汽车工程 ›› 2021, Vol. 43 ›› Issue (10): 1419-1426.doi: 10.19562/j.chinasae.qcgc.2021.10.001

• •    下一篇

基于HS⁃FCM模糊聚类的快速多目标车辆跟踪算法

章军辉1,2,3,4(),付宗杰3,4,郭晓满3,4,李庆1,2,陈大鹏1,2,3,赵野1   

  1. 1.中国科学院微电子研究所,北京 100029
    2.江苏物联网研究发展中心,无锡 214135
    3.无锡物联网创新中心有限公司,无锡 214135
    4.昆山微电子技术研究院,苏州 215347
  • 收稿日期:2020-04-07 修回日期:2021-05-26 出版日期:2021-10-25 发布日期:2021-10-25
  • 通讯作者: 章军辉 E-mail:zhangjunhui@ime.ac.cn
  • 基金资助:
    江苏省博士后科研资助计划项目(2020Z411);国家重点研发计划“新能源汽车”重点项目(2016YFB0100516)

A Fast Multiple Maneuvering Vehicle Tracking Algorithm Based on Half Suppressed Fuzzy C⁃means Clustering

Junhui Zhang1,2,3,4(),Zongjie Fu3,4,Xiaoman Guo3,4,Qing Li1,2,Dapeng Chen1,2,3,Ye Zhao1   

  1. 1.Institute of Microelectronics of Chinese Academy of Sciences,Beijing 100029
    2.Jiangsu R&D Center for Internet of Things,Wuxi 214135
    3.Wuxi Internet of Things Innovation Center Co. ,Ltd. ,Wuxi 214135
    4.Institute of Microelectronic Technology of Kunshan,Suzhou 215347
  • Received:2020-04-07 Revised:2021-05-26 Online:2021-10-25 Published:2021-10-25
  • Contact: Junhui Zhang E-mail:zhangjunhui@ime.ac.cn

摘要:

为了提高复杂交通环境下多目标数据关联的实时性与可靠性,本文中基于半抑制式模糊聚类(half suppressed fuzzy c?means clustering, HS?FCM)发展了一种快速多目标车辆跟踪算法。首先对多目标车辆跟踪问题进行了数学描述,并建立了相机像素坐标系与世界坐标系的空间映射关系;其次基于模糊理论将点迹-航迹关联问题转换成量测模糊聚类问题,通过求解各候选量测与聚类中心的模糊隶属度,间接计算出联合概率数据关联(joint probability data association, JPDA)算法中不确定性量测与各目标的关联概率,再利用概率加权融合对多目标状态进行滤波估计;再次在车辆密集工况下通过合理调整卡尔曼增益对量测更新进行抑制,以克服车辆跟踪中目标短暂跟丢问题。实车试验与仿真结果验证了该跟踪算法的可行性与有效性。

关键词: 智能车辆, 数据融合, 感兴趣目标, 状态估计, 模糊C均值聚类

Abstract:

In order to improve the real-time performance and reliability of multi-target data association in complex traffic scenarios, a fast multi-target vehicle tracking algorithm based on half suppressed fuzzy c-means clustering (HS-FCM) is thus proposed in this article. Firstly, the multi-target vehicle-tracking problem is described mathematically, and the spatial mapping relationship between the camera pixel coordinate system and the world coordinate system is established. Secondly, the fuzzy clustering approach based on fuzzy theory is employed to solve the plot-track association problem. The probability of a feasible joint event in the joint probability data association (JPDA) algorithm is indirectly calculated by solving the fuzzy membership function defined by the distance between a sample and its cluster center. The multi-objective state is filtered and estimated by the probability weighted fusion method. Thirdly, in the dense vehicle environment the measurement update is suppressed by adjusting Kalman gain reasonably to solve the problem of short-term target losing. The real vehicle test and simulation results validate the feasibility and effectiveness of the proposed fast multi-vehicle tracking algorithm.

Key words: intelligent vehicle, data fusion, object of interest, state estimation, fuzzy C?means