Dextrous Assembler Robot Working with Embodied Intelligence SEVENTH FRAMEWORK PROGRAMME

Estimating the pose of objects from range data is a prob-lem of considerable practical importance for many visionapplications. This paper presents an approach for accu-rate and efficient 3D pose estimation from 2.5D range im-ages. Initialized with an approximate pose estimate, theproposed approach refines it so that it accurately accountsfor a sensed range image. This is achieved by using ahypothesize-and-test scheme that combines randomized op-timization and graphics-based rendering to minimize an ob-jective function that quantifies the misalignment betweenthe acquired and a rendered range image. Extensive ex-perimental results demonstrate the superior performance ofthe proposed approach compared to the commonly used It-erative Closest Point (ICP) algorithm.