Gathering Data for Processing and Generation of a 3D Model Using Pix4D Software

Data Collection:

The purpose of this is to compare different data acquisition methods. There were a total of 7 crews flying a different pattern at different altitudes over the same shed building. The goal was to compare the different data collection methods and see which resulted in the best 3D model.

I was a member of crew #7, along with my partners: Cameron Dine & Nick Phillips. We were tasked with flying the shed at the following parameters:

Using Pix4D flight planner at 40m NADIR and at 40m with a lens angle of 60°

For planning the mission

Launch Pix4D Capture
Select Your Drone
Select a Mission from the Plan New Mission Page (Figure 15)
Move, Resize, and Rotate the Mission so it Covers the Intended Area
Set Your Altitude
Customize Any Other Desired Mission Settings
Save the Mission for Future Use

Figure 15: Shows Pix4D New Mission Page

Field Notes:

When we originally set out to fly this mission, the LAANC was down, and we were unable to get authorization and had to postpone our operation. We went back out and collected the data on 10/13/21. Nick Phillips provided a Mavic 2 Pro for this operation and he would be acting as Pilot in Command (PIC).

On-Site Conditions:

Heavy cloud cover
Potential rain in the forecast but we had an hour window for the operation
Wet ground conditions - We brought proper landing pad to counter
Light wind out of the East

Hazards in the Area:

Light poles
Few birds
Potential rain
Civilians running
A lot of metal equipment that could cause interference

We were the only flight crew present at the time so we didn't have to worry about other UAVs in the area. The shed is located in a grassy sporting practice field (Figure 16).

Figure 16: Shed Where Operation Was Conducted

Flight 1:

PIC: Nick Phillips
VO: Hunter Donaldson
Flight 1 was a grid flight at 40m NADIR
Takeoff: 4:37pm
Landing: 4:40pm
Aircraft: DJI Mavic 2 Pro

Flight 2:

PIC: Nick Phillips
VO: Hunter Donaldson
Flight 2 was a grid flight at 40m with a lens angle of 60°
Takeoff: 4:41pm
Landing: 4:43pm
Aircraft: DJI Mavic 2 Pro

* After we finished both flight operations, we checked the SD card to ensure all the data was collected correctly. We noticed that our data did not fully capture the shed like we wanted it to. Upon further investigation, we realized that the overlap was not properly set. We corrected the overlap and set it to 80% to ensure all the data would be collected properly. We re-flew both missions for better data collection. Therefore, flight 3 is a repeat of flight 1, and flight 4 is a repeat of flight 2. *

Flight 3:

PIC: Nick Phillips
VO: Hunter Donaldson
Flight 3 was a grid flight at 40m NADIR
Takeoff: 4:50pm
Landing: 4:54pm
Aircraft: DJI Mavic 2 Pro

Flight 4:

PIC: Nick Phillips
VO: Hunter Donaldson
Flight 4 was a grid flight at 40m with a lens angle of 60°
Takeoff: 4:55pm
Landing: 4:59pm
Aircraft: DJI Mavic 2 Pro

Processing the Data:

The first thing you want to do is upload your imagery data to Pix4D mapper. You do this by selecting all the images you want and then uploading them to the software (Figure 17).
The next step is to set the image properties. This includes the coordinate system, geolocation options, and camera model/settings (Figure 18).
After that, you'll want to select the output coordinate system you want to use. A coordinate system allows for the measurement of distances and to determine direction. Except for the geographic grid, all coordinate systems are based on some form of map projection (Figure 19). (See GCP post for more information)
The final step is to select the type of product you'd like Pix4D to create. In this case, it's a 3D model (Figure 20).

Figure 17: Uploading Images To Pix4D Mapper

Figure 18: Image Properties Page

Figure 19: Coordinate System Page

Figure 20: Selecting The Type Of Project

We repeated these steps for both the data sets: One flown NADIR and one with 60° lens angle. The results for the 3D models are posted below.

40m NADIR 3D Model Results:

Figure 21: Top-Down View

Figure 22: Front View

Figure 23: Side View

Figure 24: Back View

40m With 60° Camera Angle 3D Model Results:

Figure 25: Top-Down View

Figure 26: Front View

Figure 27: Side View

Figure 28: Back View

Conclusion:

After the 3D models finished processing, the results didn't look very good. I think that is largely contributed to the data acquisition methods our crew was assigned. Having your only data collection be a NADIR and 60° camera angle at 40m, won’t allow for a proper 3D model. The front view of the NADIR model is mushed together on the top and right side of the shed (Figure 22). This happened because the drone was flying directly overhead looking straight down. It couldn’t collect data to generate underneath the overhang of the roof. This caused that distortion at the top of the shed at every angle. However, the top-down view of the shed looks good in the NADIR view (Figure 21). As expected, that is the only aspect of the NADIR model that turned out well. This can be explained because that is the only area that data was accurately collected.

On the other hand, the 3D model that was generated with the flight at 40m with a camera angle of 60° didn't turn out much better either. Due to the improper oblique imagery, the software was not able to accurately compute the shed. The model depicts the shed as an anvil-type shape due to the distortion that was caused by the 60° angle the camera was set to, best visible on the side view (Figure 27). Furthermore, the front view (Figure 26) is completely distorted and doesn't resemble the actual shape at all, this can also be contributed to the improper oblique imagery.

Ultimately, I think that the best data collection method for this type of project would be to combine the operations we did with circular orbits above, at the top level, and base level around the shed. This way, we would be able to get good quality on the top view, along with the details around the outside. This would be the most accurate data acquisition method.

Search This Blog

Hunter Donaldson UAS