基于FPGA的双目实时测距算法设计

Design of Binocular Real-Time Distance Measurement Algorithm Based on FPGA

为了克服软件层面实现双目测距算法实时性差的缺点,提出了一种基于FPGA的高速的双目测距算法的硬件架构,该架构使用并行流水线的结构来实现兴趣点的检测与描述,通过乒乓操作来实现描述向量的缓存与处理,同时通过WTA(Winner Takes All)电路和三角测量法来实现视差的搜索和距离的测量.实验结果表明:在25MHz的时钟频率下,处理640×480的图像对只需12.5ms,处理速度约为软件上的68倍,满足了高速场景下实时性的需求.同时,处理速度和资源消耗也优于一些相关硬件实现的文献.

In order to overcome the shortcomings of poor real-time performance of the binocular distance measurement algorithm on the software level, we present a high-speed hardware architecture of binocular distance measurement algorithm based on FPGA, the architecture uses a parallel pipeline structure to detect and describe interest points, and description vectors caching and processing is achieved by the ping-pong operation, disparity search and distance measurement are realized by WTA （Winner takes All） circuit and triangulation method simultaneously. The experimental results show that it only takes 12. 5ms to deal with a pair of 640X480 images at the clock frequency of 25 MHz, and the processing speed is about 68 times of the software, it meets the real-time needs of some high-speed scenes. Simultaneously, the processing speed and resource consumption are also better than some related hardware implementations.