Structure From Motion Algorithm Research Articles

SEMANTIC ROAD SCENE KNOWLEDGE FOR ROBUST SELF-CALIBRATION OF ENVIRONMENT-OBSERVING VEHICLE CAMERAS

Abstract. Environment-observing vehicle camera self-calibration using a structure from motion (SfM) algorithm allows calibration over vehicle lifetime without the need of special calibration objects being present in the calibration images. Scene-specific problems with feature-based correspondence search and reconstruction during the SfM pipeline might be caused by critical objects like moving objects, poor-texture objects or reflecting objects and might have negative influence on camera calibration. In this contribution, a method to use semantic road scene knowledge by means of semantic masks for a semantic-guided SfM algorithm is proposed to make the calibration more robust. Semantic masks are used to exclude image parts showing critical objects from feature extraction, whereby semantic knowledge is obtained by semantic segmentation of the road scene images. The proposed method is tested with an image sequence recorded in a suburban road scene. It has been shown that semantic guidance leads to smaller deviations of the estimated interior orientation and distortion parameters from reference values obtained by test field calibration compared to a standard SfM algorithm.

UFPR CampusMap: a laboratory for a Smart City developments

UFPR CampusMap: a laboratory for a Smart City developments

TREE SPECIES CLASSIFICATION BASED ON 3D SPECTRAL POINT CLOUDS AND ORTHOMOSAICS ACQUIRED BY SNAPSHOT HYPERSPECTRAL UAS SENSOR

Abstract. In recent years, there has been a growing number of small hyperspectral sensors suitable for deployment on unmanned aerial systems (UAS. The introduction of the hyperspectral snapshot sensor provides interesting opportunities for acquisition of three-dimensional (3D) hyperspectral point clouds based on the structure-from-motion (SfM) workflow. In this study, we describe the integration of a 25-band hyperspectral snapshot sensor (PhotonFocus camera with IMEC 600 – 875 nm 5x5 mosaic chip) on a multi-rotor UAS. The sensor was integrated with a dual frequency GNSS receiver for accurate time synchronisation and geolocation. We describe the sensor calibration workflow, including dark current and flat field characterisation. An SfM workflow was implemented to derive hyperspectral 3D point clouds and orthomosaics from overlapping frames. On-board GNSS coordinates for each hyperspectral frame assisted in the SfM process and allowed for accurate direct georeferencing (< 10 cm absolute accuracy). We present the processing workflow to generate seamless hyperspectral orthomosaics from hundreds of raw images. Spectral reference panels and in-field spectral measurements were used to calibrate and validate the spectral signatures. This process provides a novel data type which contains both 3D, geometric structure and detailed spectral information in a single format. First, to determine the potential improvements that such a format could provide, the core aim of this study was to compare the use of 3D hyperspectral point clouds to conventional hyperspectral imagery in the classification of two Eucalyptus tree species found in Tasmania, Australia. The IMEC SM5x5 hyperspectral snapshot sensor was flown over a small native plantation plot, consisting of a mix of the Eucalyptus pauciflora and E. tenuiramis species. High overlap hyperspectral imagery was captured and then processed using SfM algorithms to generate both a hyperspectral orthomosaic and a dense hyperspectral point cloud. Additionally, to ensure the optimum spectral quality of the data, the characteristics of the hyperspectral snapshot imaging sensor were analysed utilising measurements captured in a laboratory environment. To coincide with the generated hyperspectral point cloud data, both a file format and additional processing and visualisation software were developed to provide the necessary tools for a complete classification workflow. Results based on the classification of the E. pauciflora and E. tenuiramis species revealed that the hyperspectral point cloud produced an increased classification accuracy over conventional hyperspectral imagery based on random forest classification. This was represented by an increase in classification accuracy from 67.2% to 73.8%. It was found that even when applied separately, the geometric and spectral feature sets from the point cloud both provided increased classification accuracy over the hyperspectral imagery.

Optimal flight parameters for unmanned aerial vehicles collecting spatial information for estimating large-scale waste generation

ABSTRACT Large-scale waste and debris from disasters and building construction can cause secondary damage and environmental pollution. Satellite images and the Global Positioning System are being used to calculate the amounts of such waste. However, efficiency in terms of accuracy and cost remains problematic. To solve this problem, this study used an unmanned aerial vehicle (UAV) to build three-dimensional (3D) spatial information for large-scale waste generation. Appropriate parameters were selected to quickly and accurately estimate the amount of waste. Fifty-six scenarios were distinguished according to parameters such as flight altitude, image overlap, and number of ground control points (GCPs). Images were then registered by implementing the SIFT (Scale Invariant Feature Transform) and SfM (Structure from Motion) algorithms to build 3D spatial information and estimate the amount of waste. The accuracy was generally high when flight altitude and image overlap were high, and the number of GCPs was specific. Scenario A, with the lowest flight altitude and overlap, did not yield any registrable images. The results of the flight altitude showed a different tendency from those of existing studies. This could be attributable to the difference in the distance covered on the ground by an image in the width direction, which depended on the characteristics of study sites. Among all the scenarios, D-7 had the highest accuracy, with a root-mean-square error of about 0.08 m. In this scenario, the amount of waste was estimated to be approximately 767,571.3 m3. The top 10 scenarios in terms of the accuracy of x, y, and z locations displayed similar tendencies with regard to the amount of waste. In contrast, the bottom 15 scenarios showed a variety of results. When the accuracy and flight time of each scenario were considered, scenarios H-3, H-4, and H-5 achieved the highest accuracy in a short time. Thus, these scenarios were the most effective and applicable for estimating the amount of waste. Consequently, parameters such as flight altitude, image overlap, and GCPs need to be considered when estimating large-scale waste associated with a difference in elevation using a UAV.

Comparison between point cloud and mesh models using images from an unmanned aerial vehicle

Structure from motion (SfM) is a well-known algorithm used for the generating of three-dimensional (3D) spatial information using images. The objective of this study is to compare the measurements of objects ascertained from point cloud and mesh models derived from the SfM algorithm. In particular, we analyze a single tree to determine the correlation between the number of acquired images from the UAVs and the object measurement for each model. The results indicate that the number of images does not have a critical impact on surveys and the point cloud is approximately 2% more accurate than mesh models for individual tree measurement. Our results will be useful in terms of selecting the data acquisition method as well as the data itself for measuring objects based on SfM 3D data.

Joining multi-epoch archival aerial images in a single SfM block allows 3-D change detection with almost exclusively image information

Archival aerial imagery is a worldwide resource for documenting past 3-D change at very high-resolution. However, external information is normally required so that accurate 3-D models can be computed from archival aerial imagery. In this research, we propose and test a new method which joins multi-epoch images in a single block in the first steps of the structure from motion (SfM) processing. It allows for computing coherent multi-temporal digital elevation models (DEMs) using just image information. This method is based on the invariance properties of the feature detection procedures that are at the root of the SfM algorithms.On a test site covering 170 km2, we applied SfM algorithms to a single image block consisting of all images captured at four different epochs and spanning a forty year period. We compared this approach to the more classical methods which imply a separation of epochs in different processing blocks. We tested different densities of ground control points derived simply and cheaply from a recent orthophoto and DEM, different ways of image preprocessing and different autocalibration procedures. By determining which choice most affected the final result through this extensive testing procedure, we evaluated the potential of the proposed method for detecting 3-D change.Our study showed that the proposed method resolves the problem of registration between epochs, so allowing the production of informative DEMs of difference using almost exclusively image information and limited photogrammetric expertise and human intervention. As the proposed method can be automatically applied using just image information, our results pave the way to more systematic processing of archival aerial imagery with very large spatio-temporal windows, which should greatly help document of past 3-D change.

Using Laser Measuring and SFM Algorithm for Fast 3D Reconstruction of Objects

Effective measurement of the reflective or transparent surface of an object has always been a disadvantage in laser scanning modeling. We propose a fast and complete three-dimensional (3D) reconstruction method for small static objects using laser scanning and the structure from motion (SFM) algorithm. Meanwhile, a complete reconstruction workflow is designed and a multi-angle 3d reconstruction system is set up. To generate the complete point cloud model of the object, the SFM algorithm is used to reconstruct the surface part of the object, the data for which cannot be obtained by the laser measuring instrument. The experimental results show that this method not only improves the speed, accuracy, integrity, and visual effect of 3D reconstruction of small objects, but also extends the scope of 3D reconstruction of laser measurement.

The reproducibility of SfM algorithms to produce detailed Digital Surface Models: the example of PhotoScan applied to a high-alpine rock glacier

ABSTRACTIn geomorphology, PhotoScan is a software that is used to produce Digital Surface Models (DSMs). It constructs 3D environments from 2D imagery (often taken by Unmanned Aerial Vehicles (UAV)) based on Structure-from-Motion (SfM) and Multi-View Stereo (MVS) principles. However, unpublished computer-vision algorithms used, contain random elements which can affect the accuracy of the outputs. For this letter, ten model runs with identical inputs were performed on UAV imagery of a rock glacier to analyse the magnitude of the variation between the different model outputs. This variation was quantified calculating the standard deviation of each cell value in the respective DSMs and derivatives (curvature). Places with steep slope gradients have considerably more DSM variation (up to 10 cm) but stay within the range of the model’s accuracy (10 vertical cm) for 88 – 96% of the area. The edges of the model also show a larger variability (0.10 – 3 m), related to a lower number of overlapping images. These results should be accounted for when performing a geomorphological research at centimetre scale using PhotoScan, especially in areas with a complex relief. Using medium-quality runs, additional oblique viewpoints and respecting a minimum of five overlapping images can minimize the software’s variations.

An Indoor Location-Based Positioning System Using Stereo Vision with the Drone Camera

Stereo vision is used to reconstruct 3D information of the space by estimating the depth value from the simulation of human eyes. Spatial restoration can be used as a means of location estimation in an indoor area, which is impossible to accomplish using the relative location estimation technology, GPS. By mapping the real world in virtual space, it is feasible to clear the boundary between real space and virtual space. This paper presents a method to control the drone indoors through a positioning system using Structure from Motion algorithm (SfM). SfM calculates the relative relationship between cameras based on images to be acquired from various locations and obtains disparity to enable restoration of 3D space. First, the 3D virtual space is reconstructed using several photographs taken from an indoor environment. Second, the real-time drone position is determined by comparing the 3D virtual space camera with the image displayed on the drone camera. In this case, if the direction of the virtual camera used for 3D virtual space construction is the same as the amount of yaw rotation of the drone, it is possible to quickly find the same position as the image seen in the real drone camera in the virtual space. As a result, if the scale of the actual camera image and the virtual camera image is 1 : 1 matched, then it is possible to know that the drone is in the position of the virtual camera. The proposed indoor location-based drone controlling method can be applied to various drone applications such as group flight in an indoor environment because of its ability to fly the drone without the use of the traditional remote-control and flight trajectory programming.

Accuracy of Unmanned Aerial Vehicle (UAV) and SfM Photogrammetry Survey as a Function of the Number and Location of Ground Control Points Used

The geometrical accuracy of georeferenced digital surface models (DTM) obtained from images captured by micro-UAVs and processed by using structure from motion (SfM) photogrammetry depends on several factors, including flight design, camera quality, camera calibration, SfM algorithms and georeferencing strategy. This paper focusses on the critical role of the number and location of ground control points (GCP) used during the georeferencing stage. A challenging case study involving an area of 1200+ ha, 100+ GCP and 2500+ photos was used. Three thousand, four hundred and sixty-five different combinations of control points were introduced in the bundle adjustment, whilst the accuracy of the model was evaluated using both control points and independent check points. The analysis demonstrates how much the accuracy improves as the number of GCP points increases, as well as the importance of an even distribution, how much the accuracy is overestimated when it is quantified only using control points rather than independent check points, and how the ground sample distance (GSD) of a project relates to the maximum accuracy that can be achieved.

Using Drones and 3D Modeling to Survey Tibetan Architectural Heritage: A Case Study with the Multi-Door Stupa

Tibet has an average altitude of 4900 m, a complex terrain, and unique climatic conditions. The technologies used to survey this country’s local architectural heritage must be portable, efficient, and versatile. Low-cost Unmanned Aerial Vehicles (UAVs) and Structure from Motion (SfM) algorithms can help satisfy these demands. Recent studies employing nadir images from low-cost UAVs and SfM algorithms have reported positive metric results (with centimeter-level accuracy) when modeling 2D objects (e.g., land, roofs, and facades). In Tibet, however, forming a complete 3D model of architectural heritage is highly preferable and doing so requires a camera network that can create nadir and oblique images with various baselines. This study compared the accuracies of surveying a Tibetan stupa using the UAV-SfM method compared with Ground Control Points (GCP) and Terrestrial Laser Scanning (TLS). The results indicated that the UAV-derived model is accurate enough for most surveying purposes (RMSE = 2.05 cm; 1/2000 of the stupa’s dimension). The accuracy and completeness of the 3D model allowed Historic Building Information Modeling (HBIM) and structural deformation analysis to also be undertaken. In addition, the stupa was integrated with geographic data (terrain and infrastructure) for visualization, management, and evaluation purposes at a larger scale. Considering the low cost, portability, and completeness offered by UAV and SfM, this tool offers promise for surveying Tibet’s architectural heritage.

Three-Dimensional Modeling of Weed Plants Using Low-Cost Photogrammetry.

Sensing advances in plant phenotyping are of vital importance in basic and applied plant research. Plant phenotyping enables the modeling of complex shapes, which is useful, for example, in decision-making for agronomic management. In this sense, 3D processing algorithms for plant modeling is expanding rapidly with the emergence of new sensors and techniques designed to morphologically characterize. However, there are still some technical aspects to be improved, such as an accurate reconstruction of end-details. This study adapted low-cost techniques, Structure from Motion (SfM) and MultiView Stereo (MVS), to create 3D models for reconstructing plants of three weed species with contrasting shape and plant structures. Plant reconstruction was developed by applying SfM algorithms to an input set of digital images acquired sequentially following a track that was concentric and equidistant with respect to the plant axis and using three different angles, from a perpendicular to top view, which guaranteed the necessary overlap between images to obtain high precision 3D models. With this information, a dense point cloud was created using MVS, from which a 3D polygon mesh representing every plants’ shape and geometry was generated. These 3D models were validated with ground truth values (e.g., plant height, leaf area (LA) and plant dry biomass) using regression methods. The results showed, in general, a good consistency in the correlation equations between the estimated values in the models and the actual values measured in the weed plants. Indeed, 3D modeling using SfM algorithms proved to be a valuable methodology for weed phenotyping, since it accurately estimated the actual values of plant height and LA. Additionally, image processing using the SfM method was relatively fast. Consequently, our results indicate the potential of this budget system for plant reconstruction at high detail, which may be usable in several scenarios, including outdoor conditions. Future research should address other issues, such as the time-cost relationship and the need for detail in the different approaches.

MULTISEASONAL TREE CROWN STRUCTURE MAPPING WITH POINT CLOUDS FROM OTS QUADROCOPTER SYSTEMS

Abstract. OTF (Off The Shelf) quadro copter systems provide a cost effective (below 2000 Euro), flexible and mobile platform for high resolution point cloud mapping. Various studies showed the full potential of these small and flexible platforms. Especially in very tight and complex 3D environments the automatic obstacle avoidance, low copter weight, long flight times and precise maneuvering are important advantages of these small OTS systems in comparison with larger octocopter systems. This study examines the potential of the DJI Phantom 4 pro series and the Phantom 3A series for within-stand and forest tree crown 3D point cloud mapping using both within stand oblique imaging in different altitude levels and data captured from a nadir perspective. On a test site in Brandenburg/Germany a beach crown was selected and measured with 3 different altitude levels in Point Of Interest (POI) mode with oblique data capturing and deriving one nadir mosaic created with 85/85 % overlap using Drone Deploy automatic mapping software. Three different flight campaigns were performed, one in September 2016 (leaf-on), one in March 2017 (leaf-off) and one in May 2017 (leaf-on) to derive point clouds from different crown structure and phenological situations – covering the leaf-on and leafoff status of the tree crown. After height correction, the point clouds where used with GPS geo referencing to calculate voxel based densities on 50 × 10 × 10 cm voxel definitions using a topological network of chessboard image objects in 0,5 m height steps in an object based image processing environment. Comparison between leaf-off and leaf-on status was done on volume pixel definitions comparing the attributed point densities per volume and plotting the resulting values as a function of distance to the crown center. In the leaf-off status SFM (structure from motion) algorithms clearly identified the central stem and also secondary branch systems. While the penetration into the crown structure is limited in the leaf-on status (the point cloud is a mainly a description of the interpolated crown surface) – the visibility of the internal crown structure in leaf-off status allows to map also the internal tree structure up to and stopping at the secondary branch level system. When combined the leaf-on and leaf-off point clouds generate a comprehensive tree crown structure description that allows a low cost and detailed 3D crown structure mapping and potentially precise biomass mapping and/or internal structural differentiation of deciduous tree species types. Compared to TLS (Terrestrial Laser Scanning) based measurements the costs are neglectable and in the range of 1500–2500 €. This suggests the approach for low cost but fine scale in-situ applications and/or projects where TLS measurements cannot be derived and for less dense forest stands where POI flights can be performed. This study used the in-copter GPS measurements for geo referencing. Better absolute geo referencing results will be obtained with DGPS reference points. The study however clearly demonstrates the potential of OTS very low cost copter systems and the image attributed GPS measurements of the copter for the automatic calculation of complex 3D point clouds in a multi temporal tree crown mapping context.

What is the Point? Evaluating the Structure, Color, and Semantic Traits of Computer Vision Point Clouds of Vegetation

Remote sensing of the structural and spectral traits of vegetation is being transformed by structure from motion (SFM) algorithms that combine overlapping images to produce three-dimensional (3D) red-green-blue (RGB) point clouds. However, much remains unknown about how these point clouds are used to observe vegetation, limiting the understanding of the results and future applications. Here, we examine the content and quality of SFM point cloud 3D-RGB fusion observations. An SFM algorithm using the Scale Invariant Feature Transform (SIFT) feature detector was applied to create the 3D-RGB point clouds of a single tree and forest patches. The fusion quality was evaluated using targets placed within the tree and was compared to fusion measurements from terrestrial LIDAR (TLS). K-means clustering and manual classification were used to evaluate the semantic content of SIFT features. When targets were fully visible in the images, SFM assigned color in the correct place with a high accuracy (93%). The accuracy was lower when targets were shadowed or obscured (29%). Clustering and classification revealed that the SIFT features highlighted areas that were brighter or darker than their surroundings, showing little correspondence with canopy objects like leaves or branches, though the features showed some relationship to landscape context (e.g., canopy, pavement). Therefore, the results suggest that feature detectors play a critical role in determining how vegetation is sampled by SFM. Future research should consider developing feature detectors that are optimized for vegetation mapping, including extracting elements like leaves and flowers. Features should be considered the fundamental unit of SFM mapping, like the pixel in optical imaging and the laser pulse of LIDAR. Under optimal conditions, SFM fusion accuracy exceeded that of TLS, and the two systems produced similar representations of the overall tree shape. SFM is the lower-cost solution for obtaining accurate 3D-RGB fusion measurements of the outer surfaces of vegetation, the critical zone of interaction between vegetation, light, and the atmosphere from leaf to canopy scales.

Dense depth maps from correspondences derived from perceived motion

Many computer vision applications require finding corresponding points between images and using the corresponding points to estimate disparity. Today’s correspondence finding algorithms primarily use image features or pixel intensities common between image pairs. Some 3-D computer vision applications, however, do not produce the desired results using correspondences derived from image features or pixel intensities. Two examples are the multimodal camera rig and the center region of a coaxial camera rig. We present an image correspondence finding technique that aligns pairs of image sequences using optical flow fields. The optical flow fields provide information about the structure and motion of the scene, which are not available in still images but can be used in image alignment. We apply the technique to a dual focal length stereo camera rig consisting of a visible light—infrared camera pair and to a coaxial camera rig. We test our method on real image sequences and compare our results with the state-of-the-art multimodal and structure from motion (SfM) algorithms. Our method produces more accurate depth and scene velocity reconstruction estimates than the state-of-the-art multimodal and SfM algorithms.

Efficient Non-Consecutive Feature Tracking for Robust Structure-From-Motion.

Structure-from-motion (SfM) largely relies on feature tracking. In image sequences, if disjointed tracks caused by objects moving in and out of the field of view, occasional occlusion, or image noise are not handled well, corresponding SfM could be affected. This problem becomes severer for large-scale scenes, which typically requires to capture multiple sequences to cover the whole scene. In this paper, we propose an efficient non-consecutive feature tracking framework to match interrupted tracks distributed in different subsequences or even in different videos. Our framework consists of steps of solving the feature "dropout" problem when indistinctive structures, noise or large image distortion exists, and of rapidly recognizing and joining common features located in different subsequences. In addition, we contribute an effective segment-based coarse-to-fine SfM algorithm for robustly handling large data sets. Experimental results on challenging video data demonstrate the effectiveness of the proposed system.

An SfM Algorithm With Good Convergence That Addresses Outliers for Realizing Mono-SLAM

Monocular simultaneous localization and mapping (mono-SLAM) is a key component of autonomous robot visual navigation. Recently, the structure from motion (SfM) approach has become an attractive means of implementing mono-SLAM because of its high accuracy; although, in this application, the SfM approach must be operable in real time and robust to outliers. However, because of strong nonlinearity, conventional SfM methods, such as bundle adjustment, must consider multiple initial values to obtain the globally optimal result, which is time consuming. In this paper, a novel iterative SfM algorithm based on the object-space objective function is proposed. To improve the method’s robustness to outliers and to incorporate the information obtained from other types of sensors, an approach to closely integrate the proposed SfM algorithm with extended Kalman filter-based SLAM is proposed. Experimental results using both synthesized and real data are consistent with our theory, and verify that the main advantage of the proposed SfM algorithm is its good convergence. The algorithm is therefore particularly appropriate for realizing mono-SLAM, because when rotations can be obtained approximately using gyroscope information, the algorithm is globally convergent from any initial value.

Onboard monocular pedestrian detection by combining spatio-temporal hog with structure from motion algorithm

In this paper, we brought out a novel pedestrian detection framework for the advanced driver assistance system of mobile platform under the normal urban street environment. Different from the conventional systems that focus on the pedestrian detection at near distance by interfusing multiple sensors (such as radar, laser and infrared camera), our system has achieved the pedestrian detection at all (near, middle and long) distance on a normally driven vehicle (1---40 km/h) with monocular camera under the street scenes. Since pedestrians typically exhibit not only their human-like shape but also the unique human movements generated by their legs and arms, we use the spatio-temporal histogram of oriented gradient (STHOG) to describe the pedestrian appearance and motion features. The shape and movement of a pedestrian will be described by a unique feature produced by concatenating the spatial and temporal histograms. A STHOG detector trained by the AdaBoost algorithm will be applied to the images stabilized by the structure from motion (SfM) algorithm with geometric ground constraint. The main contributions of this work include: (1) ground constraint with monocular camera to reduce the computational cost and false alarms; (2) preprocessing by stabilizing the successive images captured from mobile camera with the SfM algorithm; (3) long-distance (maximum 100 m) pedestrian detection at various velocities (1---40 km/h). Through the extensive experiments under different city scenes, the effectiveness of our algorithm has been proved.

THE METRIC DOCUMENTATION OF CHAM TOWERS IN VIETNAM BY SPHERICAL PHOTOGRAMMETRY

Abstract. The Università Politecnica delle Marche is charged of the study, survey and documentation of the Chan towers in Vietnam. The towers are spread near the coastline from latitude 16° 30', to latitude of 10° 56'. The Champa civilization flourished from 8th century till 17th century produced the Cham towers, places of religious worship. The ancient Cham people created a developed culture expressed in the architecture in the towers, which were Buddhist temples. Roughly there are six different architectonic styles and two different layouts in plan. One lay-out is that one main tower is surrounded by smaller service buildings, like walls, doors, gates; the other layout is a set of three aligned towers. We carried out the documentation with three different techniques, laser scanning, dense point clouds generation with SfM (Structure from Motion) algorithm, and mainly by Panoramic Spherical Photogrammetry (PSP). Laser scanning is indeed the best instrument but sometimes its availability is very limited for different practical and logistic reasons. The point cloud generation produced good results, but it has some problems and limitations of management too. On the other hand PSP is very quick technique in the taking phase, but the restitution is rather slow. PSP is suitable for a good metric documentation, making the Cham tower project the main project of documentation executed with spherical photogrammetry.

USING A MICRO-UAV FOR ULTRA-HIGH RESOLUTION MULTI-SENSOR OBSERVATIONS OF ANTARCTIC MOSS BEDS

Abstract. This study is the first to use an Unmanned Aerial Vehicle (UAV) for mapping moss beds in Antarctica. Mosses can be used as indicators for the regional effects of climate change. Mapping and monitoring their extent and health is therefore important. UAV aerial photography provides ultra-high resolution spatial data for this purpose. We developed a technique to extract an extremely dense 3D point cloud from overlapping UAV aerial photography based on structure from motion (SfM) algorithms. The combination of SfM and patch-based multi-view stereo image vision algorithms resulted in a 2 cm resolution digital terrain model (DTM). This detailed topographic information combined with vegetation indices derived from a 6-band multispectral sensor enabled the assessment of moss bed health. This novel UAV system has allowed us to map different environmental characteristics of the moss beds at ultra-high resolution providing us with a better understanding of these fragile Antarctic ecosystems. The paper provides details on the different UAV instruments and the image processing framework resulting in DEMs, vegetation indices, and terrain derivatives.