Object Tracking Algorithm Based On Adaptive Scale Change Suddenly

In the process of target tracking,the sudden change of target scale in a short period of time will lead to the loss of tracking elements,the accumulation of tracking errors,and finally the target tracking drift.In recent years,deep learning method has made great progress in target detection.In order to better solve the problem of low tracking accuracy caused by target scale mutation,this.paper studies this problem,designs and puts forward an adaptive scale mutation tracking algorithm(KCF_YOLO),which uses deep learning network to detect the target first and then uses kernel correlation filtering method to track The validity of the model is verified by experiments.The specific work of this paper is as follows:First of all,a lot of experiments are carried out on the public data set of target tracking.After analysis,it is concluded that the common feature of video sequence when the tracking accuracy is low is that the target scale changes,and the video sequences with scale changes in the OTB and VOT public data set are selected for subsequent experiments.Secondly,an improved target tracking algorithm(KCF_YOLO)is proposed.The improved point is to change the mode of detecting and tracking while the traditional kernel correlation filtering algorithm is detecting first and tracking later.The combination of deep learning and traditional kernel correlation filtering tracking is applied in the process of target tracking.The addition of deep learning network can not only learn more accurate feature representation,but also can In order to deal with the low resolution of video sequence effectively,the algorithm can get more accurate target tracking in the case of scale mutation.At the same time,the same video sequence is applied to five tracking algorithms:KCF,SAMF,fdsst,dsst and TLD,and the results are analyzed and compared.Thirdly,in order to verify the effectiveness of this method in scale mutation,four evaluation criteria,namely,average accuracy,cross and parallel accuracy,time robustness and space robustness,are integrated to prove the effectiveness of kcf_yoloin scale mutation.The accuracy is increased by 31.74%.Finally,in order to verify whether the test results need to have contingency,this paper further exists in the actual tracking process The experiment shows that the tracking model can still track the target effectively when the target is lost for a long time and reappears,which further proves the effectiveness of the algorithm.At last,the idea of using kernel correlation filter and neural network to detect before tracking in the process of target tracking is proved,which improves the adaptability of the algorithm to the scale mutation in the process of target tracking.The experimental results show that adding detection strategy can correct the following target scale mutation and lead to tracking drift,and the adaptive template update strategy has Effectiveness.