Study Of Geometry Detection Algorithm Based On Improved Hough Transform

The detection and recognition of geometry is a critical issue in digital image processing.The outline of most objects in life could be abstracted as geometry shapes.Fast and accurate detection of geometry shapes is significant in theory and practical applications.The main works of this paper include three aspects: study of circle detection algorithm,study of general geometry shape detection algorithm,image acceleration algorithm based on GPU.The work in this paper begins with the algorithms of detecting circular objects,then it is generalized to the detection of general geometry shapes,finally,the algorithm is accelerated by GPU parallel computing.In the aspect of circle detection algorithm,after analyzing and summarizing previous methods,making use of look-up table and bitwise center accumulator,this thesis proposes a new circle detection algorithm which is based on Hough transform.The new algorithm stores bias coordinates of every edge point related to the circle center.When detecting circles,the direction of edge point can be used to index bias coordinates,and they are voted in the bitwise center accumulator.Different from traditional accumulator,both of direction information and the number of votes are recorded in the bitwise center accumulator.The application of look-up table and bitwise center accumulator accelerate the algorithm and make the detection results more accurate.In the aspect of general geometry detection algorithm,based on template convolution algorithm and generalized Hough transform,this thesis generalizes look-up table and bitwise center accumulator,and proposed a new general geometry detection algorithm.The new algorithm build template of object geometry previously,when detecting objects,the direction information of edge point can be used to index bias coordinates,and they are voted in the bitwise center accumulator.The local maximum in the bitwise center accumulator indicates the presence of object geometry.In the aspect of image acceleration with GPU,based on the analysis of parallel processing technology with CUDA platform,especially the image parallelization in CUDA,including the edge extraction algorithm,circle detection algorithm.After analyzing the parallelism of the proposed algorithm,the thesis implemented the circle detection algorithm on CUDA platform.Compared with the program executes in CPU,the parallelized program is accelerated obviously.