Multi-object Tracking Algorithm

Object tracking algorithm is the core of intelligent surveillance technology. Object trajectory tracking can be realized through the object tracking technology in surveillance video sequences which provides important information for special event detection and behavior analysis. In recent years, with the development of computer technology and intensive study of target tracking algorithm, the global association multi-target tracking algorithm has gradually become a hot issue for the limitations of traditional single target tracking algorithm. The global association multi-target tracking algorithm takes the association of all tracking data in consideration to track objects in the scene. Object tracking techniques have been widely used in the industrial production, public places and family security, public traffic, public security system and military fields. There are broad prospects for development and the urgent market demand of object tracking techniques.Multi-target tracking algorithm is the main contents of this paper and the key and difficulty of intelligent surveillance technology.For the research of multi-target tracking algorithm, an improved global association multi-target tracking algorithm system is proposed based on the study of existing technology. There are 4 main parts of the global association multi-target tracking algorithm system, moving object detection, optimization of target trajectory points in the integrity and accuracy, global association optimization of target trajectory. The main steps of improved global association multi-target tracking algorithm system is as follow: Firstly,the binary image of the moving object in video is calculated through the moving target detection algorithm. Secondly, an effective mathematical model is established to track all the moving targets which can track the occluded objects independently to get the initial tracking results. Thirdly, an optimization model is proposed to optimism the data of all the moving targets trajectory points in integrality and accuracy of the whole video. Finally, the global association of target trajectory is calculated though the global association multi-target tracking model proposed in this paper according to the trajectory data calculated above. Then the trajectory is obtained though the calculation above. Overall, the main researches and innovations are summarized as follows:1 Image pre-processing techniques. In this paper, image pre-processing is in order to get a stable moving object detection image. The study of image pre-processing techniques mainly focus on image de-noising algorithms in this paper. Some basic and important image pre-processing techniques, such as median filtering, Gaussian filtering, morphological processing, are introduced in this paper base on the study of related literatures.2 Moving object detection. The research of moving object detection has made a lot of progress. There are many moving object detection algorithms, such as background difference by average background model, the classic single Gaussian moving object detection model and Gaussian mixture model, Vi Be and so on. A improved Vi Be algorithm is proposed in this paper. The initialization method of ViBe algorithm is improved by combining with the idea of the frame difference in the proposed algorithm. The weird shadow problem of moving object detection image can be well disposed by the proposed algorithm.3 Single-object tracking. Tracking is the main content in this paper. Firstly, some classic single-object tracking algorithms, such as Online-boosting, Mean Shift and Kalman filtering and particle filtering, is introduced briefly in this paper. Based on the study of existing research, a simple single-object tracking method is proposed. The proposed method is a bidirectional match algorithm which achieved the tracking and object status determination by a simple objective measure of similarity defined in this paper.4 Multi-object tracking. Multi-object tracking is the most important research content in this paper. Firstly, a tracking optimization model based on objects’ scale features is proposed in this paper. The optimization model tracked moving objects according to the initial moving object detection and bidirectional matching Single-object tracking result data. Object losing caused by adhesion and cover problem can be well handled which ensured the accuracy and completeness of the object tracking data. Secondly, an improved trajectory optimization model based on the object association is proposed, which effectively guarantee the stability and continuity of global tracking.Finally, we summarized the mainly work of this paper and present future work in multi-target tracking algorithms research.