Image Target Recognition System's Design&Realization Based On Class Specified Hyper Graphic

The attacked target used to be selected arbitrarily in the live battle ground,which runs short of transcendental knowledge.However,the common object detection algorithm needs to prepare for constructing the detecting model a great number of object's knowledge in advance.But the environment of live war is so complicate that the target presents various visible appearances under different surroundings.It is difficult for a fixed model which is unable to learn by itself to detect and recognize object.If a recognized system without ability of self-learning solely depend on image pixels and image feature processing,it is hard to track and recognize the arbitrary selected target.Moreover,It is separated for target tracking and recognizing data in the common recognition system.Thus,the precious recognized result can't work for tracking.The visible target recognition takes advantage of computing resource of machine to simulate the recognizing capability of human.The eyes work as the object sampling device and brain is employed to analyze and resolve the object's trait.The brain deals with the object's various appearances through self-learning and updating object's knowledge and the special associated mechanism.This thesis study the self-learning and updating of object's trait knowledge according to the feature knowledge continuity in the process of object's imaging.And the image object based on CSHG(class specified hyper graph)recognition system is designed after interpreting the theory of object's detecting,tracking and recognition based on class specific hyper graph emphatically.The following is my work and accomplishment.In this dissertation,firstly,the task of object recognition is emphasized as result of the instant interesting object under complicate surrounding.This paper brings in and takes advantage of the robustness of SIFT descriptors for illumination,affine variation and unsupervised self-learning trait of CSHG model.Starting from both the description of object trait and object model updating,it makes use of Haar-Adaboost specified object detection algorithm and achieves the stable recognition of object under complicate environment.Meanwhile,the adaptability of CSHG model for varied imaging condition is proved.However,the selection of instant interesting object used to be uncertain.So the Haar-Adaboost object detecting method which is based on the samples statistic model cannot work well.As we known,this method needs to prepare a lot of object samples and background samples for stable detection in advance.This paper comes up with a fresh object detection method which combine with the fast matching trait and the self-learning trait of CSHG model.Consequently,continuous learning from blank object's knowledge,the system can realize fast detection and recognition of multi-object.At last,Multi-computer cooperative recognition system are designed according to variant targets recognition task' performance requirement.Its framework and sub-function modules are illustrated.And the whole system has been run with certain software and hardware.