Research And Implementation Of Passenger Flow Detection System On Train Carriages Based On Video Analysis

Railway passenger transport is one of the most principal means of transportation in our country. It is closely related to people’s daily lives. At present, due to the characteristics such as the big number of passenger number, great fluidity, complex personnel, and so on, there are all kinds of safety accidents in railway passenger transportation. There are more important practical significance to build a rail passenger car video monitoring system to channel the flow of the people in the car in time, and optimize the car space resources configuration. Because of the passengers on the train and off in the midway frequently, the traffic rapidly increased during the holidays, and the actual carriage of passenger flow in the actual statistics tend to lack accuracy, so, in order to achieve the goal of the railroad passenger traffic statistics, decision making supporting for railway transportation and management departments to provide important basis, it is need to use information technology to the automatic identification and statistics of passenger number. In this thesis, based on train video surveillance images, using computer image recognition technology, we identify and statistics human head of the images to achieve the goal of traffic statistics of carriage. The main contributions of this thesis are as follows.This thesis is mainly based on the basic theory of video image processing to identify and statistics human head. In the specific train environment, because of the head contour as one of head characteristics with better robustness, this thesis adopts several edge detection algorithms to extract the edge information of the head. After analyzing the result of various edge detection algorithms, this thesis proposes an improved Canny edge detection algorithm to extract the edge information, then uses the Hough transform algorithm to identify the human head from the edge information which detected by Canny edge detection algorithm and an improved Canny edge detection algorithm. At last, according to the characteristics of the train carriage seat distribution, this thesis puts forward a statistical method which is based on spatial grid in a region to statistics the head. The experimental results show that the improved Canny and Hough transform algorithm of detecting effects better than the traditional detection algorithm for human head detection.Further based on multiple characteristics combined with the mind, the head is identified by the comprehensive characteristics such as head hair, skin color and so on(?)e put forward a head detection and statistics method based on characteristic information. Based on image gray level image, image reverse transform and interference area optimization, we eliminate some noise, remove hair shadow area and reduce the light interference. So that it provides better head candidate area. We use head length breadth ratio, head area local threshold points to detect the head. Finally, we compare our work with two kinds of statistical results to eliminate the repeat count. The experimental results of this method show that the detection rate increases and miss detection rate decreases. However, with the increase in the total number of detected, the error detection rate increases a little bit. In total, the detection result is better.Finally, in order to meet the requirements of practical engineering application, we design and implement the automatic identification and statistics module based on the human head characteristics. As one of the core module of train video surveillance system, it plays an important role for traffic statistics of carriage.