Illegal Driving Behavior Analysis Based On Machine Vision

Real-time supervision of the driving is an important means of reducing traffic accidents,which has become the forefront and hotspots on Intelligent Transportation System (ITS) area.In recent years, machine vision-based approaches are becoming pervasive and have beenregarded as the most promising technique. In this paper, a novel approach is proposed todetect illegal driving behaviors by analyzing the driver’s hand posture and motion base onimage processing and pattern recognition technique. The proposed approach began speakingof image pre-processing and the establishment of region of interest, extraction of driver’sdriving posture, the driver’s driving posture tracking, and finally the classification to achievethe detection method of driving behavior, Furthermore, numerical experiments are carriedout to verify effectiveness of the proposed approach. The main work and research include:i. A method of image pre-preprocessing and the establishment of regions of interestthat is suitable for this research is introduced, Firstly, the image pre pre-processingprocedures are designed by comparing the experimental results of image preprocessing.Then, fitting the contours of the steering wheel by using direct least squares ellipse fittingarithmetic and the establishment of the region of interest.The Experimental results showthat the method is effective in reducing noise before extracting the imagefeatures.ii. A common hand extraction method is constructed in this chapter. Firstly, introducingthe color space, The skin tone regions are detected via a Gaussian skin color model built byexperiment in YCbCr color space. Then, connecting the extracted skin color and denoising,hand can be identified and tracked. Finally, the experimental results are given to verify thecorrectness of the method.iii. Some adaptive hand extraction methods are constructed in this chapter. including twoadaptive methods. The first adaptive method is proposed to skin color model based onconsidering the brightness. The dynamic adaptive method is applied to real-time access topictures, using some similar characteristics between the pictures and some features of theproposed picture.iv. The Naive Bayes model based driving behavior classification method is studied. Touse the Naive Bayes model, the continuous feature values of the data set which consists of thedriver’s hand posture and hand motion instances must bediscretized first. After that, thetraining set is used to train the Naive Bayes Classifier, and then the classification experimentis carried out by using the test set. Experimental results indicate Naive Bayes model is effective in classifying the driving behaviors.v. An illegal driving behavior detection system is developed in winCE using C++language; practical use of the system implies it has vast prospect in application.