Target Vehicle Tracking And Recognition Based On Multi-camera Correlation

Intelligent traffic and road intelligent monitoring system is one of the hotspots of information technology research.With the increase of vehicle ownership and traffic problems becoming increasingly prominent,it has been widely used and concerned.Vehicle tracking and recognition is an important basic technology in related fields.With the gradual development and maturity of image processing technology,more and more researchers pay attention to it.In this paper,vehicle tracking and recognition in road environment is taken as the research object.By means of theoretical research and experimental analysis,we devote ourselves to a thorough study of vehicle tracking and recognition,strive to improve the accuracy and robustness of vehicle tracking and recognition.This paper comprehensively analyses the research progress of vehicle tracking and recognition technology at home and abroad,and expounds the existing technology and its corresponding characteristics from two aspects of target tracking and target recognition.This paper focuses on the problem of candidate target search in existing target tracking methods,as well as the problem of target omission and target deformation caused by local search distance.Aiming at the limitation of single camera method in road scene,the following improved vehicle tracking and recognition scheme based on multi-camera is proposed.Firstly,a global search method based on the idea of target detection is proposed to solve the problem of losing target locally and being insensitive to deformation.This scheme combines the idea of one-stage and two-stage depth target detection methods,combines tracking target information to search candidate targets in current frame,so as to improve the search accuracy.Secondly,aiming at the disadvantage that a single camera can not handle the change of vehicle angle,this paper proposes a vehicle tracking method based on multi-camera,which updates the classifier online by using multi-camera tracking results,so as to ensure that the classifier can learn the multi-angle feature information of the tracking target and improve the tracking accuracy.On the basis of multi-camera tracking a space/scale suppression method is proposed to suppress non-target objects by using the smoothness of target motion law in a very short time,eliminating the interference of similar vehicles.,Finally,aiming at the problem of high similarity between different types of vehicles in vehicle recognition,a fine-grained vehicle recognition algorithm is proposed.Combining with multi-angle information of vehicles tracked by multi-cameras,a weighted voting method is used to eliminate the influence of false recognition in a single camera.The experimental results show that although the proposed method sacrifices some time efficiency,it can achieve better tracking and recognition accuracy.