Asia Pacific Journal of Engineering Science and Technology 3 (1) (2017) 16-21 Asia Pacific Journal of Engineering Science and Technology journal homepage: www.apjest.com Full length article On-road obstacle detection system for driver assistance Shivani Godha * Automation & Robotics, Department of Mechanical Engineering, Ajay Kumar Garg Engineering College, Ghaziabad-201009, India (Received 19 February 2017; accepted 17 March 2017; published online 24 March 2017) ABSTRACT Obstacle detection is an active research area in the field of image processing. On road obstacle detection is an important problem in the area of intelligent transportation systems, driver assistance systems and automated guided vehicles. The proposed algorithm detects all obstacles on road using a single onboard camera which is mounted on a travelled vehicle in real-time for driver assistance. When the obstacle found in front of travelled vehicle using morphological filtering then the driver assistance system gives warning to the driver for avoiding collision. Experimental results are given for real time on road obstacle detection. Keywords: Obstacle detection; Image processing; Morphological filtering; Otsu s method 1. Introduction Object detection and tracking are important in many computer vision applications using image processing techniques. Image processing is a process where input image is processed to get output in form of information. The main objective of all the image processing techniques is to recognize the object under considerations. Today, all the images used in the world are in digital format. In this paper, we propose a purely vision-based approach for real time obstacle detection for an automotive vehicle designed to provide automotive safety purpose. Our proposed system uses as input the video streams from a synchronized, forward-looking camera. To analyze this data, the system combines visual object detection and warning system. According to the present invention, the object scene captured by low cost camera which is located at the head of an automotive vehicle. Camera captured the images as RGB images which is converted in to the grayscale images by some image processing algorithms. These grayscale images are used to process the object scene using Otsu s method, for predicting as an obstacle is approaching the vehicle using MATLAB environment. As a result, Warning system provides an alert message and indication to the driver on recognizing the object. According to this, the * E-mail address: shimmysg@gmail.com 2017 Author(s)
17 Shivani Godha Asia Pacific J. Eng. Sci. Tech. 3 (1) (2017) 16-21 obstacle passing through the vehicles surely and stably detected and avoid any traffic accident using MATLAB environment. The purpose of this work is to reduce driver error by providing information about the task situation using obstacle detection. 2. System development and implementation The system is designed for an automotive vehicle on road equipped with forward looking camera on vehicle. It can be broken into two main modules as (a) Obstacle detection module (b) Warning system module. The task of Obstacle detection module used to analyze the frames of video captured by the USB camera is shown in Fig. 1. An image in MATLAB is stored as a 2D matrix (of size m x n) where each element of the matrix represents the intensity of light/color of that particular pixel. Hence, for a binary image, the value of each element of the matrix is either 0 or 1 and for a grayscale image each value lies between 0 and 255. A color image is stored as an (m x n x p) matrix where each element is the RGB value of that particular pixel for each frame. It calculate the object by finding boundary as right and left both side of image, apply some set point on these boundaries compared with it to threshold level. According to recognized object, the warning system module provides an alert message, a warning indication as shown in Fig. 2. Fig. 1 Obstacle detection module Fig. 2 Warning system module
18 Shivani Godha Asia Pacific J. Eng. Sci. Tech. 3 (1) (2017) 16-21 Warning system module alerts vehicle driver on the basis of object recognition. Firstly obstacle detected because their luminance substantially differs from background, therefore two module algorithms formed the system as (i) Boundary detection algorithm is used to reduce search area of obstacle (region of interest ROI) where obstacle can be found. (ii) Obstacle validation algorithm is used to check the correctness of recognized obstacles. (i) Boundary detection algorithm This algorithm performs a region growing approach which as starting from the region of image just in front of vehicle and extends up to end on the road surface in Fig. 3. The boundary detection used to define area of interest where obstacle appears. For region of interest (ROI) find the mean pixel luminance on each row of image between boundaries left and right borders respectively. Fig. 3 The region of interest definition Fig. 4 Processing of image frame (ii) Obstacle validation algorithm This algorithm process frames of image to detect the obstacle and its validation. Obstacles are perceived because their luminance differs from the road surface or background. For this, firstly image enhances using grayscale conversion. Second the feature extraction techniques applied to validate object in Fig. 4. The luminance of pixel inside the boundary is used to predict region of interest (ROI) points as road or no road. The area of object is larger due to presence of object is nearest to the centre of line and lie in the region of interest (ROI) in Fig. 3. If object is lying far to the centre of line and having a maximum distance of camera s view range, then the area of object is smaller as compared to approximated area range. When obstacle validation algorithm formulated with in region of interest, the obstacle found using morphological filtering
19 Shivani Godha Asia Pacific J. Eng. Sci. Tech. 3 (1) (2017) 16-21 in terms of size and a line fitted in region of interest to validate the object. If the object lie in region of interest, then fitted line deviate from assumed position and its deviation of data vector x, calculated as G= { (x i x) 2 } where, x= x i. Here, x is an image matrix, then G returns a row vector containing the standard deviation of the elements of each column of x. On comparing G to threshold, warning system modules works. 3. Results and discussions The proposed system has been tested on real time on road vehicle detection. The video images captured by CCD camera (2/3 sensor & 16mm lens) placed on the head of automotive vehicle. To perform the experimental results, we firstly capture the input from camera continuously in form of video. This original image in RGB color and image size as similar of captured video image is 160x120 pixel. When the vehicle approaching obstacles on road, obstacle detection algorithm starts when vehicle is at 70m, while the warning system module works and provides message at 65m. The obstacle detection algorithm processed about 50 frames /second, being relative speed of 40km/h. This algorithm is quite simple in order to satisfy real time requirements and this criteria is sufficient to warn the driver about conditions. When a vehicle approaches a car as obstacle shown in Figure, then obstacle recognized as following: Convert the given RGB image frame into grayscale image as in Fig. 5a.
20 Shivani Godha Asia Pacific J. Eng. Sci. Tech. 3 (1) (2017) 16-21 Compute the image array value where each output pixel contains the range value as (max value- min value) of the 3-by-3 neighborhood around the corresponding pixel in the input image. The value of each pixel in term of local range as shown in Fig. 5b. Performs morphological filtering on the local values grayscale image with the structuring element SE. The argument SE must be a single structuring element of disk shaped. These morphological filtering shows in Figs. 6a, 6b, 6c. Trace the boundary using size of array of image as row element and column element s values. If the first and last array index of row and column in image array is empty and then we find the next array element and trace the boundary of both right and left side using fitted line. The boundary plot deviate from its original position towards both left and right side. We find out the standard deviation as shown in Fig. 7. If the deviation of boundaries left and right side with in the region of interest is higher than a limit which is decided by user, then the warning system provides warning to the driver using red indicator, alarm and displays the message ATTENTION on screen as shown in Fig. 8. If there is no obstacle on road then driver assistance system displays the message OK as Fig. 9. Fig. 9 Output of driver assistance system in case of no obstacle
21 Shivani Godha Asia Pacific J. Eng. Sci. Tech. 3 (1) (2017) 16-21 4. Conclusions We can see that the proposed algorithm processes video captured by a single monocular camera. We make the assumption for object detection, camera placed at front of vehicle. The user should start the MATLAB program on screen and according to the algorithm obstacle detects, then the warning system indicated the driver. Implementation of warning system for the driver gives improper results and failure sometimes due to some processing delay. Practically, this project needs a proper lightening and object in background when we run this project on the road. In such practical conditions this technique is giving exact results. There are some threshold values used in MATLAB program to recognize the objects and provides the alertness to the driver. There are many aspects related to object recognition which is not considered like if object is not in the camera range. References [1] S. Badal, S. Ravela, B. Draper, A Practiacal obstacle detection and Avoidance system, Proceedings of the Second IEEE Workshop on applications of computer vision, 1994. [2] S. Denasi, G. Quaglia, Early obstacle detection using a Region segmentation and model based edge grouping, International Conference of the IEEE on intelligent vehicles, 1998. [3] I. Ulrich, I. Nourbakhsh, Appearance-based obstacle detection with monocular color Vision, AAAI Conference on Artificial Inteligence, 2000. [4] R. Collins, A. Lipton, H. Fujiyoshi, T. Kanade, Algorithms for cooperative multisensor surveillance, Proceedings of the IEEE, 89 (10) (2001) 1456-1477. [5] Dr. Bekiaris, A. Amditis, A. Polychronopo, Multiple - Sensor - Collision avoidance system for automotive applications using an IMM approach for obstacle tracking, IEEE Conference, 2002. [6] E. Morimoto, Masahiko, M. Umeda, Obstacle Avoidance System for Autonomous Transportation Vehicle based on Image Processing, CIGR Journal of Scientific Research and Development, 4 (2002) 1-11. [7] Z. Sun, G. Bebis, R. Miller, On- Road Vehicle Detection: A review, IEEE transactions on pattern analysis and machine intelligence, 28 (5), 2006. [8] T.H. Nguyen, Jordan, D.M. Pham, H.T. Nguyen, Real-Time Obstacle Detection for an Autonomous Wheelchair Using Stereoscopic Camera, Annual International Conference of the IEEE EMBS Cité Internationale, 2007. [9] J. Fasola, M. Veloso, Real time object detection using segmented and Gray scale images, International Conference of the IEEE on intelligent transportation systems 2009. [10] I. Kostavelis, L. Nalpantidis, A. Gasteratos, Real-Time Algorithm for Obstacle Avoidance Using a Stereoscopic Camera, Workshop on New Trends in Robotics, 2010. [11] H. Kyutoku, D. Deguchi, T. Takahashi, On-road Obstacle Detection by Comparing Present and Past In-vehicle Camera Images, MVA2011 IAPR Conference on Machine Vision Applications, Japan, 2011. [12] S. Qian, J.K. Tan, H. Kim, S. Ishikawa, Road region estimation and obstacles extraction using a monocular camera, International journal of innovative computing, Information and Control, 9 (9) (2013) 3561-3572.