Available online at ScienceDirect. Procedia Computer Science 50 (2015 )

Transcription

1 Available online at ScienceDirect Procedia Computer Science 50 (2015 ) nd International Symposium on Big Data and Cloud Computing (ISBCC 15) Monitoring Driver Head Postures to Control risks of Accidents Pranoti Meshram a, Nisha Auti b, Himanshu Agrawal c, * a M.Tech Student, Symbiosis Institute of Technology, Pune, Gram-Lavale, Tal-Mulshi, , India a Research Guide, Symbiosis Institute of Technology, Pune, Gram-Lavale, Tal-Mulshi, , India b Research Guide, Symbiosis Institute of Technology, Pune, Gram-Lavale, Tal-Mulshi, , India Abstract Continuous monitoring driver's head dynamics during driving under varying lighting conditions is important yet a challenging problem. Head monitoring becomes even more difficult, when it is done on real-time. Over the past few years, there is growing research interest from research labs, Industry and academic Institutions to develop algorithms for the Advance Driver Assistance System (ADAS).In this paper we present a continuous head monitoring mechanism to detect any drowsiness of driver while driving to avoid the risk of road accidents The The Authors. Published by by Elsevier Elsevier B.V. B.V. This is an open access article under the CC BY-NC-ND license ( Peer-review under responsibility of scientific committee of 2nd International Symposium on Big Data and Cloud Computing Peer-review (ISBCC 15). under responsibility of scientific committee of 2nd International Symposium on Big Data and Cloud Computing (ISBCC 15) Keywords: Driver distraction; Risk control; Image Processing; Eye Region Extraction; Blink pattern; Head Movements; Alarm Generation. 1. Introduction Road safety is one of the major issues to be resolved to prevent mishaps and accidents. According to WHO report, by 2020 traffic related accidents will be the 3rd leading cause of number of deaths. In India alone, half a million accidents occur in a year as per Road Traffic Injuries data. To add further, approximately 2/3rd of these accidents are caused by the driver's fatigue or drowsiness condition during driving. Continuous monitoring driver's head dynamics during driving under varying lighting conditions within is important yet a challenging problem. Head monitoring becomes even more difficult, when it is done on real-time. Over the past few years, there is growing research interest from research labs, Industry and academic Institutions to develop The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( Peer-review under responsibility of scientific committee of 2nd International Symposium on Big Data and Cloud Computing (ISBCC 15) doi: /j.procs

2 618 Pranoti Meshram et al. / Procedia Computer Science 50 ( 2015 ) algorithms for the Advance Driver Assistance System (ADAS). Many solutions have been developed involving both intrusive and nonintrusive methods on head monitoring. Despite many attempts, there is no promising work under multi-sensor data fusion framework In past, there has been a great interest in driver's assistance that use computer vision technology to develop risk-free automobiles. Focus on driver's drowsiness attracted many researchers which led to the developments of various efficient tracking methods and algorithms on monitoring head movement, face expression and eye movement. Recently, in [1] a distributed framework is proposed to analyse the head dynamics and the estimation of head pose is given on the basis of facial features geometric configuration using 3D model. Unlike distributed camera for continuous head movement estimations, a SVM based approach classifies a sequence of video segment into alerts or non-alert driving events by making use of special features such as eye index(ei), pupil activity(pa) and headpose(hp)(2). EI determines if the eye is closed, open or half-open. PA measures the rate of deviation of pupil centre from the eye centre and HP finds the amount of head movement by 3 Euler angles of HP. Furthermore new algorithms using 3D visual tracking has been introduced to monitor the head which consist of three interconnected modules; one detects the driver's head called Head detection module - array of Haar-wavelet Adaboost cascades; secondly Initial pose estimation module using localized orientation histograms and finally a tracking module to estimate the 3D motion of head using appearance based particle filter. Based on the existing state of art for head pose estimation, this paper instantiate an algorithm which is largely motivated by geometric and feature based method for the detection of drowsiness condition. Earlier approaches on the driver's assistance are based on either solely head pose estimation or tracking head pose, eye blinks, yawn detection. Major drive in this paper is develop a novel algorithm to improve the accuracy and robustness of the continuous head and eye monitoring for the driver's assistance. Through the research work, we are trying to address the following questions: 1. To investigate the effectiveness of computer vision based approach and multi-sensor data fusion for continuous head and eye monitoring. 2. How does the varying lighting conditions affect. 3. How reactive the algorithm is. 4. To study the robustness in the head movement and eye blink condition. 2. Basic problems of face detection and eye region extraction: 2.1. Problems in face detection: Researchers working on driver alertness face challenges in this field involves: Camera-pose: In real time scenarios, position of camera with respect to face and image orientation may significantly vary such as frontal face, degree variations, upside down. Moving backgrounds: Existence of more than one face instead of driver can root to diversion from actual face detection. Occlusion: Face may be partially occluded by various objects such as hands of another person sitting next to the driver. Lighting conditions: Data collected for the process is solely dependent on factors like intensity of light that affects the appearance of face. Computational efficiency: To avoid having a false positive in every image, our false positive rate has to be less than 10.

3 Pranoti Meshram et al. / Procedia Computer Science 50 ( 2015 ) Problems in eye detection: Another important aspect is the localization of eyes on face. The system should be able to focus attention on eyes as well for detailed feature extraction such as eye region. At the same time it should not be prone to occlusions. The limitation which hampers the eye gaze detection process is presence of frontal face only, can be advanced to any orientation. Under challenging environment, illumination changes can source to insufficiency in data gathering and robustness of system. 3. Overview of Proposed Method The proposal consists of face detection, eye region extraction, eye blink rate pattern, head postures. In face detection phase a sub window identifies the face region and discards the noise from the background. In eye region extraction, the center nodal point is extracted to localize eyes on the face. In eye blink rate all the three parameters are taken into consideration such as spontaneous, reflex, voluntary and based on that patterns are recognized to define drowsy state of driver. In head postures, for specific position along with blink pattern are noticed to prove drowsiness. Fig. 1 depicts this observation. Fig1: D rowsiness justification makes use of head and eye blink pattern to identify the state and alert before accident occurs

4 620 Pranoti Meshram et al. / Procedia Computer Science 50 ( 2015 ) Face Detection The technique used for this purpose is Viola Jones using Haar-like features for Adaboost algorithm. The process is dividing into 4 stages: Haar feature selection, Creating Integral image, Adaboost algorithm, Cascade classifier. The workflow of the Adaboost algorithm is: Given examples images, y Initialize weights - For t=1,...,t: x 1 1,... xn, y n where y 1 =0,1 for negative and positive examples. 1 1 w1, i, 2m 2l for y 1 =0,1 where m and l are the numbers of positive and negative examples. wt, i 1) Normalize the weights, wt, i n wt, i j1 2) Select the best weak classifier with respect to the weighted error: t min f, p, w h xi, f, p, y i i 3) Define ht x hx, f t, p t, t where f t, p t and t are the minimizers of t 4) Update the weights: wt iwt i1 1,, et, where e t =0 if example xi is classified correctly and e t =1 otherwise, and t t 1 t -The final strong classifier is: Where, 1 t log t 3.2. Eye Region Extraction x r tht t1 = 0 Otherwise C 1 if 1 r x 2 t t1 For this motive JEER method is used to advance the Adaboost face detection algorithm. It enhances the previous process at the same time assigns the eye region which will help out to observe the blink rate. This procedure is divided into two steps: Possible areas of eye region. Necessary condition for existence of the eye region areas Possible areas of eye region In this process, first human eyes on human face are analyzed and nose tip is assumed to be in the center. Straight lines are drawn from the center to identify frontal bone highlighted as brightest bone among others.

5 Pranoti Meshram et al. / Procedia Computer Science 50 ( 2015 ) Necessary condition for existence of the eye region areas This step is used to determine whether eyes might be present in possible areas of the eye region chosen by Step 1. First, characteristics of the eye region on human face are analyzed, the summation of the gray pixel values in each part in eye region is similar regardless whether glasses are worn or not. Thus JEER subsequently chooses the second brightest pixel among the four pixels as the front bone and decides the second possible eye region. It satisfies the symmetry of the summation of the gray values in eye region area characteristics in detected window. 4. Result The results were carried out on Yale university Dataset where 5 different subjects in 7 different conditions are shown with their detection time in msec. Comparisons were made between them. Figures 2 to 5 show the results observed. Fig 2: Comparison between normal and glass Fig 3: Comparison between normal and wink Fig 4: Comparison between normal and sleepy eye Fig 5: Comparison between light conditions The data collected by the University in Yale Database A is comprised of 165 grayscale images in GIF format of 15 individuals and focused on parameters that happen to occur as a reason of biological event. The considered criterion are normal eye, eye with glasses, wink state, sleepy eyes and 3 lighting conditions viz. Left light, Right light and Center light. We provide a collation for determining the detection rate with these parameters using Viola Jones algorithm. Fig 2 represents the difference in detection rate under normal and eye with glasses condition. While Fig 3 represents the difference in detection rate under wink state and normal eye. Counterpart s detection rate of sleepy eye as compared to normal eye is found to be more. And lastly, Fig 4 shows the detection rate under varying lighting conditions, resembles center light to be more effective than left and right light for detection.

6 622 Pranoti Meshram et al. / Procedia Computer Science 50 ( 2015 ) Conclusion In this paper we are proposing a technique which utilizes the hybrid of geometric and feature based algorithm for head pose estimation, so that both head and eye blink pattern are enough to provide information about the driver's drowsy condition. Above results shown where supported over small data set which generalizes that face and eye detected in normal, glasses and center light equips better accuracy than rest. This work will be further advanced with complete grounds as indicating conjugation of eye blink pattern and head posture respectively. References 1. A. Tawari, S. Martin, and M. M. Trivedi, Continuous Head Movement Estimator for Driver Assistance: Issues, Algorithms and On-road Evaluation, IEEE Trans. Intell. Transp. Syst., vol. 15, no. 2,pp , Apr R. Oyini Mbouna, S.G. Kong and M.-Geun Chun, Visual Analysis of Eye State and Head Pose or Driver Alertness Monitoring, IEEE Trans. Intell. Transp. Syst., vol. 14, no. 3,pp , Sept E. Murphy-Chutorian and M. M. Trivedi, Head pose estimation and augmented reality tracking: An integrated system and evaluation for monitoring driver awareness, IEEE Trans. Intell. Transp. Syst., vol. 11, no. 2, pp , Jun K.Dwivedi, K. Biswaranjan and A.Seth, Drowsy Driver Detection using Representation Learning, IEEE Trans. Intell. Transp. Syst., P. Smith,M. Shah, and N.da Vitoria Lobo, Determining Driver Visual Attention With One Camera, IEEE Trans. Intell. Transp. Syst., vol. 4, no. 4,pp , Dec