2015-467 – ROCK : Robust Object Constellation for Kinematic Pose

ROCK is a computer program designed to detect rigid objects in color and depth images and estimate their pose relative to a fixed camera. Applications include robotic perception and state estimation. Central to this work is a 3D model of visual appearance that is used to rapidly evaluate candidate object poses. Some experiments have shown that this method is successful in certain contexts, but more experiments are necessary to evaluate its performance relative to existing methods. Scott McEvoy smcevoy@andrew.cmu.edu 412-268-6053