Designing an omnidirectional vision system for a goalkeeper robot. In Andreas Birk, Silvia Coradeschi, and Satoshi Tadokoro, editors, RoboCup-2001: Robot Soccer World Cup V, pages Springer Verlag, Berlin, [11] Tom Mitchell. Machine Learning. McGraw-Hill, [12] Rein-Lien Hsu Mohamed. Face detection in color images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(5): ,

Horus: Object Orientation and Id without Additional Markers

Horus: Object Orientation and Id without Additional Markers Computer Science Department of The University of Auckland CITR at Tamaki Campus (http://www.citr.auckland.ac.nz) CITR-TR-74 November 2000 Horus: Object Orientation and Id without Additional Markers Jacky

More information

(a) (b) (c) Fig. 1. Omnidirectional camera: (a) principle; (b) physical construction; (c) captured. of a local vision system is more challenging than

(a) (b) (c) Fig. 1. Omnidirectional camera: (a) principle; (b) physical construction; (c) captured. of a local vision system is more challenging than An Omnidirectional Vision System that finds and tracks color edges and blobs Felix v. Hundelshausen, Sven Behnke, and Raul Rojas Freie Universität Berlin, Institut für Informatik Takustr. 9, 14195 Berlin,

More information

using an omnidirectional camera, sufficient information for controlled play can be collected. Another example for the use of omnidirectional cameras i

using an omnidirectional camera, sufficient information for controlled play can be collected. Another example for the use of omnidirectional cameras i An Omnidirectional Vision System that finds and tracks color edges and blobs Felix v. Hundelshausen, Sven Behnke, and Raul Rojas Freie Universität Berlin, Institut für Informatik Takustr. 9, 14195 Berlin,

More information

Nao recognition and coordination

Nao recognition and coordination Nao recognition and coordination Sander Nugteren August 24, 2012 1 Project In RoboCup soccer, it is important to have robots communicate and position themselves properly. This prevents them from getting

More information

Toward an Undergraduate League for RoboCup.

Toward an Undergraduate League for RoboCup. Toward an Undergraduate League for RoboCup. John Anderson, Jacky Baltes, David Livingston, Elizabeth Sklar, and Jonah Tower Department of Computer Science Columbia University New York, NY, 10027, USA sklar,jpt2002@cs.columbia.edu

More information

Robust and Accurate Detection of Object Orientation and ID without Color Segmentation

Robust and Accurate Detection of Object Orientation and ID without Color Segmentation 0 Robust and Accurate Detection of Object Orientation and ID without Color Segmentation Hironobu Fujiyoshi, Tomoyuki Nagahashi and Shoichi Shimizu Chubu University Japan Open Access Database www.i-techonline.com

More information

Visual object classification by sparse convolutional neural networks

Visual object classification by sparse convolutional neural networks Visual object classification by sparse convolutional neural networks Alexander Gepperth 1 1- Ruhr-Universität Bochum - Institute for Neural Dynamics Universitätsstraße 150, 44801 Bochum - Germany Abstract.

More information

Towards a Calibration-Free Robot: The ACT Algorithm for Automatic Online Color Training

Towards a Calibration-Free Robot: The ACT Algorithm for Automatic Online Color Training Towards a Calibration-Free Robot: The ACT Algorithm for Automatic Online Color Training Patrick Heinemann, Frank Sehnke, Felix Streichert, and Andreas Zell Wilhelm-Schickard-Institute, Department of Computer

More information

Traffic Signs Recognition using HP and HOG Descriptors Combined to MLP and SVM Classifiers

Traffic Signs Recognition using HP and HOG Descriptors Combined to MLP and SVM Classifiers Traffic Signs Recognition using HP and HOG Descriptors Combined to MLP and SVM Classifiers A. Salhi, B. Minaoui, M. Fakir, H. Chakib, H. Grimech Faculty of science and Technology Sultan Moulay Slimane

More information

Notes on Multilayer, Feedforward Neural Networks

Notes on Multilayer, Feedforward Neural Networks Notes on Multilayer, Feedforward Neural Networks CS425/528: Machine Learning Fall 2012 Prepared by: Lynne E. Parker [Material in these notes was gleaned from various sources, including E. Alpaydin s book

More information

Brainstormers Team Description

Brainstormers Team Description Brainstormers 2003 - Team Description M. Riedmiller, A. Merke, M. Nickschas, W. Nowak, and D. Withopf Lehrstuhl Informatik I, Universität Dortmund, 44221 Dortmund, Germany Abstract. The main interest behind

More information

Realtime Object Recognition Using Decision Tree Learning

Realtime Object Recognition Using Decision Tree Learning Realtime Object Recognition Using Decision Tree Learning Dirk Wilking 1 and Thomas Röfer 2 1 Chair for Computer Science XI, Embedded Software Group, RWTH Aachen wilking@informatik.rwth-aachen.de 2 Center

More information

Robust Color Choice for Small-size League RoboCup Competition

Robust Color Choice for Small-size League RoboCup Competition Robust Color Choice for Small-size League RoboCup Competition Qiang Zhou Limin Ma David Chelberg David Parrott School of Electrical Engineering and Computer Science, Ohio University Athens, OH 45701, U.S.A.

More information

A New Omnidirectional Vision Sensor for Monte-Carlo Localization

A New Omnidirectional Vision Sensor for Monte-Carlo Localization A New Omnidirectional Vision Sensor for Monte-Carlo Localization E. Menegatti 1, A. Pretto 1, and E. Pagello 12 1 Intelligent Autonomous Systems Laboratory Department of Information Engineering The University

More information

Evaluation of Moving Object Tracking Techniques for Video Surveillance Applications

Evaluation of Moving Object Tracking Techniques for Video Surveillance Applications International Journal of Current Engineering and Technology E-ISSN 2277 4106, P-ISSN 2347 5161 2015INPRESSCO, All Rights Reserved Available at http://inpressco.com/category/ijcet Research Article Evaluation

More information

UAV Position and Attitude Sensoring in Indoor Environment Using Cameras

UAV Position and Attitude Sensoring in Indoor Environment Using Cameras UAV Position and Attitude Sensoring in Indoor Environment Using Cameras 1 Peng Xu Abstract There are great advantages of indoor experiment for UAVs. Test flights of UAV in laboratory is more convenient,

More information

Ball tracking with velocity based on Monte-Carlo localization

Ball tracking with velocity based on Monte-Carlo localization Book Title Book Editors IOS Press, 23 1 Ball tracking with velocity based on Monte-Carlo localization Jun Inoue a,1, Akira Ishino b and Ayumi Shinohara c a Department of Informatics, Kyushu University

More information

Locality through Modular Network Topology

Locality through Modular Network Topology Locality through Modular Network Topology Oliver Lipton and Harry Huchra December 18, 2017 Abstract Our experiment concerns designing a robot to accomplish the task of locality grounding. Through a dataset

More information

Static Gesture Recognition with Restricted Boltzmann Machines

Static Gesture Recognition with Restricted Boltzmann Machines Static Gesture Recognition with Restricted Boltzmann Machines Peter O Donovan Department of Computer Science, University of Toronto 6 Kings College Rd, M5S 3G4, Canada odonovan@dgp.toronto.edu Abstract

More information

Using Layered Color Precision for a Self-Calibrating Vision System

Using Layered Color Precision for a Self-Calibrating Vision System ROBOCUP2004 SYMPOSIUM, Instituto Superior Técnico, Lisboa, Portugal, July 4-5, 2004. Using Layered Color Precision for a Self-Calibrating Vision System Matthias Jüngel Institut für Informatik, LFG Künstliche

More information

Notes 9: Optical Flow

Notes 9: Optical Flow Course 049064: Variational Methods in Image Processing Notes 9: Optical Flow Guy Gilboa 1 Basic Model 1.1 Background Optical flow is a fundamental problem in computer vision. The general goal is to find

More information

Acquiring Observation Models through Reverse Plan Monitoring

Acquiring Observation Models through Reverse Plan Monitoring Acquiring Observation Models through Reverse Plan Monitoring Sonia Chernova, Elisabeth Crawford, and Manuela Veloso Computer Science Department, Carnegie Mellon University, Pittsburgh PA 15213, USA Abstract.

More information

Convolution Neural Networks for Chinese Handwriting Recognition

Convolution Neural Networks for Chinese Handwriting Recognition Convolution Neural Networks for Chinese Handwriting Recognition Xu Chen Stanford University 450 Serra Mall, Stanford, CA 94305 xchen91@stanford.edu Abstract Convolutional neural networks have been proven

More information

A robust method for automatic player detection in sport videos

A robust method for automatic player detection in sport videos A robust method for automatic player detection in sport videos A. Lehuger 1 S. Duffner 1 C. Garcia 1 1 Orange Labs 4, rue du clos courtel, 35512 Cesson-Sévigné {antoine.lehuger, stefan.duffner, christophe.garcia}@orange-ftgroup.com

More information

CS231A Course Project Final Report Sign Language Recognition with Unsupervised Feature Learning

CS231A Course Project Final Report Sign Language Recognition with Unsupervised Feature Learning CS231A Course Project Final Report Sign Language Recognition with Unsupervised Feature Learning Justin Chen Stanford University justinkchen@stanford.edu Abstract This paper focuses on experimenting with

More information

Real-time target tracking using a Pan and Tilt platform

Real-time target tracking using a Pan and Tilt platform Real-time target tracking using a Pan and Tilt platform Moulay A. Akhloufi Abstract In recent years, we see an increase of interest for efficient tracking systems in surveillance applications. Many of

More information

NuBot Team Description Paper 2013

NuBot Team Description Paper 2013 NuBot Team Description Paper 2013 Zhiwen Zeng, Dan Xiong, Qinghua Yu, Kaihong Huang, Shuai Cheng, Huimin Lu, Xiangke Wang, Hui Zhang, Xun Li, Zhiqiang Zheng College of Mechatronics and Automation, National

More information

Object Recognition in Robot Football Using a one Dimensional Image

Object Recognition in Robot Football Using a one Dimensional Image Object Recognition in Robot Football Using a one Dimensional Image Hatice Köse and H. Levent Akin Bogaziçi University, Department of Computer Engineering, 80815 Bebek, Istanbul, TURKEY {kose,akin}@boun.edu.tr

More information

COMBINING NEURAL NETWORKS FOR SKIN DETECTION

COMBINING NEURAL NETWORKS FOR SKIN DETECTION COMBINING NEURAL NETWORKS FOR SKIN DETECTION Chelsia Amy Doukim 1, Jamal Ahmad Dargham 1, Ali Chekima 1 and Sigeru Omatu 2 1 School of Engineering and Information Technology, Universiti Malaysia Sabah,

More information

Multi-Cue Localization for Soccer Playing Humanoid Robots

Multi-Cue Localization for Soccer Playing Humanoid Robots Multi-Cue Localization for Soccer Playing Humanoid Robots Hauke Strasdat, Maren Bennewitz, and Sven Behnke University of Freiburg, Computer Science Institute, D-79110 Freiburg, Germany strasdat@gmx.de,{maren,behnke}@informatik.uni-freiburg.de,

More information

FutBotIII: Towards a Robust Centralized Vision System for RoboCup Small League

FutBotIII: Towards a Robust Centralized Vision System for RoboCup Small League RoboCup-99 Team Descriptions Small Robots League, Team FutBotIII, pages 43 47 http: /www.ep.liu.se/ea/cis/1999/006/07/ 43 FutBotIII: Towards a Robust Centralized Vision System for RoboCup Small League

More information

A Symmetry Operator and Its Application to the RoboCup

A Symmetry Operator and Its Application to the RoboCup A Symmetry Operator and Its Application to the RoboCup Kai Huebner Bremen Institute of Safe Systems, TZI, FB3 Universität Bremen, Postfach 330440, 28334 Bremen, Germany khuebner@tzi.de Abstract. At present,

More information

Character Recognition

Character Recognition Character Recognition 5.1 INTRODUCTION Recognition is one of the important steps in image processing. There are different methods such as Histogram method, Hough transformation, Neural computing approaches

More information

Advanced Vision Guided Robotics. David Bruce Engineering Manager FANUC America Corporation

Advanced Vision Guided Robotics. David Bruce Engineering Manager FANUC America Corporation Advanced Vision Guided Robotics David Bruce Engineering Manager FANUC America Corporation Traditional Vision vs. Vision based Robot Guidance Traditional Machine Vision Determine if a product passes or

More information

Machine Learning 13. week

Machine Learning 13. week Machine Learning 13. week Deep Learning Convolutional Neural Network Recurrent Neural Network 1 Why Deep Learning is so Popular? 1. Increase in the amount of data Thanks to the Internet, huge amount of

More information

Chapter 9 Object Tracking an Overview

Chapter 9 Object Tracking an Overview Chapter 9 Object Tracking an Overview The output of the background subtraction algorithm, described in the previous chapter, is a classification (segmentation) of pixels into foreground pixels (those belonging

More information

Computing the relations among three views based on artificial neural network

Computing the relations among three views based on artificial neural network Computing the relations among three views based on artificial neural network Ying Kin Yu Kin Hong Wong Siu Hang Or Department of Computer Science and Engineering The Chinese University of Hong Kong E-mail:

More information

Real time game field limits recognition for robot self-localization using collinearity in Middle-Size RoboCup Soccer

Real time game field limits recognition for robot self-localization using collinearity in Middle-Size RoboCup Soccer Real time game field limits recognition for robot self-localization using collinearity in Middle-Size RoboCup Soccer Fernando Ribeiro (1) Gil Lopes (2) (1) Department of Industrial Electronics, Guimarães,

More information

Techniques for Dealing with Missing Values in Feedforward Networks

Techniques for Dealing with Missing Values in Feedforward Networks Techniques for Dealing with Missing Values in Feedforward Networks Peter Vamplew, David Clark*, Anthony Adams, Jason Muench Artificial Neural Networks Research Group, Department of Computer Science, University

More information

IRIS SEGMENTATION OF NON-IDEAL IMAGES

IRIS SEGMENTATION OF NON-IDEAL IMAGES IRIS SEGMENTATION OF NON-IDEAL IMAGES William S. Weld St. Lawrence University Computer Science Department Canton, NY 13617 Xiaojun Qi, Ph.D Utah State University Computer Science Department Logan, UT 84322

More information

Canny Edge Based Self-localization of a RoboCup Middle-sized League Robot

Canny Edge Based Self-localization of a RoboCup Middle-sized League Robot Canny Edge Based Self-localization of a RoboCup Middle-sized League Robot Yoichi Nakaguro Sirindhorn International Institute of Technology, Thammasat University P.O. Box 22, Thammasat-Rangsit Post Office,

More information

Visual Attention Control by Sensor Space Segmentation for a Small Quadruped Robot based on Information Criterion

Visual Attention Control by Sensor Space Segmentation for a Small Quadruped Robot based on Information Criterion Visual Attention Control by Sensor Space Segmentation for a Small Quadruped Robot based on Information Criterion Noriaki Mitsunaga and Minoru Asada Dept. of Adaptive Machine Systems, Osaka University,

More information

Detecting Spam with Artificial Neural Networks

Detecting Spam with Artificial Neural Networks Detecting Spam with Artificial Neural Networks Andrew Edstrom University of Wisconsin - Madison Abstract This is my final project for CS 539. In this project, I demonstrate the suitability of neural networks

More information

Pedestrian Detection Using Correlated Lidar and Image Data EECS442 Final Project Fall 2016

Pedestrian Detection Using Correlated Lidar and Image Data EECS442 Final Project Fall 2016 edestrian Detection Using Correlated Lidar and Image Data EECS442 Final roject Fall 2016 Samuel Rohrer University of Michigan rohrer@umich.edu Ian Lin University of Michigan tiannis@umich.edu Abstract

More information

2. Basic Task of Pattern Classification

2. Basic Task of Pattern Classification 2. Basic Task of Pattern Classification Definition of the Task Informal Definition: Telling things apart 3 Definition: http://www.webopedia.com/term/p/pattern_recognition.html pattern recognition Last

More information

Line Based Robot Localization under Natural Light Conditions

Line Based Robot Localization under Natural Light Conditions Line Based Robot Localization under Natural Light Conditions Artur Merke 1 and Stefan Welker 2 and Martin Riedmiller 3 Abstract. In this paper we present a framework for robot selflocalization in the robocup

More information

Gesture Recognition using Neural Networks

Gesture Recognition using Neural Networks Gesture Recognition using Neural Networks Jeremy Smith Department of Computer Science George Mason University Fairfax, VA Email: jsmitq@masonlive.gmu.edu ABSTRACT A gesture recognition method for body

More information

Motion Control in Dynamic Multi-Robot Environments

Motion Control in Dynamic Multi-Robot Environments Motion Control in Dynamic Multi-Robot Environments Michael Bowling mhb@cs.cmu.edu Manuela Veloso mmv@cs.cmu.edu Computer Science Department Carnegie Mellon University Pittsburgh, PA 15213-3890 Abstract

More information

AN ADAPTIVE COLOR SEGMENTATION ALGORITHM FOR SONY LEGGED ROBOTS

AN ADAPTIVE COLOR SEGMENTATION ALGORITHM FOR SONY LEGGED ROBOTS AN ADAPTIE COLOR SEGMENTATION ALGORITHM FOR SONY LEGGED ROBOTS Bo Li, Huosheng Hu, Libor Spacek Department of Computer Science, University of Essex Wivenhoe Park, Colchester CO4 3SQ, United Kingdom Email:

More information

AN APPROACH OF SEMIAUTOMATED ROAD EXTRACTION FROM AERIAL IMAGE BASED ON TEMPLATE MATCHING AND NEURAL NETWORK

AN APPROACH OF SEMIAUTOMATED ROAD EXTRACTION FROM AERIAL IMAGE BASED ON TEMPLATE MATCHING AND NEURAL NETWORK AN APPROACH OF SEMIAUTOMATED ROAD EXTRACTION FROM AERIAL IMAGE BASED ON TEMPLATE MATCHING AND NEURAL NETWORK Xiangyun HU, Zuxun ZHANG, Jianqing ZHANG Wuhan Technique University of Surveying and Mapping,

More information

Video Inter-frame Forgery Identification Based on Optical Flow Consistency

Video Inter-frame Forgery Identification Based on Optical Flow Consistency Sensors & Transducers 24 by IFSA Publishing, S. L. http://www.sensorsportal.com Video Inter-frame Forgery Identification Based on Optical Flow Consistency Qi Wang, Zhaohong Li, Zhenzhen Zhang, Qinglong

More information

Applying Synthetic Images to Learning Grasping Orientation from Single Monocular Images

Applying Synthetic Images to Learning Grasping Orientation from Single Monocular Images Applying Synthetic Images to Learning Grasping Orientation from Single Monocular Images 1 Introduction - Steve Chuang and Eric Shan - Determining object orientation in images is a well-established topic

More information

Component-based Face Recognition with 3D Morphable Models

Component-based Face Recognition with 3D Morphable Models Component-based Face Recognition with 3D Morphable Models B. Weyrauch J. Huang benjamin.weyrauch@vitronic.com jenniferhuang@alum.mit.edu Center for Biological and Center for Biological and Computational

More information

Face Quality Assessment System in Video Sequences

Face Quality Assessment System in Video Sequences Face Quality Assessment System in Video Sequences Kamal Nasrollahi, Thomas B. Moeslund Laboratory of Computer Vision and Media Technology, Aalborg University Niels Jernes Vej 14, 9220 Aalborg Øst, Denmark

More information

Automated Canvas Analysis for Painting Conservation. By Brendan Tobin

Automated Canvas Analysis for Painting Conservation. By Brendan Tobin Automated Canvas Analysis for Painting Conservation By Brendan Tobin 1. Motivation Distinctive variations in the spacings between threads in a painting's canvas can be used to show that two sections of

More information

Fast Vehicle Detection and Counting Using Background Subtraction Technique and Prewitt Edge Detection

Fast Vehicle Detection and Counting Using Background Subtraction Technique and Prewitt Edge Detection International Journal of Computer Science and Telecommunications [Volume 6, Issue 10, November 2015] 8 ISSN 2047-3338 Fast Vehicle Detection and Counting Using Background Subtraction Technique and Prewitt

More information

Analytical model A structure and process for analyzing a dataset. For example, a decision tree is a model for the classification of a dataset.

Analytical model A structure and process for analyzing a dataset. For example, a decision tree is a model for the classification of a dataset. Glossary of data mining terms: Accuracy Accuracy is an important factor in assessing the success of data mining. When applied to data, accuracy refers to the rate of correct values in the data. When applied

More information

Hybrid Feature Selection for Modeling Intrusion Detection Systems

Hybrid Feature Selection for Modeling Intrusion Detection Systems Hybrid Feature Selection for Modeling Intrusion Detection Systems Srilatha Chebrolu, Ajith Abraham and Johnson P Thomas Department of Computer Science, Oklahoma State University, USA ajith.abraham@ieee.org,

More information

Particle-Filter-Based Self-Localization Using Landmarks and Directed Lines

Particle-Filter-Based Self-Localization Using Landmarks and Directed Lines Particle-Filter-Based Self-Localization Using Landmarks and Directed Lines Thomas Röfer 1, Tim Laue 1, and Dirk Thomas 2 1 Center for Computing Technology (TZI), FB 3, Universität Bremen roefer@tzi.de,

More information

Predicting Messaging Response Time in a Long Distance Relationship

Predicting Messaging Response Time in a Long Distance Relationship Predicting Messaging Response Time in a Long Distance Relationship Meng-Chen Shieh m3shieh@ucsd.edu I. Introduction The key to any successful relationship is communication, especially during times when

More information

A *69>H>N6 #DJGC6A DG C<>C::G>C<,8>:C8:H /DA 'D 2:6G, ()-"&"3 -"(' ( +-" " " % '.+ % ' -0(+$,

A *69>H>N6 #DJGC6A DG C<>C::G>C<,8>:C8:H /DA 'D 2:6G, ()-&3 -(' ( +-   % '.+ % ' -0(+$, The structure is a very important aspect in neural network design, it is not only impossible to determine an optimal structure for a given problem, it is even impossible to prove that a given structure

More information

Human Detection. A state-of-the-art survey. Mohammad Dorgham. University of Hamburg

Human Detection. A state-of-the-art survey. Mohammad Dorgham. University of Hamburg Human Detection A state-of-the-art survey Mohammad Dorgham University of Hamburg Presentation outline Motivation Applications Overview of approaches (categorized) Approaches details References Motivation

More information

Dynamic Routing Between Capsules

Dynamic Routing Between Capsules Report Explainable Machine Learning Dynamic Routing Between Capsules Author: Michael Dorkenwald Supervisor: Dr. Ullrich Köthe 28. Juni 2018 Inhaltsverzeichnis 1 Introduction 2 2 Motivation 2 3 CapusleNet

More information

Testing omnidirectional vision-based Monte-Carlo Localization under occlusion

Testing omnidirectional vision-based Monte-Carlo Localization under occlusion Testing omnidirectional vision-based Monte-Carlo Localization under occlusion E. Menegatti, A. Pretto and E. Pagello Intelligent Autonomous Systems Laboratory Department of Information Engineering The

More information

Neuron Selectivity as a Biologically Plausible Alternative to Backpropagation

Neuron Selectivity as a Biologically Plausible Alternative to Backpropagation Neuron Selectivity as a Biologically Plausible Alternative to Backpropagation C.J. Norsigian Department of Bioengineering cnorsigi@eng.ucsd.edu Vishwajith Ramesh Department of Bioengineering vramesh@eng.ucsd.edu

More information

Face Detection System Based on MLP Neural Network

Face Detection System Based on MLP Neural Network Face Detection System Based on MLP Neural Network NIDAL F. SHILBAYEH and GAITH A. AL-QUDAH Computer Science Department Middle East University Amman JORDAN n_shilbayeh@yahoo.com gaith@psut.edu.jo Abstract:

More information

Single Pass Connected Components Analysis

Single Pass Connected Components Analysis D. G. Bailey, C. T. Johnston, Single Pass Connected Components Analysis, Proceedings of Image and Vision Computing New Zealand 007, pp. 8 87, Hamilton, New Zealand, December 007. Single Pass Connected

More information

Video Alignment. Literature Survey. Spring 2005 Prof. Brian Evans Multidimensional Digital Signal Processing Project The University of Texas at Austin

Video Alignment. Literature Survey. Spring 2005 Prof. Brian Evans Multidimensional Digital Signal Processing Project The University of Texas at Austin Literature Survey Spring 2005 Prof. Brian Evans Multidimensional Digital Signal Processing Project The University of Texas at Austin Omer Shakil Abstract This literature survey compares various methods

More information

In this assignment, we investigated the use of neural networks for supervised classification

In this assignment, we investigated the use of neural networks for supervised classification Paul Couchman Fabien Imbault Ronan Tigreat Gorka Urchegui Tellechea Classification assignment (group 6) Image processing MSc Embedded Systems March 2003 Classification includes a broad range of decision-theoric

More information

Bayesian Learning Networks Approach to Cybercrime Detection

Bayesian Learning Networks Approach to Cybercrime Detection Bayesian Learning Networks Approach to Cybercrime Detection N S ABOUZAKHAR, A GANI and G MANSON The Centre for Mobile Communications Research (C4MCR), University of Sheffield, Sheffield Regent Court, 211

More information

Estimating the wavelength composition of scene illumination from image data is an

Estimating the wavelength composition of scene illumination from image data is an Chapter 3 The Principle and Improvement for AWB in DSC 3.1 Introduction Estimating the wavelength composition of scene illumination from image data is an important topics in color engineering. Solutions

More information

CS 231A Computer Vision (Fall 2012) Problem Set 3

CS 231A Computer Vision (Fall 2012) Problem Set 3 CS 231A Computer Vision (Fall 2012) Problem Set 3 Due: Nov. 13 th, 2012 (2:15pm) 1 Probabilistic Recursion for Tracking (20 points) In this problem you will derive a method for tracking a point of interest

More information

3D object recognition used by team robotto

3D object recognition used by team robotto 3D object recognition used by team robotto Workshop Juliane Hoebel February 1, 2016 Faculty of Computer Science, Otto-von-Guericke University Magdeburg Content 1. Introduction 2. Depth sensor 3. 3D object

More information

Visual localization using global visual features and vanishing points

Visual localization using global visual features and vanishing points Visual localization using global visual features and vanishing points Olivier Saurer, Friedrich Fraundorfer, and Marc Pollefeys Computer Vision and Geometry Group, ETH Zürich, Switzerland {saurero,fraundorfer,marc.pollefeys}@inf.ethz.ch

More information

A Behavior Architecture for Autonomous Mobile Robots Based on Potential Fields

A Behavior Architecture for Autonomous Mobile Robots Based on Potential Fields A Behavior Architecture for Autonomous Mobile Robots Based on Potential Fields Tim Laue and Thomas Röfer Bremer Institut für Sichere Systeme, Technologie-Zentrum Informatik, FB 3, Universität Bremen, Postfach

More information

DEEP LEARNING REVIEW. Yann LeCun, Yoshua Bengio & Geoffrey Hinton Nature Presented by Divya Chitimalla

DEEP LEARNING REVIEW. Yann LeCun, Yoshua Bengio & Geoffrey Hinton Nature Presented by Divya Chitimalla DEEP LEARNING REVIEW Yann LeCun, Yoshua Bengio & Geoffrey Hinton Nature 2015 -Presented by Divya Chitimalla What is deep learning Deep learning allows computational models that are composed of multiple

More information

MATRIX: A force field pattern matching method for mobile robots

MATRIX: A force field pattern matching method for mobile robots MATRIX: A force field pattern matching method for mobile robots Felix von Hundelshausen, Michael Schreiber, Fabian Wiesel, Achim Liers, and Raúl Rojas Technical Report B-8-3 Freie Universität Berlin, Takustraße

More information

Fourier analysis of low-resolution satellite images of cloud

Fourier analysis of low-resolution satellite images of cloud New Zealand Journal of Geology and Geophysics, 1991, Vol. 34: 549-553 0028-8306/91/3404-0549 $2.50/0 Crown copyright 1991 549 Note Fourier analysis of low-resolution satellite images of cloud S. G. BRADLEY

More information

Fabric Defect Detection Based on Computer Vision

Fabric Defect Detection Based on Computer Vision Fabric Defect Detection Based on Computer Vision Jing Sun and Zhiyu Zhou College of Information and Electronics, Zhejiang Sci-Tech University, Hangzhou, China {jings531,zhouzhiyu1993}@163.com Abstract.

More information

Post-Classification Change Detection of High Resolution Satellite Images Using AdaBoost Classifier

Post-Classification Change Detection of High Resolution Satellite Images Using AdaBoost Classifier , pp.34-38 http://dx.doi.org/10.14257/astl.2015.117.08 Post-Classification Change Detection of High Resolution Satellite Images Using AdaBoost Classifier Dong-Min Woo 1 and Viet Dung Do 1 1 Department

More information

Texture Classification by Combining Local Binary Pattern Features and a Self-Organizing Map

Texture Classification by Combining Local Binary Pattern Features and a Self-Organizing Map Texture Classification by Combining Local Binary Pattern Features and a Self-Organizing Map Markus Turtinen, Topi Mäenpää, and Matti Pietikäinen Machine Vision Group, P.O.Box 4500, FIN-90014 University

More information

Classification Lecture Notes cse352. Neural Networks. Professor Anita Wasilewska

Classification Lecture Notes cse352. Neural Networks. Professor Anita Wasilewska Classification Lecture Notes cse352 Neural Networks Professor Anita Wasilewska Neural Networks Classification Introduction INPUT: classification data, i.e. it contains an classification (class) attribute

More information

Task Driven Perceptual Organization for Extraction of Rooftop Polygons. Christopher Jaynes, Frank Stolle and Robert Collins

Task Driven Perceptual Organization for Extraction of Rooftop Polygons. Christopher Jaynes, Frank Stolle and Robert Collins Task Driven Perceptual Organization for Extraction of Rooftop Polygons Christopher Jaynes, Frank Stolle and Robert Collins Abstract A new method for extracting planar polygonal rooftops in monocular aerial

More information

A Developmental Framework for Visual Learning in Robotics

A Developmental Framework for Visual Learning in Robotics A Developmental Framework for Visual Learning in Robotics Amol Ambardekar 1, Alireza Tavakoli 2, Mircea Nicolescu 1, and Monica Nicolescu 1 1 Department of Computer Science and Engineering, University

More information

A Multi-Scale Focus Pseudo Omni-directional Robot Vision System with Intelligent Image Grabbers

A Multi-Scale Focus Pseudo Omni-directional Robot Vision System with Intelligent Image Grabbers Proceedings of the 2005 IEEE/ASME International Conference on Advanced Intelligent Mechatronics Monterey, California, USA, 24-28 July, 2005 WD2-03 A Multi-Scale Focus Pseudo Omni-directional Robot Vision

More information

A Method for Identifying Irregular Lattices of Hexagonal Tiles in Real-time

A Method for Identifying Irregular Lattices of Hexagonal Tiles in Real-time S. E. Ashley, R. Green, A Method for Identifying Irregular Lattices of Hexagonal Tiles in Real-Time, Proceedings of Image and Vision Computing New Zealand 2007, pp. 271 275, Hamilton, New Zealand, December

More information

ECE 172A: Introduction to Intelligent Systems: Machine Vision, Fall Midterm Examination

ECE 172A: Introduction to Intelligent Systems: Machine Vision, Fall Midterm Examination ECE 172A: Introduction to Intelligent Systems: Machine Vision, Fall 2008 October 29, 2008 Notes: Midterm Examination This is a closed book and closed notes examination. Please be precise and to the point.

More information

Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah

Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Reference Most of the slides are taken from the third chapter of the online book by Michael Nielson: neuralnetworksanddeeplearning.com

More information

Using Geometric Blur for Point Correspondence

Using Geometric Blur for Point Correspondence 1 Using Geometric Blur for Point Correspondence Nisarg Vyas Electrical and Computer Engineering Department, Carnegie Mellon University, Pittsburgh, PA Abstract In computer vision applications, point correspondence

More information

Image Processing Strategies for Artefact Classification

Image Processing Strategies for Artefact Classification Peter Durham Paul Lewis Stephen J. Shennan Image Processing Strategies for Artefact Classification 1 Introduction It need hardly be stated that the identification of objects (usually artefacts) is a fundamental

More information

An Illumination Identification System for the AIBO Robot

An Illumination Identification System for the AIBO Robot An Illumination Identification System for the AIBO Robot Michael S. Zehmeister, Yang Sok Kim and Byeong Ho Kang School of Computing, University of Tasmania Sandy Bay, Tasmania 7001, Australia {yangsokk,

More information

Flexible Lag Definition for Experimental Variogram Calculation

Flexible Lag Definition for Experimental Variogram Calculation Flexible Lag Definition for Experimental Variogram Calculation Yupeng Li and Miguel Cuba The inference of the experimental variogram in geostatistics commonly relies on the method of moments approach.

More information

AIBO Goes to School Basic Math

AIBO Goes to School Basic Math AIBO Goes to School Basic Math S. RADHAKRISHNAN and W. D. POTTER Artificial Intelligence Center University of Georgia 111 Boyd GRSC, Athens, GA 30602 USA http://www.ai.uga.edu Abstract: -. One of the objectives

More information

Ceiling Analysis of Pedestrian Recognition Pipeline for an Autonomous Car Application

Ceiling Analysis of Pedestrian Recognition Pipeline for an Autonomous Car Application Ceiling Analysis of Pedestrian Recognition Pipeline for an Autonomous Car Application Henry Roncancio, André Carmona Hernandes and Marcelo Becker Mobile Robotics Lab (LabRoM) São Carlos School of Engineering

More information

Wave front Method Based Path Planning Algorithm for Mobile Robots

Wave front Method Based Path Planning Algorithm for Mobile Robots Wave front Method Based Path Planning Algorithm for Mobile Robots Bhavya Ghai 1 and Anupam Shukla 2 ABV- Indian Institute of Information Technology and Management, Gwalior, India 1 bhavyaghai@gmail.com,

More information

Texture Segmentation by Windowed Projection

Texture Segmentation by Windowed Projection Texture Segmentation by Windowed Projection 1, 2 Fan-Chen Tseng, 2 Ching-Chi Hsu, 2 Chiou-Shann Fuh 1 Department of Electronic Engineering National I-Lan Institute of Technology e-mail : fctseng@ccmail.ilantech.edu.tw

More information

6. NEURAL NETWORK BASED PATH PLANNING ALGORITHM 6.1 INTRODUCTION

6. NEURAL NETWORK BASED PATH PLANNING ALGORITHM 6.1 INTRODUCTION 6 NEURAL NETWORK BASED PATH PLANNING ALGORITHM 61 INTRODUCTION In previous chapters path planning algorithms such as trigonometry based path planning algorithm and direction based path planning algorithm

More information

Implementation of a Library for Artificial Neural Networks in C

Implementation of a Library for Artificial Neural Networks in C Implementation of a Library for Artificial Neural Networks in C Jack Breese TJHSST Computer Systems Lab 2007-2008 June 10, 2008 1 Abstract In modern computing, there are several approaches to pattern recognition

More information

DESIGNING A REAL TIME SYSTEM FOR CAR NUMBER DETECTION USING DISCRETE HOPFIELD NETWORK

DESIGNING A REAL TIME SYSTEM FOR CAR NUMBER DETECTION USING DISCRETE HOPFIELD NETWORK DESIGNING A REAL TIME SYSTEM FOR CAR NUMBER DETECTION USING DISCRETE HOPFIELD NETWORK A.BANERJEE 1, K.BASU 2 and A.KONAR 3 COMPUTER VISION AND ROBOTICS LAB ELECTRONICS AND TELECOMMUNICATION ENGG JADAVPUR

More information

Eagle Knights: RoboCup Small Size League. Dr. Alfredo Weitzenfeld ITAM - Mexico

Eagle Knights: RoboCup Small Size League. Dr. Alfredo Weitzenfeld ITAM - Mexico Eagle Knights: RoboCup Small Size League Dr. Alfredo Weitzenfeld ITAM - Mexico Playing Field and Robot Size camera 15 cm 4 m 18 cm 4 m 5.5 m Distributed Robot Control Eagle Knights (2003-2006) 2003 2004

More information