汽车工程 ›› 2021, Vol. 43 ›› Issue (9): 1322-1327.doi: 10.19562/j.chinasae.qcgc.2021.09.008

• • 上一篇    下一篇

基于AMMFO优化的MAFPF车辆跟踪方法

黄鹤1,2(),吴琨1,2,李昕芮1,王珺2,王会峰1,茹锋1,2   

  1. 1.长安大学电子与控制工程学院,西安 710064
    2.西安市智慧高速公路信息融合与控制重点实验室,西安 710064
  • 收稿日期:2021-02-01 修回日期:2021-05-23 出版日期:2021-09-25 发布日期:2021-09-26
  • 通讯作者: 黄鹤 E-mail:huanghe@chd.edu.cn
  • 基金资助:
    国家重点研发计划项目(2018YFB1600600);陕西省重点研发计划项目(2021SF-483);陕西省自然科学基础研究计划项目(2021JM-184);长安大学中央高校基本科研业务费专项资金项目(300102329401);西安市智慧高速公路信息融合与控制重点实验室(长安大学)开放基金项目(300102321502)

A Multi⁃feature Particle Filter Vehicle Tracking Algorithm Based on Improved Moth⁃flaming Optimization

He Huang1,2(),Kun Wu1,2,Xinrui Li1,Jun Wang2,Huifeng Wang1,Feng Ru1,2   

  1. 1.School of Electronic and Control Engineering,Chang’an University,Xi’an 710064
    2.Xi’an Key Laboratory of Intelligent Expressway Information Fusion and Control,Xi’an 710064
  • Received:2021-02-01 Revised:2021-05-23 Online:2021-09-25 Published:2021-09-26
  • Contact: He Huang E-mail:huanghe@chd.edu.cn

摘要:

针对粒子滤波算法的在车辆跟踪应用中跟踪精度较差及样本贫化问题,提出了一种利用自适应变异更新策略飞蛾扑火优化的多特征粒子滤波车辆跟踪算法。首先,利用目标纹理与颜色特征的互补性,融合两种特征来提高粒子滤波算法在复杂场景跟踪的稳定性。其次,改进飞蛾扑火算法的更新机制,将自适应权值引入飞蛾的螺旋更新策略,并使随机变异策略与其交替更新,增大算法的搜索空间,使得算法更快速地搜索到全局最优。最后,根据阈值分层样本粒子,并使适应变异更新策略飞蛾扑火优化低权值粒子的分布状态,避免样本贫化现象发生。实验表明,本文算法能够在低样本粒子数下有效地提升粒子滤波算法的性能,并在车辆受到阴影、遮挡、尺度和角度变化等复杂背景下,仍能稳定精确地跟踪目标车辆。

关键词: 目标跟踪, 颜色特征, 纹理特征, 粒子滤波, 飞蛾扑火优化

Abstract:

To solve the problems of poor tracking accuracy and sample impoverishment of particle filter algorithm in vehicle tracking application, a multi-feature particle filter vehicle tracking algorithm based on adaptive mutation update strategy and moth-flaming optimization is proposed. Firstly, the tracking stability of the particle filter algorithm in complex scene is improved by combining the complementary characteristics of target texture and color features. Secondly, the adaptive weight is introduced into the spiral update strategy of the moth by changing the update mechanism of the moth-flaming optimization, and then the random mutation strategy is updated alternately to increase the search space of the algorithm, so that the algorithm can search the global optimum more quickly. Finally, the sample particles are layered according to the threshold and the improved moth-flaming optimization is used to optimize the distribution of low weight particles to avoid sample impoverishment. Experiments show that the proposed algorithm can effectively improve the performance of particle filter algorithm in the case of low sample particle number, and can still track the target vehicle stably and accurately in the complex background of shadow, occlusion, scale and angle change.

Key words: target tracking, color feature, texture feature, particle filter, moth?flaming optimization