Understanding Carlson Photo Capture’s new Project Report

By Steve Cummings • May 22nd, 2019

Y’all know that guy. Heck – you might be that guy – the one who, at closing time, reminds the band that they haven’t played “Freebird”1 yet, and really need to, before laying down the sticks and axes for the evening…

Only to be told “Maybe next time.”

Well, here at Carlson, we do take requests. We listen, track, rank, and implement them. Like this one:

While Carlson Photo Capture has always had a report called “Bundle Adjustment” showing the accuracy of data points (pixels) for every data set, it is sometimes difficult to read and understand.

Customers wanted something better, spoke to their Account Reps and Tech Support, and provided some input on what they were looking for. The result is the “Project Quality Report.” You’ll find it at the bottom of your “Output Products” page of every new job you process in PDF format.

The first information grouping is Project Overview:

You see your Project Name, time and date, how many images you uploaded, the number of Ground Control Points (GCPS) you shot and uploaded. Average GSD stands for Ground Sampling Distance, which is the distance between pixel centers on the ground, or what area on the ground each pixel on your camera’s sensor represents. The first NASA sensors, Landsats, had a GSD of 98 feet in the visible band, here the GSD is .0354 ft – about 7/16 of an inch.

Orbital Dataset means “Did you fly in a circle around a structure?” True or False (you set this in Advanced Fields, see below) and, finally, the area the images cover – in this case 891901/43560 = 20.5 acres.

The second information grouping, Coordinate System, lists the projection, earth model, and output units used for your Output Products.

The third grouping, Accuracy Check, details the accuracy of your project and provides a visual symbol and color for a quick look. Green checkmarks are deemed “Good”, Yellow caution triangles indicate possible data problems while Red caution triangles indicate that there may be a severe problem.

Here we see that several images were not able to be used – 4 of 53 could not be included in the dataset output. They might be takeoff or landing shots, taken way-over-there shots, etc. The residual, or error, is less than half a pixel. To understand this, imagine you are just one pixel in one image on your camera’s sensor, one of 17 million. From there you drop a ball down into the math of the optimization algorithms. After the best placement of you is optimized from all the images, the ball bounces back to the sensor. How far off is it from where you dropped it? In this case, 0.466 pixels. Average Camera Position Error and Standard Deviation shows the location (in)accuracy of your drone’s GPS. The Average GCP Position Error and the GCP Error RMSE above shows poor positioning with large scatter. The error here was caused by using a projection for the GCP input different from the output projection selected for the point cloud and other products. Here’s one that looks much better:

Note: The units of the report are the same as the Output Projection units.

The next information grouping, Camera Parameters, is for those who understand cameras much better than I – this is a check of the camera’s quality in much the same way pixels and GCPS are checked. Some camera software allows the user to calibrate their camera with these parameters.

If you are going to fly a drone with RTK/PPK geolocation, you can further improve your accuracy by inputting the Adjusted Focal Length you got from a previous flight into the highlighted box in Carlson Photo Capture’s Project creation Advanced Fields dialog along with the RTK’s accuracy from the manufacturer’s specs, also highlighted.

The next two pages are the Orthoimage and the Digital Elevation Model snapshots for quick reference.

If you have included GCPS in your project, you next see the Ground Control Error Review, a closeup image of each GCP with its error and a Green, Yellow, or Red symbol indicating our opinion of the GCP. A bad shot might cause the error, as might mixing projections or having a lot of GCPS over here and almost none over there.

Note that this is the same information you see in the Sparse Point Cloud in CPC, but the units in the report are what you want while the units in the Sparse Point Cloud are always in meters.

Next in line is the Image Connectivity Graph. This shows where every image was taken. Image locations with Green dots are highly correlated with images around them: Yellow not as highly matched, Red means difficulty matching pixels to other images. A line between two images means that they share correlated pixels, the darker the line, the more correlations exist between the two images.

The last information group is Dense Point Cloud. The Total 3D points is important especially to Photo Capture customers who don’t use Precision 3D or Carlson Point Cloud and have to worry about their software crashing while loading just a few million points. Point Density is another way of looking at the GSD. Minimum Stereo Pairs Per Point now defaults to 2, but you can adjust it to 3 or more during Project creation in the Advanced Fields dialog shown above. The image resolution is a value that, as far as I can tell, matters only to photogrammetry programmers and maybe statistics professors.

To learn more about Carlson Photo Capture and score a free 5 GB demo in lieu of listening to “that dead band’s song”2 one more time, follow this link:


1 Freebird – Lynyrd Skynyrd

2 Bad Luck Streak in Dancing School – Warren Zevon

« Previous | Home | Next »