Multiple Object Detection And Tracking In Complex Scene

Multiple object detection and tracking has been one of the most popular topic in computer vision.A large number of algorithms have been proposed with the effort of numerous science researchers.However,most of the current algorithms can only fit specific scene,which work poorly in complex scene.In this thesis,I studied the problem of multiple object detection and tracking,the contribution of this thesis can be summarized as follows:(1)In this thesis,I studied the problem of moving object detection and analyze the advantages and disadvantages of current algorithms.I proposed an adaptive moving object detection method called AMOD to against the most common difficulties of moving object detection,such as camera jitter,dynamic background and intermittent object motion.Based on the sampling background method,I introduced a lot of modules to enhance the accuracy of this method.First of all,dynamic thresholder and dynamic subsample rate are introduced to handle with the problems caused by dynamic background.Then,the proposing of update state and blinking pixel detection help a lot to reduce noise.Besides,I proposed a ghost elimination module to handle with the problem of intermittent object motion.(2)In this thesis,I studied the traditional object tracking algorithms and analyze the problem of multiple object tracking.To handle the problem of object merge and object split,I proposed a detection-based multiple object tracking algorithm called DBMOT.DBMOT algorithm combines the AMOD and FREAK(Fast Retina Kepoint)feature descriptor,and fixs object information when object merge or object split happens.Consequently,DBMOT can track multiple objects robustly in complex scene without any priori information while needing only a few intuitive parameters.