@ Mark Thanks a lot for this step by step guide. I will give it a try ASAP.
I’ve always been fine using Drone Deploy and auto-focus. Are these tips on camera settings requirements, or options? What if your site slopes a lot, and has trees and so on. How does manual focus work then?
Its like “it works now for P4P V2, but you have to do all these things…”
So I am wondering, is the update fully done, and all these things are the same you had to do with previous P4D?
Also, is P4D somehow not as good as drone deploy’s camera handling? thx
These are all just my tips based on my (limited) experience with mapping, some comments from a PIX4D trainer, and my general photography experience.
A number of early P4Pv2 users were trying to use PIX4D Capture and were getting blurry images. My approach here was to address that problem. If you have the camera in autofocus that way may well work too, but the camera lense properties are changing with each focus which may hinder or reduce the quality of processing later. I prefer to control this variable at the time of shooting.
If you have a steep terrain slope, you can assure large depth of field using a small aperture like f/11. If really height varied I have heard it recommended (again by a trainer) to break the project into separate mapping projects at different altitudes and to merge them later.
If want to confirm DOF there are a variety of on-line calculator sites that will calculate DOF for a given camera, lens, and aperture. For wide angle lenses like on drones if you do have the ability to stop down to say f/11 your range of focus is HUGE. (Not an option with fixed aperture cameras like on the new Mavic Pro Zoom which only has f/2.8).
I have only worked a little with DD and had better processing results and control with P4D mapper (proccessing). In theory you should be able to use ANY flight control app (DD / P4D Capture / Maps Made Easy, DJI GS Pro) and collect the images and then use ANY processing package for map construction. I suspect many early P4Pv2 owhere have been experimenting with other flight apps that currently support the P4Pv2.
I am iOS based, so I am still waiting for official support of the P4Pv2 in PIX4D Capture.
@Mark
Thanks for sharing all your experiences and taking time to explain all the details AND supporting our queries. Pix4D should really compensate you / give you a medal for all the work you’re putting in to this forum!
Pix4D guys should really be providing their expertise here but unfortunately, its not the case.
Great job Mark!
DEPTH OF FIELD
I did a little more research to determine depth of field for the P4Pv2 camera to determine what f/stops will work for mapping.
The P4Pv2 has a 1" CMOS 20MP sensor, and a 8.8mm lens (angle of view is equivalent to a 28mm lens on a 35mm (full frame) DSLR).
I used the DOFMaster.com web site to compute DOF values. The site does not know the DJI drone cameras, but the specs for the P4Pv2 are about the same as many 1" CMOS 20MP cameras (perhaps the sensor is actually made by Sony?). I found the “circle of confusion” or CoC for this sensor which is basically the distance between pixels. I used CoC = 0.011mm which is the same as the Sony RX100 which has a similar size and resolution sensor.
For the DOF calculator, the sensors CoC, the focus subject distance, and the f/stop and it tells you the range of distances that will be in focus. Here are some examples:
f/2.8, subject at 50’ (focus plane) => DOF is 7.7’ to Infinity
f/4.0, subject at 50’ => DOF is 5.3’ to infinity
f/8.0, subject at 50’ => DOF is 2.8’ to infinity
What I have learned from this it that it should be possible to get sufficiently sharp images at any f/stop as long as the camera on the drone is focused at about the flight height. Trees, slope, etc. should not be an issue. Autofocus is not likely to be needed for mapping. Keeping a fixed focus also assures the best consistency of camera parameters for map processing.
Also - no need to stop this way down to f/11. Wider apertures such as f/4 should be just fine and will allow faster shutter speeds.
Based on this I am going to do some more experiments. If I set the camera at f/4, focus at 50’, on a typical day I might be able to get sharp images even at max flight speed. This will be simple to check - fly a short capture sequence at different speeds and see if I can detect any motion blur in the original images.
I just tested the new IOS update, 3.4.0, with my P4Pro2.0 and it still lands at the mission endpoint and not the home point. I was running with the manual camera settings described above and the images came out a little better. I think the white balance was adversely affecting things as I was testing over my office which has a white roof.
Has anyone else had an issue with the landing point? I know I can just take control and return to home, but it’s a tad frustrating the app still does this to me.
@ Mark Muntean
RE: Thank you & Safe Mode / Fast Mode
First off - thanks for all the great advice! There’s some really good stuff in you above posts! I fully plan on printing your step-by-step and experimenting with it. When I was young I shot a lot of 35mm, but I’ve forgotten almost everything I knew.
Regarding the safe vs. fast modes, I’m really upset that they took this option away. Looks like I’ll have to turn back to Litchi for the waypoint shooting. I need to be able to run analysis on raw images in GIS. Being able to return the sensor to the same location at a later time is crucial. As far as the positional accuracy of the GPS is concerned, this is my territory - I’m fully aware of the DJI capabilities. It is what we call “map-grade” as opposed to “survey-grade”. When shooting from 250+ feet AGL, the positional accuracy of the p4p is sufficient for my needs - nothing that a few GCPs wont fix.
Ironically, that’s the main reason I’ve been sitting on the edge of my seat waiting for Pix4D to release support for the P4PV2. At this point, I don’t really see how Pix4DCapture offers anything more (in terms of an autonomous flight and photographing) than what Drone Deploy had already delivered. In fact, I think I prefer the DD app, although, I haven’t made up my mind for certain yet. Even though I haven’t used Pix4D’s post-processing software yet, from my research, it appears to be far superior to Drone Deploy.
One last question (for now)…to all:
What if I simply want to stitch nadir images that I’ve captured from the drone. i.e. I don’t necessarily want to go through the whole 3D processing. My hope is that this will cut out all the edge blurring/blobbing introduced during 3D post-processing. Similar to a stitched panoramic view, but with nadir images as opposed to a landscape, for example. Anyone have experience with this?
Cheers
@ Richard Rogers
I did a direct comparison recently of the same images being processed by PIX4D and DD. The PIX4D map was very accurate - basically limited by the resolution or GSD of the original images. Checkpoint accuracies reported abut 5mm laterally, and 10-20 vertically (used Aeropoint GPCs which reported to be good to about 1-2 cm). It seemed that all limiting factors - GSD, map construction, and GPCs accuracy were in balance at about 1cm.
I ran the same images though DD. The processed map was clearly much lower resolution and had more artifacts (samples below). In the first I could clearly read license plates on cars, on the second not so. DD seemed like it was downsampled by a factor of 4 to 8 from the original images. I went around and around with DD support who kept saying perhaps issues with images. The images, overlaps, grid plus nadir collection was text book however, and of course PIX4D’s map was similar in resolution to the original images. DD was not. What bothered me as much as the resolution issue is that several different folks in DD support simply did not seem to understand my question. I took the time to submit reference maps, screen shots, etc. and the support replies seemed canned and ignorant. This was my first experience / test / evaluation of DD and the mapping and the support did not impress. Perhaps I had a bad first date, but I spent a lot of time and could not solve it. I have also heard from other experienced mappers that PIX4D’s mapping engine is more accurate that DD - perhaps that is what I observed. DD just came out with a new mapping engine too - which may be the issue too.
Note that there may be still some artifacts around edges in the mapping orthomosaic. PIX4D’s Desktop Mapper (but not Cloud Mapper) allows you to go in and edit these to clean up your 2D map and 3D mesh. This is what I am experimenting with now.
Here are maximum screen zoom extracts from the orthomosatic maps processed by DD and then PIX4D. Both worked from exactly the same images. The DD is first. Note the clarity pink 2" wide duct tape X in both images, and the artifacts from the trash can edges in the DD construction. To me the PIX4D map is clearly sharper and has fewer artifacts…
@ Richard Rodgers
Now - as to simplest stitching images together - will not work well at all. That was my first naive attempt at mapping. If you take a set of shots on the ground with a camera from a fixed position and stitch with Photoshop or Lightroom, you rely on the fact that the images all have the same angle of view of all subjects. Consider for example a coke can on the ground appearing in adjacent overlapped image taken from the air slightly to the left and slightly to the right. Both images see the top, but also the sides (different sides) of the can. Panorama stitching in Lightroom cannot match the image of the top of the can with a clean circular edge with the ground cut up all around it. Mapping can. Mapping can because the middle step if creating a 3D point cloud of the ground and can, and then constructing a what if view from above of that point cloud. This process is completely different than photo panorama stitching, and what we pay the big bucks for for mapping software.
Note that there are still some artifacts around edges in the mapping orthomosaic. PIX4D’s _ Desktop _ Mapper (but not PIX4D Cloud Mapper) allows you to go in and edit these to clean up your 2D map and 3D mesh. This is what I am experimenting with now.
@Mark,
Note that the trial version of DD always does the lower resolution product. I ran into that too, and worked with the sales people to get temp real version level. They did that for me.
Again, not to advertise competitors on the P4D forum, as I think P4D is an excellent product, but I was impressed with the DD person I talked to. It was not the sales person. DD has developed their own processing engine, and seems very active in fixing and improving it. Given that DD got their flight control tool working way before P4D, that tells you something about their responsiveness.
I personally went with agisoft PS, as it gives control over more, and costs the same to buy as the others to rent for 1 year. The admin may edit this post, I’m likely really getting on the edge of the forum policies so no worries if you must, Admin.
@Richard Rogers
It seems like someone should come up with a common xml formatted flight control script schema. Like 3d printers all take gcode, no matter the stuff before that stage. Then the flight planning part would be decoupled from the “run the drone” part which needs the DJI sdk part.
As far as just stitching images, I think that does need some 3d element, so you can’t get around that for any kind of decent results.
Not trying to push other products on P4D forum, but an example is in photoscan, the ortho construction step needs either a point cloud or DEM derived from it. I think you can use the sparse point cloud PS makes upfront (10 minutes processing) to use for the ortho. Classic case where control of the steps allows potential short cuts. It would not be in proper place, as GCP’s would not be incorporated in sparse cloud, but its something.
I now have downloaded the new PIX4D Capture for iOS from the App Store. It is version 3.4.0 and reports to support the P4Pv2. Will give it a try this week. Thanks PIX4D!
@ Richard Rogers - I note that there is still the FAST/SAFE mode choice under the settings page in this new version so you should be able to take your precise shots. Let me know however if this is limited to 100 (or ?) maximum images due to waypoint limits in the drone.
Also, I am curious as to what the matching position is needed for? In PIX4D Desktop Mapper, you can point to any point on the processed map and readily find and “inspect” all individual images that include that point (from all perspectives taken). This allows you to zoom in on any of the original images in full detail. If you place two browser windows side by side for two different mapping days it would seem you can clearly inspect the same object in much detail, albeit not necessary with identically positioned cameras.
Another question for you with your testing - does SAFE MODE actually result in fixed and repeatable GPS positions for each shot? What about the inherent uncertainty in the drone GPS system (which may be many feet) - is this a limiting factor?
Just purchased the Phantom 4 Pro v2.0 and didn’t realize any of these issues… Doh! On top of that, we got the remote with the integrated screen. I finally figured out how to side-load apps onto it but now I found out PIX4D doesn’t work with Android. Going to have to purchase a new remote, and use my iphone to do any mapping. Unleeeeessssss, PIX4D announces a new release of Capture this weekend… And I’m ready to fly Monday morning 
@Caleb Anderson Pix4DCapture with support for the Phantom 4 Pro v2.0 is available for Android now. I sideloaded a previous version of Capture on a CrystalSky… should work with the integrated one.
@ Mark Muntean:
Let me start with…
- I’ve not actually had time to fly with Pix4DCapture yet - just played with the app.
“I note that there is still the FAST/SAFE mode choice under the settings page in this new version so you should be able to take your precise shots. Let me know however if this is limited to 100 (or ?) maximum images due to waypoint limits in the drone.”
I am unable to find any safe/fast setting in the app I have installed. Could you please give clear instructions for finding this. I do have a slow-fast slider, once I have a mission open.
“Also, I am curious as to what the matching position is needed for? In PIX4D Desktop Mapper, you can point to any point on the processed map and readily find and “inspect” all individual images that include that point (from all perspectives taken).”
This is for use in GIS software, not photo processing software such as Pix4D. Using image classification, the RGB value can be used to detect change between time-series images. e.g. delineation of landslide scarp, wetland expansion/contraction, paved/impervious surfaces. I do not want to simply, visually compare two images. I want to (and have) created raster data processing models to mine data from the RGB values of each individual pixel. These images will likely be captured from higher altitudes - pushing 400-ft limit.
“Another question for you with your testing - does SAFE MODE actually result in fixed and repeatable GPS positions for each shot?”
This is what I’m trying to figure out. It was my understanding that the two options were to shoot timed images (the timing would be based on aircraft speed) or use waypoints, which I assumed were just that (a fixed coordinate). I believe Litchi uses waypoints.
If so, I should be able to calculate the positional error of the GPS between missions using GCPs.
“What about the inherent uncertainty in the drone GPS system (which may be many feet) - is this a limiting factor?”
Yes, it is. However, I can compensate for lack of location horizontal and vertical location repeatability in post-processing within GIS. It would be a painstaking process without a good orthophoto which brings us full-circle back around to my question regarding image pixel stitching vs. full-on 3D model creation, then conversion to a “blobby” orthophoto. Again, I’ve not tried Pix4D for processing yet. All research I’ve done indicates that it produces much better orthophotos.
@ James Maeding
“Note that the trial version of DD always does the lower resolution product. I ran into that too, and worked with the sales people to get temp real version level. They did that for me.”
I find this particularly interesting as they (DD) wouldn’t even extend my trial in order for me to use the software for the first time. I could not use it as there was no android application available for image capture for the P4Pv2 at the time. The trial ran out before the bird was supported…never got to try the processing software during my trial period… I’ve been using WebODM in the meantime while waiting for Pix4D. DD just assumed I was a freeloader and said NO! Needless to say, we’re not friends. My overall experience with DD is similar to Mark Muntean’s. My questions were often met with canned answers from a person that seemed to be more of a public relations representative than a technical person.
Just for giggles, I did run one quick data-set through DD’s processor and it worked (to my surprise), but it produced an extremely “blobby” orthophoto. I didn’t bother to look at the model or point-cloud (not sure the option was even there). I had seen enough. I was getting much better results from other processors using identical data-sets. Processors that ran on my desktop, so there’s no question about data security or data ownership (which is a big deal to me).
Thanks for the conversation and shared knowledge… I will hopefully get to fly with P4Pcapture this evening if the weather holds out. I will post anything I learn.
@Michael Kemper
This is great news thanks! I sideloaded Ctrl+DJI and Capture but it won’t connect to the Phantom. Maybe the problem is that DJIs app is always running in the background on the integrated screen/remote combo. If I can figure out a way to force-quite DJI Go, maybe Pix4D could connect.
@ Richard Rogers
I am using PIX4D Capture 3.4.0 for iOS (released earlier this week).
The choice for SAFE MODE appears here. Setting up a nadir or grid flight, hit the small gear to go into setting options:
@Richard Rogers
Yah, I had the same thing where DD original trial ran out right when the V2 was supported. I asked them if they thought it “might” be a good idea to extend that. Then they did, and I found out about the low res in demo issue. Then I pointed out that was not a real trial, and they gave me a couple weeks of real as I really was a potential customer, not someone trolling for free tools. That turned me off a bit, having to push for it. On top of that, my industry (civil) seems new at using drones, so guys like me that figure things out tend to steer others. They lost out for now. P4D is still in the running, as we may end up needing the processing power on their end.
@ James Madeling
My experiments with DD were with the licensed version or “PRO” level, so I assume I am getting the best maps they can process. Also, it would make no sense for them to reduce the quality of maps for trial users - the whole purpose of a trial is to see what it can do. Why offer a lower quality as your first opportunity to show your capabilities?
Also - I had some issues during my first trial with PIX4D, and they extended my trial period. Nice.
@Mark Muntean
“My experiments with DD were with the licensed version or “PRO” level, so I assume I am getting the best maps they can process. Also, it would make no sense for them to reduce the quality of maps for trial users - the whole purpose of a trial is to see what it can do. Why offer a lower quality as your first opportunity to show your capabilities?”
EXACTLY - lets show off the product by making it look bad.
Having said that, at first I thought they were degrading at time of download, but then realized you have to reprocess after they upgrade you to real version.