Abstract: The feature fusion based particle filter algorithm can make tracking system more robust, by fusing different features. However, currently these algorithms have some drawbacks, such as insignificant feature differences, poor in real-time process, and confined fusion tactics. Based on that, an improved fusion algorithm for preceding vehicles tracking system is proposed, which adopts new intensified edge information SULBP feature, and enhances the real-time feature extraction through adaptive dimensionality reduction method. In addition, this fusion tactic appears more universal by designing the adaptive fusion algorithm based on the distribution state of the particle set. Indicated by the experiments result, this multi-feature fusion particle filter  algorithm is improved in both tracking performance and feasibility.

Key words: preceding vehicles tracking, particle filter, SULBP feature, adaptive dimensionality reduction, adaptive fusion method

摘要： 基于特征融合的粒子滤波算法可以将多个不同的特征进行融合，增强跟踪系统鲁棒性，但是现有的算法存在着特征显著性差，算法实时性不强以及融合策略不具备通用性的缺点。针对上述问题提出了一种适用于前车追踪系统的改进融合算法，采用增强边缘信息的SULBP新特征，并通过自适应降维方法提升特征提取的实时性；利用粒子集的分布状态设计自适应融合算法解决了融合策略的通用性问题。实验结果表明，所提出的多特征融合粒子滤波算法在跟踪性能和算法实时性上均有显著地提升。

关键词: 前车追踪, 粒子滤波, SULBP特征, 自适应降维, 自适应融合策略