Reach M+ and DJI Phantom 4 Integration

For 3000 images, with an area of 1425ft x 1500ft, I get these numbers of GCP’s (not including CP’s) and spacings.

RMSe*GSD #GCP Spacing
2,25 rmse*GSD 45 GCP’s 71 m
2 rmse*GSD 60 GCP’s 61 m
1,7 rmse*GSD 90 GCP’s 50 m
1,5 rmse*GSD 120 GCP’s 44 m

Wow, that’s amazing! 90 GCP’s seems about right at 100ft due to the seriously decreased coverage per image. It’s interesting that with PPK it was still more accurate across the board.

excellent. I will have the opportunity to conduct this test in 2-3 weeks. I will put all the data on the cloud


I am using 1.5.9.
Will the initial and final text file with coordinates be enough for you?

You are using too much transverse overlap. if you try 60%, you will see that the accuracy will increase

Nevermind. I found the program install files. Thanks.

1 Like

Why is that?

Because the excessive number of images leads to an increase in noise due to the small angle of the photogrammetric notch.
We fly with 55-60 / 85 % overlaps.


Guys, please stop. That’s not even how photogrammetry works. You are seeing that noise because you are flying too high thus decreasing the angle of attack to the point. While there is a diminishing point of return it is in more of the order of 85 percent flying at a standard altitude. We’re not doing photogrammetry with airplanes or fixed wings flying at 600 feet anymore. I can assure you that if I brought one of the engineers on here it would be pretty obvious that you don’t know how the algorithms work. I’m going to go ahead and forward these messages on to them and should have an engineer’s perspective back early next week.

…and since we are on a public forum please use the terms front and side overlap so that non-technical users can understand what you’re talkin about.


Hi Konstantin,

Please define “Photogrammetric notch”. I’ve not come across the term.

Are your flight overlap numbers based on experience, literature, or both? And could you not decrease your forward overlap to increase your accuracy?


Only scale depends on height. But no nois.
Sorry for my bad English. Sometimes I’m translate with a google :slight_smile: But I’m working on it

Hm, if you only had 2 images with a shallow angle, or many image-pairs with a shallow angle between them, then yes, I can see the point. However, that should only be the instance if only comparing to the previous image? Most modern Photogrammetry applications also do reference-based comparisons.

Having more data-points (more images) with common tie-points should bring more confidence to the mean, not the opposite.

A small angle has a bad effect on the final result. With a decrease in the basis, the matrix of exterior orientation elements becomes degenerate and this introduces errors into angular quantities.

1 Like

We make more then 1500 flight per summer.
A lot of mapping projects.
Last summer I had 1 project 60 sq km and made it by Teodrone for 10 days. It was 5 cm/px.

1 Like

Anton, is this the case where constraints (small error on camera location) are introduced with PPK flights? This should lessen exterior orientation sigmas for PPK flights when compared to the ground control only case.

Konstantin flies with an 85% frontlap. Couldn’t his accuracy be increased with less frontlap then?

I’m glad someone else is understanding the relationship of overlaps with altitude. What I was trying to point out is that as your altitude increases the shallower the angle is. You also have less overlaps in scenarios where you have structures or trees blocking the ground. Obviously we need more overlap as we get closer to the ground because of the reduced field of view. If they’re running 60 at 400 feet then you have to increase by percentage right?

Here’s the response I got back from engineering this morning.

“With today’s photogrammetric engines and AI assisted algorithms noise is greatly reduced by outlier filtering. The initial product of the process are the tie-points sometimes referred to as the ray cloud or ray-tracing cloud. The tie-points are generated from the images with the most matches under a criteria of their proximity to the rule-of-thirds positioning. This creates the maximum amount of angular variation while ensuring the maximum accuracy of focus from a center-weighted or matrix focusing configuration. In any sense of photogrammetic mapping for 3D purposes the side overlaps should be no less than 65 percent and front overlaps no less than 70 percent. Overlaps of 60 percent and 55 percent are bare minimums for stitching a simple yet accurate 2D orthomosaic.”

1 Like

…and a few other industry resources. All of them State at least 60% minimum and other sources recommend 65 so the recommendation to run 60 is pretty shallow.

Flight Altitude – The lower the flight altitude, the better the resolution. If paired with an appropriate number of ground control points as well, this often means better accuracy. But be careful, lower flight altitude, with insufficient ground control points, can actually lead to worse accuracy than a higher flight altitude with the right amount of ground control.

Overlap – For a project to be accurate, it needs to have sufficient overlap. We recommend 75/75 for the vast majority of projects, though there are a few exceptions. Insufficient overlap causes bad accuracy. However, too much overlap can cause problems too. Anything above 85/85 can actually cause matching errors in photogrammetry algorithms and should be avoided where possible.

The recommended minimum overlap should be:
75% frontal and 60% side overlap in general cases.
85% frontal and 70% side overlap for forests, dense vegetation and fields.
85% frontal overlap for single track corridor mapping. Use 60% side overlap if the corridor is acquired using two flight lines.

The biggest culprit we see that makes aerial photos unusable is lack of overlap and it’s often the easiest issue to fix. In general you need a minimum of 70% overlap between photos to ensure proper alignment for processing. This means that for each subsequent photo taken in a survey, at most a third of the features its contains should be new.

What is the minimum overlap I can use for mission planning?
To ensure good quality data, we recommend flying with at least 75% front lap and side lap. If the overlap is insufficient, the outputs are likely to have defects such as blotchy artifacts or errors in the image alignment.

Another aspect of this is redundancy! I have had tried to have the focus set incorrectly for an entire row, or have single images not taken.
I’d rather shoot take a bit too many, than the opposite!

1 Like

Closing comments and I promise I am done with this.

From my own experience of creating 3D models over the last 20 years triangulation was chose because of it efficiency in recreating non-standard formations. A cellular or grid method of stitching can be more accurate, but the cell size has to be much smaller to achieve abnormal more particularly round shapes which leads to an increased vertex (point) count that is not efficient and too burdensome for most modern hardware in these applications. Using the rule of thirds in photography makes sense as does triangulation in geometry. Therefore 9 matches per tie-point is optimum for triangulation to each image. Due to the aspect ratio of the images used the sidelap can be less, but 65% in combination with the 75% front overlaps ensures that this will produce at least those 9 matches and in a perfect run as many as 12, but is far from the point of diminishing returns. Great info all!