A fuzzy-based approach for approximating depth information in RGB-D images

Robot navigation has several security and defence applications. The major technical challenges include measuring the distance between a robot and its surrounding obstacles and modelling the sensing environment. Existing methods using stereo cameras, laser sensors, and low-cost MS Kinect cameras have been suggested for the problems. In this paper, we propose a fuzzy-based approach for approximating the missing depth values in RGB-D images collected from a MS Kinect camera. By investigating different noise models, the missing information, and the relations between the depth and colour images, the proposed approach produces an accurate approximation for missing depth values, which enhances results of subsequent steps in RGB-D image processing.