Research And Application Of Long-time Target Tracking Algorithm

The rapid development of science and technology leads to the arrival of the era of intelligence.Computer vision plays an important role in the field of artificial intelligence.The realization of image understanding is the ultimate goal of computer vision.Many classical target tracking algorithms have been proposed by many scholars,and more excellent tracking algorithms have emerged on the basis of the classical algorithms,but these algorithms in the many challenges of scenarios,such as illumination change,target be sustained,movement environment complex interweave together,still do not have a satisfactory good adaptability,cannot satisfy the long in the practical application scenario is more steady tracking target.In order to satisfy the practical application of target tracking,it is necessary to be real-time.The correlation filter tracking algorithm has obvious advantages in running speed.However,it also has some problems such as complex scenes,weak robustness under multi-interference,and easy to be interfered by similar targets.In this paper,aiming at the shortcomings of correlation filtering algorithm,two improved tracking algorithms are proposed,compared with the relevant excellent algorithms,and the data set of autonomous aerial refueling cone set is made,which is verified by experiments in the tracking application.The main research and work of this paper are summarized as follows:(1)proposed a target tracking algorithm based on model update supervision.Due to the staggered influence of various challenging interference factors in the tracking process and the existence of various uncertainties in the actual application process,especially the appearance of similar targets in a complex environment,the accuracy and success rate of the algorithm will be greatly reduced.The general algorithm updates the template even when the accuracy performance is poor and continues the target tracking of the next frame,which leads to the contamination of the template and further reduces the accuracy and success rate.To solve above problems,this paper add template updating supervision mechanism,response information obtained by this frame analysis,found that when the stability of target tracking,a single peak,the rest of the flat response and the maximum response of proposed award template and response figure of turbulence degree combined with the comprehensive evaluation,after setting the threshold value judgment to decide whether to update.Comparison experiments show that the improved algorithm can effectively protect the template and improve the anti-interference ability of the algorithm in the tracking environment with many complex challenges.(2)put forward an improved long time target tracking algorithm integrating ORB.In practical application,in addition to the above mentioned interference factors,the existing algorithm is also faced with the problem of long-term tracking instability.In the process of long-term tracking,it is inevitable for the target to temporarily leave the field of vision and extreme interference,so when the target reappears,the existing algorithm is difficult to find the target again,thus losing the target,and the performance is not satisfactory in the case of long-term tracking.To solve above problems,this paper on the basis of the first algorithm,increased the target weight detection mechanism,put forward based on the model updating and rapid detection of long-term tracking algorithm framework,when the target is lost after the timing for target detection based on the extraction characteristic of child weight detection method,feature extraction of the dimension unity to dimensional optimization,512 bits to speed up the detection rate.The experimental results show that the algorithm can meet the long-term tracking stability in complex situations such as interference interleaving and the target temporarily leaving the field of vision,and the performance of the algorithm is further improved.(3)the application of autonomous aerial refueling cone sleeve tracking was selected as the simulation experiment.For this purpose,the self-made cone sleeve was used for video self-collection,video processing and data set production.Real aerial refueling cone sleeve video is selected and the data set is formed after video processing.The two algorithms proposed in this paper and the classical correlation filter algorithm KCF were tested in six sets of self-made data sets.The results show that the two algorithms proposed in this paper have good adaptability in the application of autonomous aerial refueling cone tracking,and show better performance than before the improvement.In this paper,The algorithm presented in this study selects 20 representative sequences from OTB100 dataset for testing in the test,by comparison with good correlation filter algorithm and the algorithm experiment?The results show that the proposed algorithm is able to strong changes in illumination,scale changes,similar to the target jamming,shade under complex conditions such as meet long-term tracking accuracy and real-time performance.