Accurate garment surface analysis using an active stereo robot head with application to dual-arm flattening

2016 IEEE International Conference on Robotics and Biomimetics (ROBIO)

et al. 2016

Self Cite

Abstract-We present an interactive perception model for object sorting based on Gaussian Process (GP) classification that is capable of recognizing objects categories from point cloud data. In our approach, FPFH features are extracted from point clouds to describe the local 3D shape of objects and a Bag-of-Words coding method is used to obtain an objectlevel vocabulary representation. Multi-class Gaussian Process classification is employed to provide and probable estimation of the identity of the object and serves a key role in the interactive perception cycle -modelling perception confidence. We show results from simulated input data on both SVM and GP based multi-class classifiers to validate the recognition accuracy of our proposed perception model. Our results demonstrate that by using a GP-based classifier, we obtain true positive classification rates of up to 80%. Our semi-autonomous object sorting experiments show that the proposed GP based interactive sorting approach outperforms random sorting by up to 30% when applied to scenes comprising configurations of household objects.

Section: Discussionmentioning

confidence: 99%

Interactive perception based on Gaussian Process classification for house-hold objects recognition & sorting

2016 IEEE International Conference on Robotics and Biomimetics (ROBIO)

et al. 2016

Self Cite

“…This paper provides a refined flattening strategy and details more comprehensive experiments than those reported in [6]. Accordingly, the novel contribution made by this paper, extending the work reported in [7], is that we now focus on verifying the utility and implementation of appropriate feature extraction methods within a repeatable simulated environment, while [7] provides the evaluation of the performance of various depth sensors and manipulation protocols (single-arm or dual-arm flattening) on flattening real garments.…”

Section: Introductionmentioning

confidence: 98%

A Precise Method for Cloth Configuration Parsing Applied to Single-Arm Flattening

International Journal of Advanced Robotic Systems

et al. 2016

Self Cite

In this paper, we investigate the contribution that visual perception affords to a robotic manipulation task in which a crumpled garment is flattened by eliminating visually detected wrinkles. In order to explore and validate visually guided clothing manipulation in a repeatable and controlled environment, we have developed a hand-eye interactive virtual robot manipulation system that incorporates a clothing simulator to close the effector-garment-visual sensing interaction loop. We present the technical details and compare the performance of two different methods for detecting, representing and interpreting wrinkles within clothing surfaces captured in high-resolution depth maps. The first method we present relies upon a clustering-based method for localizing and parametrizing wrinkles, while the second method adopts a more advanced geometrybased approach in which shape-topology analysis underpins the identification of the cloth configuration (i.e., maps wrinkles). Having interpreted the state of the cloth configuration by means of either of these methods, a heuristicbased flattening strategy is then executed to infer the appropriate forces, their directions and gripper contact locations that must be applied to the cloth in order to flatten the perceived wrinkles. A greedy approach, which attempts to flatten the largest detected wrinkle for each perception-iteration cycle, has been successfully adopted in this work. We present the results of our heuristic-based flattening methodology which relies upon clustering-based and geometry-based features respectively. Our experiments indicate that geometry-based features have the potential to provide a greater degree of clothing configuration understanding and, as a consequence, improve flattening performance. The results of experiments using a real robot (as opposed to simulated robot) also confirm our proposition that a more effective visual perception system can advance the performance of cloth manipulation.

“…For our active binocular robot head, we employed relatively inexpensive and commercially available components in order to allow us to capture high-quality 3D depth maps and dense point clouds for deformable object recognition [3] and manipulation [2]. Hence, our robot head comprises two off-the-shelf Nikon DSLR cameras (D5100) that capture 16 MegaPixels images (MP) every 400ms.…”

Section: B Active Binocular Robot Headmentioning

confidence: 99%

“…This robot head is capable of changing its gaze under computer control. Our robot head has been used successfully for clothes perception and manipulation research [2], [3], [4] because of its ability to provide high resolution imaging for 3D mapping and range sensing. Due to the ability of the robot head to target different parts of a scene, it is required to maintain accurate calibration of its intrinsic and extrinsic parameters with respect to the robot's reference frame.…”

Section: Introductionmentioning

confidence: 99%

On the calibration of active binocular and RGBD vision systems for dual-arm robots

2016 IEEE International Conference on Robotics and Biomimetics (ROBIO)

et al. 2016

Self Cite

Abstract-This paper describes a camera and hand-eye calibration methodology for integrating an active binocular robot head within a dual-arm robot. For this purpose, we derive the forward kinematic model of our active robot head and describe our methodology for calibrating and integrating our robot head. This rigid calibration provides a closedform hand-to-eye solution. We then present an approach for updating dynamically camera external parameters for optimal 3D reconstruction that are the foundation for robotic tasks such as grasping and manipulating rigid and deformable objects. We show from experimental results that our robot head achieves an overall sub millimetre accuracy of less than 0.3 millimetres while recovering the 3D structure of a scene. In addition, we report a comparative study between current RGBD cameras and our active stereo head within two dual-arm robotic testbeds that demonstrates the accuracy and portability of our proposed methodology.