Design And Implementation Of Multi-target Tracking Based On High-order Appearance Feature Fusion

Multi-Object Tracking(MOT)is a computer vision task and mainly used in the field of video surveillance,that aims to analyze videos,then identify and track objects such as pedestrians,cars,animals and inanimate objects.MOT has a myriad of other applications ranging from video surveillance to driverless cars,from motion recognition to crowd behavior analysis,from general security systems to tracking cells in microscope images,the analysis and resolution of many problems will benefit from the development of target tracking algorithms.However,it is easy to cause tracking failure due to the influence of many factors such as occlusion,deformation,so MOT is still a very challenging task.How to design features that can accurately track the object without being affected by various external diversity factors has become the focus of research in this fieldIn this paper,to solve the problem that the target is prone to occlusion and deformation,we propose a framework to obtain the appearance features of a target in an end-to-end fashion.In the feature extraction stage,it can fully extract object features from different convolution layers and different sizes,and fuses high-level and low-level semantic information.The high-order feature map is abstracted using the high-order apparent relationship for each target between the current frame and the previous frames.To increase the discriminative ability of the appearance model,residual blocks are used to add the proposed framework,which can fuse non-local information and capture the dependencies between pixels at distant locations.In the data association stage,the similarity matrix is used to describe the high-order features of the target.The best matching relationships between targets are obtained using hierarchical data association and the Hungarian algorithm.The proposed method is called Multi-target tracking Based on High-order Appearance Feature Fusion(MTT-HAFF),which can handle a large number of input sequences,local association failures that result from unreliable detections.The algorithm is tested on the MOT 15 and MOT 17 datasets,respectively,results suggest that the proposed algorithm has a good robustness for large changes in deformation and long-term occlusion.