First look at the Phantom 4 RTK

I had the privilege yesterday of helping Jonathan from Ferntech to un-box the new Phantom 4 RTK from DJI. This is a summary of the experience.

First Impression

It is the same as the Phantom 4 Pro that most surveyors now are familiar with, it just has its little top hat.

The big difference is the controller, where there is now a hidden USB slot which comes with a vodem stick for connecting to the internet. All that you need to supply is the SIM card and a data plan. There are also slots for a memory card and comes with the built-in flight screen. This drone must be used with this controller to take advantage of the GPS functionality.

The DJI base station had still not arrived, but we had a few options for an NTRIP connection.

Take Flight

For the first flight we connected to the free LINZ RTK service. A simple flight plan was made where the front overlap was set to 80% and the side overlap 60%. The camera was set to shutter priority mode with Auto ISO and a shutter speed of 1/800. The "Distortion Correction" was disabled, which is normally enabled by default with other DJI drones. There were a couple of times where the flight planning software froze up while planning the mission, but after the mission was saved, this was not a problem as a restart of the device was quick. The drone took off and flew the mission as expected and returned to land. All seemed to go well.

The second flight we connected to Synergy's Skynet-RTK VRS service. We flew the same mission with the same overlaps and camera settings. This time though, we put the controller behind a building during the middle of the mission. This meant that the drone stopped getting the RTK corrections for each image from the controller. The drone continued to fly its mission and take photos.

The third and final flight we set it off without RTK corrections, which means it logs PPK.

Processing

Using Pix4D version 4.3.27, we started a new project and added the images from the first flight. Immediately, the columns for horizontal and vertical accuracy of the images were populated, each photo varied slightly between 10-30mm as expected of an RTK position.

The elevation of the images was set to 0.000 above the ellipsoid, and the output coordinate system was changed to match our site. The camera calibration was set as per the parameters in in the "xmp" metadata which is contained in each image, and is constant.

The ground control points were imported as check points, and immediately we saw that the GCPs were within about 20mm horizontally. The verticals were a bit further out because our geoid model is not in Pix4D, but that was remedied by holding one GCP fixed, and leaving the rest as check points. See for yourself the results that we got against the check points:

Then, processing the second flight which had the loss of RTK corrections, the results were similar. This is thanks to the EXIF data of each photo identifying the accuracy of each photo, so Pix4D knows which photos to hold fixed.

The PPK photo folder comes with 4 files: a rinex file, a nav file in "bin" format, a time of exposure or event log in "bin" format and the same in ASCII format. I have yet to figure out how to combine these files to get the accurate event locations because it does not come with software for processing these files.

Summary

On the face of it, this process seems very easy, but this is were I must say that a qualified surveyor is best person to use the tool. A surveyor will have a full understanding of the process and have the confidence to assure the accuracy achieved.

This test took a few hours and there needs to be more time spent on this, but the Phantom 4 RTK is everything I expected. This drone system will take drone surveys to the next level and is at a very competitive price. I'd suggest you still need at least one or two GCPs, but they can be near the ground control station, therefore eliminating the need to enter the site to be surveyed. Together with the latest version of Pix4D, the processing is very fast compared with PPK where some time is spent geo-referencing the images.

What's next???