Long-term Tracking Of Infrared Visual Object Based On Improved BACF Algorithm

With the rapid development of visual target tracking technology,a large number of infrared target tracking algorithms have been proposed in recent years,which play indispensable roles in various fields.Nevertheless,the infrared target tracking has become a challenging task in machine vision due to its low resolution,poor contrast,and the lack of texture information coupled with complex background environmental effects.The background-aware correlation filtering algorithm(BACF)shows good tracking performance in visible target tracking tasks while it performs poorly in infrared target tracking tasks.Aiming at the problem that the BACF algorithm is prone to algorithm drift under occlusion,fast motion and motion blur,a long-term infrared target tracking based on the improved BACF algorithm is proposed in this paper.The specific research work and results are as follows:(1)In order to solve the problem of tracking failure of the BACF algorithm under occlusion,fast motion and motion blur conditions,a long-term infrared target tracking based on multi-feature algorithm is proposed.First,HOG feature and motion feature are combined to enhance the tracker's ability to represent the appearance information of the infrared target.Second,the spatial window weighting is used instead of the original cosine window weighting to give higher weight to the prominent feature of the target center.Meanwhile,the edge effect can be suppressed.Then,an evaluation response quality function is proposed to determine whether the target is occluded.When the quality value is lower than the set threshold,the filter template will not be updated to prevent the filter from learning the background information.Finally,the structured support vector machine is used to re-detect the infrared target after the tracking failure,thereby achieving long-term tracking.The experimental results show that the accuracy of the long-term infrared target tracking based on multi-feature algorithm is 0.788 and the success rate is 0.712 under the OPE evaluation standard,which is 1.3% and 8.2% higher than the benchmark algorithm.The EAO score is 0.241,and the A and R scores are 0.662 and 0.845 under the VOT evaluation standard.(2)A long-term infrared target tracking based on the improved BACF algorithm is proposed to solve the problem that the BACF algorithm cannot achieve the target relocation after failure in the infrared target tracking process.The target is relocated when the algorithm tracking fails by introducing a re-detection module track.Firstly,determine if the target has failed to track through the re-detection mechanism.When the quality of the response graph and the number of feature point matches are lower than the thresholds set,it indicates that the target has failed to track.Then,the target is relocated using the relocation strategy.Finally,long-term tracking of infrared targets is achieved.The experimental results show that the accuracy of long-term tracking of infrared targets based on improved BACF is 0.806 and the success rate is 0.726 under the OPE evaluation standard,which are 3.6%and 10.3% higher than the benchmark algorithm.Under the VOT evaluation standard,the EAO score is 0.241,and the A and R scores are 0.662 and 0.845.