UAV image processing: Orthorectified image

Processing UAV data

The final project consists of learning how to operate the software program Pix4Dmapper, and learning more information about the processing of UAV data. For this project, a point cloud will be constructed, which can be used later to produce a orthomosaic. The pix4D is a highly developed software program, that allows for the advanced processing of 3D datasets.

Overview of software

               
The program Pix4D gives users the ability to view, procress, and extract information from images taken using an Unmanned Aerial System (UAS). The software works by locating points in an image that are common between the two images. If the two images have a confirmed point that is similar bwtween the two images, are called keypoints. In orerder to create a 3D point, two keypoints much match.  

What is the overlap needed for Pix4D to process imagery?

For this process to be facilitated, a certain level of overlap is required. In the most general sense, Pix4D recommends at least 75% frontal overlap, with no less then 60% side overlap. If the AOI is more complex, like a dense forest, the overlap should increase to about 85% overlap in the front, and 70% on the sides.

What if the user is flying over sand/snow, or uniform fields?

Another environment which needs special consideration is if the AOI is snow, sand. These types of land surfaces have very little distinguishing content, and are very uniform. To produce a quality rendering, Pix4D suggest increasing the overlap to 80% frontal, and 70%, and adjusting the ISO/exposure settings to increase the contrast of the images. Finally, water bodies present a problem for accurate mapping, and can confuse the Pix4D programming. It is suggested that when flying a UAS over a water body, that the image contain some level of land surface to help the program calculate the surface. Pix4D states that oceans are impossible to reconstruct, because the suns refelction and the waves on the water cannot be used for visual matching.

What is rapid check?
This processing extent allows the user to process the data quicker, but the increase in processing speed is comes with a decrease in image quality and accuracy. Rapid check would is useful for quick calculations where accuracy and image quality are not important.

Can Pix4D process multiple flights? What does the pilot need to maintain if so?

 The Pix4D also allows the user to fly multiple flight paths. For this, as much of the information must be kept as constant as possible. Variables like flight height at image capture, general time of day (suns position, shadows), weather conditions (sunny vs cloudy) and contain as much overlap as possible, at minimal 8-% frontal, and 60% side overlap. The imaging software also allows users to process oblique images, but recommends the maximum amount of overlap, with multiple flights, varying the distance above ground with each flight to gain the most amount of information about the terraign.

Are GCPs necessary for Pix4D? When are they highly recommended?

 For the production of a 3D image, many other programs require the prescence of ground control points (GCP) to essentially ‘tie down’ the image to a real ground surface, and dramatically improving the accuracy. In Pix4D, GCP’s are not inherently required to produce ortholog images, but are highly recommended. This cannot be understated how highly GCP’s are recommended. For this project, as an informative guide to learning the Pix4D software, no GCP will be used for the processing of the images.

What is the quality report?

One thing that is very helpful with Pix4D, is the program produces a ‘quality report’, which entails a great deal of information about the images that are going to be processed, and if anything in particular stands out that may produce an error, or inaccuracy in the mosaic image. The quality report informs the user of the details of how the images are processed, and gives a quick preview of what the data may look like.

Methods

               
For the purposes of this lab, the instructor Dr. Hupy provided the class with UAV imagery that was taken from flights performed at a sand mining facility a few miles south of the City of Eau Claire. First, the data is transferred into the students folders. The complimation of images is then imported into the Pix4D mapper by connecting to the folder. Once the images are imported into Pix4, the AOI is specified, the flight path can be visualized on the main Pix4D mapper view. For this project, a AOI is created inside the image by specifying the extent of the processing, and hand drawing a polygon of the desired area. Next, the images can be processed by clicking on the processing tab in the bottom right of the screen. To save on processing time, the initial processing can be ran individually, instead of running the entire process at once. Once the initializing processing has taken place, the quality report is derived from the information about the images. The quality check provides a detailed report about the specifics of the images.

Figure 1: Summery of the initial quality report

Importantly, is the initial summery which gives the basic information about, like the camera used, the sampling distance and project name. The quality check then describes the Images, dataset, camera optimization, matching, and georeferencing.

Figure 2: Preview of the dataset image mosaic
provided by the quality report.

More information about overlap extent, and an image preview is produced, and can inform the user about any problems that may occur in further processing. Figure 3 displays an image provided by the quality report that shows the areas of overlap between the images. The areas of red and yellow have relatively poor overlap, and the area in green is a high level of overlap. The edges of the image are the areas with the least overlap, and may have a poor quality of the image. Its important to keep in mind the area the study is interested in, and make sure the entirety of the area is in high image overlap.

Figure 3: Areas of image overlap

If all the quality check criteria are green, then the user proceeds to the next step of processing, which is production of a point cloud mesh, a raster DSM, and finally an orthomosaic. All of these steps are processed by the computer, by clicking on the processing button, same as before. Once the project is complete, further processing can be done to explore the power of the pix4D mapper.

Results

The final product of an ortholomosaic can be visualized before once the processing has been completed. An orthomosaic can produce many functionally useful products. One of which is the ability to measure the volume of a given object. 

Figure 4: Orthomosaic image

The orthomosaic image is a spatially accurate image, and can be used to calculate volumes of 3D objects. Using a volumetric measuring tool in Pix4D, the volume of one of the mounds is calculated. This powerful tool is a great analysis tool that can add a great deal of quality data to any project.

Figure 5: Volume calculation using Pix4D. Calculating the volume of the
mound highlighted in red. 

A final product of this project, is a video 'fly by" of the orthomosaic image that was made previously. The program in Pix4D facilitates the collection of images that are compiled together to create a video where the viewer gets a tour of the image created in 3D space.

Conclusion

This project offered a quick, and easy way to get an introduction to working with UAV datasets, and the Pix4D mapper program. The Pix4D presents a great deal of complex methods to process and visualize 3D data, and is a dynamic tool that can be used to enhance the quality of a project. Many of the techniques of processing are dynamic, and can be used in many other programs like Esri ArcMap. Although this lab worked through the process of creating a orthomosaic image, Pix4D offers many more capabilities and output processing methods.

Field Survey

Introduction

The project for this week’s field exercise involves taking data points a highly accurate GPS receiver. The goal of the project is to produce a continuous surface that illustrates the elevation change in a small patch of grass on the University of Wisconsin Eau Claire campus mall. To facilitate the collection of data points in the small area, a survey grade GPS is utilized to collect data with accuracy close to a centimeter. Survey GPS offers an incredibly accurate source of data collection that can be utilized to carry out a field survey. Like any advanced technology, there are a few downsides. Survey grade GPs units are often very expensive, and can be rather cumbersome. For this exercise, the entire class had to take turns working in pairs to collect points, because the university only owns one Survey Grade GPS unit. For many, this was not the first time using the Survey GPS. The GPS was also used during the mapping of the Hadleyville Grave yard to gather data on the head stone locations.

Study Area

Figure 1: Blue represents the study Area on the
University of Wisconsin Eau Claire Campus

            The area of interest for this project is a small portion of grass that is located in the middle of the University of Wisconsin Eau Claire campus mall (Figure 1). The grass is surrounded by sidewalk, and contains 3 small trees, and 3 benches. The area is located a decent distance away from any large buildings, so there should be little expected distortion of the GPS signal, especially since the collection will be carried out on a Survey grade GPS. The grass area has a gentle curve in elevation, with the lower portions of the area located on the northern portion, and the higher elevation located in the southern portion.

Methods

            To collect the data, the survey grade GPS unit (receiver and tablet) was deployed to the location on campus. Since there was only one unit to offer, students in pairs of two took turns setting up the receiver in different locations and collecting data. The grass area was sampled using a random sampling method. The collection period was limited to 20 points.

Once data collection was complete, the text file is brought into the computer. The text is then converted to an excel file, along with a changing in the attribute headings for import into Esri's Arc maps. To display the now XY data, the Excel file is changed to table and added to the map using the ‘Display as XY data’ tool. The data point are projected into UTM Zone 15N, so fit the study area.

To complete the project goal, 5 different interpolations were completed to illustrate a continuous surface of the elevation of the grass area. The techniques used are Inverse Distance Weight (IDW), Kriging, Natural Neighbor, Spline, and a TIN surface.

Results

 IDW

Figure 2: IDW Interpolation

The equation for the Inverse Distance Weighted technique takes the cells closely surrounding a sample cell at a greater weight mathematically then cells further away from the sample cell. This method provided a very real to life representation of the slope of the grass area (Figure 2). The high point of the area is located fairly central, in the south east corner. The lowest points are located along the western border.

Kriging

Figure 3: Kriging

The Kriging method uses the height attribute to calculate the continuous surface. The data we collected, the Kriging did a rather poor job of capturing the slope and direction of the grass area (Figure 3). The interpolation over generalized the high point, and the relief in the northwest corner of the surface.

Natural Neighbor

Figure 4: Nearest Neighbor Interpolation

The Nearest Neighbor Technique uses a proportionality to take cells surrounding a sample point to be weighted differently. This technique is quite complex and involved a fair amount of modification (Figure 4). This is probably a reason behind the product being a complex representation of the actual surface of the grass area.

Spline

Figure 5: Spline Interpolation

The technique using spline interpolation uses a mathematical formula that creates a very smooth output by estimating the minimal curve of the elevation points. The spline interpolation did give a very smooth output, but over compensated and added a seeming bump in the slope in the west central side of the surface that is not present in the actual grass area (Figure 5). The spline technique also placed the high point to far in the southeastern corner then it presents in actuality. The technique would improve with further customization and more data points.

TIN

Figure 6: TIN 

This method produced the best result for the illustration of the elevation of the grass area. The TIN method did not capture the relief with complete accuracy (Figure 6). Like the Nearest Neighbor technique, the TIN tool is very programmable and the result could change quite a bit with more investment of manipulating the tool parameters.

Conclusions

The products of the interpolations from the data collected with the survey grade GPS did not accurately reflect the actual surface of the grass area that was surveyed. There could be several different reasons for the lack of strength in the final product. Initially, the sampling method was carried out in a disorganized fashion by students, which resulted with no information being captured in the central western, and north eastern portions of the grass area Even though the sampling method was ‘random’, some care could have been taken to assure that at least some points for every area is collected. Another issue encountered was the data was originally uploaded from the GPS unit text file as UTM ZONE 16N, not UTM 15N (faculty error). The error was rectified quickly and replaced with the correct coordinate system UTM 15N, and placed into a temporary drive to be accessed by students. However when the time when the interpolations were done the Temporary folder was empty. The only file left was the original upload, which was in UTM Zone 16N. This meant that a basemap was unable to be loaded behind the surface to provide context to the results. More organization at all levels could have increased the accuracy of the results.