Multi-object Tracking Based On Discriminative Correlation Filter

In the field of computer vision,Multiple Object Tracking(MOT)has consistently been difficult and challenging research.In this field,the amount of computing data is large and the processing speed of the algorithm requires high so often the tracking speed and efficiency and accuracy difficult to be well balanced.Due to this reason,the Discriminative Correlation Filter with Channel and Spatial Reliability(DCF-CSR)in single object tracking is used for MOT.To assure high tracking efficiency while maintaining the tracking speed.This literature is classified into the following four sections:(1)The target detection.Due to the missed detection of 2D MOT 2015 and MOT16 dataset targets,applying a Faster Region-based Convolutional Neural Networks(Faster R-CNN)to detection is proposed,and provided data sets are displaced with the results of network detection.(2)Target tracking.To add target appearance data in MOT,and to combine speed and accuracy factors,this article uses the DCF model.However,it has been found that the tracking results are not ideal when performing multi-target tracking data sets,because the DCF algorithm does not support multi-scale.To overcome this difficulty,this article solves the method of calculating the Intersection-Over-Union(IOU)between the detection target box and the prediction target box and using the detection target box instead of the predict.(3)The Data association.For the calculation of the IOU of the part(2),it is a mutual calculation problem between a batch of detection boxes and a batch of prediction boxes,which is essentially the task assignment and optimization problem.To this end,the Hungarian algorithm is used to determine the optimal correlation between the targets.(4)Object redisplays experience.In the process of target tracking,there will be a situation in which the target disappears quickly and reappears.To improve the ability of the model to deal with these problems,it is proposed to exercise the correlation analysis method to find the similarity of the emerging "new target",if the value less than a particular range,it is determined to be the same target.Because both the DCF and the IOU model have fast response,the algorithm can meet the requirements of processing data online.Besides,the DCF applies to the model of the position search decision target,and the IOU is an overlap region calculation model.The combination of the two can efficiently solve the target non-inertial motion and camera movement.