Multimodal cue integration through Hypotheses Verification for RGB-D object recognition and 6DOF pose estimation

Abstract

This paper proposes an effective algorithm for recognizing objects and accurately estimating their 6DOF pose in scenes acquired by a RGB-D sensor. The proposed method is based on a combination of different recognition pipelines, each exploiting the data in a diverse manner and generating object hypotheses that are ultimately fused together in an Hypothesis Verification stage that globally enforces geometrical consistency between model hypotheses and the scene. Such a scheme boosts the overall recognition performance as it enhances the strength of the different recognition pipelines while diminishing the impact of their specific weaknesses. The proposed method outperforms the state-of-the-art on two challenging benchmark datasets for object recognition comprising 35 object models and, respectively, 176 and 353 scenes.

Autonomous Learning of Object Models on a Mobile Robot

“…To identify learnt objects within the environment, we use the object recognizer from [20]. To exploit the strengths of different feature descriptors, this recognizer extracts multiple features in parallel pipelines, generates object hypotheses by a graph-based grouping of merged feature correspondences, and verifies these object hypotheses by finding a global optimum solution that best explains the scene in terms of number of metrics.…”

Data association for semantic world modeling from partial views

“…Object recognition and pose estimation has received widespread attention from the computer vision and robotics communities. With the recent advances in RGB-D cameras, several systems have been developed to detect object types/instances and their 6-D poses from 3-D point clouds (Rusu et al, 2010;Glover et al, 2011;Lai et al, 2012;Aldoma et al, 2013;Marton et al, 2014). We will use one such detector (Glover and Popovic, 2013) as our black-box attribute detector, but we emphasize that our methods are agnostic to the detector used.…”

3D Object Detection and Recognition for Robotic Grasping Based on RGB-D Images and Global Features

“…In addition, an Iteractive Closest Point (ICP) algorithm for pose refinement introduced in [12] and extended in [48] was applied to point clouds. Finally, to reduce the number of false positives, a Hypotheses Verification (HV) step introduced in [3] was extended in [4] by additional hypotheses generation pipelines with the use of SIRF and OUR-CVFH descriptors. Aldoma et al adapted the solution from [4] to a multi-view recognition system [7].…”

“…Finally, to reduce the number of false positives, a Hypotheses Verification (HV) step introduced in [3] was extended in [4] by additional hypotheses generation pipelines with the use of SIRF and OUR-CVFH descriptors. Aldoma et al adapted the solution from [4] to a multi-view recognition system [7]. In [43] Tang et al demonstrated advantages of simultaneous segmentation, object detection and 3D pose recovery in extracted global and local feature models acquired from multi-view color images and point clouds.…”