The Algorithm Research For Localization And Detection Of Vehicles' Windows In HOV Occupancy Counting System

With the rapid development of modern economy, traffic jam becomes a serious problem in the world. Building roads is becoming more and more costly, and rebuilding and reconstructing roads are also becoming more and more difficult. How to efficiently deal with the traffic jam is very essential to economic sustainable development.HOV (High Occupancy Vehicle) is the vehicles in which there is more than one occupant. America and Canada set special HOV lanes in roads to get efficient and economical transport and reduce traffic jams. In China, many specialists are also advocating for implementing HOV. In order to ensure legal use of HOV lanes, it is important to count vehicle occupants in HOV implementation.If we set cameras along the lanes, and capture the front pictures of coming vehicles, and employ image-processing algorithms, we can recognize and count the occupants in the vehicles. If we first detect and locate vehicles' windows from the images, then recognize and count the occupants only in the windows, we can greatly reduce the calculation work and enhance the accuracy and speed of the algorithm.After studying the methods for locating vehicles windows and the algorithms for detecting lines, we propose an algorithm to detect and locate the vehicles windows. We put forward a background abstraction method to remove the background and keep vehicles in images. We employ edge detection and template filter to keep horizontal lines in images. We use phase groups of pixels to divide the image into several blocks, and make use of Hough transform to detect horizontal lines of the vehicle window in the blocks, thereby we can locate the window horizontal band. We employ the integral projection method to locate the two sides of vehicle windows. We use MATLAB to study and simulate the algorithm. Experiments show that the algorithm can accurately locate vehicle windows in most situations.