Depth image denoising and key points extraction for manipulation plane detection

The handling of twist-locks has been a heavy burden for the container industry. To address this challenge, we are developing a customized mobile manipulator for handling the twist-locks. In this paper, we propose a fast normal computation algorithm for depth image, which is able to use normal deviation along eight directions to extract key points for segmenting points into objects on manipulation support plane in an unstructured table top scene. Before further processing point clouds, a bilateral filter is used to denoise depth images. To evaluate the effectiveness of the bilateral filter, eight direction angles are also used to observe the effectiveness of filter. To further evaluate the proposed approach, a median filter is also used for comparison with the bilateral filter. Experimental results show that the fast surface normal computation based on depth image and eight directions to determine a point are feasible for plane detection.