Without their continued moral support and motivation, this dissertation would not have been completed within the deadline. 23 III.3 Example of a back-and-forth flight path (solid line) to cover a convex polygon area.
Mawchu Llacta Archaeological Site
Brief History of UAVs
UAVs must be competitive in terms of economic application compared to other measurement technologies. UAVs are more time efficient, while other low-altitude systems may provide more accurate and higher resolution results, but as sensors decrease in size and increase in efficiency, UAVs will be the system of choice for photogrammetric applications (Eisenbeiss and Zhang, 2006; Lambers et al., 2007; Sauerbier et al., 2011).
UAV Classification
Types
Coaxial rotocraft UAVs have two counter-rotating blades mounted on the same axis, allowing them to carry heavier payloads and fly at higher altitudes. Fixed-wing UAVs cannot fly as close to objects as drones and are not as agile in maneuverability (Bendea et al., 2007), but a significant advantage is their ability to fly at higher altitudes (Eisenbeiss, 2009).
Sizes
Multi-rotor rotocraft UAVs are usually designed with either six or eight rotors and are capable of carrying larger payloads than quadrotors and continuing flight unaffected even if a motor fails due to the multiple rotor redundancy. Another significant advantage is that fixed-wing UAVs are capable of remaining in flight for much longer periods of time without the need to refuel.
Levels of Autonomy
These systems are capable of performing missions of the highest complexity in the most extreme environments, without any human interaction. Manual mode corresponds exactly to level one of the autonomy levels framework, where all movements of the UAV are controlled remotely by a human pilot with the option for system status information (such as battery/fuel levels and connection status) to be monitored by the pilot. or a secondary operator.
UAV Path Planning
Point to Point Planning
- Avoiding Threats and Obstacles
- Harnessing Wind
Semi-automatic or assisted mode roughly corresponds to the second or third level of the autonomy level framework. Theta∗ is an extension of the A∗ algorithm, as it allows the generation of paths in three dimensions.
Surveillance
However, if the ground target being observed is large enough (such as a bridge), it is possible that special wind patterns (such as the Venturi effect or Karman eddies) occur that affect the UAV's path (Guerrero and Bestaoui, 2012). The UAV must generate a flight plan that allows it to photograph all sides of the structure, calculating the position of the sun.
Simultaneous Localization And Mapping
GPS has an accuracy of less than fifteen meters in the latitude and longitude directions, but between twenty-five and fifty meters accuracy in the vertical (elevation) direction (Eynard et al., 2012). However, all the landing parameters are pre-programmed into the system and do not adapt to different environments, such as the higher altitudes at Mawchu Llacta.
Photogrammetric Applications
Overview
Hardware specifications and algorithmic improvements in the near future may enable these calculations to be performed in real time. In addition, the poor resolution GPS in the altitude direction makes it effectively useless for take-off and landing.
Archaeological Photogrammetry
- Path Planning
- Image Acquisition
- Image Processing
Eisenbeiss (Eisenbeiss, 2009) developed a 2/2.5D planning tool, but it is described as taking the necessary parameters and then outputting a file specifying all the waypoints using in-house software. This match reduction process dramatically speeds up and increases the accuracy of the image matching process. The resolution of the images and the accuracy of the multi-view stereopsis algorithm determine the density of points in the point cloud and ultimately the resolution of the final DSM.
The resolution of the 3D model is limited only by the point cloud density and the amount of polygon faces that the graphics card can display.
Conclusion
LPS and Photoscan generate the point cloud while matching the images, as both processes require knowledge of shared points between images. The final step is to build a 3D model around the point cloud and fit the image mosaic to this model if desired. Depending on the density of the point cloud, this step is fairly straightforward – connect the points in the point cloud to polygon surfaces, then decimate the polygon mesh by removing redundant polygons to fit the specified resolution parameters.
Introduction
UAV Platform
The Skate has an auto-landing function that, when activated, causes the UAV to fly in a downward spiral until it lands. a) The Skate 2 from Aurora Flight Sciences. The camera used to capture footage of the Skate was an HD Hero2 GoPro with an 11 MP sensor and an adjustable field of view setting at 90◦ (narrow field of view), 127◦ (medium field of view), or 170◦ (wide field of view). The camera is mounted under the payload pod on the front of the skate (indicated by the yellow arrow in Figure III.1a), pointing downwards in the lowest direction, and can take up to ten photos per second.
Due to the nature of how the skate flies (at a slight upward angle as shown in Figure III.1c), the camera lens points slightly forward as the skate flies.
Flight Planning Algorithm
- Flight Altitude
- Area Geometry
- Wind
- Combining Geometry and Wind
- Optimum Time-of-Day
- The Algorithm
Flying into the wind will use more battery power if that is the selected direction, but the UAV will only fly into the wind half the time - the other half will be spent flying into the wind, allowing the motors to throttle. back. If we simply take the average of the two vectors, we only consider the direction of each vector. The UAV flies in semi-circles outside the polygon as it turns to fly the next trajectory line as shown in Figure III.3.
Once all the algorithm parameters are specified by the user, a flight plan can be generated.
Simulation Data
- Rotated 0 ◦
- Rotated 30 ◦
- Rotated 60 ◦
- Rotated 90 ◦
- Simulation Summary
The oscillations seen when the rectangle was rotated 30◦ in combination mode (Figure III.8c) are missing from the simulation in Figure III.10c, but are most likely due to the fact that the wind speed in that test was only 2 m/ s was. Although not all flight lines are shown, the reader can see that there are clearly more total flight lines in Figure III.10b than, for example, in Figure III.10a. Even under these conditions, the geometry mode is the best in terms of flight time and path accuracy (see Figures III.11 and III.12).
The simulations for wind set at 4 m/s and the rectangle rotated 90◦ in each of the three flight modes are presented in Figure III.12.
Introduction
Lens Distortion
In Figure IV.3a, at the top half of another row of three grid blocks is present, which is missing in Figure IV.3b. This is due to the loss of information when the images are cropped after the fisheye distortion is removed. It is unclear why using the distorted images produces more accurate results, but that phenomenon is most likely caused by the pixel compression during the fish eye removal process.
That compression and loss of information may be responsible for the loss of visual information at the top edge of Figure IV.3b, as well as the inaccuracies in point matching.
Less is More
Images that have a mostly uniform texture—for example, images that are only grass or dirt (see Figure IV.5a). Images that contain ambiguous structural details, such as images of only one flat wall (see Figure IV.5b). Images that are further away from the lowest axis than usual – such as images that contain the horizon or even the sky2 (see Figure IV.5c).
It is important to emphasize that every single one of the 634 images used in Figure IV.6 was also used in the generation of Figure IV.4, but Figure IV.4 used approx. 58.75% more images, where many of these images were either redundant or of poor quality compared to the specified criteria.
Ideal Performance
However, there are still many anomalies in the generated orthophoto, as indicated by the yellow rectangles in Figure IV.6. However, additional images that were not part of the back-and-forth section (the cluster of images at the top of Figure IV.7d) were part of this sequence, causing some artifacts to appear. Figures IV.8a and IV.8b show the top view of the same structure from the DSM, generated using images captured in a linear path and a 3D perspective view of the DSM, showing how the structure geometry is planar.
The orthoimage of that particular section of the DSM in Figure IV.8c is slightly fuzzy, suggesting that there may have been too much or not enough image information for that section.
Image Analysis Summary
Overall, the resulting DSM/orthophoto is close to what would be expected if the UAV were to fly in autonomous mode using the path planning algorithm presented in Chapter III. The area covered by these images is not the same area covered by the straight line path, so it is not possible to make a comparison of the same structure, but the structure with similar geometry is still present and a more detailed DSM can still be seen. Finally, to get a sense of the quality of results that can be achieved with Photoscan under ideal image capture conditions, images of a medium-sized rock were taken from all sides while moving around it twice to capture it from different angles .
Very few images were used (compared to the hundreds of images presented in the other experiments presented) and although the images were not captured in a perfect pattern, they were captured close to what is the ideal sequence and it seemed it was good enough for Photo Scanning to produce a high quality result.
Contributions
Conclusion
Future Work
In The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Pecora 15/Land Satellite Information IV/ISPRS Commission I/FIEOS, Denver, CO, USA. In Proceedings of the IEEE International Conference on Robotics and Automation, volume 2, St. Method for photogrammetric survey of archaeological sites with light aerial platforms.Journal of Archaeological Science. InProceedings of the International Society for Photogrammetry and Remote Sensing Conference on Unmanned Aerial Vehicles in Geomatics, Zurich, Switzerland.
InProceedings van die International Society for Photogrammetry and Remote Sensing Commission V Simposium, Tyne, VK.
Classification of Tactical UAVs
RUAS Size Classifications
Categorization with respect to price and payload of UAV systems
UAVs used in 3D aerial mapping
Therefore, minimizing the number of turns involves calculating the direction of the polygon that is the smallest (also known as the width of the polygon) and determining the flight path that is perpendicular to the direction of latitude. The current version of the path planning algorithm in this combined mode calculates the new flight vector (C) as the average of G~ and W~. The flight path shows the UAV trying to fly into the wind towards the corner of the range to begin surveying.
All optimization modes show anomalies in the half-circles that the UAV is supposed to fly when rotating out of the polygon. The geometry mode simulations performed surprisingly well with only small oscillations at the zone boundary. All other sets of simulations had the combination mode performing either slightly better or about the same as the wind mode.
Geometry mode performed best in each of the rotation sets, even when the proposed flight vector was perpendicular to the wind vector. However, the overshoot seen in the simulation when the UAV turns around causes the UAV to be consistently off-path at the boundaries of the flight region, potentially destroying the required overlap.
Path Planning Algorithm Simulation Data
Manual flights flown at Mawchu Llacta to capture images
Parameter settings for all of the image processing trials presented in this chapter
Orthophoto generated from every third image of the longest flight after removing im-
Straight-line and back-and-forth image capturing pattern orthophoto results
DSMs generated from straight-line image path and back-and-forth image path
Example DSM generated under ideal imaging conditions
