Progress 08/15/16 to 04/14/17
Outputs
Target Audience: Throughout the Phase I effort, we have found an ever-expanding group of interested parties that have expressed interest in implementing a system like VIA. The initial target audience was scientists working on land management issues such as erosion control and watershed management. It has since expanded out to those working on other land use issues including invasive species management, cattle grazing land management, as well as private interests in infrastructure management such as railroad, powerline and pipeline inspection. Changes/Problems: A major change during the project was the FAA's implementation of Part 107 operations. While we had planned for this to occur, the details within the new regulations were not finalized until its release. As the governing restrictions were released and clarified, and other organizations pushed for various exemptions, we were able to learn from their successes and failures. We were particularly interested in exemptions regarding flight over non-participating persons. With the CNN pathfinder project being met with much resistance from the FAA, we have come to realize that our initial μQuadcopter design may still face regulatory scrutiny, despite it falling under the 250 g registration limit. This will drive us towards a slightly larger design (≤ 2 kg) for Phase II, enabling us more capability along with increased safety features, facilitating exemptions from overflight requirements as well as operating beyond line of site and without the traditional vehicle operator supervising the flight. While this did not affect the outcome of the Phase I effort, it will drive the most significant redesign of the system moving forward into Phase II and commercialization efforts. A significant problem that prevented the completion of all of the intended objectives was the delay in initial funding. Because two of the objectives were contingent upon operating during the monsoon season (July through September) and funding was delayed beyond that time frame, we were not able to fly the same area repeatedly after successive rain fall events to measure erosion amounts or evaluate the effects of vegetation regrowth. However, both of these objectives will be carried forward into a Phase II effort if it is awarded. Some smaller technical challenges we have encountered consist of the details on how to capture the most effective data. Currently, our system only collects nadir pointed images. However, in a highly-sloped environment, it will be useful to incorporate imagery collected at oblique angles, allowing for more detail under canopies and around objects. Furthermore, steep slopes can cause errors in reconstruction of the point cloud, as the distance between the camera and the surface changes significantly across each image. Future vehicle designs will incorporate the capability for multiple cameras at different angles, ensuring imagery is captured that will yield the highest quality data. Similarly, we noted that surveys taken in highly sunny and bright conditions create point cloud errors within the much darker shadows of solid objects. The human eye's dynamic range is many times that of a typical camera, and when imaging the light tan dirt of the desert, details within the shadows are quickly lost. Therefore, we have begun exploring how taking multiple surveys of the same area, separated by a few hours (enough for the shadows to move significantly) and merging these chunks of data together. This should eliminate the dead spots of the shadows seen in any one survey. While results on this strategy are preliminary, but we continue to explore this method in future work. What opportunities for training and professional development has the project provided?
Nothing Reported
How have the results been disseminated to communities of interest? During the duration of the project, we continually engaged interested parties to gain an understanding of their specific needs regarding this type of data gathering. We were able to quickly turn around data sets and analysis to each party for them to weigh in on their confidence in the results and their interest in pursuing this strategy for further analysis. This included a local mining company, Modular Mining, the several groups at the University of Arizona including their mining department, Arizona Geological Society (AZGS), and the USDA Agriculture Research Service (ARS). All of these parties have received data of the parts of the project that they are most interested in the form requested. This constituted raw point clouds, full DEM models, simple volume measurements of the generated models and images demonstrating detectability and measurement capabilities. All of these groups were reached through leveraging Latitude's network of researchers at the University of Arizona and prior customers or collaborators. This network further connected us to more interested parties and facilitated further collaboration. Latitude's attitude toward collaboration and willingness to experiment with ideas it aligns with has led to partners and customers becoming strong advocates, bringing new clients and interested parties to us. Through this model, we have been able to reach far beyond our initial target audience, and despite the project having been completed, we continue to build new relationships with parties who have heard of this work and are asking how they can help. This puts us in a strong position for leveraging a phase II opportunity and quickly transitioning the technology to a commercial product. What do you plan to do during the next reporting period to accomplish the goals?
Nothing Reported
Impacts
What was accomplished under these goals?
Goal 1 - successful flights from PERCH within a burned area This objective was not met due to a major delay in receipt of funding. While the original schedule would have allowed for work to commence prior to the fire season beginning, the award funds were not distributed until after the fire season ended and areas that were identified by interested parties were no longer of value. In lieu of this setback, we contacted other groups that we identified with similar objectives that did not require a time critical schedule. The AZGS facility near Tombstone, AZ is currently using LIDAR to map various terrain features for several purposes. One such use is to determine the volume of ponds used for grazing cattle and to verify the hydrology of the areas around the ponds to maximize the catchment areas and measure over flow during monsoon events. Modular Mining routinely uses LIDAR to scan their operations to evaluate amounts of extracted material and are also interested in more extensively mapping their facilities for a variety of reasons. Due to the constantly changing landscape of their properties due to their operations, regular, repeated mappings must be conducted. This is another use case of the VIA system that fits well within the capabilities of the system. The University of Arizona's mining department is also interested in more accurately mapping their facility, including the potential to map within an underground mine. While this use is beyond the technology demonstrated during this project, it has the capability to grow into this industry, potentially providing safer alternatives for tasks dangerous to humans. Goal 2 - tie flights to triggering rainfall events This was successfully accomplished using the SCHOLAR software and a rain gauge mounted to the PERCH. Due to the seasonal limitations and impracticalities of waiting for a suitable rain event in the desert climate of southern Arizona, we simulated a rain event of approximately 0.02 inches, which triggered the SCHOLAR software to command the microquad to initiate a flight. Goal 3 - Assess best resolution of DEMs created from photos This may have been the most challenging of the goals and is still a source of academic research. While the standard metric to compare this type of data is ground based LIDAR, even leading scientists in the field are unsure of its limitations and potential weaknesses, especially when compared to such a new technology. The point cloud density difference between LIDAR and what is possible with the airborne structure from motion photogrammetry (1-3 orders of magnitude more) drives differences in the way the data is analyzed and measurements are taken. Typical LIDAR work flows involve creating a mesh, which removes the finer structures this work is aimed at capturing. Therefore, it becomes challenging to create a compelling comparison of data to validate against. Furthermore, as the survey area increases and the measurements being evaluated are more intricate and complex, it is more difficult to ground truth the results, resulting in LIDAR data that disagrees significantly from the photogrammetry result, but with no way of knowing which is more accurate. Furthermore, the use case proposed in this SBIR is one in which LIDAR surveys are difficult or impossible to gather. This further complicates the comparison and validation of the accuracy of the generated DEM. However, simple change detection, as proposed in the SBIR is a simpler problem to validate, as it is somewhat straightforward to calculate a measurement error and generate a quantitative understanding of noise and signal strength, allowing for more confidence in the resulting answers. Goal 4 - Evaluate if repeated flights can be used to estimate volumes of eroded material This goal was not completed due to the delay in funding to start the project. However, we did conduct extensive analysis evaluating how well our methods matched LIDAR measurements of the volume of an empty pond being studied by the USDA ARS group out of the University of Arizona. During this work, a significant understanding was gained in the methods used and acceptable error. Furthermore, through this work, it has become clear that there is not a great deal of confidence in the accuracy of these measurements and despite the high accuracy of LIDAR as a measurement instrument, the resulting point clouds and meshes used to compute the desired measurements are subject to significant errors due to vegetation, stitching, meshing and other factors. In the end, however, we were able to measure the volume of the same ponds measured by LIDAR and were within 1% of the LIDAR method at best, and 9% at worst. According to the scientists studying this feature, they expect approximately 10% error in their final pond volume calculations based on LIDAR measurements. Therefore, even with worst case estimates, this provides a significant reduction in time and cost to complete a full survey within the expected error, having performed the data collection flight in 15 minutes compared to the 3+ hours it took for the LIDAR measurements to be taken. Goal 5 - Evaluate how effective technology is as vegetation begins to recover and grow in Again, this task was not completed due to the delay in funding. This prevented us from being able to gather data in a freshly burned area and compare to flights after new growth returns. However, in working with the ARS researchers measuring the pond volumes, we did have the opportunity to explore methods in removing vegetation from the point clouds to refine the volume measurements and reduce noise in the processed measurements. Goal 6 - Evaluate feasibility of gathering and processing data from a research and land management agency perspective Given the one-time installation effort and only periodic maintenance of the PERCH and associated µQuads, the feasibility of gathering the data is quite high. The current state of the art in automatic flight path execution is well within the requirements for conducting this type of survey, even over varied and steep terrain. The PERCH can be positioned near any area of interest, but far enough to ease the task of installation and maintenance. Processing the data is a bit more challenging of an effort. This can take place in one of two ways, and the acreage and required resolution of the resulting data would determine the best option to maximize efficacy. The first option is to send the raw image data back to a central computing system, which can perform the necessary data computation and product generation. This requires a relatively robust datalink to transmit the large amount of images the µQuads gather. However, this solution allows for the use of large computing platforms that can process the data extremely fast. The second option is to do the majority of the computations locally on the PERCH. This would then require only the relatively small point cloud data to be transmitted, significantly reducing the need for a high bandwidth datalink. However, it significantly increases the PERCH's complexity, cost, and power requirements. While both of these methods have their own advantages and disadvantages, neither is technically challenging, but could affect the cost and operation of a VIA system significantly.
Publications
|