Research Of Passenger Counting Algorithm Based On Binocular Stereo Vision

Because of advantage of the bus, it has been accepted by the general public and as the preferred means of transport to travel the city. But now there are many public transport problems needed to be solved urgently, such as unreasonable urban route planning, unscientific dispatch during the morning and evening peak, impertinent station distribution of sparse or tightness, frequent traffic rub accidents in flow traffic etc, so development of modern intelligent transportation systems for solving the problems of these buses has a very important practical significance. In the scientific comparatively transportation system, how to get high accuracy and real-time traffic information is particularly important. With accurate and real-time passenger information, not only scheduling the line bus can be more scientific, but also helping the bus company to make a more reasonable arrangement about lines and sites. In addition, these accurate and real-time traffic information can also create special economic and social benefits in the era of big data.Currently we are increasingly able to reach this consensus: Since the image contains a huge amount of information, the use of image processing techniques to count traffic statistics in the real-time public transport system has become a important development direction. One-dimensional monocular vision and multi-dimensional or two-dimensional binocular vision are two branches of passenger statistics which are based the image processing technology.In consequence monocular vision technology for processing image is easy to be affected by the intensity changes of color light and cover and other factors, now we use extensively stereoscopic technology to make image processing. This study utilizes stereo vision technology to design algorithm of getting the transit passenger flow information. Finally the algorithm is achieved on SBC3730 platforms.This research firstly elaborates stereo matching technique, and analyzes several matching algorithms, and selects dynamic programming algorithm to obtain the disparity map according to the real-time and accurate requirements. The dense disparity map which is obtained by the method of processing globally is smoother than the other methods Then this research use the way which combining background subtraction model and optical flow residual to detect moving objects, and segment the target, and use the associated field marks to locate and extract target in moving state. Multi-feature fusion tracking algorithm is applied for target tracking count in this paper, and this algorithm is made appropriate improvements. The improved algorithm can do a more accurate count when it tracks the target of motion state. Finally, this research builds the accounting system for state of passenger flow through build the SBC3730 platform. The algorithm in the paper is ported to the system and tested. The final conclusion illustrate that the accuracy and real-time of the system is relatively high in tracking and counting moving target aspects.