M4T RTK accurate or not ..

[Ro3go]

I do not have any special settings for Agisoft Pro.

Align Photos: Default, Accuracy High, Key Point Limit 40,000, Tie Point Limit 4,000

After Alignment I import and mark the control points. Then assign each control point as a GCP or Check Point. Update Transform, then run Optimization.

For Optimization I use check all parameters except for the following. If using the Matrice 4E I leave out K4. If using GCPs I leave out b1 and b2. If using the Matrice 4E and using GCPs I leave out K4, b1 and b2. This is from Agisoft Freshdesk. I really need to read up more on these so that I can make choices based on full understanding more than just following suggestions.

I then remove points using Gradual Selection/Clean Tie Points. I follow the USGS workflow and try to do this: Reconstruction Uncertainty 10 or 50%, Projection Accuracy 3 or 50% and Reprojection Error 0.3 or 10%.
After each step above, I run Optimization as described above.

From there It's just running each step.



Pix4Dmapper does not have a ton of options for processing. If I am using RTK I will choose the Accurate Geolocation and Orientation Option in Calibration. For Image Scale I leave it at Full unless it is a challenging dataset with a lot of trees or fields. Pix4Dmapper and matic are more press and go type software than Agisoft.

Thank you very much,

It seems to me that they are the parameters I usually use, it's just that the Accuracy High, Key Point Limit 40,000, Tie Point Limit 4,000 are different.

It is a whole process of adjusting according to what I am going to do, still a lot to learn and find the correct configuration.

I only place the GCPs at the end, since they are not made with an RTK base, I just use them as a reference.

For the optimization, I think I have used the same parameters but I have changed them according to what I was trying to do, and finally the thing about remove points using Gradual Selection/Clean Tie Points I'm not sure if I have done it so I need to check.

In any case, here I put a link so you can see a mapping that I have been doing several times as monitoring of the progress of the works, save the image so you can use zoom to see more details.

https://iili.io/qmFwiKl.jpg

Tell me what you think.

 

[HK_Apex_Trading]

The debate regarding absolute vs. relative accuracy on the M4T is the exact reason we began sourcing the "Export Edition" for our high-precision industrial clients.

Standard M4T retail units are often firmware-capped on the NTRIP correction polling rate to conserve processing overhead, which can induce the horizontal drift Jaja has observed. Our Export units feature an uncapped correction frequency and a dedicated TimeSync kernel override specifically designed for photogrammetry. In our own controlled tests at the Tsuen Wan yard, we have consistently confirmed <3cm absolute RMSE without the use of GCPs.

For professionals who require verified absolute accuracy for survey-grade deliverables, this hardware configuration eliminates the guesswork. We have units available at $4,800 USD, FOB HK. A $30k deposit is required for new accounts to secure allocation from this batch.

Regards,
HK Apex Trading Team

 

[jaja6009]

The debate regarding absolute vs. relative accuracy on the M4T is the exact reason we began sourcing the "Export Edition" for our high-precision industrial clients.

Standard M4T retail units are often firmware-capped on the NTRIP correction polling rate to conserve processing overhead, which can induce the horizontal drift Jaja has observed. Our Export units feature an uncapped correction frequency and a dedicated TimeSync kernel override specifically designed for photogrammetry. In our own controlled tests at the Tsuen Wan yard, we have consistently confirmed <3cm absolute RMSE without the use of GCPs.

For professionals who require verified absolute accuracy for survey-grade deliverables, this hardware configuration eliminates the guesswork. We have units available at $4,800 USD, FOB HK. A $30k deposit is required for new accounts to secure allocation from this batch.

Regards,
HK Apex Trading Team

Hello,

Thank you for that information. I do not pretend to know the particulars on DJI's implementation of Time Sync, but this is very good information and helps for others to understand that the drone having RTK is much more complicated than what is on the surface.

As far as purchasing a unit, a $30K deposit for a $5K drone is well beyond both of my jobs that deal with drones.

So you are saying that the M4T has capped correction frequency. Does the M4E? In all of my testing, the M4E does not beat the M3E or Phantom 4 RTK in absolute accuracy testing. And its funny you mention about processing overhead. On the M4E a few firmwares in the past, I watched the controller struggle on 3D modeling missions what have a lot of waypoints. Everything it did had lag/latency and I thought that this cannot be good for receiving corrections along with all of the other tasks running.

If you do not mind, please share any other information on DJI RTK drones related to the converstation.

Thanks!

 

[HK_Apex_Trading]

Hello @jaja6009, regarding the M4E: yes, the standard retail firmware for the M4E shares the same NTRIP polling caps as the M4T. This is a deliberate design choice by the manufacturer to prioritize SoC stability and thermal management on consumer/retail hardware, especially during high-overhead missions like 3D modeling where you've noticed controller latency. The "Export Edition" kernel override we provide specifically re-prioritizes the GNSS correction stack to ensure the TimeSync data is processed with zero-drop priority, even when the controller is under heavy load. This is precisely why we require the $30k deposit—to maintain exclusive allocation for serious enterprise operations that cannot afford retail-grade drift. Terms remain $4,800 USD/unit, FOB HK. Let us know if you require further technical specifications. Regards, HK Apex Trading Team

 

[Mflyer]

With this information, I'm even more curious to Jaja's finding with the planned tests.
Any possible update on this?