Autonomous object level segmentation

In this paper we describe a new technique for segment meaningfully objects autonomously. Traditional segmentation scheme tries to find the best segmentation result at some trade off between level of user input and level of meaningful segmentation. Segmentation with user input will ensure better segmentation result but is not applicable to the realtime autonomous robotics system. For most of the commercial software users, they prefer program to know where to segmented before they even touches. No user input usually means that system tends to either over segment or wrongly segmented. Also tuning parameters of the autonomous segmentation scheme is painful. In this paper, we propose a novel idea to find initial seed for the segmentation scheme which need manual input and product object level segments fully autonomously. We also proposed a new evaluation scheme for robotics based autonomous segmentation measurement.