Depth sensing with coding-free pattern based on topological constraint

Structured light depth sensing with a single-shot pattern is widely employed to capture depth maps for dynamic scenes. For conventional structured light techniques, the projected pattern has to be coded delicately in regards to color, shape, and intensity, in order to assign each pixel with a unique label. However, using such a complicated pattern is a double-edged sword, as although it is effective in labelling pixels, it is also sensitive to environmental noise such as: ambient illumination, textures, uneven albedos, or colors of objects in a scene. In contrast, a coding-free pattern is simply constructed and also insensitive to various environmental noise. Therefore, the coding-free pattern method is capable of robustly sensing the depth for complex scenes. The main challenge in coding-free depth sensing is the 'correspondence retrieval' between the projected and captured pattern (i.e. matching pixels between the projected and captured pattern). In this study, we focused on evaluating the correspondence retrieval in a coding-free binary grid pattern. A graph based topological labelling (GBTL) algorithm is proposed to determine the topological coordinates of the intersections of the grid. Then we retrieved the correspondence by using the topology of the grid and the epipolar constraint. We also demonstrated the upper bounds of depth variance by employing the proposed method. The proposed technique alleviates many of the limitations faced with traditional correspondence retrieval. Experimental results showed that the proposed technique performed better (i.e. in terms of precision) than the popular RGB-D cameras Kinect v1 and Kinect v2. Compared with the traditional single-shot techniques, which require complicated patterns, the proposed technique significantly improved the robustness and ease of work, while achieving comparable precision. Additionally, this proposed technique could also be used for both the binary coding-free and the traditional chromatic grid patterns.