摘要
利用视觉传感器信息丰富的特性,提出了一种基于边缘对称性的行人检测方法。利用Sobel算子和Hough变换确定车辆前方的感兴趣区域(AOI),然后提取感兴趣区域图像的垂直边缘,根据行人腿部的垂直边缘对称性确定垂直边缘对称轴,并结合行人形态特征以确定行人初始候选区域,最后采用灰度对称性和局部熵对行人候选区域进行目标识别验证。道路试验结果表明,该检测方法识别有效、可靠,并具有良好的鲁棒性。
A pedestrian detection method based on edge symmetry according to the abundant features of optical sensors was put forward. Firstly, the area of interest (AOI) ahead of vehicle was determined by using Sobel operator and Hough transform. Secondly, as pedestrian legs had prominent vertical edge symmetry, the paper managed to figure out the column, which had the most prominent edge symmetry through vertical edge extraction in the AOL Combined with pedestrian shape features, the candidate pedestrian could be locked. Thirdly, the candidate pedestrian was validated based on the gray symmetry and local entropy. The experiment results show that the algorithm is effective, reliable and robust.
出处
《交通与计算机》
2007年第1期40-43,共4页
Computer and Communications
关键词
行人检测
安全辅助驾驶
机器视觉
对称性
pedestrian detection
safety assistant driving
machine vision
symmetry