User Tools

Site Tools


Fast and Robust Object Detection in Household Environments Using Vocabulary Trees with SIFT Descriptors

In this paper we describe the ODUfinder, a novel perception system for autonomous service robots acting in human living environments. The perception system enables robots to detect and recognize large sets of textured objects of daily use. Efficiency, robustness, and a high detection rate are achieved through the combination of modern text retrieval methods that are successfully used for indexing huge sets of web pages and state-of-the-art robot vision methods for object recognition. The result is a robot object detection and recognition system that, with an accuracy rate of more than 80%, can recognize thousands of objects by learning and using vocabulary trees of SIFT descriptors. [ pdf ] [ code ]

Voxelized Shape and Color Histograms for RGB-D

Real world environments typically include objects with different perceptual appearances. A household, for example, includes textured, textureless and even partially transparent objects. While perception methods exist that work well on one such class of objects at a time, the perception of various classes of objects in a scene is still a challenge. It is our view that the construction of a descriptor that takes both color and shape into account, thereby fostering high discriminating power, will help to solve this problem. In this paper we present an approach that is capable of efficiently capturing both the geometric and visual appearance of common objects of daily use into one feature. We showcase this feature’s applicability for the purpose of classifying objects in cluttered scenes with obstructions, and we evaluate two classification approaches. In our experiments we make use of Kinect, a new RGB-D device, and build a database of 63 objects. Preliminary results on novel views show recognition rates of 72.2%.
[ pdf ] [ code ]

Contracting Curve Density Algorithm for Applications in Personal Robotics

This paper investigates an extended and optimized implementation of the state-of-the-art local curve fitting algorithm named Contracting Curve Density (CCD) algorithm, originally developed by Hanek et al. In particular, we investigate its application in the field of personal robotics for the tasks such as the mobile manipulation which requires a segmentation of objects in clutter and the tracking of them. The developed system mainly consists of the two functional parts, the CCD algorithm to fit the model curve in still images and the CCD tracker to track the model in the videos. We demonstrate algorithm’s working in various scenes using handheld camera and the cameras from the Personal Robot 2 (PR2). Achieved results show that the CCD algorithm achieves robustness and sub-pixel accuracy even in the presence of clutter, partial occlusion, and changes of illumination.
[ pdf ] [ code ]

General 3D Modelling of Novel Objects from a Single View

In this paper we present a method for building models for grasping from a single 3D snapshot of a scene composed of objects of daily use in human living environments. We employ fast shape estimation, probabilistic model fitting and verification methods capable of dealing with different kinds of symmetries, and combine these with a triangular mesh of the parts that have no other representation to model previously unseen objects of arbitrary shape. Our approach is enhanced by the information given by the geometric clues about different parts of objects which serve as prior information for the selection of the appropriate reconstruction method. While we designed our system for grasping based on single view 3D data, its generality allows us to also use the combination of multiple views. We present two application scenarios that require complete geometric models: grasp planning and locating objects in camera images. [ pdf ] [ video ]
people/pangercic/recognition.txt · Last modified: 2011/11/19 22:13 by pangerci