Design Of Signal Intersection Pedestrian Flow Detection System

Pedestrian detection and tracking based on the video is a research focus in thefield of computer vision which is widely used in intelligent video surveillance,intelligent driving vehicles, robotics and other fields. At present, the application ofvideo-based pedestrian detection and tracking technology in the field of IntelligentTransportation are also concerned for coordinating the relationship between peopleand cars, solving traffic congestion, ensuring that the road is smooth, and protectingthe safety of pedestrians. In the paper, we study pedestrian detection and tracking ofsignal intersection or pedestrian passageway, develop the detection system of signalintersection pedestrian flow based on video.We test on the performance of the systemthrough the actual video, and complete detection, recognition, tracking and countingof pedestrians.The main research contents of this paper are as follows:(1) The paper describes the purpose and significance of the research, analysesthe development and research status of pedestrian flow detection, and outline thetheoretical basis about image of the relevant subject. Then the paper has completedthe overall design of signal intersection pedestrian flow detection system which wasdivided into two types of tilting camera and vertical camare.(2) For tilting camera situation which can shoot to the overall outline ofpedestrians, we have completed the design of pedestrian flow detection algorithmbased on motion model which is divided into moving object detection, pedestrianrecognition, tracking and counting of pedestrians. In the moving object detection, thispaper uses ViBe background modeling to extract the moving object foreground. Ituses morphological erosion and dilation treatment to remove noise, extract contour ofmoving foreground to draw a rectangle outside and merge it. In the pedestrianrecognition, the first we extract HOG features from pedestrian samples to train SVMclassifier below the line. When detecting pedestrian video, we use the SVM classifierto detect and recognize the foreground image, to determine whether it is pedestriansor not. In the tracking stage, we use Kalman filter to track pedestrians. In thepedestrian counting, we set checkbox where pedestrians on the outside and inside are labeled as1and-1, and we count the pedestrian number by the two change of markedsymbol.(3) For vertical camera situation which can shoot to the pedestrian’s headbetween which there is no shelter, we have completed the design of pedestrian flowdetection algorithm based on the head which is divided into pedestrian’s headdetection, head tracking and counting. In the pedestrian’s head detection, werespectively train for head samples below the line, and get two kinds of AdaBoostclassifier based on HOG features and based on the LBP features. At the time ofdetection, the frist we use AdaBoost classifier based on LBP feature to exclude somenon-head regions from the head image. And then we use AdaBoost classifier based onHOG feature to select the head again. The head tracking stage, we use Kalman filtermethod to tracking the head. In pedestrian counting, we set the polygon checkbox tocount the number of pedestrian’s head.(4) We use Visual Studio2008and OpenCV2.4.2to achieve the pedestrian flowdetection system by programming.