Object Categorization in Clutter Using Additive Features and Hashing of Part-Graph Descriptors

Abstract

Abstract. Detecting objects in clutter is an important capability for a household robot executing pick and place tasks in realistic settings. While approaches from 2D vision work reasonably well under certain lighting conditions and given unique textures, the development of inexpensive RGBD cameras opens the way for real-time geometric approaches that do not require templates of known objects. This paper presents a part-graph-based hashing method for classifying objects in clutter, using an additive feature descriptor. The method is incremental, allowing easy addition of new training data without recreating the complete model, and takes advantage of the additive nature of the feature to increase efficiency. It is based on a graph representation of the scene created from considering possible groupings of over-segmented scene parts, which can in turn be used in classification. Additionally, the results over multiple segmentations can be accumulated to increase detection accuracy. We evaluated our approach on a large RGBD dataset containing over 15000 Kinect scans of 102 objects grouped in 16 categories, which we arranged into six geometric classes. Furthermore, tests on complete cluttered scenes were performed as well, and used to showcase the importance of domain adaptation.

Tracking-based interactive segmentation of textureless objects

“…Fig. 2 and demonstrated in an accompanying video 4 . In the first step we obtain an RGBD point cloud from the Kinect sensor.…”

“…In order to achieve a pre-segmentation we make use of the classification method presented in [4] based on part-graphbased hashing. The basic idea is that segmenting objects accurately in a cluttered scene does not always yield the expected result, as seen in Fig.…”

“…3 column 4, and can lead to classification failures, but over-segmenting is easily realizable [18]- [20]. We use the classification approach described in [4] for categorizing over-segmented object parts in cluttered scenes by considering combinations of these parts to compute features and classify these efficiently with the aid of hashing. The result is a set of labeled parts with geometric categories that can be grouped in order to obtain object hypotheses.…”

“…Top-left: The service robot PR2 aiming to segment the scene consisting of textureless object. Results of the scene segmentation using Region Growing method [3] (top-right, NW), Part-Graph-based Hashing [4] method (top-right, NE) and Graph-based segmentation method [5] (top-right, SW). These methods work in depth, RGB and RGBD space respectively and all underachieve due to the complexity of this challenging task.…”

“…Similar to Katz et al [6], Bergstrom et al [7], and our earlier work [8] we propose a system that uses induced motions in a scene to enable effective object segmentation. Our system employs a combination of the following complementary techniques: presegmentation of a raw point cloud of a given scene from a single camera view using part-graph-based hashing [4], estimation of a contact point and a push direction of the robot's end effector [8], RGBD feature extraction and tracking using particle filtering-based tracking, graph-based feature trajectory clustering algorithm, and dense model reconstruction based on region growing in normal space. There are three important assumptions in our system.…”