基于卡尔曼滤波的多特征加权最近邻数据关联与跟踪算法

针对传统最近邻数据关联算法正确率较低且容易出现漏关联的问题,提出一种多特征加权的最近邻关联算法。根据智能车环境感知系统获得的障碍物特征数据,定义了一种相似度函数,提出基于生命周期计算有效关联度的方法,从而判定目标是否关联;基于卡尔曼滤波对关联目标进行迭代更新,实现对目标的跟踪;通过实验对比了静止目标、无交互的低速运动目标和有交互的低速运动目标的跟踪轨迹。结果表明,与传统的最近邻数据关联算法相比,所提出的改进算法可以实现对低速运动目标准确连续的关联跟踪,不会出现目标丢失或位置突变的现象,且跟踪目标的交互与遮挡对跟踪效果影响较小,具有较高的有效性与实用性。研究结果可为智能车辆的目标跟踪设计提供参考。

Multi-feature weighted nearest neighbor data association and tracking algorithm based on Kalman filter

Aiming at the problems of low accuracy and missing association in traditional nearest neighbor data association algorithm , a multi-feature weighted nearest neighbor association algorithm was proposed. A similarity function was defined according to the obstacle data obtained by the intelligent vehicle environment perception system, and a method was proposed to calculate the effective correlation degree based on the life cycle, so as to determine whether the objects are related. Based on Kalman filter, the associated target was updated iteratively to realize the tracking of the target. The tracking trajectory of stationary target, low-speed moving target without interaction and low-speed moving target with interaction were compared through experiments. Experimental results show that compared with the conventional nearest neighbor data association algorithm, the improved algorithm proposed in this paper can realize the accurate and continuous associated tracking of low-speed moving targets, and there will be no the phenomenon of target loss or position mutation. With high effectiveness and practicability, the interaction and occlusion of tracking targets have less effect on the tracking performance. The research results may provide reference for the target tracking design of intelligent vehicles.