We propose a novel and specially designed method for piecewise dense monocular depth estimation in dynamic scenes. We utilize spatial relations between neighboring superpixels to solve the inherent relative scale ambiguity (RSA) problem and smooth the depth map.
In this paper we propose a novel and specially designedmethod for piecewise dense monocular depth estimation indynamic scenes. We utilize spatial relations between neighboring superpixels to solve the inherent relative scale ambiguity (RSA) problem and smooth the depth map. However,directly estimating spatial relations is an ill-posed problem. Our core idea is to predict spatial relations basedon the corresponding motion relations. Given two or moreconsecutive frames, we first compute semi-dense (CPM)or dense (optical flow) point matches between temporallyneighboring images. Then we develop our method in fourmain stages: superpixel relations analysis, motion selection, reconstruction, and refinement. The final refinementprocess helps to improve the quality of the reconstructionat pixel level. Our method does not require per-object segmentation, template priors or training sets, which ensuresflexibility in various applications. Extensive experimentson both synthetic and real datasets demonstrate that ourmethod robustly handles different dynamic situations andpresents competitive results to the state-of-the-art methodswhile running much faster than them.