How to inspect quality of flat roofs

[MGee]

Hi colleagues, I have a request to inspect the quality of the roofing of 7 big roof surfaces by drone: to check the edges, borders close to installations and the connections of the roofing. How would you suggest to handle this?
1) I can take pics using pix4D or drone deploy to make one big picture for each roof but wonder if the quality wil be enough to inspect the roofing quality and if stitching will be possible since the pics will be very very similar if you take them from a close distance (the higher the easier to stitch but quality is less)
2) Another possibility is to film the whole stuff using a specified path and then to inspect the movie. Problem I see here is how to react (this is: what is the fact position) when you see a defect?
3) live check but that will be very time consuming and will need lots of battery power
Other possibilities?
Thanks for helping me out with this!

 

[BigAl07]

Welcome to the forum

It depends on what your CLIENT is expecting from you.

I'd say if the area is fairly clear of obstacles you could do an Automated Grid and get high detail images and stitch together but again it depends on what the client is expecting and willing to pay for.

I'd put this back into the client's lap and find out what the "person writing the check" is wanting in terms of Actionable Data.

 

[MGee]

Well the possible client wants a report were is stated where possible defects of roofing are located (so I have to analyse the pics) My worry (or rather question) is if stitching will work for a more or less uniform gray surface. Does anyone has a sample of such a roofing check with stitched images where one can see the connections of the roofing clearly enough?

 

[BigAl07]

Well the possible client wants a report were is stated where possible defects of roofing are located (so I have to analyse the pics) My worry (or rather question) is if stitching will work for a more or less uniform gray surface. Does anyone has a sample of such a roofing check with stitched images where one can see the connections of the roofing clearly enough?

I'd suggest doing an experiment/sample. Pick a roof and do an test flight taking pics etc. Review the final product and then do an actual "in person" inspection and compare your aerial data to your human eyes data. If they don't match you have a problem.

 

[MGee]

I'll do that testing - thanks for the suggestion.
What about the stitching? Is it feasible and should I use the Pix4D or another solution

 

[LUIS MARTINEZ]

Hi colleagues, I have a request to inspect the quality of the roofing of 7 big roof surfaces by drone: to check the edges, borders close to installations and the connections of the roofing. How would you suggest to handle this?
1) I can take pics using pix4D or drone deploy to make one big picture for each roof but wonder if the quality wil be enough to inspect the roofing quality and if stitching will be possible since the pics will be very very similar if you take them from a close distance (the higher the easier to stitch but quality is less)
2) Another possibility is to film the whole stuff using a specified path and then to inspect the movie. Problem I see here is how to react (this is: what is the fact position) when you see a defect?
3) live check but that will be very time consuming and will need lots of battery power
Other possibilities?
Thanks for helping me out with this!

Equipment?

 

[AH-1G]

Here's my suggestion and I've done similar on construction projects.
I video the whole project, then use an app called GOM Player (free) while watching the video I can select any frame of that video and it auto save it to a Gom File in Documents. I would also write down the time of the frame you snapped, example - (snapped at 1 minute 31 seconds).

 

[R.Perry]

I did home and commercial inspections, grandson took over the business. At the time I was using a Phantom 4. I would first shoot a video about 50 feet over the roof. Pick out the areas that need a closer look and take photos of those areas. If Possible I would take one photo of the entire roof then mark any problem areas and reference photos of the area.
If you are doing inspections are you certified by someone? If not always call for a professional roof inspection if you see anything, and I mean anything that is suspect. Reason being if you tell the client the roof looks good, and problems are found later, guess what, you get sued. On flat roof pay very close attention to edges where water can accumulate. Get close ups of all flashing and make sure they are installed properly.

 

[Dave Pitman]

Here's my suggestion and I've done similar on construction projects.
I video the whole project, then use an app called GOM Player (free) while watching the video I can select any frame of that video and it auto save it to a Gom File in Documents. I would also write down the time of the frame you snapped, example - (snapped at 1 minute 31 seconds).

Just a note that MPC_HC also will print out a frame and automatically adds the timestamp to the file name. (also free).

 

[MGee]

LUIS MARTINEZ: I got an inspire 2 with x5s as well as a yuneec h520 with e90 to do the job. Maybe even using my older mavic pro is a possibility

AH-1G and Dave Pitman: those frame capturing apps do they capture also the gps coordinates? How do you know otherwise the exact location on the roof?

R.Perry: the inspections are for a company who is doing repairing of roof and renovations of it. But good point that I should be carefull about the statements in my report

 

[Dave Pitman]

AH-1G and Dave Pitman: those frame capturing apps do they capture also the gps coordinates? How do you know otherwise the exact location on the roof?

On the dji craft's video, you should be able to turn on subtitles in playback which includes gps coordinates.

 

[BigAl07]

On the dji craft's video, you should be able to turn on subtitles in playback which includes gps coordinates.

Great point Dave. I totally forgot that was an option even though we've used it before LOL! Good reminder sir.

 

[clolsonus]

On the dji craft's video, you should be able to turn on subtitles in playback which includes gps coordinates.

For whatever it's worth, I wrote a python script that can parse the dji subtitle file and automatically frame grab every "n" meters of movement, geotag the image, and save them all out. The set of images could then be processed through any of the popular map stitching software tools. (The script is part of this larger software project for anyone who doesn't mind a little DIY-ing): UASLab/ImageAnalysis

My subjective commentary: the standard process of generating sparse mesh -> dense mesh -> orthophoto + 3d model often leads to lower detail in the final result compared to the original images. If your use case is achieving/preserving the most details possible (or the most detail + angles possible) you may have to give up some ground on your expectations. I've made a couple pretty nice maps out of a DJI movie (with my above mentioned script) but you do give up potential detail/resolution if you capture video vs. images. The video is probably more compressed than the pictures as well.

In my case the reason I captured the area as a movie was because I needed to fly super low (i.e. 30-40' AGL) and the area was right next to a treeline (~100' tree tops). So I flew the grid manually and followed the tree line by hand ... but in manual flight there's no way to autotrigger the camera, so I just pointed the gimbal 90 degrees down and recorded the flight as a movie.

Curt.

 

[AH-1G]

AH-1G and Dave Pitman: those frame capturing apps do they capture also the gps coordinates? How do you know otherwise the exact location on the roof?

Unfortunately no. Needing GPS coordinates, this must be one massive roof?

 

[MGee]

AH-1G: massive roof -compared to mine yes but not massive overall . But if you take pics/video and see a defect on the roof, you need to trace back the exact spot where it should be repared. Just giving pics where you see defects is not enough if you cannot give the spot on the roof. If I can see enough details on a stiched picture of the overall roof, then of course I know where to look on the actual/physical roof - then there's no problem and I can deliver what was requested. The only uncertainty I have now (but will test this as suggested) is if flying high enough so stichting should be no problem (eg enough differences visible besides roofing material to stitch automatically) I can see de defects on the roof. If that works, my problem of locating individual pictures or need for video is not needed anymore.
The possibility of geotagging a video is a very interesting feature however I was not aware of. This could mean another way of inspecting large surfaces: you fly low (lots of detail) without worrying to see the overall 'picture' and @home you can analyse the video and where you see anomalities, you now the exact spot through GPS coordinates. Lot of time saved no?

 

[MGee]

of course I agree with what clolsonus says about giving up on quality if one compares video versus pictures.

@clolsonus: you show a very interesting link to stitch images. As a it-dumb person as me, how can I use this? I download the whole stuff (I have a Mac) and somewhere (how??) use the commands below? Help would be appriciated to increase my knowledge a bit on this

• run "process.py /path/to/images" This will process all the images and create the map.
• run "explorer.py /path/to/images" This will launch the interactive explorer tool to view your map.

 

[clolsonus]

of course I agree with what clolsonus says about giving up on quality if one compares video versus pictures.

@clolsonus: you show a very interesting link to stitch images. As a it-dumb person as me, how can I use this? I download the whole stuff (I have a Mac) and somewhere (how??) use the commands below? Help would be appriciated to increase my knowledge a bit on this

• run "process.py /path/to/images" This will process all the images and create the map.
• run "explorer.py /path/to/images" This will launch the interactive explorer tool to view your map.

I'm happy to answer questions for anyone that wants to experiment with the software. Recently I created a virtual box image with all the software and prerequisites and libraries all installed and configured. For smaller data sets, running the software in a virtual machine isn't too bad.

To go down this path, you would first need to download and install oracle virtualbox software (free) for your Mac (or PC). The download link should pop up with a quick google search. Then you would need to download and import the actual virtual machine image from here: ImageAnalysis v20200214.ova

It may all sound a little daunting at first (especially compared to the ease of uploading images to drone deploy in the cloud) but my software is all open-source/free so depending on your use case it might help you out -- or you may decide paying for commercial software is the easier path in the long run. Every one's needs and projects are different and we are all here trying to find the best tools and straightest paths to get things done.

Once you have the virtual machine image up and booted, then we could probably talk about how to actually proceed from there. It's not that hard, but the virtual machine is running Linux and the tools are command-line style, so you have to be willing to read some instructions and ask a few questions. The instructions always lag behind the code in these types of projects ... sorry about that.

Disclaimer: this software has been developed out of a university research lab to support our research projects. I have been writing software for > 35 years so in my view these tools are really solid and outperform commercial tools in several areas, but they are presented in a "guts out" style. I use a command line tool model versus the polished GUI you would get from a typical commercial package. The benefit to me is that I can focus on the core algorithms and strategies and I don't get bogged down in look-and-feel and all the endless details that go along with developing a modern GUI app. I also make zero money from other people using the software, and most likely it just burns up my time answering questions, but I do think it would be cool if a few people started experimenting with it, and I'm happy to support that as best as I can.

 

[PhantomFandom]

Hi colleagues, I have a request to inspect the quality of the roofing of 7 big roof surfaces by drone: to check the edges, borders close to installations and the connections of the roofing. How would you suggest to handle this?
1) I can take pics using pix4D or drone deploy to make one big picture for each roof but wonder if the quality wil be enough to inspect the roofing quality and if stitching will be possible since the pics will be very very similar if you take them from a close distance (the higher the easier to stitch but quality is less)
2) Another possibility is to film the whole stuff using a specified path and then to inspect the movie. Problem I see here is how to react (this is: what is the fact position) when you see a defect?
3) live check but that will be very time consuming and will need lots of battery power
Other possibilities?
Thanks for helping me out with this!

Many great responses from the other members already. I have done this as well and the biggest factor really does come down to the exact needs of the client. Big flat gray roofs aren't exciting but they are easy to fly and inspect. I have done the grid pattern with Drone Deploy and end up with one large stitched photograph. Even at 50 feet above the roof surface the detail is amazing. This was with a P4P.
If you need to get in closer for any specific detailed shots, then follow up and get those individual pictures too. This was necessary in some cases to inspect seams and such.

 

[HammerMissions]

Really depends on what the expected outcome is. But from what we've noticed - operations typically err towards capturing more details when it comes to inspection missions. For roof inspections specifically, we've seen data being capturing in a standard grid mission with the following options:

1. Stop and Capture: Instead of flying continuously and capturing pictures, pictures are captured by stopping at each picture point and then taking a picture. This allows for greater camera control on every picture point - for e.g. camera can be auto-focused before each picture to ensure sharp images. Likewise, camera gimbal can be adjusted (oblique images) to capture vertical details on the roof.

2. Drone Orientation: Unlike mapping missions, where the drone always faces the direction of travel, we've noticed inspection missions being conducted with the drone facing a constant direction throughout the mission and travelling along the longest edge of the roof polygon. This allows for easier assembly of images in the post, specially if the goal is stitching vs processing.

3. Facade / Gutter Inspection: Alongside capturing nadir imagery, data is also being captured in a vertical mission following the boundary of the roof. This is sometimes done as a completely facade inspection, or a linear mission covering just the gutter for details.

Hope this helps!
- Hammer (Drone Missions, from A to Z | Hammer)

PS: Sample roof inspection:

 

[Jobee]

1) I can take pics using pix4D or drone deploy to make one big picture for each roof but wonder if the quality wil be enough to inspect the roofing quality and if stitching will be possible since the pics will be very very similar if you take them from a close distance (the higher the easier to stitch but quality is less)

I suggest mapping with DroneDeploy. Their roofing AI will help to detect issues. I use it for roof inspection reports. It does work better on pitched roofs, but will work on flat.

For flat roofs I use a thermal image camera. It can detect water infiltration. For very large like your pictures, I divide the mapping up into logical sections and provide several maps vs. just one large map. This way the swipes are smaller and you get more detail.

Talk to the building owner or your contact and ask what they suspect are the issues so you can concentrate on them after a full run. Get extra shots of drains, pools of water.

2) Another possibility is to film the whole stuff using a specified path and then to inspect the movie. Problem I see here is how to react (this is: what is the fact position) when you see a defect?

3) live check but that will be very time consuming and will need lots of battery power
Other possibilities?
Thanks for helping me out with this!