The vision system of the mobile robot is a low-level function that provides the required target information of the current environment for the upper vision tasks. The real-time performance and robustness of object segmentation in cluttered environments is still a serious problem in robot visions. In this paper, a new real-time indoor scene segmentation method based on RGB-D image, is presented and the extracted primary object regions are then used for object recognition. Firstly, this paper accomplishes the depth filtering by the improved traditional filtering method. Then by using improved depth information, the algorithm extracts the foreground and implements the object segmentation of color image at the resolution of 640×480 from Kinect camera. Finally, the segmentation results are applied into the object recognition in indoor scene to validate the effectiveness of scene segmentation. The results of indoor segmentation demonstrate the real-time performance and robustness of the proposed method. In addition, the results of segmentation improve the accuracy of object recognition and reduce time of object recognition in indoor cluttered scene.
展开▼
