Research On Target Tracking And Identification Technology In Vehicle-mounted Viewing System

Targets tracking and recognition technology in vehicle-mounted imaging system is widely applied and researched in military and civilian areas.The robust tracking and recognition technology is studied in this thesis aiming at the situation of vehicle-mounted infrared and visible light circular imaging systems with common aperture.Compared with conventional image tracking procedures for fixed visual angle videos,the disparate key points in our project include: First,the long image sampling period results in long moving distance between same target in neighbor frames;Second,platforms machine swing,shaking and other factors result in targets trajectory's regularity;Third,front-view terrestrial environment result in complex background.So,the algorithms need abilities such as flexibility to imagery quality,rapid searching,anti-false alarms and other interferences.According above states,the major works in this thesis are:1)The characteristics of the typical vehicle-mounted images were analyzed.The tracking requirement was confirmed for multi-imaging systems circular viewing with common aperture.The imagery features and complementary attributes were compared between infrared and visible light images according target types.And some compatible preprocess algorithms were organized;For the visible light video image sequences with fixed visual angle,the classical Mean shift tracking algorithms was designed and implemented to region targets.2)An extended Mean Shift tracking algorithm was proposed for infrared targets tracking under the circular viewing situation.Through the searching strategy from rough to fine,the problem of wide range searching was solved.A set of soft matrices based on Bhattacharyya similarity were presented to accommodate multi-character templates matching and improve targets description for anti-false alarms.Another more,to avoid the degeneration of tracking template,a feasible mechanism of dual template updating was studied which can altered by tracking result and similarity measures.3)To enhance the complementary sensing superiority of the two types of imagery,a fusion tracking approach was explored for the multi-source and asynchronous system.Based on the geometry registration information of common aperture,the tracking box from infrared image was projected to the visible light image to guide tracking.A correlative degree measure method was proposed to portray concordant regions from two images by mutual information of the histogram of gradient.Combined with the correlative degree and the similarity matrices among tracking,the algorithm realized fusion tracking procedure step by step.The above methods were verified by sets of real videos.The tracking results is good and real-time.