Research On Recognition System For Driver’s Seatbelt Unfasten

Study on traffic violation recognition based on machine vision has been a hot topic in the domain of transportation at home and abroad which greatly promotes the development of urban intelligent traffic system(ITS). As the most effective passive protection in car, seatbelt fasten while driving is defined by traffic laws. But the detection of seatbelt unfasten while driving is still dealt with manual inspection which is too time-consuming and working hard to meet the need of traffic management. And not only that, but there are many shortcomings such as human intervention, high enforcement cost and poor efficiency, and so on. In order to improve the automatic processing level of seatbelt unfasten detection, the recognition system for drivers’ seatbelt unfasten is designed. This system focuses on three-point belt and recognizes the driver’s seatbelt by analyzing the road monitor image based on image processing, optimum control and pattern recognition. It’s important of the system to realize the automatic processing of seatbelt unfasten detection and enhance the urban traffic management level. Paper is divided into three parts:(1) License plate location based on haar-like cascaded classifierIn the research, the license plate is as the first step of seatbelt identification. Relative to seatbelt location, license plates have a relatively fixed position relationship and are drab colors. Compared with the common methods, haar-like cascaded classifier is selected to recognize the license plate. Firstly, the monitor image is preprocessed and the car position is obtained by gray-level integration projection in vehicle edge image which is used to establish the region of license plate recognition and then the license plate detector is gotten with the statistics learning method. Finally, the region of license plate identification is searched to obtain the position of license plate by the detector.(2) Seatbelt identification area extraction based on gradient transformAll information in the car are expressed through the car window. By analyzing, the seatbelt on driver’s shoulder is as the region to identify seatbelt because the area is minimally affected by occlusion of steering wheel and driver’s arm. In the paper, referring to the plate, coordinate system is established to get the driver’s image. And then the image is processed again. Finally, edge coordinates of car window are calculated by gradient transform which confirm the specific location of seatbelt so that the features of seatbelt could be extracted quickly to simplify the algorithm and raise identification efficiency.(3) Identification of seatbelt based on texture featuresHaar-like features and Line detection based on Hough transform are the common methods in traditional identification of seatbelt which are complexity and low accuracy. So the texture features of driver’s shoulder region are extracted as eigenvectors that are used to identify the seatbelt by support vector machine optimized by genetic algorithm. Result shows that the accuracy of algorithm is 86.12% and the value of AUC is 0.9141 which shows that the texture features are very effective and feasible for seatbelt unfasten identification compared with traditional algorithms.