Design And Implementation Of Real Time Target Recognition System Based On CUDA

Robot visual servoing is a complex system which can be applied in different fields.This paper focuses on the fast target recognition based on CUDA.The camera information is collected,and finally the position deviation of the target is given,which is output to the visual servo system for control applications.First of all,this paper focuses on the related problems of target recognition algorithm,including tracking method,identification method and so on.Secondly,we focus on the parallel optimization problem,and optimize the implementation of the CUDA platform for the basic modules,such as SIFT,CAMSHIFT and so on.Finally,the feasibility and performance of the proposed method are verified by the robotic visual servo system.The content of this paper includes four parts,each part is summarized as follows:The first part focuses on the content of several basic algorithms involved in the thesis,including the explanation,understanding and explanation of some principles.First of all,a brief introduction to the algorithm process under the project background is given,and the actual input and output are explained.Secondly,the basic knowledge of SIFT is introduced,including scale space construction,extreme point detection,feature point gradient calculation,feature descriptor calculation,feature matching and so on.Thirdly,the basic knowledge of CAMSHIFT is introduced,including histogram generation,reverse probability projection,image moment calculation,histogram intersection and so on.Finally,the related contents of parallel optimization are introduced,including parallel specification,Amdahld theorem,Gustafson theorem,parallel optimization principle,and so on.The second part focuses on the specific design of the fast target recognition algorithm.Including concrete application realization,concrete cooperation way and so on.Firstly,SIFT feature matching is introduced to realize stable feature matching,which is used to provide stable feature reference.Secondly,the fast target ROI acquisition based on CAMSHIFT tracking is introduced.Finally,the evaluation mechanism and recognition strategy of the algorithm are introduced.The third part focuses on the actual parallel optimization design of the fast target recognition algorithm.From the point of view of parallel optimization,the specific application of correlation principle is presented.First of all,the design and implementation of the actual CUDA framework is based on the SIFT related modules.Secondly,the design and implementation of the actual CUDA framework is based on the CAMSHIFT related submodule.The fourth part has carried on the concrete experiment.Separately from the single module recognition effect,the overall recognition effect and so on,shows the actual effect of this method.And the results are briefly analyzed and introduced.