基于非极大值抑制的SUSAN算法改进及硬件实现

Improvement and implementation of SUSAN algorithm based on non-maximum suppression

基于FPGA的嵌入式双目立体视觉系统中要求其边缘检测算法具备抗噪能力强、定位精度高、实时性好、易于硬件实现的特点。本文在SUSAN算法的基础上,提出了一种能够满足上述要求的改进边缘检测算法,先用SUSAN算法计算图像的初始响应和法线方向,再进行基于USAN重心与中心点距离阈值的非极大值抑制,最后进行边缘连接,得到边缘图像。该算法各模块采用Verilog HDL语言编程,并在FPGA中验证实现;边缘检测结果用于立体匹配,匹配效果良好,完全满足设计要求。

The method of edge detection binocular stereo-vision system based on FPGA must have characters of strong noise-rejection, high position accuracy, good real-time performance and hardware implementation easily. This paper proposes a modified SUSAN algorithm for edge-detection which is well suited to the FPGA-based binocular stereovision system. Firstly, the original response and normal direction of each pixel is calculated by SUSAN algorithm. Then non-maximum suppression is applied according to the distance threshold between USAN barycenter and center. Finally, connecting the edges and getting the edge images. The algorithm is programmed by Verilog HDL and implemented in FPGA, The result indicated that the design performance is good when applied to stereo matching.