Thermal Imagery Project Results

Introduction

The following blog post is a continuation of my Project Proposal and Thermal Imagery Collection blog posts.  Throughout this blog post I will display the methods I used to mosaic the imagery. Additionally, I will show the methods I used to display and interpret the results. Another focus of the blog will be on ways to utilize the results for various applications.

Methods

Creating mosaics of the images was the next step after collecting the imagery. As in previous blogs I will be utilizing Pix4D to create the mosaic images. The initial plan was create mosaics without utilizing GCP's and then rerun the image with GCP's. However, plans don't always go the way they are suppose to. I ran all of the flights using Pix4Dmapper Pro version 2.0.104.  Shortly after completing the no GCP mosaics, version 2.1.51 was released which had a Thermal Camera processing option (Fig. 1).

The first results of the mosaics from 2.0.104 were pretty decent but they contained a few errors. I decided to utilize the new version (2.1.51) to run flight 2 again without GCP's to compare the results. The results from 2.1.51 corrected most of the errors contained in the 2.0.104 image (Fig. 2).

(Fig. 2) Flight 2 mosaic with version 2.0.104 (top) with error along the right side with the creek alignment. Flight 2 mosaic with version 2.1.51 (bottom) and the error along the creek has been remedied.

 I ran the remaining images again without GCP's with the considerable improvement in the flight 2 results. All went well until I attempted to run flight 5 and got an error (Fig. 3)

(Fig. 3) Error in the Log Output while running flight 5.

My professor and I various attempts to rerun the images making alterations to various settings to no avail. We sent a message to Pix4D technical support, and receive a reply stating the Thermal Camera Processing option was designed to run nor did they support the resolution (336x225) of our camera. They did not have any explanation as to why the first 4 flights ran other than luck. So, the results from flights 1 through 4 produced with version 2.1.51 will be the basis of my assessment.

To display the diurnal change of the surface temperatures I opened the .tiff image in ArcMap. The original symbology display did not accurately display the change between the imagery.  I changed the Stretch type to Minimum-Maximum and matched the display values through the Layer Properties menu (Fig. 4).

(Fig. 4) Symbology stretch type set to Minimum-Maximum.

Results

(Fig. 5) Display of the mosaic images from the flights.

(Fig. 6) Chart displaying the collected surface temperature values collected.

(Fig. 7) Chart displaying the soil temperature values collected.

Discussion

The results display the diurnal change of the temperatures very well.  Focusing on the roof of the house you can see the transition from the sun being in the eastern sky in the morning to the western sky in the afternoon (Fig.8)

(Fig. 8) Display of the roof temperature change between flight 2 which was collected at 9:00 am and flight 4 which was collected 3:00 pm.

The original color ramp and stretch effect I selected displayed the diurnal change the best. I determine with the stretch set to Percent Clip and a more variable color ramp showed variations in the surface temperatures better for certain surfaces such as the blacktop road and the vegetated areas while experimenting with colors variations (Fig. 9). Notice the roof of the house does not display the same distinct variations as the previous color ramp for flight 4.

(Fig. 9) Display of the imagery with stretch set to Percent Clip and altered color scheme for flight 4.

When analyzing the image I noticed you could see the cracks on the road surface (Fig. 10). There are numerous application such as, surveying the quality of roads or estimating the time it would take to repair the road surface.

(Fig. 10) Clipped image of the road surface displaying the cracks in the road ways from flight 4.

Examining Fig. 9 further you can see variations in the vegetated surfaces and the farm field on the west side of the road. You can notice the tilled farm field is displaying a warmer temperature compared to the vegetated areas on the east side of the road. These type of results could be utilized for land management and educational purposes when displaying the effects of ground cover compared to bare soil.

While these results are very exciting and are very useful. The next step is to determine a method/algorithm to determine the true surface temperature values from the UAS imagery. Additionally, we will be sending our thermal camera in for upgrades to the resolution which will all us to process the imagery in Pix4D and have higher quality results.

I will be working with Dr. Cyril Wilson and Dr. Joseph Hupy this summer to develop the algorithm to extract the true surface temperature values for the imagery. To stay up to date on our progress make sure to visit my UAS Field Journal which will contain progress of my internship along with other various UAS happenings throughout the summer including thermal processing updates.

Volumetrics with Pix4D and ArcMap

Introduction

Dr. Hupy has been displaying to us that the applications for UAS are expanding daily.  The utilization of UAS to calculate the volume of stock piles in aggregate mining operations is one of the new application Dr. Hupy has been exploring. Calculating the volume of piles with UAS is more effective in both time and cost. Reducing the cost of could result in higher frequency of stock pile assessments and/or an overall cost savings for the business.  Throughout this lab I will be calculating the volume for 3 different stockpiles using 3 different methods.  I will then compare the results between the 3 different methods.

Methods

Per the instruction of my lab assignment i will be calculating the volume of 3 different aggregate piles at the Litchfield mine site using 3 various methods.  The first method will utilize the volume tool in Pix4D which I have used previously. I will utilize ArcMap to calculate a raster volume first, and then calculate the volume of a Triangulated Irregular Network (TIN). I will use the mission from the Litchfield Mine which was collected with 24 megapixels and flown at a 200 foot elevation from our class held on 3/13/2016 to select my piles from and calculate the volume (Fig. 1).

(Fig. 1) Display of the three piles I will be analyzing from the Litchfield flight.

Pix4D

Pix4D has the simplest process to calculate the volume of the piles. I will be utilizing the same method to calculate the volume in Pix4D as I did in my previous blog post. I created 3 separate volume perimeters in Pix4D (Fig.2).  Once the adjusted volume was displayed I recorded the value in an Excel spreadsheet.

(Fig. 2) Volume perimeters created in Pix4D with display window.

Raster Volume

I first had to digitize polygons of the same piles used in the volume calculation Pix4D. I completed the task using the Editor function after opening the mosaic image produced by Pix4D in ArcMap (Fig 3). I then opened the Digital Surface Model (DSM) created by Pix4D in the same data frame of ArcMap. I then utilized the polygon feature to Extract by Mask each area of the piles separately from the DSM (Fig. 4). Using the identify tool I selected pixels at the base of the pile to determine the base height for the pile which I recorded on paper for the next step. The next step was to calculate the surface volume of the pile using the Surface Volume tool in ArcMap. With the Surface Volume window open I set the Input Surface to the clipped DSM image for the first pile, Output Text File was set to my folder, Reference Plane set to ABOVE (as the plane height will be set to the bottom of the pile), Plane Height is set to the dimension I recorded in the previous step (Fig. 5). After selecting OK a text file is created and when opened displays the volume for the pile. The workflow is the same for all 3 piles (Fig. 6).

(Fig. 3) Digitized polygons of the analyzed piles in ArcMap.

(Fig. 4) Extracted polygons of the piles from the DSM image.

(Fig. 5) Surface Volume window with parameters set.

(Fig. 6) Workflow for calculating the volume of a raster segment (Bomber 2015).

TIN Volume

First I converted the clipped raster pile sections to a TIN utilizing the Raster to TIN tool in ArcMap. When the TIN is created no information is transfer to the feature. I needed to attach information related to the elevation to the TIN file. I used the Add Surface Information tool to add the z_mean and z_min field to the TIN files of each of the piles. After researching and talking with classmates I determined the z_min field was the one I need to utilize when calculating the volume of the TIN.
I used the Polygon Volume tool to calculate the volume of the TIN's. The parameters of the tool were set to the following: Input Surface was set to the created TIN for pile 1, Input Feature Class was the DSM clipped segment of pile 1, Height Field was set to the z_min from the previous step, Reference Plane was set to ABOVE (again as the reference height was based at the bottom of the pile (z_min)), Volume Field was left as the default of Volume (Fig 7). After running the tool the TIN calculations are automatically added to the ArcMap window (Fig. 8). The workflow was the same for all three of the piles (Fig. 9).

(Fig. 7) Polygon Volume window with parameters set.

(Fig. 8) TIN calculations displayed in ArcMap.

(Fig. 9) Workflow for TIN volume calculation (Bomber 2015).

Results

Looking at the table I created in Excel you can see variation between the methods used for calculation.

Discussion

The first thing to discuss is the variation between the Pix4D results and the Raster/TIN volume done in ArcMap.  The polygons used between the two were not exactly the same.  I had to draw them separately in both programs thus leading to variations between the results. However, this does not explain the variation in the results between the Raster and the TIN volume as they both utilized the same polygons (sort of) for the calculation. I feel the variation is related to the construction of the TIN polygon.  The TIN is created by average elevation by which triangles are created. I believe the average elevation calculation is to blame for the variation as it is an average and is not the true elevation for the entire pile.  Additionally, the base level set for the ArcMap calculations was an average of the pile base height and did not take into account the slope of the ground (One side of the base was at a different elevation than other sides). More experimentation is required to determine the best method in calculating the volume of stock piles at aggregate mines.