Research On Multi-Object Tracking Via Discriminative Appearance Modeling

Moving target detection and tracking is an important and active research focus in the field of computer vision,image processing and pattern recognition.It plays an indispensable role in a wide range of application involving smart video surveillance,industrial inspection,medical analysis and military.There are enormous commercial value and great potential in it.As the application scenario continues to expand,tracking multiple objects simultaneously is important for automatic video content analysis and virtual reality.Multi-object tracking aims at inferring trajectories for each object from video sequence,which can be considered as a dynamic incremental spatiotemporal clustering problem.Recently,how to formulate data association optimization more effectively to overcome ambiguous detected responses and how to build more effective data association affinity model have attracted more concerns.To address these issues,based on multiple hypotheses tracking framework,this paper proposes a metric learning and multi-cue fusion based hierarchical multiple hypotheses tracking method(MHMHT),which conducts data association more robustly and incorporates more temporal context information.The association appearance similarity is calculated using the distances between feature vectors in each associated tracklet and the salient templates of each track hypothesis,which is then fused with the dynamic similarity calculated according to Kalman filter to get association affinity.To make appearance similarity more discriminative,the spatial-temporal relationships of reliable tracklets in sliding temporal window are used as constraints to learn the discriminative appearance metric which measures the distance between feature vectors and salient templates.This work mainly has three contributions:(1)A hierarchical MHT association framework,which has the merits of hierarchical manner and MHT,is proposed to gradually link the short tracklets to obtain object track by considering more context information;(2)A robust appearance model is proposed to calculate the similarity of the salient template set and consider the high order temporal context of the generated tracks.The appearance similarity is then fused with the dynamic similarity by logistic regression with reliable tracklets;(3)The discriminative appearance metric is learned using the spatial-temporal relationships of tracklets as constraints to measure the similarity between feature vectors and salient templates for appearance modeling.The salient templates of generated track hypotheses are updated using an incremental clustering method.At last,this paper evaluates the MHMHT tracker proposed on challenging benchmark datasets and use common evaluation metrics to make the evaluation persuasive.Qualitative and quantitative evaluations demonstrate that the proposed tracking algorithm performs favorably against several state-of-the-art methods.