Abstract: People tracking for mobile robot is an important application which reflects the intelligence of robot, and it has wide prospect and practical value. However, there exists a big challenge because of the complexity of environment and the uncertainty of people moving. This paper improves the particle filter for mobile robot people tracking based on the analysis of particle filter framework. On the hand, the color information, depth information and social force are combined to estimate the similarity between the template and candidate, which increases the tracking precision. On the other hand, a concept of secondary particle is proposed to overcome the loss of particle??s diversity, which increases the tracking accuracy. In the last, the improved particle filter, sequential importance resampling (SIR) particle filter and extended Kalman filter (EKF) algorithm are compared on the turtlebot robot and the public data IAS-Lab. Results prove the superiority of improved particle filter.

Key words: mobile robot, people tracking, particle filter, depth information, social force, secondary particle

摘要： 移动机器人对行人进行跟踪，是体现机器人智能的一个重要方面，具有广阔的发展前景和应用价值。然而环境的复杂性和行人运动的不确定性给行人的跟踪带来了极大的挑战。为此，在分析粒子滤波框架的基础上，对基本粒子滤波算法进行了两方面改进，提出了适用于移动机器人的行人跟踪方法。一方面在相似性估计阶段结合颜色信息、深度信息和社交力概念，提高了跟踪的精度；另一方面提出了二级粒子的概念，解决了粒子多样性缺失问题，提高了跟踪的准确度。在移动机器人turtlebot和公开数据集IAS-Lab上对改进的粒子滤波、序贯重要性重采样（sequential importance resampling，SIR）粒子滤波和扩展卡尔曼滤波（extended Kalman filter，EKF）算法进行对比，实验结果表明，改进的粒子滤波算法明显优于其他两种算法。

关键词: 移动机器人, 行人跟踪, 粒子滤波, 深度信息, 社交力, 二级粒子