Heightmap

Abstract: Mobile ground robots operating on unstructured terrain must predict which areas of the environment they are able to pass in order to plan feasible paths. We address traversability estimation as a heightmap classification problem: we build a convolutional neural network that, given an image representing the heightmap of a terrain patch, predicts whether the robot will be able to traverse such patch from left to right. The classifier is trained for a specific robot model (wheeled, tracked, legged, snake-like) using simulation data on procedurally generated training terrains; the trained classifier can be applied to unseen large heightmaps to yield oriented traversability maps, and then plan traversable paths. We extensively evaluate the approach in simulation on six real-world elevation dataset, and run a real-robot validation in one indoor and one outdoor environment.

Abstract: We present a novel method for 3D reconstruction of urban scenes extending a recently introduced heightmap model. Our model has several advantages for 3D modeling of urban scenes: it naturally enforces vertical surfaces, has no holes, leads to an efficient algorithm, and is compact in size. We remove the major limitation of the heightmap by enabling modeling of overhanging structures. Our method is based on an an n-layer heightmap with each layer representing a surface between full and empty space. The configuration of layers can be computed optimally using a dynamic programming method. Our cost function is derived from probabilistic occupancy, and incorporates the Bayesian Information Criterion (BIC) for selecting the number of layers to use at each pixel. 3D surface models are extracted from the heightmap. We show results from a variety of datasets including Internet photo collections. Our method runs on the GPU and the complete system processes video at 13 Hz.

Abstract: We propose a boundary-aware multi-task deep-learning-based framework for fast 3D building modeling from a single overhead image. Unlike most existing techniques which rely on multiple images for 3D scene modeling, we seek to model the buildings in the scene from a single overhead image by jointly learning a modified signed distance function (SDF) from the building boundaries, a dense heightmap of the scene, and scene semantics. To jointly train for these tasks, we leverage pixel-wise semantic segmentation and normalized digital surface maps (nDSM) as supervision, in addition to labeled building outlines. At test time, buildings in the scene are automatically modeled in 3D using only an input overhead image. We demonstrate an increase in building modeling performance using a multi-feature network architecture that improves building outline detection by considering network features learned for the other jointly learned tasks. We also introduce a novel mechanism for robustly refining instance-specific building outlines using the learned modified SDF. We verify the effectiveness of our method on multiple large-scale satellite and aerial imagery datasets, where we obtain state-of-the-art performance in the 3D building reconstruction task.

Abstract: Underwater range scanning techniques are starting to gain interest in underwater exploration, providing new tools to represent the seafloor. These scans (often) acquired by underwater robots usually result in an unstructured point cloud, but given the common downward-looking or forward-looking configuration of these sensors with respect to the scene, the problem of recovering a piecewise linear approximation representing the scene is normally solved by approximating these 3D points using a heightmap (2.5D). Nevertheless, this representation is not able to correctly represent complex structures, especially those presenting arbitrary concavities normally exhibited in underwater objects. We present a method devoted to full 3D surface reconstruction that does not assume any specific sensor configuration. The method presented is robust to common defects in raw scanned data such as outliers and noise often present in extreme environments such as underwater, both for sonar and optical surveys. Moreover, the proposed method does not need a manual preprocessing step. It is also generic as it does not need any information other than the points themselves to work. This property leads to its wide application to any kind of range scanning technologies and we demonstrate its versatility by using it on synthetic data, controlled laser scans, and multibeam sonar surveys. Finally, and given the unbeatable level of detail that optical methods can provide, we analyze the application of this method on optical datasets related to biology, geology and archeology.