Recent advances in robotic control and autonomous systems have driven the need for robust, fast and accurate estimations of surrounding environments and motions of objects within. Stereo depth estimation has been a widely adopted method to estimate the geometry of a scene from a pair of cameras. From the disparity in features across two sets of images, it is possible to estimate the distance for each respective feature and create a depth map. Autonomous vehicles benefit from frequent estimates of the surroundings, and as such video- rate stereo depthmaps are a common target throughout literature. The driving limitation in existent implementations is the lack resolution that is achieved at these frame rates when the computation is completed in software on general purpose processors, namely low-power ARM processors. GPU accelerations have demonstrated higher resolutions depth estimates for the same frame-rate, and recently dedicated hardware architectures have been proven to even further accelerate such tasks. This project proposes a hardware accelerated dense stereo depth camera with a targeted operation of 1920x1080 pixels at 30 frames per second, operating on a fully integrated Xilinx MPSoC, with parts of the capture and computation accelerated on the programmable logic and the rest running on the quad-core ARM A53 processors. This High Level Synthesis driven project highlights the ease of integration for complex embedded systems, while maintaining optimal performance and customizability.

The goal of this project was to develop visualization methods for the interpretation and analysis of time-varying voxel concentrations from a particle propagation simulation. A simulated plume provided 3d spatial concentration values at 1 minute intervals for 16 hours in proximity to the Diablo Power Plant. From this plume, we wanted to model with representative colors how this plume propagated in relation to the coastal environment at different altitudes. With a high resolution 3D terrain model, we were able to visualize the plume evolution at a desired playback speed. Future works will focus on sensory measurement analysis of this plume in the environment and the visualization of such nodes.

Geometry, appearance and context are essential aspects to capture in the digital documentation of cultural heritage sites. Geometry must be accurate and should provide a level of precision necessary for quantified diagnostics. Visual appearance should capture the “as-is” state, while site specific context is important for correlation, interpretation and analysis. Light detection and ranging (LiDAR) has established itself as the premier laser scanning modality for the acquisition of trusted geometry, while photogrammetry techniques like structure from motion (SfM) are used to construct visually compelling models. A common challenge of these line-of-sight techniques is that the imaging equipment must be systematically moved throughout the target environment to assure that the data captures the entire target and allows for the removal of occlusions in the final model. By combining terrestrial and airborne imaging techniques using unmanned aerial vehicles (UAV), also frequently referred to as drones, it is possible to streamline the acquisition of the target data sets. This research focuses on the fusion of full resolution three-dimensional data streams generated from laser scanning, ground based photogrammetry and drone based photogrammetry. Maintaining full resolution of the data sets allows for diagnostic analysis of very subtle deformations and defects like erosion and cracks. In a presented case study in Mexico, terrestrial laser scanning serves as a geometric scaffold that the photogrammetry data is registered to in order to generate a holistic model of a one hectare site containing two historic structures. The laser scanning and photogrammetry data sets have sufficient overlap to enable fusion, and more importantly the individual sets can supplement each other, providing geometry, photorealism and context that the other set lacks.

Arma Veirana is a Middle/Upper Paleolithic cave site of the Maritime Alps of Liguria, Italy, which has the potential to unfold knowledge on the interaction between Modern Humans and the Neandertals. Preliminary excavations have shown a continuous occupation between the Middle and Upper Paleolithic time periods, yet the complexity of the cave morphology and geology have made it difficult to isolate erosion, environmental and non-natural factors to understand the full image of hominin interaction and prehistoric life. We propose a novel method to rapidly combine and visualize timelapsed Photogrammetric and LIDAR data of the cave excavations. This enables archaeologists to evaluate and analyze the layers and artifacts of the excavation within the context of the cave. Through a better understanding and visualization of the cave excavation over time, it is expected to further link micro and macro factors of the site which would otherwise not have been as apparent.

With the emergence of low-cost multicopters on the market, archaeologists have rapidly integrated aerial imaging and photogrammetry with more traditional methods of site documentation. Unmanned Aerial Vehicles (UAVs) serve as simple yet transformative tools that can rapidly map archaeological sites with increased efficiency and higher resolution than manual measurements while contextualizing the site within the landscape at costs significantly cheaper than plane-based aerial LiDAR systems and total stations. UAV drones serve as a cost-effective platform, and Structure From Motion (SFM) an ideal entry point for further adoption of diagnostic imaging to facilitate efficient site mapping and offer archaeologists innovative ways to visualize and analyze data. The aim of the aerial surveying was to obtain an accurate Digital Elevation Model (DEM) of multiple sites which could be compared to models which were created using ground total stations as well as plane based LIDAR.

Recent advances in robotic control and autonomous systems have driven the need for robust, fast and accurate estimations of surrounding environments and motions of objects within. Stereo depth estimation has been a widely adopted method to estimate the geometry of a scene from a pair of cameras. From the disparity in features across two sets of images, it is possible to estimate the distance for each respective feature and create a depth map. Autonomous vehicles benefit from frequent estimates of the surroundings, and as such video-rate stereo depthmaps are a common target throughout literature. The driving limitation in existent implementations is the lack resolution that is achieved at these frame rates when the computation is completed in software on general purpose processors, namely low-power ARM processors. GPU accelerations have demonstrated higher resolutions depth estimates for the same frame-rate, and recently dedicated hardware architectures have been proven to even further accelerate such tasks. This project proposes a hardware accelerated dense stereo depth camera with a targeted operation of 1920x1080 pixels at 30 frames per second, operating on a fully integrated Xilinx MPSoC, with parts of the capture and computation accelerated on the programmable logic and the rest running on the quad-core ARM A53 processors. This High Level Synthesis driven project highlights the ease of integration for complex embedded systems, while maintaining optimal performance and customizability.

The goal of this project was to develop visualization methods for the interpretation and analysis of time- varying voxel concentrations from a particle propagation simulation. A simulated plume provided 3d spatial concentration values at 1 minute intervals for 16 hours in proximity to the Diablo Power Plant. From this plume, we wanted to model with representative colors how this plume propagated in relation to the coastal environment at different altitudes. With a high resolution 3D terrain model, we were able to visualize the plume evolution at a desired playback speed. Future works will focus on sensory measurement analysis of this plume in the environment and the visualization of such nodes.

Geometry, appearance and context are essential aspects to capture in the digital documentation of cultural heritage sites. Geometry must be accurate and should provide a level of precision necessary for quantified diagnostics. Visual appearance should capture the “as-is” state, while site specific context is important for correlation, interpretation and analysis. Light detection and ranging (LiDAR) has established itself as the premier laser scanning modality for the acquisition of trusted geometry, while photogrammetry techniques like structure from motion (SfM) are used to construct visually compelling models. A common challenge of these line-of-sight techniques is that the imaging equipment must be systematically moved throughout the target environment to assure that the data captures the entire target and allows for the removal of occlusions in the final model. By combining terrestrial and airborne imaging techniques using unmanned aerial vehicles (UAV), also frequently referred to as drones, it is possible to streamline the acquisition of the target data sets. This research focuses on the fusion of full resolution three-dimensional data streams generated from laser scanning, ground based photogrammetry and drone based photogrammetry. Maintaining full resolution of the data sets allows for diagnostic analysis of very subtle deformations and defects like erosion and cracks. In a presented case study in Mexico, terrestrial laser scanning serves as a geometric scaffold that the photogrammetry data is registered to in order to generate a holistic model of a one hectare site containing two historic structures. The laser scanning and photogrammetry data sets have sufficient overlap to enable fusion, and more importantly the individual sets can supplement each other, providing geometry, photorealism and context that the other set lacks.

Arma Veirana is a Middle/Upper Paleolithic cave site of the Maritime Alps of Liguria, Italy, which has the potential to unfold knowledge on the interaction between Modern Humans and the Neandertals. Preliminary excavations have shown a continuous occupation between the Middle and Upper Paleolithic time periods, yet the complexity of the cave morphology and geology have made it difficult to isolate erosion, environmental and non-natural factors to understand the full image of hominin interaction and prehistoric life. We propose a novel method to rapidly combine and visualize timelapsed Photogrammetric and LIDAR data of the cave excavations. This enables archaeologists to evaluate and analyze the layers and artifacts of the excavation within the context of the cave. Through a better understanding and visualization of the cave excavation over time, it is expected to further link micro and macro factors of the site which would otherwise not have been as apparent.

With the emergence of low-cost multicopters on the market, archaeologists have rapidly integrated aerial imaging and photogrammetry with more traditional methods of site documentation. Unmanned Aerial Vehicles (UAVs) serve as simple yet transformative tools that can rapidly map archaeological sites with increased efficiency and higher resolution than manual measurements while contextualizing the site within the landscape at costs significantly cheaper than plane-based aerial LiDAR systems and total stations. UAV drones serve as a cost-effective platform, and Structure From Motion (SFM) an ideal entry point for further adoption of diagnostic imaging to facilitate efficient site mapping and offer archaeologists innovative ways to visualize and analyze data. The aim of the aerial surveying was to obtain an accurate Digital Elevation Model (DEM) of multiple sites which could be compared to models which were created using ground total stations as well as plane based LIDAR.