The Algorithm Study Of Vehicle Detection In HOV System

In recent years, traffic jams, traffic accidents and environmental pollution have had a significant impact on the development of social economy, production and life. Although building more roads can meet the growth of road transportation, the growth of roads has been greatly limited with today's conflicts of resources and environment. Therefore, it is necessary to search new ways to meet the demand. Building HOV (High Occupancy Vehicle) lanes is one of these solutions.We put cameras along the lanes to capture front pictures of coming vehicles and use image-processing algorithms to count vehicle occupants. Firstly, the coming vehicle is detected in the video frame, and then the vehicle's window is located. Secondly, occupants' faces are detected and counted in the window region. Combing the algorithms of counting the passenger and license plate recognition, we can automatically monitor the HOV lanes. The paper mainly studies two aspects.First, we detect and locate vehicles' windows from the videos to reduce the calculation, enhance the accuracy and speed of face detection. We use a background subtraction method and change detection method to remove the background and locate vehicles in video frames. We employ template matching and Hough transform to detect horizontal lines of vehicles' window and integral projection method to locate two sides of vehicles' window. The experiments' results show that the algorithm can detect and extract the windows correctly in most situations.Second, we study the algorithm of vehicle license recognition, including location of license plates, segmentation and recognition of characters. We use border feature extraction and fuzzy template matching to locate vehicles' license plate. We use projection segmentation method to divide license plates into single characters. Experiments show that the algorithm can accurately find plates and segment characters in most situations. Finally, we do a preliminary study on the character recognition with the method of template matching and the method of neural network, and compare the advantages and disadvantages of the two methods.