Object Segmentation And Motion Estimation Based On RGBD

With the development of robotics, locomotive serive-oriented robots have been widely applied in our daily lives. Robots have worked well for human in many domains. To prevent accidents, we need to locate and make motion estimate for robots accurately. It is insufficient to utilize robots’sensors only. For this reason, we need to monitor the entire scene where robots shall work in real time, and detect and segment the objects of the scene to understand the environment and predict the possible motions. Currently, RGBD sensing has experienced a rapid development. The point cloud captured by RGBD sensors gives well description of the structure, texture and color information of the 3D scene. The research on object detection and segmentation based on RGBD data has attracted lots of attention from academic and industrial society. We apply RGBD sensors to capture the scene, and give object detection and segmentation, as well as motion estimation to monitor the scene.We conduct our work on the basis of 2D image segmentation methods, and employ the RGBD data to overcome the complicated 3D object segmentation and motion estimation problem. A novel method which works on RGBD point cloud to detect objects and estimate motions is proposed in this paper. Since locomotive robots work on grounds, we first extract the support planes from the scene, where an improved RANSAC algorithm is proposed to remove the redundant planes from RGBD data. Then we construct the remaining RGBD point cloud in the form of KD-trees, and perform the nearest neighbor algorithm to detect and segment scene objects. We study object motion estimation object motion estimation based on continuous multi frame of the object. We further make a motion estimation based on the segmented objects from several sequential RGBD frames. Specifically, we propose to employ SVM algorithm to classify the motion types, and estimate the motion parameters of moving objects. Experimental results show that, based on the RGBD point cloud data object segmentation effect is obvious, the movement of the object is accurate.At the end of this paper, a prototype system of object segmentation and motion estimation based on RGBD is presented. The system integrates the results of technology research, and it can realize the positioning function of the robot in the scene.