DepthComp: Real-time Depth Image Completion Based on Prior Semantic Scene Segmentation

We address plausible hole filling in depth images in a computationally lightweight that leverages recent advances in semantic scene segmentation. Firstly, we such segmentation over a co-registered color image, commonly available from depth sources, and non-parametrically fill missing depth values based on a multipass within each semantically labeled scene object. Within this formulation, we a bounded set of explicit completion cases in a grammar inspired context that be performed effectively and efficiently to provide highly plausible localized depth via a case-specific non-parametric completion approach. Results demonstrate this approach has complexity and efficiency comparable to conventional interpolation but with accuracy analogous to contemporary depth filling approaches. , we show it to be capable of fine depth relief completion beyond that of contemporary approaches in the field and computationally comparable interpolation .