Research On Target Detection And Tracking Based On Infrared Image

Infrared target detection and tracking is an important research direction in the field of computer vision,and it has a wide range of applications in autonomous driving and intelligent security.Compared with visible light images,infrared images are not affected by the intensity of light and external environmental factors,and can realize all-weather applications.However,due to the small number of infrared weak targets and the lack of internal features,the infrared target detection and tracking system applied in the project is not ideal in the face of small targets and is not applicable.Aiming at the above problems,this paper proposes an infrared target detection and tracking system,which can perform robust detection and tracking on both pedestrian targets and weak targets,so that the applicability of the infrared target detection and tracking system is improved.The specific work is as follows:(1)Infrared target detection.To improve the detection rate of infrared target detection systems: Firstly,in the image preprocessing stage,the Lo G filtering process is improved based on the human vision mechanism to obtain a more robust image.Secondly,in the candidate region generation stage,an information fusion algorithm is proposed to perform thermal analysis and motion analysis on the infrared image,making full use of the infrared image target brightness information and moving information to generate target candidate regions.Thirdly,the OCS-LBP feature and the three-line interpolation HOG feature are introduced.They use the texture information and gradient amplitude information of the infrared target,respectively.Finally,the above two features are fused to obtain a more descriptive fused feature for the infrared image target,and a random fern classifier based on the fused feature is used to classify and extract the target to obtain the final target detection result.In order to improve the applicability of the infrared target detection system,an infrared weak small target detection module is added: for the small target in the infrared image,an improved local contrast algorithm is proposed,which can make full use of the contrast information between the target and its surrounding background to effectively detect the infrared small targets.(2)Infrared target tracking.After the position of the target is obtained in the detection phase,the target is tracked in the infrared video image.To improve the tracking accuracy and applicability of infrared target tracking systems: Firstly,context information is introduced during the training phase,and a high-confidence model update strategy is introduced during the model update phase,which solves the problem of target loss due to severe occlusion of the target and improves the performance of the algorithm.Secondly,When tracking infrared weak small targets under a complex sky background,Gaussian curvature filtering is used to reduce the influence of the edge of the background cloud in the infrared image on the tracking effect.Finally,when multiple targets need to be tracked in the image,an improved multitarget tracking algorithm is proposed.The algorithm uses weighted fusion of the tracking results of two improved features,establishes a correlated filter tracker for each detected target,and correlates the target's motion trajectory step by step with the confidence of the tracking slice,achieving the effect of accurately tracking multiple targets.Strengthen the applicability of infrared target detection and tracking systems.