Question about mapping results using Maps Made Easy

[GsquaredAerials]

So I did my first test with my M30T using MME and the overall results were good from far away but zoomed in, the edges of roofs or really any contrasting edge were pixelated and missing sections as well as a few swirly trees. I have a mission at the end of the month for a corn maze but it's not the typical mapping mission where the client wants all the metadata results or incredibly sharp detail of each corn stalk. All they want is a nice aerial image of the maze.

Last year I was hired to provide them an aerial photo of the maze for the farm's social media and for the client as it was their design. The field is 1362ft long and 200ft wide so at 400ft altitude, my Air 2 obviously couldn't get the maze in one shot. Even if I wanted to break the the ceiling and get up high enough with my Air 2, the property is barley (and I mean barely) outside the perimeter of Class C airspace and about 1/2 mile east of the flight path so that wasn't happening. At the time, I didn't know about mapping missions so I took a series of photos that I stitched together in Photoshop and it was a nightmare for a few reasons. So my idea this year was to fly a mapping mission over the maze and let a company like MME do the stitching.

So the test in my neighborhood was 65 images over a 3.80 acre area at 150ft AGL with an 80% overlap (see breakdown below). If it matters, I was running with RTK using PointOneNavigation but it was only to make the drone more accurate in its flight and not for any measurement.


When I inquired about the jaggies on the roofs, Zane over at MME replied with the following:

Greg,
Try to correlate the areas that didn't turn out well to the same spot in the Overlap Report at the bottom. For something to turn out well you really need to have 16+ views at it. If not more. This is why we recommend using 80% overlap. There is some filtering on the model that goes on and vertical features are always going to be slightly affected. It doesn't really have much to do with the speed the images were taken at as long as there isn't streaking in the source images.

Thanks,
Zane

Here is the overlap report he mentioned from my mission and here is a explanation from their site on how the report works:


Here is the RGB render from the mission (Google Drive)


To me, the houses in the center of the image are in the blue area of the overlap report so I guess I'm not understanding why I got those results. The explanation page above states "9 or more views (blues) is required to have the detail necessary to reconstruct complex objects like buildings, trees or other vegetation." and that area in my report is in the 14+ range. Again, this is for a corn maze and probably won't have these issues but I'd like know what I can adjust on my side before the mission at the end of the month. I am headed out to the maze this weekend to show them aerials of a particular section they had issues with last season and will also be doing a mapping test run just to run though the process and make sure the results are good enough. If there's something I should adjust like lowering the altitude to increase the picture count or maybe increase the overlap to 90% or something, let me know. Zane did say speed or lack of mechanical shutter wouldn't have mattered in this case.

Unfortunately I can't post this weekend's results as the farm and the client don't want anything released on the internet until a week before the opening.

 

[BigAl07]

In order to get really good data on roof edges etc you'll need images from lower altitudes etc. I'm not sure if MME can handle that processing or not.

I use Drone Deploy and I use the following "settings" to improve the models.

Crosshatch pattern
Perimeter Flight
Additional orbits at various angles to capture eaves & overhangs

You aren't going to get really good edges and overhangs with nothing but NIDAR (90deg) images. You need some DATA in the images for the model to build those areas correctly. My last model (and my best one so far but still not what I want yet) I captured at 65deg camera angle and it made a TON of difference.

 

[GsquaredAerials]

I was flying parallel lines with the test so I modified tomorrow's mission to this angle, dropped to 120ft instead of 150ft and will stay at 80% overlap. It added 300 photos but that's fine. I'll look into Drone Deploy but thought their monthly costs were way outside my affordability range right now.

 

[BigAl07]

I was flying parallel lines with the test so I modified tomorrow's mission to this angle, dropped to 120ft instead of 150ft and will stay at 80% overlap. It added 300 photos but that's fine. I'll look into Drone Deploy but thought their monthly costs were way outside my affordability range right now.

I don't think that will make a whole lot of difference. The camera is still going to be pointing straight down. Maybe MME has improved their AI processing but straight down doesn't capture the angle of DATA you need for good edges IMHO. Below is my "Grid" and after it flies going both directions it then flies the perimeter. Then I have it programmed to fly 2 additional orbits at lower and lower altitudes (camera facing more towards center in a POI type maneuver).

 

[BigAl07]

Also, make sure your area of flight/data acquisition is larger than the desired mapping area. I do mine larger and then crop during processing to get the area saturated with Data if I'm looking for finer detail.

I use Drone Deploy for Flight Control (it's FREE) but I process my maps myself locally using MetaShape. I feel like DD gives me more control and flexibility over the flight and it has Terrain Following to help prevent Flight Into Terrain (help being the key word).

 

[GsquaredAerials]

I don't think that will make a whole lot of difference. The camera is still going to be pointing straight down. Maybe MME has improved their AI processing but straight down doesn't capture the angle of DATA you need for good edges IMHO. Below is my "Grid" and after it flies going both directions it then flies the perimeter. Then I have it programmed to fly 2 additional orbits at lower and lower altitudes (camera facing more towards center in a POI type maneuver).

View attachment 4154

Thanks for this. I'll go do some research.

 

[GsquaredAerials]

I'm messed with the Drone Deploy map for the corn maze and with the cross-hatch and perimeter options added, it says it needs to capture 4900 photos over a 113 minute flight. Haha! Now I couldn't find my drone in the planning camera section but I used the Zenmuse X7 profile as it has the same focal range and aperture of 24mm @ f/2.8. Regardless, tomorrow I'm just going fly the mission I created in the DJI Pilot 2 app and see how the quality turns out. They just want a nice aerial image of the maze layout so I may be overthinking this because of my edge results in the neighborhood test. Well see. I did download MetaShape and started their 30 day trial. If MME produces similar artifacting, I'll go watch some MetaShape tutorials and give that a try.

 

[TreeLineView]

I am not sure I am following your end game. If you are looking for something the client can put on their Facebook page to generate customers, I would suggest a 360 image. Not Sure if the 30T does it natively, but the 3T creates one in 20 seconds. You can then create a free account on Momento360 and share it out to Facebook. Easy, quick and FREE. If the goal is to generate traffic for your client I find these work well. Example that was created and posted to Facebook in less that 5 minutes. This is a High School that I have been providing monthly updates for the public for 2 years. Just a thought. DJI 0001

 

[GsquaredAerials]

That 360 is perfect! Thank you for thinking of this. I just checked and the M30T does it with one button. Woohoo!

Still, I'd like to also provide the farm and client with a flat aerial image for whatever other purpose they desire so I have a few missions to run with much image counts to see what stitches best.

On a side note, I ran my 65 images through Photoshop's Merge feature and it produced an image with no flaws anywhere. Now I'm not sure it can handle 500-1000+ pictures without crashing but it's good to know it'll work for smaller capture counts.

 

[TreeLineView]

You can also get a free 14 day trial for DD. Create and export the map you create. Another free option for a one off, but keep in mind a file you create, either a Tiff or PDF is going to be very large and not easily transferrable or accessible. Not sure how much good it would be to anyone other than someone that is importing into another program for analysis. BUT, another free option for testing

 

[GsquaredAerials]

You can also get a free 14 day trial for DD. Create and export the map you create. Another free option for a one off, but keep in mind a file you create, either a Tiff or PDF is going to be very large and not easily transferrable or accessible. Not sure how much good it would be to anyone other than someone that is importing into another program for analysis. BUT, another free option for testing

Thanks. Photoshop is very good at opening PDFs and Tiffs unless it's some really weird proprietary codec.

 

[BigAl07]

We've been shooting a Corn Maze for about 10 years now. I go to the ~396' height and then back away ever so slightly and shoot with a slight oblique angle. This might be a possible work around.

MetaShape is a VERY powerful program but it takes time and LOTS of tweaking. However, a user over at Mavic Pilots did a great video tutorial recently. I just looked at he took the video down. I'll reach out to him and see if we can get a new link to it.

 

[BigAl07]

I am not sure I am following your end game. If you are looking for something the client can put on their Facebook page to generate customers, I would suggest a 360 image. Not Sure if the 30T does it natively, but the 3T creates one in 20 seconds. You can then create a free account on Momento360 and share it out to Facebook. Easy, quick and FREE. If the goal is to generate traffic for your client I find these work well. Example that was created and posted to Facebook in less that 5 minutes. This is a High School that I have been providing monthly updates for the public for 2 years. Just a thought. DJI 0001

Great suggestion. I've never heard of Momento360 but I just created an account and I'm going to give it a try. Seems super simple to say the least.

Thanks for sharing

Allen

 

[BigAl07]

I'm messed with the Drone Deploy map for the corn maze and with the cross-hatch and perimeter options added, it says it needs to capture 4900 photos over a 113 minute flight. Haha! Now I couldn't find my drone in the planning camera section but I used the Zenmuse X7 profile as it has the same focal range and aperture of 24mm @ f/2.8. Regardless, tomorrow I'm just going fly the mission I created in the DJI Pilot 2 app and see how the quality turns out. They just want a nice aerial image of the maze layout so I may be overthinking this because of my edge results in the neighborhood test. Well see. I did download MetaShape and started their 30 day trial. If MME produces similar artifacting, I'll go watch some MetaShape tutorials and give that a try.

Here's a good MetaShape tutorial. It's Mavic Mini3 specific on the Capture side so you can probably just skip to around the 3:50 mark if you only want to watch the MetaShape portion of the video.

 

[adm_geomatics]

I'll second Al's guidance! In order to capture vertical faces and get clean edges on your ortho you have to include oblique imagery. The downside to this is that terrain accuracy will start to suffer because of the extended field of view and number of points out of focus on the edges. A 65 degree gimbal pitch is preferred in order to localize the data from the obliques to the structures. I fly a nadir, a standard oblique (not crosshatch) and then a crabwalk of each structure making sure to capture the surface a square as possible and maintaining overlaps. This usually amounts to 2 second interval shutter at 5 mph. If you are trying to capture intricate detail just remember that you need at least 5-6 images of it from varying angles in order to give the processing software a chance to reconstruct. Personally I have never had any luck with cloud processing producing cinematic or game quality meshes and use Metashape. Reality Capture is probably the best but it suffers from a lack of georeferencing functionality if you care where the subject is on the planet.

 

[BigAl07]

I'll second Al's guidance! In order to capture vertical faces and get clean edges on your ortho you have to include oblique imagery. The downside to this is that terrain accuracy will start to suffer because of the extended field of view and number of points out of focus on the edges. A 65 degree gimbal pitch is preferred in order to localize the data from the obliques to the structures. I fly a nadir, a standard oblique (not crosshatch) and then a crabwalk of each structure making sure to capture the surface a square as possible and maintaining overlaps. This usually amounts to 2 second interval shutter at 5 mph. If you are trying to capture intricate detail just remember that you need at least 5-6 images of it from varying angles in order to give the processing software a chance to reconstruct. Personally I have never had any luck with cloud processing producing cinematic or game quality meshes and use Metashape. Reality Capture is probably the best but it suffers from a lack of georeferencing functionality if you care where the subject is on the planet.

I used your suggestions on my project last weekend and the results were a LOT better but I need to keep tweaking my work and getting more images. THANK YOU for your input.