Joseph L. Mundy

Scott Sorensen, Wayne Treible, Robert Wagner et al.

Physics driven image simulation allows for the modeling and creation of realistic imagery beyond what is afforded by typical rendering pipelines. We aim to automatically generate a physically realistic scene for simulation of a given region using satellite imagery to model the scene geometry, drive material estimates, and populate the scene with dynamic elements. We present automated techniques to utilize satellite imagery throughout the simulated scene to expedite scene construction and decrease manual overhead. Our technique does not use lidar, enabling simulations that could not be constructed previously. To develop a 3D scene, we model the various components of the real location, addressing the terrain, modelling man-made structures, and populating the scene with smaller elements such as vegetation and vehicles. To create the scene we begin with a Digital Surface Model, which serves as the basis for scene geometry, and allows us to reason about the real location in a common 3D frame of reference. These simulated scenes can provide increased fidelity with less manual intervention for novel locations on earth, and can facilitate algorithm development, and processing pipelines for imagery ranging from UV to LWIR $(200nm-20μm)$.

Joseph L Mundy, Hank Theiss

This paper describes an investigation of the source of geospatial error in digital surface models (DSMs) constructed from multiple satellite images. In this study the uncertainty in surface geometry is separated into two spatial components; global error that affects the absolute position of the surface, and local error that varies from surface point to surface point. The global error component is caused by inaccuracy in the satellite imaging process, mainly due to uncertainty in the satellite position and orientation (pose) during image collection. A key result of the investigation is a new algorithm for determining the absolute geoposition of the DSM that takes into account the pose covariance of each satellite during image collection. This covariance information is used to weigh the evidence from each image in the computation of the global position of the DSM. The use of covariance information significantly decreases the overall uncertainty in global position. The paper also describes an approach to the prediction of local error in the DSM surface. The observed variance in surface position within a single stereo surface reconstruction defines the local horizontal error. The variance in the fused set of elevations from multiple stereo pairs at a single DSM location defines the local vertical error. These accuracy predictions are compared to ground truth provided by LiDAR scans of the same geographic region of interest.

Maxim Bazik, Brien Flewelling, Manoranjan Majji et al.

Automated 3D pose estimation of satellites and other known space objects is a critical component of space situational awareness. Ground-based imagery offers a convenient data source for satellite characterization; however, analysis algorithms must contend with atmospheric distortion, variable lighting, and unknown reflectance properties. Traditional feature-based pose estimation approaches are unable to discover an accurate correlation between a known 3D model and imagery given this challenging image environment. This paper presents an innovative method for automated 3D pose estimation of known space objects in the absence of satisfactory texture. The proposed approach fits the silhouette of a known satellite model to ground-based imagery via particle filtering. Each particle contains enough information (orientation, position, scale, model articulation) to generate an accurate object silhouette. The silhouette of individual particles is compared to an observed image. Comparison is done probabilistically by calculating the joint probability that pixels inside the silhouette belong to the foreground distribution and that pixels outside the silhouette belong to the background distribution. Both foreground and background distributions are computed by observing empty space. The population of particles are resampled at each new image observation, with the probability of a particle being resampled proportional to how the particle's silhouette matches the observation image. The resampling process maintains multiple pose estimates which is beneficial in preventing and escaping local minimums. Experiments were conducted on both commercial imagery and on LEO satellite imagery. Imagery from the commercial experiments are shown in this paper.

Wayne Treible, Scott Sorensen, Andrew D. Gilliam et al.

Digital Surface Model generation from satellite imagery is a difficult task that has been largely overlooked by the deep learning community. Stereo reconstruction techniques developed for terrestrial systems including self driving cars do not translate well to satellite imagery where image pairs vary considerably. In this work we present neural network tailored for Digital Surface Model generation, a ground truthing and training scheme which maximizes available hardware, and we present a comparison to existing methods. The resulting models are smooth, preserve boundaries, and enable further processing. This represents one of the first attempts at leveraging deep learning in this domain.

Hasnain Vohra, Maxim Bazik, Matthew Antone et al.

Monocular visual SLAM has become an attractive practical approach for robot localization and 3D environment mapping, since cameras are small, lightweight, inexpensive, and produce high-rate, high-resolution data streams. Although numerous robust tools have been developed, most existing systems are designed to operate in terrestrial environments and at relatively small scale (a few thousand frames) due to constraints on computation and storage. In this paper, we present a feature-based visual SLAM system for aerial video whose simple design permits near real-time operation, and whose scalability permits large-area mapping using tens of thousands of frames, all on a single conventional computer. Our approach consists of two parallel threads: the first incrementally creates small locally consistent submaps and estimates camera poses at video rate; the second aligns these submaps with one another to produce a single globally consistent map via factor graph optimization over both poses and landmarks. Scale drift is minimized through the use of 7-degree-of-freedom similarity transformations during submap alignment. We quantify our system's performance on both simulated and real data sets, and demonstrate city-scale map reconstruction accurate to within 2 meters using nearly 90,000 aerial video frames - to our knowledge, the largest and fastest such reconstruction to date.

Daniel Crispell, Octavian Biris, Nate Crosswhite et al.

The performance of modern face recognition systems is a function of the dataset on which they are trained. Most datasets are largely biased toward "near-frontal" views with benign lighting conditions, negatively effecting recognition performance on images that do not meet these criteria. The proposed approach demonstrates how a baseline training set can be augmented to increase pose and lighting variability using semi-synthetic images with simulated pose and lighting conditions. The semi-synthetic images are generated using a fast and robust 3-d shape estimation and rendering pipeline which includes the full head and background. Various methods of incorporating the semi-synthetic renderings into the training procedure of a state of the art deep neural network-based recognition system without modifying the structure of the network itself are investigated. Quantitative results are presented on the challenging IJB-A identification dataset using a state of the art recognition pipeline as a baseline.