Active Pose Estimation of Daily Objects

Abstract

Caring for people who are elderly is now a global challenge given the shortage of nursing and healthcare providers. This makes life support robots that could assist people to live independently highly significant. A frequent task during life support is the fetching of daily containers, which requires accurate six-dimensional pose estimation. In this paper, we develop an active pose estimation pipeline that is capable of providing such estimation even under partial occlusions caused by the limited camera position. We improve our pre-proposed basic pose estimation approach by immigrating a viewpoint control module which consists of two loops. The viewpoint update loop enables efficient camera pose adjustment while the evaluation loop allows the judgment of whether a new viewpoint is qualified for producing accurate pose estimation. Eventually, through experiments involving actual daily scenarios, we show that the pose estimation performance under partial occlusions could be dramatically increased.