交通运输系统工程与信息 ›› 2015, Vol. 15 ›› Issue (3): 49-55.

• 智能交通系统与技术 • 上一篇    下一篇

一种基于激光与视频信息时空数据融合的行人检测方法

张荣辉1,李福樑2,周喜1,蒋同海1,游峰*2,徐建闽2,杨三强3   

  1. 1. 中国科学院新疆理化技术研究所,乌鲁木齐830011;2. 华南理工大学土木与交通学院,广州510640; 3. 新疆交通科学研究院,乌鲁木齐830000
  • 收稿日期:2014-10-29 修回日期:2015-03-28 出版日期:2015-06-25 发布日期:2015-06-29
  • 作者简介:张荣辉(1981-),男,江西广丰人,博士、副研究员.
  • 基金资助:

    国家自然科学基金(51208500,51408237,51108192);中国博士后科学基金(2012M521824,2013T60904);新疆维吾尔自治区自然科学基金(2013211B36,2013211B43).

A Pedestrian Detection Method under Time-space Data Fusion Based on Laser and Video Information

ZHANG Rong-hui1,LI Fu-liang2,ZHOU Xi1,JIANG Tong-hai1,YOU Feng2,XU Jian-min2, YANG San-qiang3   

  1. 1. Xinjiang Technical Institute of Physics & Chemistry, Chinese Academy of Sciences, Urumqi 830011, China; 2. School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510640, China; 3. Xinjiang Academy of Communication Science, Urumqi 830000, China
  • Received:2014-10-29 Revised:2015-03-28 Online:2015-06-25 Published:2015-06-29

摘要:

针对城市交通行人安全问题,本文提出了一种基于激光与视频数据融合的行人检测方法.通过激光与视频数据空间和时间上的融合,将激光数据映射到图像坐标;在激光聚类过程中,采用K-means 聚类算法对激光云点进行聚类分析,然后运用行人宽度模型提取候选行人区域;在基于图像的行人检测过程中,选取头肩、躯干以及腿部人体特征部位,采用Haar-like 特征集和Boosting 算法进行训练,得到部位检测器;最后,基于贝叶斯决策的组合策略对候选行人区域进行有效判定.实验结果表明,本文所述算法有较好的检测精度和实时性能.

关键词: 智能交通, 行人检测, 激光点云, 机器视觉, K-means 算法, Boosting 算法, 贝叶斯决策, 智能交通系统

Abstract:

A pedestrian detection method based on laser and video information fusion is proposed concerning the pedestrian safety problem in urban traffic. Laser data are projected to image coordinate system through spatial and temporal combination of laser and video data. In the process of laser clustering, Kmeans clustering algorithm is adopted to conduct clustering analysis on laser point clouds, while pedestrian width model is employed to extract candidate pedestrian region. In the process of pedestrian detection, characteristic part of pedestrian such as head-shoulders, body and legs are selected, and Haar-like feature is adopted and trained through Boosting algorithm. The obtained part detector is used to detect pedestrian, deciding the validity of candidate pedestrian region through composition strategy based on Bayesian decision. The result of the experiment shows that the proposed algorithm has preferable real-time and detection performance.

Key words: intelligent transportation, pedestrian detection, laser point cloud, compute vision, K-means algorithm, Boosting algorithm, Bayesian decision, ITS

中图分类号: