基于FPGA的实时SGM匹配算法研究与实现

Research and implementation of real-time semi global matching algorithm based on FPGA stereo vision

传统SGM算法,运算复杂度高,硬件资源需求量大,难以应用到实时嵌入式系统中。为此提出一种基于FPGA嵌入式平台的实时SGM(Real-Time SGM,RT-SGM)算法。RT-SGM选取三个方向作为匹配算法的优化方向;设计新的算法的结构,使该算法能运行在Pipeline状态下;提出一种新型中值滤波算法对结果进行优化。在FPGA硬件平台上完成实验。实验结果表明,RT-SGM运行速度相比于传统SGM算法提高了30%,而在资源需求上只有传统SGM算法的一半,同时其精度与传统SGM算法相当,适合应用到实时嵌入式系统中。

Traditional Semi Global Matching algorithm is not suitable for running on real-time embedded platform because of its complex structure, and high demand for hardware resources. Thus, this paper puts forward the Real-Time SGM algorithm based on FPGA embedded platform. Real-Time SGM selects three directions as the optimization direction to reduce the demand for hardware resources. Then, the algorithm can run in the condition of Pipeline by designing a new structure,and improve the operation speed. Moreover, it puts forward a new kind of median filter algorithm to optimize the results.Finally, simulation is done on the PFGA platform. Experimental results show that compared with the original algorithm,the new algorithm has lower resource requirements, and has a large increase in the speed of the algorithm, which is suitable for running in low power consumption embedded system.