Point cloud is an important type of geometric data structure. Due to its irregular format, most researchers transform such data to regular 3D voxel grids or collections of images. This, however, renders data unnecessarily voluminous and causes issues. In this paper, we design a novel type of neural network that directly consumes point clouds, which well respects the permutation invariance of points in the input. Our network, named PointNet, provides a unified architecture for applications ranging from object classification, part segmentation, to scene semantic parsing. Though simple, PointNet is highly efficient and effective. Empirically, it shows strong performance on par or even better than state of the art. Theoretically, we provide analysis towards understanding of what the network has learnt and why the network is robust with respect to input perturbation and corruption.

/pdf/pointnet-deep-learning-on-point-sets-for-3d-classification-1zge9sjodi.pdf

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

With the advent of new, low-cost 3D sensing hardware such as the Kinect, and continued efforts in advanced point cloud processing, 3D perception gains more and more importance in robotics, as well as other fields. In this paper we present one of our most recent initiatives in the areas of point cloud perception: PCL (Point Cloud Library - http://pointclouds.org). PCL presents an advanced and extensive approach to the subject of 3D perception, and it's meant to provide support for all the common 3D building blocks that applications need. The library contains state-of-the art algorithms for: filtering, feature estimation, surface reconstruction, registration, model fitting and segmentation. PCL is supported by an international community of robotics and perception researchers. We provide a brief walkthrough of PCL including its algorithmic capabilities and implementation strategies.

/pdf/3d-is-here-point-cloud-library-pcl-12x3p8s6ma.pdf

3D is here: Point Cloud Library (PCL)

Point clouds provide a flexible geometric representation suitable for countless applications in computer graphics; they also comprise the raw output of most 3D data acquisition devices. While hand-designed features on point clouds have long been proposed in graphics and vision, however, the recent overwhelming success of convolutional neural networks (CNNs) for image analysis suggests the value of adapting insight from CNN to the point cloud world. Point clouds inherently lack topological information, so designing a model to recover topology can enrich the representation power of point clouds. To this end, we propose a new neural network module dubbed EdgeConv suitable for CNN-based high-level tasks on point clouds, including classification and segmentation. EdgeConv acts on graphs dynamically computed in each layer of the network. It is differentiable and can be plugged into existing architectures. Compared to existing modules operating in extrinsic space or treating each point independently, EdgeConv has several appealing properties: It incorporates local neighborhood information; it can be stacked applied to learn global shape properties; and in multi-layer systems affinity in feature space captures semantic characteristics over potentially long distances in the original embedding. We show the performance of our model on standard benchmarks, including ModelNet40, ShapeNetPart, and S3DIS.

/pdf/dynamic-graph-cnn-for-learning-on-point-clouds-20wlcru9tg.pdf

Dynamic Graph CNN for Learning on Point Clouds

In our recent work [1], [2], we proposed Point Feature Histograms (PFH) as robust multi-dimensional features which describe the local geometry around a point p for 3D point cloud datasets. In this paper, we modify their mathematical expressions and perform a rigorous analysis on their robustness and complexity for the problem of 3D registration for overlapping point cloud views. More concretely, we present several optimizations that reduce their computation times drastically by either caching previously computed values or by revising their theoretical formulations. The latter results in a new type of local features, called Fast Point Feature Histograms (FPFH), which retain most of the discriminative power of the PFH. Moreover, we propose an algorithm for the online computation of FPFH features for realtime applications. To validate our results we demonstrate their efficiency for 3D registration and propose a new sample consensus based method for bringing two datasets into the convergence basin of a local non-linear optimizer: SAC-IA (SAmple Consensus Initial Alignment).

Fast Point Feature Histograms (FPFH) for 3D registration

Three-dimensional models provide a volumetric representation of space which is important for a variety of robotic applications including flying robots and robots that are equipped with manipulators. In this paper, we present an open-source framework to generate volumetric 3D environment models. Our mapping approach is based on octrees and uses probabilistic occupancy estimation. It explicitly represents not only occupied space, but also free and unknown areas. Furthermore, we propose an octree map compression method that keeps the 3D models compact. Our framework is available as an open-source C++ library and has already been successfully applied in several robotics projects. We present a series of experimental results carried out with real robots and on publicly available real-world datasets. The results demonstrate that our approach is able to update the representation efficiently and models the data consistently while keeping the memory requirement at a minimum.

/pdf/octomap-an-efficient-probabilistic-3d-mapping-framework-3dp42nnofi.pdf

OctoMap: an efficient probabilistic 3D mapping framework based on octrees

Towards 3D Point cloud based object maps for household environments

In this paper we investigate the usage of persistent point feature histograms for the problem of aligning point cloud data views into a consistent global model. Given a collection of noisy point clouds, our algorithm estimates a set of robust 16D features which describe the geometry of each point locally. By analyzing the persistence of the features at different scales, we extract an optimal set which best characterizes a given point cloud. The resulted persistent features are used in an initial alignment algorithm to estimate a rigid transformation that approximately registers the input datasets. The algorithm provides good starting points for iterative registration algorithms such as ICP (Iterative Closest Point), by transforming the datasets to its convergence basin. We show that our approach is invariant to pose and sampling density, and can cope well with noisy data coming from both indoor and outdoor laser scans.

/pdf/aligning-point-cloud-views-using-persistent-feature-p8rbyr645o.pdf

Aligning point cloud views using persistent feature histograms

We present a novel lossy compression approach for point cloud streams which exploits spatial and temporal redundancy within the point data. Our proposed compression framework can handle general point cloud streams of arbitrary and varying size, point order and point density. Furthermore, it allows for controlling coding complexity and coding precision. To compress the point clouds, we perform a spatial decomposition based on octree data structures. Additionally, we present a technique for comparing the octree data structures of consecutive point clouds. By encoding their structural differences, we can successively extend the point clouds at the decoder. In this way, we are able to detect and remove temporal redundancy from the point cloud data stream. Our experimental results show a strong compression performance of a ratio of 14 at 1 mm coordinate precision and up to 40 at a coordinate precision of 9 mm.

Real-time compression of point cloud streams

This paper focuses on developing a fast and accurate 3D feature for use in object recognition and pose estimation for rigid objects. More specifically, given a set of CAD models of different objects representing our knoweledge of the world - obtained using high-precission scanners that deliver accurate and noiseless data - our goal is to identify and estimate their pose in a real scene obtained by a depth sensor like the Microsoft Kinect. Borrowing ideas from the Viewpoint Feature Histogram (VFH) due to its computational efficiency and recognition performance, we describe the Clustered Viewpoint Feature Histogram (CVFH) and the cameras roll histogram together with our recognition framework to show that it can be effectively used to recognize objects and 6DOF pose in real environments dealing with partial occlusion, noise and different sensors atributes for training and recognition data. We show that CVFH out-performs VFH and present recognition results using the Microsoft Kinect Sensor on an object set of 44 objects.

Nico Blodow

Papers

Fast Point Feature Histograms (FPFH) for 3D registration

Towards 3D Point cloud based object maps for household environments

Aligning point cloud views using persistent feature histograms

Real-time compression of point cloud streams

CAD-model recognition and 6DOF pose estimation using 3D cues