Local shape descriptor selection for object recognition in range data

Local shape descriptor selection for object recognition and localization in range data is formulated herein as an optimization problem. Local shape descriptors are used for establishing point correspondences between two surfaces by way of encapsulating local shape, such that their similarity indicates geometric similarity between respective neighbourhoods. We present a generalized platform for constructing local shape descriptors that subsumes a large class of existing methods, and that allows for tuning to the geometry of specific models. Experimental analysis confirms the superiority of optimized descriptors over generic ones in object recognition tasks using real LIDAR and stereo range images.

Research highlights
► Local shape descriptor selection is expressed as an optimization problem.
► A generalized platform subsumes a large class of range data point matching methods.
► The descriptors are tuned to the geometry of specific models via feature selection.
► Object recognition experiments were performed on real LIDAR and stereo range data.
► Optimization leads to higher point matching precision in object recognition tasks.