Camera-based Night Watch

[montyp]

Are there any open source (or commercial) projects for a camera-based night watch system?

I'm imagining a system that would use cameras on your boat to detect hazards like other ships, floating containers, etc. It seems like this system might be helpful during a single-handed passage. Basically, if the system detected a potential danger an alarm would ring.

Most of the newer automation in sailing projects seem to focus on optimizing the auto-tiller. There is a whitepaper about detecting vessels at sea, but the focus seems to be on quantitative methods (vs machine learning).

[Mike OReilly]

Sounds like radar.

[Dockhead]

I think it is a good idea.

Machine vision would be a nice compliment to radar.

I actually wired my mast for a video camera at the first spreaders, but never got around to installing it.

Not for machine vision, but to enhance visual watch keeping, especially for spotting gaps in sea ice. The perspective is much different from 10 meters up.

FLIR thermal cameras are used a lot on larger vessels ; don't know if they have developed any machine vision technology to go with them, but if anyone pushes forward in this field, then I bet it will be them.

[seandepagnier]

It's possible, yes. You could use even very basic algorithms to detect lights near the horizon at night. Using IR camera can detect exhaust at night?

For daytime basic algorithms can look for some things, but eventually machine learning is needed if you want it to alert you of whales, or actually identify things. To make it work near shore and near other vessels without triggering the alarm unless there is a collision course is also more difficult.

Offshore, there are not really a lot of possible things to detect, and you only care about abnormal sightings, the processing power is not too great, it can already be done on a raspberry pi at 10-20 seconds per image.

You also either need several cameras to cover 360, or a camera which is spinning. The basic alarm if it sees anything is not too difficult compared to other machine learning problems.

I was contacted about collaborating on a similar development with cameras as well as using microphones/hydrophones for commercial ships. The idea is a cargo ship might only have 3 crew these days, and keeping a lookout is difficult. I refuse to work on something that is not free software so I cannot really contribute to the project in question.

[HankOnthewater]

In my opinion a FLIR type of camera is not much good for keeping watch. There has been a thread or two on that. Here is one link, and in it, there is a link to a previous thread as well.
http://www.cruisersforum.com/forums/...ra-211137.html

In regards to visual/camera based automated watch keeping, here are two links:

https://gcaptain.com/sea-machines-an...eid=82482d75cd

https://www.orca-ai.io/

[250224]

Originally Posted by montyp

Are there any open source (or commercial) projects for a camera-based night watch system?

https://www.oscar-system.com/index-en

But... expensive, 24 K€ iirc, and looks forward ~75 m.

[montyp]

Cool - there does seem to be some commercial efforts in this area.

The Oscar system price seems reasonable given that packaging in low volume, I'm not sure it hits right right value proposition.

boat_alexandra - you detailed basically what I was thinking Something like a pi (or whatever laptop was on-board) connected to 3 cameras (maybe two forward, one rear). Open source it, mount a training system and have someone at watch annotate (could be as simple as a button or switch) whenever something to be concerned about is seen. Train a CNN and away you go (I know its not that easy hehe).

[john61ct]

As long as it's used to supplement / assist a human watch outside, not used as an excuse to stay below

[David M]

Sorry, I have to respectfully disagree that some merchant ships (over 1600 DWT) are down to a crew of three now. Yes, I can see that with some smaller fishing boats. But if they are engaged in fishing, you have to keep clear anyway.

[montyp]

Originally Posted by john61ct

As long as it's used to supplement / assist a human watch outside, not used as an excuse to stay below

Yep, as Mike OReilly pointed out this would be something like current radar systems. The potential advantage of a camera system would be lower power and lower cost.

[David M]

Might take the same type of technology that facial recognition uses for it to work. The whales might have to smile.