Research On Target Tracking And Key Techniques Of Electro-optical Image Guidance System

Electro-Optical image guidance system is a kind of passive image tracking equipment which works in the visible light,low light level or infrared wave band.The precision guided weapon represented by this kind of system plays a decisive role in the modern war,among which,target tracking technology is the key technology.However,for the tracker of the homing photoelectric imaging guided weapons,there remain some aspects that have not been resolved fully or not very perfectly,which include the following aspects: small target detection and recognition tracking under long distance and complicated background;area target tracking under the interference factors such as small target gradually expanding,the change of the scale,the change of illumination,the affine deformation and so on which are caused by the distance decreasing between the target and missile;tracking drift,the target being occluded or out of the field of view as well as other interference of similar targets during long time tracking.The paper studied the aforementioned three problems combined with the project requirements.Furthermore,the feasibility,stability and realtime performance of the proposed algorithm were simulated and verified on a unique set of video and public video.Finally,we developed a set of tracking system using the existing technology and resources from the theoretical research and engineering practice.To further verify that the tracking system had stable performance,we did the semi physical simulation experiments and outfield tracking experiments.The main contents of the paper and the results obtained were as follows:Firstly,the research background and significance were introduced combined with the project background,the development of the classic anti tank missile which carries the missile seeker was investigated.Moreover,the development status in the field of photoelectric imaging seeker working in the process of detection and target tracking was described.And then the key and difficult problems were summed up together with the direction of the work being pointed out.Secondly,before launching the missile,with the large distance of the missile and target,the technology of detection and recognition of infrared dim target was studied and the characteristics of infrared dim small targets in the actual scene in the development process of the project were described.Several commonly used classical preprocessing techniques were summarized,and their functions were simulated and verified.According to the technical and tactical requirements before the launch of the lock,we proposed a target detection and tracking algorithm which was effective to separate the target and the background by constructing the difference map based on the statistical characteristic under the circumstance of imaging in small area and weak characteristics.Following the fusion of temporal and spatial filtering methods eliminated the suspected targets.The principle of the algorithm was introduced in detail and simulation experiments were carried using different scenarios of the data to evaluate the adaptability of the algorithm.Thirdly,a new tracking algorithm combining online learning and the correlation filter was proposed to aim at such tracking difficulties like small target gradually expanding,illumination change,scale change and other interference factors caused by distance decreasing between the target and missile,which not only paid attention to the target position changes but also on the size change of target.The new algorithm constructed the feature of Pyramid and filtered the target in 3 dimentions to focus on the change of displacement but also pay attention to the scale change.Consequently,the algorithm timely studied and updated template reasonably to avoid tracking drift and improved the real-time performance using fast Fourier transform and inverse transform technology to accelerate the calculation speed.Then the new position and size of the target were calculated by the response value to achieve stable and accurate tracking;Finally,a new correlation tracking algorithm based on discriminative was proposed to solve the problems,such as being unable to track for a long time or deal with the existing target occlusion.The new algorithm fused three mechanisms which named occlusion judgment,adaptive template update and re-detection to solve above-mentioned problems.To be more specific,the spatio-temporal context model and target model were utilized to solve the occlusion and judge the lost time to avoid excessive loss of template,with fusion of a detector searching the target when reappearing.To verify the feasibility and stability of the new algorithm,a large number of simulation experiments were made on the video sequences which existed scale changes and occlusion.And the comparison with several current popular tracking algorithms was analyzed to verify the improvement;Furthermore,the target tracking system under complex background was developed based on DSP+FPGA and applied to achieve such functions from the implementation of the image acquisition,denoising and enhancement to feature extraction,detection and tracking of the real-time processing.To further verify that the tracking system had stable performance,we did the semi physical simulation experiments and outfield tracking experiments.