Conditional Regressive Random Forest Stereo-Based Hand Depth Recovery

Rilwan Remilekun Basaru, Greg Slabaugh, Eduardo Alonso, Chris Child; Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 614-622

This paper introduces Conditional Regressive Random Forest (CRRF), a novel method that combines a closed-form Conditional Random Field (CRF), using learned weights, and a Regressive Random Forest (RRF) that employs adaptively selected expert trees. CRRF is used to estimate a depth image of hand given stereo RGB inputs. CRRF uses a novel superpixel-based regression framework that takes advantage of the smoothness of the hand's depth surface. A RRF unary term adaptively selects different stereo-matching measures as it implicitly determines matching pixels in a coarse-to-fine manner. CRRF also includes a pair-wise term that encourages smoothness between similar adjacent superpixels. Experimental results show that CRRF can produce high quality depth maps, even using an inexpensive RGB stereo camera and produces state-of-the-art results for hand depth estimation.