First, this is a game changing product. I had a few applications ideas for it, and was wondering if these were at least feasible with the current product state from a computer vision perspective. Couldn't find any specs on number of targets and performance at speed: 1. Identifying diameter / speed / direction of natural free falling hail stones during a storm in realtime. (multiple targets (say unto 20) at the same time (40-100mph). Accuracy of diameter to within a few mm's. 2. Scanning for hail after a storm has passed by passing the device at waist height over the ground. (identifying hail in grass and diameters) 3. Scanning a static foam surface with dents in (5mm - 10cm) and calculating the volume of the invdividual dents. Dents could be colored to make vision identification easier. Just wanted to get an idea if these are realistic applications before I dig in. Thanks

Hi @"clell"#75 , Sorry about the delay. I saw this last week, meant to respond, and then lost track of it. Thanks for the kind words! 1. I think so yes. So there would likely need to be some constraints on the camera doing this being fixed-location, angle, and adequate lighting during the measurement so as to be able to leverage motion estimation and background subtraction to aid the process, but I think it should be doable. It would require non-trivial experimentation though, so having some hail-simulators would be quite useful during this time. And actually doing the initial CV work in Unity is what I'd advise... so you can revise the CV pipeline techniques/etc. without having to be in a hail storm or simulate one to iterate. 2. This one is less challenging and I think totally doable. The thing I don't know on this is if say an iPhone 12 would be better suited (I don't know how active LIDAR systems do with trying to scan ice, for example). But I think disparity depth would do fine with ice (but I haven't yet tested it). 3. Yes, I think this is also doable. This ([HERE](https://devpost.com/software/roadrunner-hqwibz)) was a weekend hack to see about detecting potholes in roads (a similar effort, actually) and it seemed to do fine. So with subpixel and narrow baseline between the cameras, I think this should be quite doable. It depends on how much texture there is on the foam though. Can you share pictures? If it has no texture, you'd want to have a dot projector and IR-capable cameras (like our issue, [HERE](https://github.com/luxonis/depthai-hardware/issues/20)). Are you able to post pictures by chance? Also definitely feel free to join our slack community ([HERE](https://join.slack.com/t/luxonis-community/shared_invite/zt-emg3tas3-C_Q38TI5nZbKUazZdxwvXw) if you'd like to discuss a bit more real-ish-time. (I am, and the team is, usually pretty fast to respond there). Thanks, Brandon

Application feasibility

clell

First, this is a game changing product. I had a few applications ideas for it, and was wondering if these were at least feasible with the current product state from a computer vision perspective. Couldn't find any specs on number of targets and performance at speed:

Identifying diameter / speed / direction of natural free falling hail stones during a storm in realtime. (multiple targets (say unto 20) at the same time (40-100mph). Accuracy of diameter to within a few mm's.
Scanning for hail after a storm has passed by passing the device at waist height over the ground. (identifying hail in grass and diameters)
Scanning a static foam surface with dents in (5mm - 10cm) and calculating the volume of the invdividual dents. Dents could be colored to make vision identification easier.

Just wanted to get an idea if these are realistic applications before I dig in. Thanks

Brandon

Hi @clell ,

Sorry about the delay. I saw this last week, meant to respond, and then lost track of it. Thanks for the kind words!

I think so yes. So there would likely need to be some constraints on the camera doing this being fixed-location, angle, and adequate lighting during the measurement so as to be able to leverage motion estimation and background subtraction to aid the process, but I think it should be doable. It would require non-trivial experimentation though, so having some hail-simulators would be quite useful during this time. And actually doing the initial CV work in Unity is what I'd advise... so you can revise the CV pipeline techniques/etc. without having to be in a hail storm or simulate one to iterate.
This one is less challenging and I think totally doable. The thing I don't know on this is if say an iPhone 12 would be better suited (I don't know how active LIDAR systems do with trying to scan ice, for example). But I think disparity depth would do fine with ice (but I haven't yet tested it).
Yes, I think this is also doable. This (HERE) was a weekend hack to see about detecting potholes in roads (a similar effort, actually) and it seemed to do fine. So with subpixel and narrow baseline between the cameras, I think this should be quite doable. It depends on how much texture there is on the foam though. Can you share pictures? If it has no texture, you'd want to have a dot projector and IR-capable cameras (like our issue, HERE). Are you able to post pictures by chance?

Also definitely feel free to join our slack community (HERE if you'd like to discuss a bit more real-ish-time. (I am, and the team is, usually pretty fast to respond there).

Thanks,
Brandon