Multi-target Tracking By Tracklet Association Method

Scholars pay attention to the hot topic- Target Tracking Technology in the field of computer vision and pattern recognition research and many of them think it had a broad application prospects. The Multi-Target Video Tracking has been extensively studied in recent years, and many new tracking algorithms has been put forward. From the point of target detection and multi-target association, this paper puts a conducted in-depth research on multi-target tracking technology and proposed practical and effective solutions to frequent occlusion by clutter or other objects,similar appearances of different objects,direction changes and other tracking problems.This paper presents a statistical treatment of background modelling for use in target detection, where the global information and local information is added into the statistical framework to construct a robust background model to achieve accurate object detection results. Specifically, a novel self-adaptive Gaussian mixture model is proposed to construct a statistical background model based on the global information, which is utilized to deal with the target detection issue under illumination changes of scene; for the target detection issue under dynamic changes of background, the self-tuning spectral clustering technology is first utilized to cluster the background image, the kernel density estimation method is then utilized to construct a statistical background model based on the local information. Experimental results demonstrate that the proposed algorithm can improve the detection performance under illumination changes of scene or dynamic changes of background.This paper focuses on a multi-target tracking algorithm based on tracklet associated.The algorithm achieves the global and the local tracklets by two different association strategies for multi-target tracking. In practice, firstly for the newly input detection response,the paper uses an approach based on the scene-adaptive local tracklets generation strategy through establishing the hierarchical feature space by characteristics reliability,and the newly input detection response associated with the original tracklets in different layers of the feature space adaptively. In the phase of the global tracklets association, the paper proposes a novel discriminative appearance learning method. In the proposed learning method,online training samples are collected from tracklet objects,and a discriminative projection space is updated with the collected samples using incremental linear discriminant analysis.By projecting the appearance models of tracklets into the discriminative projection space,the algotithm makes the appearances of tracklets more discriminative. In the follow-up phase, in order to finally obtain a complete and smooth trajectory, the paper propses a trajectory filling strategy based on non-linear motion model so that trajectories which does not meet the linear motion model could associate with others. In addition, the paper also puts the method into PETS 2009/2010 benchmark and TUD-Stadtmitte video sequence database to prove that the method could achieve multiple objectives correct association in the presence of frequent occlusion by clutter or other objects,similar appearances of different objects,direction changes and other tracking problems in complex scenes,thereby obtaining stable, continuous tracking tracks.