摘要
文中介绍了一种利用移动机器人的激光信息和摄像头信息实时跟踪目标的方法。实现了对人的准确迅速的跟踪。通过大量提取照到人的双脚的激光特征作为样本集,描述了一种基于条件随机场(CRF)模型的Rao-Blackwellized particle filter(RBPF)算法,CRF的observation potential可以直接从样本数据中获得。采用类似栅格滤波方法计算样本的后验概率。RBPF算法根据后验概率进行权值的更新和采样实现对激光特征的实时跟踪,从而实现人的跟踪。根据人的位置信息可以确定人在摄像头图像窗口中的大概位置,提取该位置的SURF特征,从而获得人在图像中的精确位置。
Introduce a real- time target tracking method for a mobile robot using laser range data and camera images. It can track a person quickly and accurately. Obtained plenty of laser features which hit the person'legs as examplars set,discribe a Rao- Blackweilized particle filter(RBPF) algorithm based on CRF model, where the observation potentials are learned from data. In order to compute the posterior the grid filtering is used. The RBPF algorithm updated the weights and sample based on the posterior to realize the real -time tracking. According to the position of the person can estimate the person' position in the image window, get the precision position of the person by computing the SURF features in this area.
出处
《计算机技术与发展》
2010年第4期113-116,共4页
Computer Technology and Development
基金
广州市科技攻关项目(2007Z32D3151)