光学仪器, 2014, 36 (3): 208, 网络出版: 2014-07-10
基于FPGA的Lucas-Kanade算法优化
The optimization of the Lucas-Kanade algorithm based on FPGA
摘要
在FPGA平台上实现基于光流法的视频运动目标跟踪系统,采用Lucas-Kanade算法进行光流场的计算,在图像预处理阶段提出使用三维高斯滤波代替传统二维高斯滤波,引入相邻像素点在时间轴方向的相关性,增强图像的滤波效果。在3D导数计算阶段提出在求导方向的正交面上进行平滑滤波,并采用匹配的导数和平滑参数,提高光流场计算精度。在FPGA平台上设计多级主流水线加子流水线结构,设计了四端口RAM进行图像缓存,优化了最小二乘矩阵单元和浮点数运算单元,实现了实时视频运动目标跟踪。
Abstract
Video motion tracking based on optical flow is implemented on the FPGA, and we calculate the optical flow taking advantage of the Lucas-Kanade algorithm. In the stage of the image preprocessing, we propose that three-dimensional Gaussian filter is used to replace the conventional two-dimensional Gaussian filter and the correlation of the adjacent pixels is introduced, which improves the effect of the image filter. In the stage of 3D derivative calculation, we implement the smoothing filter on the orthogonal plane of derivative directional, and use the matching derivative parameter and smoothing parameter, so the accuracy of the optical flow is improved. Based on the FPGA platform, we design multistage pipelined architecture that includes several sub-pipeline, the four-port RAM unit that can achieve images cache, and optimize the least-square matrices and the float unit, which achieves the real-time video motion tracking.
刘东明, 刘超, 牟海维. 基于FPGA的Lucas-Kanade算法优化[J]. 光学仪器, 2014, 36(3): 208. LIU Dongming, LIU Chao, MU Haiwei. The optimization of the Lucas-Kanade algorithm based on FPGA[J]. Optical Instruments, 2014, 36(3): 208.