Research On Correlation Filter Based Target Tracking Algorithm

Photoelectric tracking and measuring equipment is an important means of performance test and function verification of aircraft.In recent years,with the complexity of test aircraft target characteristics,the diversification of test and test mission scenarios,and the improvement of target tracking accuracy requirements,the difficulty of target tracking becomes more and more serious.Stable tracking should be carried out from the large target in the initial take-off stage,the multi-characteristic target in the motion state such as cloud background,illumination change and target characteristic mutation during the flight in the loading stage,to the complex ground object background target at the end of the flight.Therefore,the long-term and stable tracking performance of the target is particularly important.Especially after the object is blocked by the ground object and the sky background disappears into the cloud,the recapture becomes the difficulty of long-term tracking.High precision target tracking performance requires high frame rate target position feedback.Improving the real-time performance of tracking algorithm is another key technology to ensure the improvement of tracking performance.In recent years,the correlation filter based tracking algorithm is a branch of the traditional tracking algorithms in the field of target tracking.It has a good balance between tracking accuracy and running speed,and is very suitable for application in the photoelectric tracking and measuring equipment.Therefore,based on correlation filter based algorithm,this paper studies the technical difficulties of long-term stable tracking of the above targets,and implements and optimizes the algorithm on the tracker of embedded FPGA+DSP architecture,which has good application value.The main work of this paper is as follows:1.The requirement of tracking algorithm for the photoelectric tracking and measuring equipment is investigated,and the correlation filter based tracking algorithm is selected as the research object in this paper.The research status of correlation filter based tracking algorithm is deeply investigated,the principle of correlation filter based tracking algorithm is elaborated,and the framework of correlation filter based tracking algorithm is summarized.2.The fast Discriminative Scale Space Tracking(f DSST)algorithm was improved according to the requirements of re-detection and running speed of the tracking algorithm in the photoelectric tracking and measuring equipment.This paper proposes a correlation filter tracking algorithm based on filter space constraint and target redetection(named f DSST＿Sr Red).The details of the improvement include: in order to reduce the negative impact of the boundary effect on the f DSST algorithm,the spatial constraint method of the position filter is proposed.Aiming at the defect that the f DSST algorithm cannot relocate the target after the target is occluded and disappeared,a target re-detection method based on zero-filling of position filter is proposed.In order to judge when to start target detection,a tracking confidence evaluation index is proposed.Based on this confidence evaluation index,a dynamic filter update method is designed to make the filter learn more accurate target information and background information.3.Test the f DSST＿Sr Red algorithm.In order to explore whether the improvement of f DSST＿Sr Red algorithm works,the improved component of f DSST＿Sr Red algorithm is tested separately.In order to explore the differences between the f DSST＿Sr Red algorithm and other long-term tracking algorithms based on correlation filter,this paper compares and tests the f DSST＿Sr Red algorithm,LCT algorithm,MUSTER algorithm and Fu Co Lo T algorithm in OTB data set and UAV-20 L data set.Finally,the tracking effect of f DSST＿Sr Red algorithm is tested in a real scenario.The test results show that the improved components achieve the expected results.The filter space constraint method can increase the success rate of the f DSST algorithm by 6.7%and the accuracy by 9.1% while the search area is enlarged.Compared with f DSST algorithm,the success rate of f DSST＿Sr Red algorithm is increased by 7.8%,the accuracy is increased by 7.8%,the running speed is up to 113 fps,and has strong longterm tracking ability.f DSST＿Sr Red algorithm is applicable to visible image data and infrared image data.4.In order to verify the feasibility of f DSST＿Sr Red algorithm in FPGA+DSP architecture tracker implementation,this paper improves the Discriminative Scale Space Tracking(DSST)algorithm from two aspects of feature extraction and scale estimation,which are suitable for hardware.The improved algorithm is successfully implemented on an 8-core DSP platform using C language.In the process of implementation,the technical problems such as multi-core task allocation,inter-core communication,memory allocation,module function implementation,optimization and acceleration are solved.Tests show that the algorithm can accurately estimate the position and follow the change of the target scale,and the running speed is up to 30 fps.The DSST algorithm is the framework of the f DSST＿Sr Red algorithm.Through the implementation and verification of the DSST algorithm,it can be predicted that the f DSST＿Sr Red algorithm can be implemented in the FPGA+DSP architecture tracker of this paper by adding DSP processing cores,and meet the real-time requirements.