基于点云协方差描述子的多机器人目标识别与编队跟踪
Multi-Robot Object Recognition and Formation Tracking Based on Point Cloud Covariance Descriptors

针对多移动机器人目标识别及编队跟踪问题, 提出一种基于点云协方差描述子的目标识别方法及编队跟踪控制方法．为降低机器人端机载处理器负荷, 基于WebSocket协议搭建网络架构．在此基础上, 通过引入点云协方差描述子进行目标检测, 并利用支持向量机完成离线建模．结合支持向量机分类器与Kullback-Leibler Divergence(KLD)-Sampling自适应粒子滤波算法, 实现目标部分遮挡下的在线识别, 得到目标点云跟踪位置信息．利用势场函数和有向刚性图论综合设计编队跟踪控制器, 实现多机器人编队目标跟踪．最后通过实物平台进行实验, 结果表明, 所提出的基于点云协方差描述子的多机器人目标识别与编队跟踪算法, 与传统方法相比, 跟踪收敛时间缩短4 s, 跟踪精度提高约2.5% , 通过搭载有限数量传感器, 可以更有效地解决多机器人编队目标跟踪问题.
To deal with the problem of object recognition and formation tracking with multi-robot，an object recognition method based on point cloud covariance descriptors was proposed，and a new formation tracking control method was designed. In order to reduce the load on the robot，the network architecture based on the WebSocket protocol was built. Firstly，a point cloud covariance descriptor was introduced to perform object detection，and the off-line modeling was performed using support vector machine(SVM). Then，combining SVM classifier with Kullback-Leibler Divergence(KLD)-Sampling adaptive particle filter，the problem of on-line matching and recognition under partial occlusion was solved effectively，and on-line recognition was accomplished to obtain object point cloud tracking position information. Finally，a multi-robot formation tracking controller was designed based on potential field function and directed rigid graph theory. The multi-robot object tracking is realized and validated by physical platform. The experimental results show that compared with the traditional method，the tracking convergence time is shortened by 4 s and the tracking precision is improved by 2.5% ，which proves that the proposed algorithm can effectively solve the problem of multi-robot formation object tracking by carrying a limited number of sensors