Research Progress of Multi-Object Tracking in Deep Learning
Multi-object tracking has broad application prospects in many industries,but it also faces problems such as object defor-mation,object overlap,object number changes,occlusion and self-occlusion,and lack of sufficient label data.Due to the rapid de-velopment of deep learning,the rapid development of multi-object tracking methods using deep learning has effectively improved the performance of multi-object tracking.Introduced the research progress of multi-object tracking in deep learning,and divided it into four types of methods based on deep features,based on end-to-end data association,based on single-object tracker exten-sion,and joint detection and tracking,and explained in detail the design principles and advantages and disadvantages of each type of method.Finally,it introduced the commonly used data sets and evaluation indicators and compares the performance of re-lated algorithms.Aiming at the shortcomings of the existing multi-object tracking algorithms,it looked forward to the future devel-opment trend,and hoping to provide theoretical support and technical guidance for the in-depth study of multi-object tracking.
Multi-Object TrackingDeep LearningFeature ExtractionData Association
万方数据
维普
