The Study On Viusal Perception Of Driver's Violation Behavior For Traffic Bayonet

With the rapid increment of vehicle in China, the consequent traffic accidents are also increasing year by year. As it is more and more difficults to manage the transportation system, intelligent transportation technology arises at this moment. Intelligent transportation system (ITS) takes advantage of advanced intelligent technology to improve road utilization ratio, safety and comfort. Traditional transportation management system mainly plays a role of supervising law-executors, whereas intelligent transportation system manages road users via advanced technology. Thus studying the driving behavior is very important establish the intelligent transportation system.Intelligent traffic bayonet plays important part in ITS, by captured passing vehicles timely, it obtains information on the vehicles and drivers, which provides strong support for public security and traffic control, etc. This dissertation using visual perception technology conducted a series studies about images captured by traffic bayonet, and offered corresponding solutions in the field of traffic image enhancement, driver's upper body location, driver's seat belt detection, driver's using mobile phone behavior detection. The methods can effectively complete the processing of traffic pictures. Specific work of the dissertation is as follows:1. In order to process traffic bayonet image under complex environment, a traffic image enhancement method is proposed based on adaptive brightness baseline drift. Firstly, changes in light and weather encountered in natural traffic bayonet scene are analyzed, especially for front-light and backlight caused by the difference direction of traffic bayonet. Then the relationship among the image, light intensity and shooting time are considered all together, in order to create the brightness benchmark model. At last the algorithm uses the brightness benchmark model to enhance the brightness of the image adaptively. This method can achieve better effects compared with other available method under different luminance conditions, which also effectively reduced the influence of the weather, and affect consequential detection processes, such as vehicle detection.2. To solve drive's upper body location problem under complex environment, an upper body detection method is proposed based on multi-level structured model. In order to get drive's upper body area accurately, the method processes traffic image step by step. The first step is vehicle detection which employs melted multi-channel HOG feature proposed, and it can achieve better effects for the vehicle that has not obvious local gradient in gray domain. The second step is drive's upper body location: The structural model is carried out on the detected vehicle, and the concept of space confidence and spatial correlation is considered for the structural vehicle parts. In our paper, space confidence is calculated by space position and space dimension of the structural vehicle parts, spatial correlation is calculated through structural parts of spatial distance and spatial orientation. Then we can get driver's upper body location based on space confidence and spatial correlation. The third step is drive's behavior region identification. The drive's behavior studied in this paper mainly refers to seat belts usage and cell phone usage. The structural model is carried out on driver's upper body area first, and then get drive's behavior region according to each behavior area's prior knowledge and structured information. This method laid a good foundation for subsequent processing. Experimental results show that this method can effectively get driver's upper body location.3. In order to detect driver's illegal behavior, a seat belt detection method is proposed. This method first enhances the driver's upper body area local adaptively, and then selects the edges as the main character of a seat belt, and it puts forward a kind of the adaptive filtering strategy for the edges of seat belt, the strategy using the concept of neighborhood pixels diagonal inclined adjacent proposed in this paper. After getting seat belt edges, perceptual organization technology is used for organizing the discrete edges; organization rules processed in this paper is the combination of edge similarity, edge connectivity and the similarity of edge neighborhood brightness value. After organizing the discrete edges, Hough transform-least squares analysis proposed in this paper is used for linear fitting. And at last, the confidence of seat belt proposed is used for validating the true and false of the fitting line. Experimental results show that this method can effectively realize the visual perception of driver's no using seat belt behavior.4. In order to detect illegal mobile phone usage during driving, I aims at extracting feature of mobile phone, proposing a new detection method for driver's using mobile phone behavior. In order to remove the interference information, ROI region is divided first according to the location of the cell phone, and then the prior probability of using cell phone is calculated by the position in the ROI. When judging using mobile phone behavior, hand skin feature is analyzed. The traditional elliptical model for skin detection is employed first, and then I put forward the probability ellipse model for skin detection. The final probability of hand skin is attained by combining skin probability and prior probability. Thus the hand skin candidate regions are selected. For each hand skin candidate regions, the histogram difference, scarce degree and concentration degree are considered to verify the significance, in order to realize the final visual perception of driver's illegal using mobile phone behavior.In addition to the achievements presented in this paper, we present the imperfect aspects in our study and introduce the task in our future work.