Research On Infrared Object Tracking Algorithm Based On Correlation Filters

The object tracking algorithm based on infrared image is an important research direction in the field of computer vision.It has been widely used in the fields of automatic driving,intelligent security,infrared guidance,and so on.Infrared image is not affected by the illumination intensity and external environment,so it can be used all-weather.However,infrared image also has some disadvantages,such as low contrast,unobvious edge features,low signal-noise ratio,etc.These disadvantages bring some challenges to the design of the object tracking algorithm based on infrared image.In recent years,the correlation filter algorithm has been widely used in the object tracking algorithm under visible image because it has advantages in accuracy and speed.However,due to the existence of many differences between infrared image and visible image,the performance of correlation filter algorithm is unsatisfied when it is applied to infrared image.This paper focuses on solving this problem.Firstly,the causes of this phenomenon have been analyzed.Then,the correlation filter algorithm based on the characteristics of infrared image has been improved.The main research works are as follows:(1)Aiming to solve the infrared object deformation and background interference caused by the low contrast and unobvious edge features of the infrared imagery,which may lead to the problem of model drift and tracking failure during tracking,the feature extraction part in correlation filter algorithm has been improved,namely the multi-feature fusion correlation filter algorithm.Firstly,the multi-cue fusion algorithm is used to fuse the three infrared features proposed in this paper,i.e.OCS-LBP feature,HSC feature and GRAY feature.Then,the distractor-aware model is introduced to suppress the interference of similar object in the infrared image.Finally,the adaptive fusion method is used to fuse the response graph results of the two methods and then to get the best object tracking location.(2)Aiming at the problem of motion blur caused by low signal-noise ratio of infrared image,rapid motion and occlusion in the tracking process,the model part of the correlation filter algorithm has been improved.Context-aware correlation filter algorithm is proposed.The infrared object tracking algorithm introduces the background information in the tracking process to cope with the tracking failure caused by the unclearness or occlusion of the infrared object,to improve the performance of the algorithm.(3)In order to solve the model pollution problem caused by the object occlusion and object missing in the tracking process,a high-confidence model update strategy to improve the model update strategy in the correlation filter algorithm is presented.Based on the proposed strategy,the object localization precision and robustness of the infrared object tracking algorithm are further improved.Meanwhile,the model migration and the model contamination during the tracking process are defused.