Robust, Unsupervised, and Ubiquitous Pose and Object Information for Robot Action Recognition