Geometry-based 3D Object Fitting and Localizing in Grasping Aid for Visually Impaired

This paper presents a geometry-based method for 3D object fitting and localization in the context of building a grasping aid service for visually impaired people using information from Kinect sensor. Given two constraints of this working application, (1) the interested object is on a table and (2) the geometrical form of the object is known in advance based on the query of the user, the proposed system consists of three steps: table plane detection, object detection, and object fitting and localization. Our work has three contributions. First, we propose to use organized point cloud representation instead of just point cloud in order to speedup the computational time and improve the accuracy of table plane detection. Second, we employ MLESAC (Maximum LikElihood SAmple Consensus) that can give better results for object fitting. Third, we introduce a new method for evaluating object localization task and make a quantitative evaluation of object localization on our captured dataset.