Study On Statistical Method Of Bus Passenger Flow Based On Binocular Stereo Vision

Bus traffic statistics has extensive application value,not only for bus lines planning as a whole to provide accurate traffic data support,configuration for traffic and public security also plays an important role,therefore,traffic statistics brought by market application value and social significance makes it become the research hotspot in the field of intelligent video surveillance at home and abroad.In the present developed bus system,it is very important to get real-time and accurate passenger flow information.Through the real-time and accurate passenger flow information,we can make a scientific and reasonable planning and scheduling of bus running routes,reduce the waste of resources and improve the operating efficiency.And these accurate passenger flow data can produce special social and economic benefits in the current era.Firstly,this thesis introduces the development history and research status of bus passenger flow statistics technology,and elaborates on the stereo matching technology based on binocular stereo vision,and points out the significance of the research of bus passenger flow statistics Based on the traditional monocular vision video analysis and processing technology,the use of binocular camera for passenger traffic statistics,first of all,the left and right cameras of the binocular camera to calibrate,find out the left and right camera internal and external parameters and correction.Secondly,the principle and mechanism of stereo matching technology are studied,because stereo matching based on convolutional neural network can better capture image information,is more robust to the luminosity difference,and in addition,the speed and accuracy have been greatly improved.Therefore,based on the previous research on stereo matching algorithm of convolutional neural network,the structure of the network was improved to generate a good effect of parallax map.Then,the improved convolutional neural network was used to stereo match the corrected image to get the parallax map,and on this basis,the real depth image in the bus was recovered.Due to stereo matching get figure than buses of the depth of the real scene depth map is the influence of such factors as noise,so using Meanshift algorithm on image preprocessing,image noise and judgment,and keep the edge of depth image features,using region growing method is used to identify the head,detect the pedestrian head,Finally,regional growth method was used to judge the head centroid trajectory for tracking and counting.Experimental results show that the bus traffic statistics method based on binocular stereo vision not only counting precision is high,and the efficiency of fast,and can adapt to the complex environment changes,and solved the passengers each shade and crowded,and the light changes,with a strong robustness.