The Design Of Passenger Flow Counting System Based On Binocular Vision

Subway which is a common mode of transport,due to its characters of speed and comfort,is favored by the public.Subway was chosed by the increasing number of citizens,and was regarded as daily transport.Because of the large number of people take the subway,which lead to congestion in many cities’ subway station at peak time.if the congestion can’t be handled in time,this phenomenon will have influence on the operation and management of the subway station.the more important isthat if the subway station is in a crowded state for long time,stampede will occurred in subway station.Therefore,in order to avoid the occurrence of similar situations,the relevant departments need to get real-time traffic information to help evacuate the crowd,and adopt the relevant emergency measuresNowadays,the main way of the subway traffic management to get the data of passenger flow is estimated by the surveillance system.This way is prone to realized,but there are still many problems.At first,the data of passenger flow that the staff of subway station gets is not an accurate value,but a proximative value.So the traffic information is inaccurate and susceptible to be impacted by human factors.Moreover,when the number of monitoring locations increases,it is difficult for the surveillant to provide accurate information to the subway station.The advantage of using the visual system to automatically count the passengers and get the data of passenger flow is incomparable.Automated counting method can provide accurate data to help the subway staff to analyze the passenger flow of subway station and deal with emergencies in time.In the mean time,it can reduce labor costs and improve the income of the subway management department.Therefore,this paper apply machine vision to the passenger flow counting system,realize the recognition and location of the pedestrian target,generate the path of pedestrians and count the number of the pedestrian.The main research work are listed as following:Real-time image acquisition and preprocessing is based on computer vision.We study the imaging principle and image preprocessing method of image.Due to the illumination,noise and other factors,the accuracy of image was decreased.Therefore,these images should be filtered through a series of image processing methods,and eliminate the influence of interference factors in the image.Recognition and location of pedestrian is based on binocular vision.Based on the computer vision,this paper utilizes binocular vision system to get the height information for pedestrians.The height of information which was regarded as constraint filter the pedestrian target.It makes the positioning of the pedestrian target more accurate.trajectory generation and pedestrian counting is based on the tracking target by feature points method.The generation of target path is mainly depending on the accuracy of target tracking.In order to satisfy the real-time requirement of passenger flow counting system and taking previous work into account,this paper adopts tracking method based on features to realize the target tracking and trajectory generation.According to the feature information acquired by binocular vision,this paper uses the feature cost function to realize the real-time tracking of moving pedestrian targets in the video.The main idea is to use the tracking method based on feature points.We take the coordinates of the feature points into the formula,calculate the similarity of feature points between adjacent frames.If the similarity of feature points is high,it means that the two feature points are from the same target.Based on traditional feature cost function,this paper proposed a feature cost function based on binocular vision and add the depth information of each feature points into feature cost function to calculate feature point similarity.The goal of generating the target trajectory can be achieved by connecting the feature points from the same target in the video frame.Finally,we need to analyze each trajectory,the use of criteria to determine whether the target is to enter or leave the carriage.According to the previous experience,most of them use counting lines to identity divided the image,and then using the relative relationship between the target cancroids and the number of lines to analyze the direction of pedestrian targets.In this paper,a new counting criterion is put forward based on the original,only one counting line was needed to judge the pedestrian area.The main idea is to record the moving direction of pedestrians in real time,and to judge the motion state of pedestrians by using the moving direction of pedestrians.This paper mainly applies computer vision to real-time traffic information system and propose a pedestrian detection method based on binocular vision and constraints.This paper proves the effectiveness by program testing and simulation results and lay the foundation for the further study of the traffic statistics system.