Background
Drones Made Easy has a special interest in how cameras place geotags into their images. Accurate geotags allow users to build a map that is created from individual aerial images that aligns with the real world. The process of aligning an image to represent space in the real world is called georeferencing.

Geotags are (or should be) the GPS coordinates and altitude of the camera at the exact time when an image is taken. These coordinates are written to the EXIF metadata that is embedded in a JPG image file. Other EXIF tags include exposure time, focus method and make/model of the camera. In general, the majority of cameras are not equipped with a GPS receiver, but almost all cameras write EXIF data embedded in the images they create.

Most of our mapping customers either use cameras that record GPS data (such as the Canon S100) or a GoPro which does not record the GPS data. For the GoPro, creating a georeferenced photo map requires an additional process of associating points in the imagery with points on a basemap or manually entered GPS coordinates. The results from doing this extra step are very accurate, but are sometimes unnecessary if super high accuracy is not required and sufficient geotags exist. For our Maps Made Easy service, this process is fully handled in the user's web browser.

The easiest way to create a georeferenced orthophoto map is to record the GPS data along with the pictures and use that information when stitching everything back together. Given good geotags, our system can give pretty well aligned results (within a few meters) without any extra steps or inputs. For a lot of applications, this is good enough.


Comparisons
Being a DJI dealer, we have a lot of experience using the various DJI multirotors to carry different camera systems for mapping.

The Phantom 2 has an internal GPS and good photogrammetry tools build into the PC Ground Station software but no integrated camera, so without using a datalogger and using the log file to insert the appropriate data into the images, Phantom 2 users were required to manually georeference their images if they were using a camera that could do so on its own. Our Deluxe Mapping Bundle was designed for users that required enough precision that they would be manually georeferencing their images anyway.

When the Vision+ came out, we got excited that its Ground Station software and geotagging camera would provide a great all-in-one platform to collect good mapping data with minimal effort or investment. While it is certainly easy to use, the Vision+ is held back from being a great mapping platform by its fisheye lens and the fact that it doesn't record the altitude (?) made the recording of the GPS coordinates useless for mapping. Not to mention the fact that the GPS coordinates are not recorded with very much precision. The coordinates are recorded as Degrees, Minutes, Seconds with no decimal precision on the Seconds portion. This limits you to 10s of meters of accuracy. The outputs of the Vision+ still require manual georeferencing and quality was limited due to the optics.

The Inspire 1 came out with its flattened lens for the camera and we thought it might be the platform we were looking for. Its imagery turned out to be some of the best mapping imagery we had processed but even though it recorded its altitude accurately in the geotags, it was still limited by the fact that it recorded the GPS coordinate data in the same manner as the Vision+. We don't know why DJI was limiting the accuracy of this measurement, but it was getting pretty annoying.

Finally, the Phantom 3 was released and it turns out it has everything we were looking for. It has a similar distortion-free lens to the Inspire 1, records the altitude information and actually records decimal precision GPS location in the geotag! Hallelujah! We got one of the first batches of the new Phantom 3 and immediately took it out into the field for a couple of test mapping missions. The results have been great.

We think the Phantom 3 is a winning combination of ease of use, optical quality, GPS accuracy and battery life that will make creating quality maps quick and easy.

Here is an example of a Phantom being used to map a 6.5 acre area:
https://www.mapsmadeeasy.com/map ... 4c93580c3acbae9ee9/

Future Development
Drones Made Easy is planning on releasing an iOS application in the coming months that will make the process of collecting quality data with a Phantom 3 as easy as it can possibly get. After collecting this data, the application will be capable of uploading the imagery directly to the Maps Made Easy map processing service for 3D model and orthophoto map processing.

Glad to hear this!  My first attempts at geotagging with the P2V+ were disappointing for exactly the reasons you described "The outputs of the Vision+ still require manual georeferencing and quality was limited due to the optics"

My P3 is supposed to arrive in 2 days and I'm looking forward to seeing if the geotags have improved!

I'm impressed with the mosaic you guys put together on this test.  

The foliage held together very well, in my work I tend to end up with lots of swirls when building foliage

Are you using photoscan on the back end then pushing it through to your mapsmadeeasy service?

If you are using photoscan, What altitude and overlap did you use for this test?  Any specific settings you can share for dealing with the dreaded swirls and artifacts when building foliage?

Is there a limitation in MapsMadeEasy that keeps an end user from exporting the raster to something like Global Mapper to clean up the edges then import that raster back into your online viewer?

Thanks

gil Posted at 2015-5-5 13:33
Glad to hear this!  My first attempts at geotagging with the P2V+ were disappointing for exactly the ...

The Vision+ has too much fisheye to ever give really great results.  We have some customers who have done some pretty awesome work, but I always feel like it is a bit of a struggle.  Even when you do the manual georeferencing, you are still going to get those swirls due to the optics unless you use the narrowest field of view and have plenty of overlap.  Our system gives all sorts of warnings about using the Vision+ images as they are.

Hooks Posted at 2015-5-5 14:48
I'm impressed with the mosaic you guys put together on this test.  

The foliage held together very  ...

Thanks Hooks.  This test was done with the Phantom 3.  That is why there aren't any swirls!  It is much better than the Vision+ from that standpoint.

We use our own fully automated processing techniques that were developed over 15 years of doing government and commercial mapping, 3D reconstruction and image processing work.  The requirements and recommendations we suggest for the imagery that gets uploaded allows us to process the imagery efficiently and quickly.  Images that don't conform do not get included in the outputs.

The automated nature of the service is what allows the jobs to be completed and shared within hours, not days.

This test was flown at 50m with 45m between passes.  This gives roughly 60% overlap.  The math for calculating pass-to-pass overlap is pretty easy with the Phantom 3 and Inspire 1 since the field of view is ~90 degrees.

You can download the created GeoTiff (or JPG, or DEM, or 3D Point Cloud) from Maps Made Easy and do whatever you want with it.  As long as you are surveying an area that is a convex hull (think rectangle or hexagon, not a star) there shouldn't be any need to trim the edges.  You can see the edges we get in this sample.

We don't have anything to allow for modification and reimporting to our system and won't likely add such a feature.  Getting all the tiles in place to work with the web viewer is a lot of work!

MapsMadeEasy Posted at 2015-5-6 00:29
Thanks Hooks.  This test was done with the Phantom 3.  That is why there aren't any swirls!  It is ...

Keeping tabs on this with great interest!!

MapsMadeEasy,
I really like the work you are doing, one of the issues I can see for me is that it seems to rely on google imagery and a net connection to do the planning. In the areas where I'm interested the google imagery is not current enough to be useful and phone service can be problematic at best.
Is there a way of doing desk based planning for a series of missions, then upload them in the field?

Cheers
Mike

Mike,
If you are using the DJI Datalink for your mapping you can use our flight planner to generate a file while you are connected and then execute the flight plan on-site with the cached map.  Our P3/Inspire mapping app will work when planned with an internet connection and the flight plan and basemap will be cached as well.

So you had no problems with the ALT numbers in the EXIF?
I took 5 photos at 394 feet according to the DJI Pilot app. When I transferred them to my PC I checked the EXIF on the photos and the ALT's ranged from 109 feet to 116 feet.

venomx15301 Posted at 2015-6-15 12:27
So you had no problems with the ALT numbers in the EXIF?
I took 5 photos at 394 feet according to th ...

GPS Altitude is notoriously bad and there are two different sources for altitude information: barometric and GPS.  It sounds like your issue is more that the EXIF tags record the altitude in meters.  116 meters is about 380 feet, so it isn't tooooo far off.  Generally you can plan on 10-15m in altitude accuracy.

If the EXIF actually did read 116 ft, you certainly have an issue.

hmm I think you are right I just looked at one photo and it says 104.6  no feet or meter

Hi, very interested in this topic and any way to automate the course of the P3 for big field photogrammetry...
Any update on your iOS app which could possibly do that?
Cheers

chamanea@gmail. Posted at 2015-7-22 17:00
Hi, very interested in this topic and any way to automate the course of the P3 for big field photogr ...

After DJI getting their MFi situation ironed out, we are waiting on Apple.  We hope to be approved for beta testing this week. Fingers crossed.

Are  you recording the elevation or the altitude ? I am generating Digital Elevation Models from the pictures recorded from a DJI Phantom3 Professional and it all works very well in XY and relative Z with respect to the take-off altitude. However I want to subtact DEMs recorded at different seasons and with different takeoff positions, so I need an absolute (WGS84) altitude. Of course I can record the takeoff altitude with a high-resolution (dual frequency) GPS receiver, but I am wondering if I am missing something and if it is posible to have the WGS84 altitude in the EXIF tag (which, with exiftool, mistakenly states
GPS Altitude                    : 34.6 m Above Sea Level
which is not possible from by 300-400 m ASL position.

jmfriedt Posted at 2015-9-21 23:32
Are  you recording the elevation or the altitude ? I am generating Digital Elevation Models from the ...

We aren't recording anything. DJI populates the Altitude EXIF tag in the image with the above ground level (take off point) altitude in meters. The Maps Made Easy processing system looks up the ground elevation using a 3rd party database and adds that offset in to the rest of the images as long as a ground reference image is included in the uploaded data.  From what we can tell, there is no way to get ASL altitude in the images.

As an update to the information we have posted in this thread about our findings...

As of the 1.5 firmware updates, the X3 and Phantom cameras all record the above-sea-level (ASL) altitude in the geotag portion of the EXIF tags. This is actually the right thing for them to do since the reference has always been marked as "Sea Level". The issue with populating the tag with the ASL value is that it is derived from the GPS which is inherently noisy. We have seen the values off by as much as 50 meters either way and have seen run to run variations of up to 20 meters. This makes photogrammetric processing difficult.

The Good News:
DJI is actually recording the above-ground-level (AGL) value in there too! It is pretty buried and most EXIF viewers won't show it but it is there.

Maps Made Easy has made changes to our processing to account for the change. Our processing system now detects the ground reference image is one is uploaded and uses that to look up the elevation using a NASA created elevation database and then uses the nice and quiet AGL values to do the stitching. This gives the most accurate outputs. In the absence of a ground reference image, the standard EXIF altitude value is used. If the firmware version is older than 1.5 the created map will show elevations relative to the ground level of the takeoff point. For firmware versions 1.5 and newer, the ASL values will be used. There is a pretty good chance that the output will be of slightly lower quality but the created map's elevation will be ASL.

Uploading a ground reference image will always yield the best results.

In the next few days we will be rolling out a new feature that allows users to view all of the data tags that DJI packs in the images.

We have added a new tool for converting the unreliable GPS altitude data that gets written into the GPS Altitude EXIF value to a corrected value based on a user provided elevation or ground reference image.

https://www.mapsmadeeasy.com/tag_fixer

This will allow for people not using our Maps Made Easy processing system to process their data elsewhere while taking advantage of the steps that we do behind the scenes to correct for this issue. This has been a big frustration for a lot of our Map Pilot users that are trying to use other software to process their images. As long as you know the elevation of the takeoff location this tool will convert all of the altitude tags to be the surveyed (or looked up from a 3rd party source) value plus the local barometric altimeter reading. This is intended to keep multiple battery flights level even when the GPS gets a different altitude with its fix.

MapsMadeEasy Posted at 2016-3-12 15:53
We have added a new tool for converting the unreliable GPS altitude data that gets written into the GPS Altitude EXIF value to a corrected value based on a user provided elevation or ground reference image.

https://www.mapsmadeeasy.com/tag_fixer

Hi, I have read most of this thread, I see you wholly recommend the Phantom 3 for geo-tagging metadata. We are about to undertake survey work in 2017. Currently we fly with two Inspires, but I am interested in replacing one of them with something that would be easy and efficient. Do you still recommend the Phantom 3 for such work, if so, which model, as there are several variations. Also, would the Phantom 4 or 4 pro, be just as good for such work. Your advice would be appreciated, thanks.