The autonomous human-target identification and tracking is a key issue in service robotics. Human body is a special type of non-rigid target for which there are many environmental factors constraining the application of target tracking. Based on the monocular vision and laser scanning, we develop a practical technique to incorporate the extended two-dimensional environment description with the indoor environment modeling of a mobile robot, constituting the basis for human target identification. In order to adapt this model to the target rotation, shielding and overlapping, a method based on the target color distribution model and Kalman filter is used to improve the multiple people tracking, in which the target initialization is implemented by a real-time laser scanning data processing algorithm. A series of experiment results with a further experiment data analysis show the validity and practicability of this method; it also meets the requirements of target rotation, partial shielding and overlapping in practical application.%移动机器人如何自主实现人体目标的检测与跟踪是服务机器人研究领域中的关键问题之一.在深入分析单目视觉和激光测距特性的基础上.文章首先针对室内场景进行扩展2维环境建模研究,并提出在该环境下的人体目标分段模型构建与自主辨识方法.为了有效实现对多个人体目标的同时跟踪,本文提出了一种基于非恒速运动模型和卡尔曼滤波对多人体目标进行有效匹配与跟踪的方法.实验表明本文所提方法能有效的克服目标旋转、部分遮挡和重叠以及光线明暗变化给人体目标跟踪带来的影响,具有较好的鲁棒性和实用性.
展开▼
