您的位置:首页  > 论文页面

仿射不变条件下基于学习算法的盲道实时检测

发表时间:2010-11-30  浏览量:2036  下载量:857
全部作者: 肖汉,王兆仲
作者单位: 北京航空航天大学图像处理中心
摘 要: 在仿射不变条件下,运用基于学习的算法实时检测盲道,以便引导盲人在盲道上行走。目前广泛使用的检测算法是基于特征的匹配算法,这类算法运算量大,很难达到实时检测。基于学习的算法通过将运算量转移到离线学习阶段,从而使得检测过程达到实时。为达到很好的鲁棒性,用少量的样本数据通过仿射变换,随机生成大量的模拟实际拍摄样本数据进行离线学习生成分类器,然后进行在线检测,最后通过形态学闭运算对检测到的区域进行后处理并统计所需信息。通过实验证明,该算法对盲道的检测结果鲁棒性强、实时性好、精度高。
关 键 词: 信息处理技术;仿射不变;学习算法;盲道检测
Title: Real-time tactile strip detection based on learning algorithm under affine invariance
Author: XIAO Han, WANG Zhaozhong
Organization: Image Processing Center, Beihang University
Abstract: A learning algorithm is proposed to detect tactile strips in real-time to guide the blind along sidewalks. At present feature-based matching algorithms are known for their robustness, but their runtime computational loads are heavy. The learning algorithm can shift much of the computational burden to a training phase, so it is real-time for detection. In order to make it affine invariant and robust enough, data used to train a classifier offline are synthesized by randomized affine transformation. Then the classifier is used to detect the tactile strips online and morphological operations are used to optimize the detected results. Results of the experiments show that this algorithm is accurate, robust and fast enough for frame-rate performance.
Key words: information processing technology; affine invariance; learning algorithm; tactile strip detection
发表期数: 2010年11月第22期
引用格式: 肖汉,王兆仲. 仿射不变条件下基于学习算法的盲道实时检测[J]. 中国科技论文在线精品论文,2010,3(22):2287-2292.
 
0 评论数 0
暂无评论
友情链接