I introduce a simple yet robust algorithm for discovering objects and level planes from depth information produced by a Microsoft Kinect. In a controlled indoor environment, this algorithm discovers objects on multiple surfaces at varying heights and distinguishes between separate objects on the same surface. The algorithm includes eliminating large planes from the input, clustering objects by spatial proximity, and by filtering results using domain knowledge of the environment.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.