Design And Implementation Of Vehicle Tracking System Based On Multi-Camera

Video and image analysis technology has now been widely used in intelligent transportation and other fields.It can provide supporting function for traffic management and improve work efficiency by analyzing the monitor videos.Multi-camera vehicle tracking is one of the focuses of research in intelligent transportation systems.But the switching of different cameras may cause the appearance change of target vehicle,so vehicle tracking under multi-camera has a poor accuracy problem.It can effectively improve the accuracy of vehicle tracking under multi-camera by combining vehicle detection and vehicle tracking.This thesis uses SSD vehicle detection to extract the vehicles from the surveillance video,then it uses global feature and local feature to realize multi-features extraction.finally it uses the extracted multi-feature combined with fully-convolutional siamese networks for object tracking(SiameseFC)to achieve continuous vehicle tracking under multi-camera.In order to reduce the false detection rate of vehicle tracking under multi-camera and improve the accuracy of tracking in the actual monitoring scenario,the thesis conducts an in-depth study.(1)The thesis has researched SSD in the application of vehicle detection.This thesis collects a large number of vehicle images,and gets a detection model by using SSD network to train the vehicle images.Then this thesis uses the detection model to achieve vehicle detection.It can solve the problems of traditional vehicle detection,such as the shadow issue,background interference.And it can improve the accuracy of detection.(2)The thesis has researched global color feature and local feature in the application of vehicle tracking.This thesis uses RGB feature in the global features and SURF in the local features to make up for their deficiencies effectively,then it achieves multi-feature vehicle tracking.(3)The thesis has researched a multi-camera tracking method based on multi-feature tracking and SiameseFC.This thesis uses SURF to locate the target vehicle firstly after SSD vehicle detection.Then it combines color score,SURF score and SiameseFC score to determine the target and to improve vehicle tracking algorithm under multi-camera.The thesis focuses on the related theories and algorithms involved in vehicle detection,feature extraction and vehicle tracking.And this thesis designs a vehicle tracking method under multi-camera,then it achieves a vehicle tracking based on multi-camera system according to the design ideas,finally,this thesis tests the software system functions and performance.The results show that: it has reduced the false detection rate to a certain degree,and improved accuracy.And it has a positive significance for the research of multi-camera vehicle tracking.