An Instance-Based Approach to 3 D Object Recognition in Open-Ended Robotic Domains

Three-dimensional (3D) object detection and recognition is increasingly used in manipulation and navigation applications in autonomous service robots. It involves clustering points from an unorganized scene into object candidates and estimating features to recognize the objects under different circumstances such as occlusions and clutter. This paper presents an approach to detect and recognize objects based on spin-image local shape descriptors. This approach starts with a preprocessing phase to remove extra information and prepare a suitable point cloud. Clustering is then applied to detect object candidates. Subsequently, the spin-images for all object candidates are calculated. Finally a specific classification rule is used to recognize the object’s category. To examine the performance of the proposed approach, a leave-one-out cross validation scheme is utilized to compute precision, recall and FMeasure. The experimental results showed that this approach performs well on different types of objects.! KeywordsOpen-ended learning; 3D object recognition; spinimage Descriptor; Autonomous robots.