# Variances in relative difference

My average relative difference is 4.7% .  Recently, I have been seeing relative differences above 5% at a higher rate than usual, and the higher relative differences seem to be intermittent.  Below is a summary of 7 flights in very similar conditions.  The first two flights are in the same area but with slightly different mission plans (one mission plan was offset from the other mission plan by about 30 meters).  The next five flights are all the same mission plan in the same spot. All seven flights were flown with the same aircraft, same camera, and no change to camera settings.

26 June 2018

|

Flight number

|

Relative difference

|

Time of day

|
|

1

|

19%

|

12:39 pm EDT

|
|

2

|

5.2%

|

1:38 pm EDT

|

27 June 2018

|

Flight number

|

Relative difference

|

Time of day

|
|

1

|

0.36%

|

11:11 am EDT

|
|

2

|

0.81%

|

11:50 am EDT

|
|

3

|

0.74%

|

12:15 pm EDT

|
|

4

|

7.13%

|

12:59 pm EDT

|
|

5

|

17.38%

|

1:30 pm EDT

|

My questions:

• Do you have any ideas as to why I am seeing such a variation in relative difference?
• What does relative difference impact?
• When looking at the orthomosaic, it does not appear that relative difference has a big impact. What is the down side to a high relative difference?
• Does a high relative difference impact the accuracy of volume calculation?

Hi Doug,

thank you for sharing the results of your tests.

First of all, when using the same camera and processing the same area, the differences are not expected to be so big.

• Which drone are you using?
• Did you by any chance change the speed of the drone between missions?
• Did you always use the same processing option in the software? The best way to do this kind of test is to always load the same processing template, e.g. 3D Maps.
• Could you share the quality reports with us? For now, the quality report of project 1 and 5 would be enough (27 June 2018).

In the end, if the difference between the initial and optimized values is big the scale of the project might be wrong. I would recommend adding GCPs or at least scale constraints in the project in order to ensure that the reconstruction is optimal for measurements. The quickest way to check the scale of the projects is to measure a distance on the field and compare it with the distance from the reconstructed project.

Best,

Hey there,

1. I fly a custom- built hexcopter with a Pixhawk autopilot. The camera images are geotagged after landing via the flight software.

2. I maintained the same speed in every flight, 6 m/s.

3. I always perform the initial processing using “3D Maps- Rapid/Low Res.”

4. Sure. Here are snips from Flights 1 and 5 from 27 June 2018. NOTE: The Relative Difference for Flight 5 is different. Since that day, I have been changing the camera settings and optimization parameters and reprocessing. Unfortunately, the original RD of 17.38% was overwritten by mistake (to 21.62%).

However, all imagery and processing was the same as Flight 1 that day.

Thanks,

To answer your question I would need more insights about your project. Could you please send, for both projects 1 and 5, the full quality reports and the log files via our request form:

• The quality report (.pdf format): …\project_name\1_initial\report\project_name_report.pdf
• The project log file (text file): …\project_name\project_name.log

Hi Doug,

I had a look at the quality reports and noticed that the flights are done over a featureless area (grass) with small elevation changes.

This kind of areas are in general challenging for the reconstruction, especially for calculating internal camera parameters. And in your case, the software estimated a wrong focal distance for some flights.

I anticipate that you will not face the same issue if you take the following recommendations for image acquisition into account:

• Enough images (50 - 150 images).
• Non-planar scene (not all objects having the same height).
• High overlap and rich texture.

For now, you can try to fix the initial camera parameters. You can do this in the setting Step 1. Initial Processing (Advanced) -> Calibration -> Camera Optimization -> Internal Parameters Optimization ->  All Prior. This way the software will force the camera optimization to stay close to the initial values.