||
本文为瑞典查尔姆斯理工大学(作者:Erik Henriksson)的硕士论文,共76页。
本文研究了利用汽车雷达传感器对动态目标进行扩展跟踪。跟踪是基于一个360度环境感知系统的数据,该系统由四个视场重叠的雷达传感器组成。本文提出了两种跟踪目标状态的方法,包括位置、速度、航向和大小。第一种算法基于检测形成的集群,并创建用于扩展目标跟踪器更新步骤的矩形框。第二种算法使用高斯混合概率假设密度(GMPHD)滤波器,对该滤波器的各个分量进行聚类,并在滤波分量周围创建一个矩形框。对采集数据的评估表明,在位置和速度精度方面,这两种方法都有很好的效果。然而,基于检测的跟踪解决方案比基于PHD的解决方案显示出更稳定的结果。当涉及到目标航向和物理尺寸扩展的估计时,所提出的解决方案有些不同,但估计的效果并不好。尤其是在远距离上,航向和尺寸估计是不稳定的。
This thesis focuses on extended targettracking of dynamic objects using automotive radar sensors. The tracking is basedon data from a 360-degree environmental perception system comprising four radarsensors with overlapping fields-of-view. Two solutions are proposed to trackthe states of the objects, which include position, velocity, heading and size.The first algorithm forms clusters based on detections and creates rectanglesthat are used in the update step of an extended target tracker. The secondalgorithm uses a Gaussian Mixture Probability Hypothesis Density (GMPHD)filter, clusters components of that filter and creates a rectangle around thefiltered components. Evaluation on logged data shows good results for bothsolutions in terms of position and velocity accuracy. However, thedetection-based tracking solution shows a slightly more stable result than thePHD-based solution. When it comes to estimation of the heading and the physicalextension of objects, the proposed solutions differ somewhat, but both producerather poor estimates. Especially at long ranges, the heading and sizeestimates are unstable.
更多精彩文章请关注公众号:
Archiver|手机版|科学网 ( 京ICP备07017567号-12 )
GMT+8, 2024-9-23 10:51
Powered by ScienceNet.cn
Copyright © 2007- 中国科学报社