基于Kinect的仿人机器人抓取目标定位

摘要

Target localization is the premise for a humanoid robot to accomplish grasping operation. Aiming at the problems that the robot monocular vision is easy to lose depth information and the binocular vision is difficult to effectively obtain the depth information of textureless objects, a target localization system is provided for humanoid robot grasping based on Kinect. First of all, Kinect and the ro-bot coordinates were established, and the Bursa coordinate transformation model which solved by the linear total least squares ( LTLS) algorithm was constructed. The grasping points sensed by Kinect were coordinated to the robot coordinate system, thus achieving the goal that the robot positioning target object. Finally, the system scheme was experimentally verified on the humanoid robot NAO. The results show that the robot can locate the target in real time and reliably within a certain space through the method, which is more accu-rate than the robot monocular vision. According to the target localization system provided, NAO robot accomplished the task of grasping different objects.%目标定位是仿人机器人实现抓取操作的前提.针对机器人单目视觉容易丢失深度信息,双目视觉难以对缺乏纹理特征的物体获得有效深度信息的问题,提供一种基于Kinect的仿人机器人抓取目标定位系统.首先建立Kinect和机器人坐标系,构建布尔沙坐标转换模型;然后利用线性总体最小二乘(LTLS)算法求解该模型;最后依据Kinect获取的抓取点坐标信息,通过坐标转换将其转换到机器人坐标系:从而实现机器人对目标物体的定位.在仿人机器人NAO平台上对该系统进行实验验证,其结果表明:利用该方法,机器人在一定空间范围内能够可靠的定位目标物体,并且较其单目视觉定位更准确;根据所提供的目标定位系统,NAO机器人实现了对不同物体的抓取操作.