Feature Extraction And Target Recognition Based On Image Information

Target recognition system which based on image feature has comprehensive applications in various fields, and its been as the study object for many institutions and organizations all over the world.And to improve the the real-time and accuracy of target detection algorithm for the successful completion of the recognition and location mission have a very important significance and the purpose of thesis is mainly concentrated on this point.At the field of UAV research, micro-unmanned helicopter flight highly favored by researchers because of its flexibility. By using on-board embedded vision systems, a variety of military and civilian mission to be completed, which greatly raised its practical value. And with its small size, simple structure, light weight, the maintenance of convenience and mobility, unmanned helicopters has great development potential and good prospects of application.Therefore, the micro unmanned helicopter systems have recently attracted considerable attention. And image processing system is a very important part of mini unmanned helicopter, which mainly dealing with how to detect the object and supply the navigation computer with location information of the object.Learning from the image recognition technology, the whole work includes design and comply target classification algorithm, which have been proved valid through emulation experiments.In order to overcome the low precision shortcomings of varied background and the moving of the canmera, the characteristics of the image information extraction and target recognition methods were studied.The proposed algorithm realized the target recognition method with the SVM classifier and the effective image features which based on the digital image preprocessing.For against the mutations background, the image detection algorithm dealed with video image with the Gaussian mixture model combining support vector machines and set up a more comprehensive system of rapid target identification.With the characteristics of support vector machine classification's accuracy and Gaussian mixture model's self-adaptive and fast speed, the proposed method is able to achieve a much better realtime and accuracy property.