Moving-Object Tracking with In-Vehicle Multi-Laser Range Sensors
Moving-Object Tracking with In-Vehicle Multi-Laser Range Sensors
复制标题
DOI：
10.20965/jrm.2008.p0367
复制
发表时间：
2008-06-01
影响因子：
1.1
通讯作者：
Takahashi, Kazuhiko
中科院分区：
文献类型：
--
作者：
Hashimoto, Masafumi;Matsui, Yosuke;Takahashi, Kazuhiko
本文提出了一种在杂乱环境中使用车载 2D 激光测距传感器 (LRS) 进行运动物体跟踪的方法。一个LRS的传感区域在方向上受到限制，因此移动机器人配备了多个LRS来进行全向传感。由于每个LRS在自己的局部坐标系上获取激光图像，因此将激光图像映射到参考坐标系上，从而可以通过多个LRS的协作来实现目标跟踪。为了进行映射，对多LRS的坐标系进行校准，即，估计多LRS的相对位置和方向。校准基于卡尔曼滤波器和卡假设检验。运动对象跟踪通过两个步骤实现：检测和跟踪。每个 LRS 通过启发式规则和基于占用网格的方法从其自己的激光图像中找到移动物体。它通过卡尔曼滤波器和基于数据关联的分配算法来跟踪移动物体。当移动物体存在于LRS的重叠感知区域时，这些LRS交换跟踪数据并以分散的方式融合它们。基于规则的跟踪管理被嵌入到跟踪系统中以增强跟踪性能。室内环境中三个步行者跟踪的实验结果验证了所提出的方法。
This paper presents a method for moving-object tracking with in-vehicle 2D laser range sensor (LRS) in a cluttered environment. A sensing area of one LRS is limited in orientation, and hence the mobile robot is equipped with multi-LRSs for omnidirectional sensing. Since each LRS takes the laser image on its own local coordinate frame, the laser image is mapped onto a reference coordinate frame so that the object tracking can be achieved by cooperation of multi-LRSs. For mapping the coordinate frames of multi-LRSs are calibrated, that is, the relative positions and orientations of the multi-LRSs are estimated. The calibration is based on Kalman filter and chi-hypothesis testing. Moving-object tracking is achieved by two steps: detection and tracking. Each LRS finds moving objects from its own laser image via a heuristic rule and an occupancy grid based method. It tracks the moving objects via Kalman filter and the assignment algorithm based data association. When the moving objects exist in the overlapped sensing areas of the LRSs, these LRSs exchange the tracking data and fuse them in a decentralized manner. A rule based track management is embedded into the tracking system in order to enhance the tracking performance. The experimental result of three walking-people tracking in an indoor environment validates the proposed method.