State-of-the-art techniques proposed for 6D object pose recovery depend onocclusion-free point clouds to accurately register objects in 3D space. Toreduce this dependency, we introduce a novel architecture called IterativeHough Forest with Histogram of Control Points that is capable of estimatingoccluded and cluttered objects' 6D pose given a candidate 2D bounding box. OurIterative Hough Forest is learnt using patches extracted only from the positivesamples. These patches are represented with Histogram of Control Points (HoCP),a "scale-variant" implicit volumetric description, which we derive fromrecently introduced Implicit B-Splines (IBS). The rich discriminativeinformation provided by this scale-variance is leveraged during inference,where the initial pose estimation of the object is iteratively refined based onmore discriminative control points by using our Iterative Hough Forest. Weconduct experiments on several test objects of a publicly available dataset totest our architecture and to compare with the state-of-the-art.
展开▼
