摘要
针对移动机器人动态背景下运动目标的检测,提出一种基于全方位视觉的检测算法。首先,改进了SIFT算法中的特征点提取方法,在将图像划分为若干网格后,再根据特征点所在位置的局部区域熵对每个网格中的候选特征点进行筛选;其次,在SIFT点匹配后采用RANSAC算法去除误匹配点,以提高背景补偿的精度;最后用帧差法检测出运动目标。实验表明,该算法减少了SIFT点的获取时间,并具有良好的鲁棒性,能准确地在机器人运动过程中检测出运动目标。
According to the detection of moving object based on mobile robot under dynamic background, an algorithm of detection based on omnidirectional vision is proposed. The method of extracting feature points is improved in the SIFT algorithm. After dividing the image as grid ceils, the feature points of each grid cell are selected according to the entropy of local region around the feature point. Then after matching SIFT points, it employs RANSAC algorithm eliminate the mismatching points to improve the accu- racy of the background compensation. Finally, frame difference method is used to detect the moving object. Experimental results show that with a better robustness, the proposed algorithm decreases the time of extracting feature points and it can detect the moving object while robot is moving.
出处
《电视技术》
北大核心
2015年第1期117-120,126,共5页
Video Engineering
基金
浙江省重点科技创新团队项目(2010R50020)
国家自然科学基金项目(50905170
61007012)
关键词
动态背景
全方位视觉
SIFT
局部区域熵
背景补偿
dynamic background
omnidirectional vision
SIFT
entropy of local region
background compensation
