Geometric Priors for Gaussian Process Implicit Surfaces

Martens, Wolfram; Poffet, Yannick; Soria, Pablo Ramón; Sukkarieh, Salah

doi:10.1109/lra.2016.2631260

Cited by 40 publications

(44 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Maybe due to miss detections of smaller curves, this first 2D then 3D approach has not been shown to work robustly in cluttered scene. While sharing the same basic segmentation idea as above, our geometric segmentation CNN performs no fitting, but only provides plausibility maps for further geometric verification if needed, or even just as shape priors to more complex scene reconstruction [20], e.g., tree trunks as cylindrical shapes.…”

Section: B Related Workmentioning

confidence: 99%

“…Such a segmentation map could already be useful for more complex tasks [20], yet for the sake of a robust primitive fitting pipeline, one cannot fully trust this segmentation map as it inevitably contains misclassification, just like all other image semantic segmentations. Fortunately, by separating pixels belonging to individual primitive classes, our original multi-model problem is converted to an easier multiinstance problem.…”

Section: Framework Overviewmentioning

confidence: 99%

See 1 more Smart Citation

Primitive Fitting Using Deep Geometric Segmentation

Feng²

2019

Proceedings of the International Symposium on Automation and Robotics in Construction (IAARC)

View full text Add to dashboard Cite

To identify and fit geometric primitives (e.g., planes, spheres, cylinders, cones) in a noisy point cloud is a challenging yet beneficial task for fields such as robotics and reverse engineering. As a multi-model multi-instance fitting problem, it has been tackled with different approaches including RANSAC, which however often fit inferior models in practice with noisy inputs of cluttered scenes. Inspired by the corresponding human recognition process, and benefiting from the recent advancements in image semantic segmentation using deep neural networks, we propose BAGSFit as a new framework addressing this problem. Firstly, through a fully convolutional neural network, the input point cloud is point-wisely segmented into multiple classes divided by jointly detected instance boundaries without any geometric fitting. Thus, segments can serve as primitive hypotheses with a probability estimation of associating primitive classes. Finally, all hypotheses are sent through a geometric verification to correct any misclassification by fitting primitives respectively. We performed training using simulated range images and tested it with both simulated and real-world point clouds. Quantitative and qualitative experiments demonstrated the superiority of BAGSFit.

show abstract

Section: B Related Workmentioning

confidence: 99%

Section: Framework Overviewmentioning

confidence: 99%

Primitive Fitting Using Deep Geometric Segmentation

Feng²

2019

Proceedings of the International Symposium on Automation and Robotics in Construction (IAARC)

View full text Add to dashboard Cite

show abstract

“…This new belief then feeds back in to the planning module to replan the observation locations, and this process continues. Traditionally, active perception problems arise while using vision sensing modalities (Chen et al, 2011), although recently there has also been a need for new formulations suitable for sensing modalities with depth information, such as 3D laser (Patten et al, 2017), RGB-D (van Hoof et al, 2014;Patten et al, 2016;Martens et al, 2017) and thermal (Cunningham-Nelson et al, 2015) modalities.…”

Section: Related Workmentioning

confidence: 99%

“…This formulation is motivated by scenarios where there are two complementary sensing modalities. For example, a long-range laser sensor (Bargoti et al, 2015) detects the presence and locations of trees on a farm, while a close-range high-resolution RGB-D sensor (Martens et al, 2017;Peng et al, 2016) performs the primary task of characterising the fruit in the trees. The planner needs to balance the use of these two modalities in order for the robots to discover as many objects as possible while also making sufficient close-range observations.…”

Section: Introductionmentioning

confidence: 99%

Online planning for multi-robot active perception with self-organising maps

2017

Self Cite

View full text Add to dashboard Cite

We propose a self-organising map (SOM) algorithm as a solution to a new multi-goal path planning problem for active perception and data collection tasks. We optimise paths for a multi-robot team that aims to maximally observe a set of nodes in the environment. The selected nodes are observed by visiting associated viewpoint regions defined by a sensor model. The key problem characteristics are that the viewpoint regions are overlapping polygonal continuous regions, each node has an observation reward, and the robots are constrained by travel budgets. The SOM algorithm jointly selects and allocates nodes to the robots and finds favourable sequences of sensing locations. The algorithm has a runtime complexity that is polynomial in the number of nodes to be observed and the magnitude of the relative weighting of rewards. We show empirically the runtime is sublinear in the number of

show abstract

“…One popular choice for object modeling are Gaussian Processes Implicit Surfaces (GPIS) 17 , that have recently been applied extensively to the field of haptic exploration 18,19,20,21,22 . Building on GPIS, an extension to incorporate prior knowledge has been proposed to enhance the probabilistic reconstruction of explored objects 23 .…”

Section: Introduction and Related Workmentioning

confidence: 99%

Active Tactile Exploration Based on Cost-Aware Information Gain Maximization

Ottenhaus¹,

Kaul²,

Vahrenkamp³

et al. 2018

Int. J. Human. Robot.

View full text Add to dashboard Cite

Active tactile perception is a powerful mechanism to collect contact information by touching an unknown object with a robot finger in order to enable further interaction with the object or grasping of the object. The acquired object knowledge can be used to build object shape models based on such usually sparse tactile contact information. In this paper, we address the problem of object shape reconstruction from sparse tactile data gained from a robot finger that yields contact information and surface orientation at the contact points. To this end, we present an exploration algorithm which determines the next best touch target in order to maximize the estimated information gain and to minimize the expected costs of exploration actions. We introduce the Information Gain Estimation Function (IGEF), which combines different goals as measure for the quantification of the cost-aware information gain during exploration. The IGEF-based exploration strategy is validated in simulation using 48 publicly available object models and compared to state-of-the-art Gaussian processes-based exploration approaches. The results show the performance of the approach regarding exploration efficiency, cost-awareness and suitability for application in real tactile sensing scenarios.

show abstract

Geometric Priors for Gaussian Process Implicit Surfaces

Cited by 40 publications

References 17 publications

Primitive Fitting Using Deep Geometric Segmentation

Primitive Fitting Using Deep Geometric Segmentation

Online planning for multi-robot active perception with self-organising maps

Active Tactile Exploration Based on Cost-Aware Information Gain Maximization

Contact Info

Product

Resources

About