Study On Vehicle Window Detection Technology

ABSTRACT:Nowadays, the urban traffic congestion problem has become a major problem in many countries including China. Urban traffic congestion has reduced the traffic transportation efficiency, increased the waste gas emissions and also raised the energy consumption. It harms human survival environment severely. The cost of roads building is improving continuously, the way to improve traffic capacity with new construction, expansion and rebuilt is also more difficult. The measure of constructing HOV (High Occupancy Vehicle) special lanes, to provide priority for high occupancy vehicle, is a very effective method to solve the problem of urban traffic congestion. Many western countries have established special HOV lanes. In order to ensure the priority and legal fair use of HOV lanes, it is necessary to detect and count internal passengers. Window detection is a crucial question in the process of researching the technology of passenger detection.1. The paper has researched the coarse position method in the vehicle video streaming. Positioning vehicles using frame difference image, effectively eliminates the complex background interference of the roads; Combining with the scale of car window and headstock to coarse locate window area, has eliminated the line interference of headstock efficiently.2. Window detection algorithm based on Hough Transformation. According to the approximate linear characteristic of the window boundary, the paper designs the vehicle window detection algorithm with template matching filter, Hough transform, integral projection and knowledge rules. The algorithm has been tested respectively on MATLAB and VC++6.0, the result shows that the algorithm can effectively detect windows of various kinds of vehicle on actual roads and the algorithm has the higher accuracy, stronger robustness and faster processing speed.3. Window detection algorithm based on Generalized Hough Transformation (GHT). Based on the isosceles trapezoidal outline feature of the vehicle window, the paper designs the GHT window detection algorithm, which maps the boundary points in the images to the parameters space, translating it into clustering problem in parameters space. We can locate the trapezoidal profile of the window if finding out the parameters of most of the boundary points that satisfied. The algorithm is tested on VC++6.0platform, the result shows that the algorithm achieved certain detection effect. But the GHT window detection algorithm is not very ideal in the detection rate and the detection speed, compared with the window detection algorithm based on Hough Transformation, this algorithm needs further improvement in the later study.