复杂光流场运动分析与特征提取-.pdfVIP

第 25 卷 第 3 期 电子测量与仪器学报 Vol. 25 No. 3 2011 年 3 月 JOURNAL OF ELECTRONIC MEASUREMENT AND INSTRUMENT · 285 · DOI: 10.3724/SP.J.1187.2011.00285 复杂光流场运动分析与特征提取* 江济良 屠大维 周许超 陈 勇 (上海大学机电工程与自动化学院, 上海 200072) 摘 要: 针对车辆行车前方场景视频图像序列这一复杂光流场, 提出一种基于光流分析的车辆行车环境特征实时提取方法: 对固定在驾驶室的摄像机所拍摄的驾驶员视野前方行车视频图像序列进行区域分割, 得到 3 个特征区域, 采用五置信点光流算 法对其中两个区域的光流场进行光流估计, 并应用数理统计方法对其作置信空间和算术平均处理, 提高了光流估计的鲁棒性, 有效提取了自身行车速度、弯道、上下坡等特征信息; 对另一区域进行帧间差分来实时提取夜间行车环境迎面车辆信息。行车 实验结果表明: 该方法抗干扰性强, 实时性高, 能有效地提取车辆行车环境特征信息,提取准确率达 82.5%。该方法在自适应前 照系统、智能交通系统中具有潜在的应用价值。 关键词: 光流场；特征提取；区域分割；置信空间；差分 中图分类号: TP391.41 文献标识码: A 国家标准学科分类代码: 510.4050 Motion analysis of complex optical flow field and feature extraction Jiang Jiliang Tu Dawei Zhou Xuchao Chen Yong (School of Mechatronics Engineering and Automation Shanghai University, Shanghai 200072, China ) Abstract: According to the complex optical flow field of video image sequences which are from frontage scene in driving environment, a new method for extracting real-time driving environment feature based on optical flow analysis was put forward. The driving video image sequences, which were taken from forward visual field in driving environment by a fixed camera in the cab, were segmented into three regions. At first, the optical flow fields of the two regions were estimated by an improved optical flow algorithm with five-confidence-point constraint. Then they were processed by confidence space and arithmetic average with the principle of mathematical statistics. So the optical flow e