Adaptive Moving Object Tracking Method Based On Multi-feature Fusion

As an important research topic in computer vision field,moving object tracking has been widely applied to intelligent monitoring,video navigation,human-computer interaction,etc.Numerous research work have made moving target tracking method get improved on the accuracy and robustness,but moving object tracking under complex scenes remains to be a challenging problem due to several factors,such as illumination changes,background clutter,pose variations.It is a remarkable fact that a robust appearance model is the key to design a high-performance tracker.However,multiple feature fusion and template matching are important methods of appearance modeling.By analyzing and studying the shortage of current target representation method and tracking algorithms of moving target tracking,this paper adaptively fuses multiple features to describle the moving object and adaptively updates the target appearance model to capture the target appearance change based on the target's history moving information.The main work of the paper is as follows:(1)In response to the issue that single visual feature cannot well describle the target object under complex scene,an object tracking algorithm based on multi-feature fusion and adaptive template matching is proposed.Firstly,we obtain motion history information of the target object by using timed motion history image(tMHI)method which is used for segmentation of moving object.Secondly,we use HSV color feature and edge orientation feature to represent the target object,respectively.In addition,we dynamically adjust the fusion weights of each feature according to the corresponding variances based on the similarities between the target template and candidate templates.According to the proposed fusion strategy,we calculate the distances between target template and candidate templates.Meanwhile,we use double templates matching(DTM)method to locate the target.Furthermore,online template is updated timely and offline template is updated according to thethreshold.Finally,experimental evaluations on challenging sequences demonstrate the effectiveness and robustness of the proposed algorithm in comparison with several state-of-the-art algorithms based on qualitative analysis,quantitative analysis and adaptive fusion strategy analysis.(2)A robust multi-feature fusion tracking algorithm is put forward to address poor robustness of moving target tracking algorithm under complex dynamic scene.In order to avoid the updating of feature weights is too sensitive to the change of scene,a more reasonable weight updating method is proposed.Meanwhile,a more effective updating method of target model is designed to better adapt to the target appearance changes.Finally,a lot of experiments are conducted to verify the good robustness and accuracy of the proposed algorithm in dealing with complex scene,such as background clutter,occlusion,pose variations.