上海交通大学学报

上海交通大学学报（自然版） ›› 2011, Vol. 45 ›› Issue (07): 1012-1016.

• 自动化技术、计算机技术 • 上一篇    下一篇

一种基于多传感器融合的车辆检测与跟踪方法

麦新晨a,杨明a,王春香b,王冰a   

  1. (上海交通大学 a. 电子信息与电气工程学院; b.机械与动力工程学院, 上海 200240)
  • 收稿日期:2010-09-17 出版日期:2011-07-29 发布日期:2011-07-29
  • 基金资助:

    教育部博士点基金(20070248097)，上海市科委世博科技行动计划(10dz0581100)

Multi-sensor Fusion Based Vehicle Detection and Tracking Method

 MAI  Xin-Chen-a, YANG  Ming-a, WANG  Chun-Xiang-b, WANG  Bing-a   

  1. (a. School of Electronic, Information and Electrical Engineering; b. School of Mechanical Engineering,  Shanghai Jiaotong University, Shanghai 200240, China)
  • Received:2010-09-17 Online:2011-07-29 Published:2011-07-29

摘要： 针对城市道路环境，提出了一种基于激光雷达和视觉的车辆检测与跟踪方法.首先，采用透视变换和多传感器联合标定，根据激光雷达数据生成包含车辆假设的兴趣区域，以提高车辆检测的可靠性和降低图像处理的计算量；然后，提出了一种基于多维特征空间马氏距离的车辆检测算法，通过提取兴趣区域内图像特征向量，并将其与标准向量间的马氏距离作为车辆状态估计；最后，采用Kalman滤波实现车辆运动跟踪.为了提高鲁棒性，将粒子滤波算法与Kalman滤波相结合，以在雷达信息不准确的情况下准确地实现目标状态估计.实验结果表明，该方法在城市环境中取得了比较理想的车辆跟踪效果.

关键词: 激光雷达, 视觉, 马氏距离, 粒子滤波

Abstract: To fulfill real-time vehicle detection and track in city environment, a method based on laser radar and vision was presented. By perspective transform, vehicle hypothesis is done through laser radar. For vehicle recognition aspect, a method based on multi-dimension space Mahalanobis distance is presented, which extracts feature vector in ROI, and using the Mahalanobis distance between it and standard vector as presence probability. In vehicle tracking, the system uses Kalman filter to fulfill target tracking, at the mean time, to improve robustness, a particle-filter method based on Kalman filter is presented, it could realize more precise target state estimation when laser radar data does not work well. The experiment shows this method can achieve better vehicle tracking in general city environment.

Key words: laser radar, vision, Mahalanobis distance, particle filter

中图分类号: