Urgent: TIA - AprilTags in large repetitive spaces. Help please12272
Pages:
1
Houston |
briangreul private msg quote post Address this user | |
| Hey gang, I'm in the process of scanning an 83,000 s/f facility. A closed hospital to be exact.. I'm North of 600 scans and about half way through it. I'm using a Pro2 and a 2020 iPad Pro with 1TB of Ram. So far it's been behaving really well. Matterport says no issues with the number of scans, I talked to support at lunch today about it when I realized it was going to be a very high scan count. I am running into two issues: - All the hospital rooms look essentially the same. - The bathrooms in the hospital rooms are nearly identical. I've done the usual tricks of putting crap in the hallway (trash cans, chairs, etc) to give the camera some differential traction. I have a set of April Tags with me but the camera isn't really grabbing them the way I was hoping. I'm putting one in the trouble area hoping that it will pick up on it and see it as unique. I've tried putting it on the floor, wall, door, etc and it doesn't seem to make much difference. Anyone care to share best practices on using April Tags in repetitive architecture? I have tags 1-200, but I would feel a whole lot more comfortable if I had 200 to 400. I wouldn't need to be as stingy with the tags which might make it go more smoothly. I'm afraid right now that I'm going to run out of tags. I'd settle for a way to generate 400 or so tags if anyone knows of an easy way to do that. Thx |
||
| Post 1 • IP flag post | ||
|
Houston |
briangreul private msg quote post Address this user | |
| I'm guessing that I'm pushing the boundaries of what the system can do. There isn't much documentation on how the system actually uses the April Tags. I did see that disabling fast scan might help, but no documentation on what fast scan actually is. I have to admit, I'm pretty impressed so far. I've done approximately 40,000 contiguous s/f. This facility is a single floor. I asked MP Support explicitly if there were any processing or app limitations that they knew of. They said it's iPad memory and processor that limit the app. I have had a few instances where the alignment took 4 or 5 minutes when I moved to another area. I've even noticed that it will queue the pending scans and I got about 4 scans deep at one point before it caught up with me. But once it figured out where I was working it caught up reasonably quickly. I will be very interested to see if the resulting model is remotely usable. I've seen some dollhouse issues when you get into long/large spaces. Battery life on the iPad Pro is around 5-6 hours with an hour recharge getting you another 5-6 hours. Camera battery is more than capable if you recharge it at the same time you are charging the iPad. Operator battery tends to need recharging more often (break for food/water/bathroom).  I'm off to explore uncharted territory......in the app and the building. |
||
| Post 2 • IP flag post | ||
 WGAN ForumFounder & WGAN-TV Podcast Host Atlanta, Georgia |
DanSmigrod private msg quote post Address this user | |
| @briangreul Thank you for the update on your large scan. Here are some WGAN Forum resources that may be helpful: ✓ Videos: How to use Matterport AprilTags for Scanning Large Open Spaces ✓ WGAN Forum discussions tagged: AprilTags | Large Please do share your Matterport digital twin, when done (and how it will be used). Best, Dan |
||
| Post 3 • IP flag post | ||
|
Houston |
briangreul private msg quote post Address this user | |
| Well, I finished. It was a very long day. Unfortunately, I'm not able to share the model for a couple of reasons. 1) It's not my client. 2) There are likely details in the model they'd rather not be public. Here's the skinny: Battery Life: iPad Pro 2020 11 - about 5.5 to 6 hours if you disable bluetooth. Will recharge in 2 hours, or 50% in 1 hour. Matterport Pro2 - It seems to be about 9 hours, but I ran it to 20% after 7 hours and charged it when I stopped for lunch. The sweet spot seems to be keeping the battery above 40% based on yesterday's experience. Performance: It was smooth sailing until about 1100 scans. I estimate that I was about 77,000 s/f at that point. At scan 1152 the app started crashing and I began having alignment issues. It's possible there was a strategy error as well, but at that point I just carved up the territory and created a second model for MP to stitch together. This is a Matterport Capture Services project so ultimately it's their issue to deal with the usability of the data. The secondary model was approximately 200 scans and I did re-do the perimeter which would have been common and a utility room (the size of a basketball court) that seemed to have issues due to large strange shelves on one wall. The app slowed way down above 900 scans. Restarting could take 5 to 7 minutes for it to align the first scan. After 1000 scans it was required to wait until the first scan aligned before doing anymore. Prior to that you could stack 4 or 5 scans and the app would catch up with you once it figured out where you were. April Tags - I used about 110. The key here is to manage complexity and then to distribute the anchor points (which is what they are) such that the camera can see two of them within about 30 feet at any time from an angle less than 30 degrees. Anymore than that and it seems to not recognize them well. so doing, patient rooms, I would put one by each door and then one on the far wall of the room. This would let the camera see anywhere from 2 to 4 of them at a time and it seemed to work well. I also skipped the individual patient room bathrooms as there were clearance issues and it would have substantially increased the number of scan points and time required. There isn't a right or wrong way as far as I can tell. You just have to evaluate the layout and if you are seeing repeating layouts then you need to deploy your objects or apriltags to provide recognizeable features. I put them at approximately the camera's height or a little lower when the object I needed to use was a counter top. Total scan time was approximately 20 hours. Had I used anchors from the beginning (April Tags) I probably could have saved 2 to 3 hours. |
||
| Post 4 • IP flag post | ||
|
Houston |
briangreul private msg quote post Address this user | |
| As for the logistics... I have no idea what they are doing with the model. I can tell you one thing - if it was my client I would have managed expectations alot differently. I seriously doubt they are going to generate a floorplan from it. They have blueprints and Facility Managers. I think they are going to use it for a Leasing Walkthrough and if that's the case I absolutely would not have scanned every room. Nobody who makes leasing decisions is going to visit every room. It's just entirely too cumbersome. They will however tour the hallways, see one of each room and it's bathroom layout. I'll personally be interested to see how useable the model is when it eventually processes. |
||
| Post 5 • IP flag post | ||
Pages:
1This topic is archived. Start new topic?