一种改进的立体图像实时相位匹配算法

An improved algorithm of real time phase matching of stereo images

为了在基于现场可编程门阵列(FPGA)的双目相机系统中实现实时立体图像匹配,提出一种改进的立体图像实时相位匹配方法。在一定视差范围内,将图像分为互相不重叠的子块,通过比较相邻块的方向值和Gabor滤波结果,以寻找合理视差值为目标,使实测相位差能够最佳匹配到理想相位差。采用流水线技术与并行处理方式对算法进行硬件结构的映射,以提高该算法在硬件系统上运行的效率。实验结果表明,改进算法能够以97.3MHz的处理速度在Stratix II型FPGA硬件实验平台上对1024×1024的灰度图像实现30fps实时匹配处理。

in order to realize real-time stereo image matching,an improved method of real-time stereo image phase is presented based on binocular camera system of field programmable gate ar-ray （FPGA）.In certain disparity range,images are divided into sub blocks which do not overlap each other,and a reasonable disparity values is chosen as target through comparison between the orientation of neighboring blocks and Gabor filtering results,therefore the measured phase differ-ence can be the best match to the ideal phase difference.Pipelining technology and parallel pro-cessing method for mapping hardware structure are adopted to improve the efficiency of the algo-rithm running on the hardware system.Experimental results show that the improved algorithm can work at the processing speed of 97.3MHz at the experimental hardware platform of Stratix Ⅱ FPGA and realize real-time matching processing of 1024×1024 30fps gray image.