Multiple Objects Tracking Based On Graph Classification

Target tracking technology is the techniques to detect,extract,identify,locate and track the target in a continuous dynamic image sequence,which can obtain the position,size,speed,acceleration and motion trajectory of the target and analysis the behavior to realize tasks like movement monitoring,diagnosis,industrial measurement,dimension reconstruction and navigation guidance.Object tracking has been a hot research field of computer vision,but the change of environment,such as the light of the scene,the occlusion of the target,and the low resolution of the test video,often lead to the discontinuity and the invalid of the tracking target.With the popularity of portable video device in recent years,the research of target tracking has gradually shifted from single target tracking to multiple target tracking.How to effectively solve the problem of tracking multiple targets in different environments,and then get the accurate and effective trajectory,has become an urgent problem.In this paper,in order to solve common problems in classical target tracking algorithm,we proposed a new tracking algorithm based on graph-clustering algorithm.Firstly,background difference method is utilized to localize individual foreground regions in each frame,which can improve the operation efficiency of the algorithm effectively.Then,we use Deformable Parts Model to determine the specific location of the tracking target in video frames.We heuristically leverage the spatio-temporal constraints to generate a set of reliable tracklets,which means short distance moving track trajectory by tracking object.Finally,the proposed graph clustering algorithm is applied to handle tracklet association problem and consequently generate the completed trajectory for individual object.In this paper,we use Deformable Parts Model to obtain the detailed local characteristic information of each detection and tracking target,which is used to calculate the similarity between tracklets.We also proposed an innovative graph clustering algorithm and formulated tracklet association problem into one clustering problem.The proposed method is suitable for different scenes.It can be used not only in single viewing scenes,but also in multiple viewing scenes for object tracking.We select three common datasets: Pets 2012 Video Datasets,Town Center Datasets and Caviar Video Sequence.Compared with other classical algorithms in target tracking field,the experimental results also prove the effectiveness of the proposed algorithm.