| 本文已被：浏览 5581次 下载 9275次 |
码上扫一扫！ |
|
|
| 结合运动信息与表观特征的行人检测方法 |
|
郭立君1,2,3, 刘曦2,3, 赵杰煜1,2, 史忠植2
|
|
1.宁波大学 信息科学与工程学院,浙江 宁波 315211;2.中国科学院 计算技术研究所,北京 100190;3.中国科学院 研究生院,北京 100049
|
|
| 摘要: |
| 提出一种结合运动信息与表观特征的行人检测方法.在对通过表观检测子获得的候选检测窗口执行分割验证的框架中,将运动信息融入到基于图像序列的对象分割算法中,通过获取更准确的分割结果来提高对候选检测窗口的检测准确率.该方法利用运动信息更新运动对象的前景/背景分布模型,将颜色信息间接地融入行人检测中, 并通过形状特征表现出来,与行人表观检测子形成互补的特性,获得更好的检测结果.上述结论在CAVIAR 视频以及行人检测视频中得到了实验验证. |
| 关键词: 行人检测 形状先验 运动信息 图割 |
| DOI：10.3724/SP.J.1001.2012.04030 |
| 分类号: |
| 基金项目:国家自然科学基金(61175026, 60903141); 国家重点基础研究发展计划(973)(2007CB311004); 浙江省自然科学基金(D1080807); 浙江省新一代移动互联网用户端软件科技创新团队项目(2010R50009); 宁波市自然科学基金(2011A610193) |
|
| Pedestrian Detection Method of Integrated Motion Information and Appearance Features |
|
GUO Li-Jun1,2,3, LIU Xi2,3, ZHAO Jie-Yu1,2, SHI Zhong-Zhi2
|
|
1.College of Information Science and Engineering, Ningbo University, Ningbo 315211, China;2.Institute of Computer Technology, The Chinese Academy of Sciences, Beijing 100190, China;3.Graduate University, The Chinese Academy of Sciences, Beijing 100049, China
|
| Abstract: |
| This paper proposes a method of pedestrian detection that takes both motion information and appearance features into account. This could be done by integrating motion information into the segmentation algorithm in the framework, which performs the validation of segmentation on candidate detection windows obtained by the appearance detector. The paper considers that better segmentation results can raise the detection accuracy. Shape features are obtained by integrating color information indirectly into pedestrian detection by using motion information to model foreground/background distribution of moving object. Better detection performance benefits from the complementary advantages between shape features and pedestrian appearance detector. The claim is supported by these experiments based on CAVIAR and the test video with pedestrians. |
| Key words: pedestrian detection shape prior motion information graph cut |
