Video-based Automatic Passenger Counting Algorithm Research

As a key technology in the intelligent transportation system, automatic passengercounting algorithm has always been the focus of domestic and foreign scholars.The techniqueis generally used in crowded places such as passenger bus,subway and mall-intensive sites.Its statistical results can provide decision makers with real-time traffic information. Also itcan provide a strong basis for the vehicle scheduling and line arrangements.This article focuses on passengers target detection algorithm and automatically trackcounting algorithm.Taking into account the passengers often block each other and this case iscomplicated to handle. The camera is installed in the entrance door so overhead shieldingfrequency can be minimized. In APC algorithm, the first is image preprocessing. This articlemainly using a histogram-based and multi-frame average mixed background extractionalgorithm to extract the video background. And use Canny edge detection algorithm toextract edge information. Also the paper uses an algorithm to get rid of most of thebackground in the video frame. Then is identification of the passengers’ head. Based on thetest of the least squares fitting circle detection and Hough transform circle detection. Thepaper selects the Hough transform circle detection algorithm based on gradient to identify thepassengers’ head outline. In order to improve the performance of the Hough transform, thepaper use cumulative array dynamic list and distance mapping table. In optimal passenger ’shead contour extraction, the paper uses the gradation degree of confidence, the arc length ofconfidence, the distribution of the degree of confidence and the matching error confidence toselect the optimal passenger ’s head from the collection of candidate passenger ’s head.Finally is passengers automatically track counting. In order to overcome the problem thatpassengers’ head is always changed and moved too fast. The paper use an algorithm thatbased on Kalman filtering prediction and CamShift passenger tracking and counting.Experiments show that the algorithm can effectively eliminate the background noise.Alsoit can identify and count the passengers accurately. But for non-circular type recognitionaccuracy passenger ’s head needs to be further improved in the subsequent work.