Ideas about a "smart" (AI-driven) aquarium monitoring camera

[FishStudentRI]

Thanks for clicking on my thread!

I’m an engineering student at Brown University, trying to gauge interest in the use of a generative artificial intelligence for our legacy aquarium monitoring systems (mainly Apex). Specifically, I am wondering if fellow hobbyists would find it useful to have a personal "aquarium AI agent" that uses a behavior-recognition camera to track fish activity and provide insights on stress events, aggression, and feeding patterns.

I think the addition of a "smart" camera could be useful for the following applications:

- Optimized feeding: With an automatic feeder, pattern recognition software can detect when fish begin to slow their feeding response and automatically stop the feeder — ensuring everyone eats and preventing excess food from sinking to the bottom.

- Aggression detection: If you suspect aggressive behavior in your tank, the smart system can isolate and record those incidents for playback, helping you identify the bullies.

- Jump alerts: If a fish jumps out while you're not watching, the camera automatically captures and saves the footage so you can review what happened.

- Remote sump monitoring: If you receive a low water-level alert while at work, you can tap to open the sump cam and visually inspect your sump from anywhere.

- Enjoyment while away: Sometimes, you just want to watch your beautiful tank remotely — and a smart cam provides more tailored insights than a standard pet camera.

Feel free to share your thoughts by replying to the thread.

I've also included a 1-2 min (anonymous) google survey to better understand our opinions on AI for improving tank monitoring. I'll be sure to share the results!

Click here for the 1-2 min Anonymous Survey

 

[jaylach]

To be honest I've never used such a setup so I don't really have an opinion other than I don't really trust AI.

 

[FishStudentRI]

Thanks for the response, and totally fair.
I hesitate when using the term AI because it's loaded with a bunch of connotations. The type of AI I am referring to is purely machine learning (probably the "dumbest" form). Basically, all the device knows is what you feed it. In short, the pattern recognition software would be trained on a bunch of common fish behavior footage and would be told what is occurring in each clip.

If you wouldn't mind answering, do you use any specific tools to keep your tank running when you leave for an extended period (automatic feeders, light timers, etc...)?

 

[MaloK]

What are you aiming for ? Detailed information on detected behaviors with fishes recognition.

Or just a simple activity monitoring, like rapid movement, fish going out of bounds, etc.

If you have already everything you need for the project and are looking for a camera that is able to provide added value to your training agent.

I would go with a deep AI learning capable camera that I would be able to use to train my AI by providing it kilometers of feed from an aquarium... That is well populated.

This idea can go really far... And deep. From automatic identification and classification, to full blown comportmental analysis and reporting.

If you can put out only one that gives good advice... You have a career.

 

[FishStudentRI]

Thanks for the advice!

As of right now, I am trying to gauge where the market would be. It's tough to deduce that from niche forums alone, since experienced hobbyists are likely the only ones contributing, so I am working on designing some wider-reaching data collection methods.

I agree, the deeper I take this, the more interesting and helpful the device could be. However, I think that making the device as capable as possible means feeding it exponentially greater amounts of data, down to the individual species level. At that point, it'd probably be more effective to market it as a custom device for commercial aquaculture, where the AI can know every metric about a specific species of salmon in a specific system (unfortunately, there are already products for this application).

So what we are left with is a spectrum of capability. On one end, it could be more of a gimmicky pet camera that features an underwater probe-like camera to immerse you in your tank while you're at work, bored. On the other end, a monitoring tool much like what you described in your second statement.

From a marketability perspective, I'd be curious to hear what you think would have more success?

 

[jaylach]

If you wouldn't mind answering, do you use any specific tools to keep your tank running when you leave for an extended period (automatic feeders, light timers, etc...)?

Can't say that I do although I DO have a light timer. As to being away I did pretty much all the travel I would ever want much earlier in my life, I am 70 years old. If an emergency came up I have a friend in my apartments that would be happy to come over to feed my fish and cockatiel.

As to monitoring I think that the fish themselves are the best indicator.

 

[MaloK]

From the marketing point of view... My first goal would be.

Simple way to see if, Light, feeder, heater, filter, water level are on line and running. With a quick connection to a good real time image feed of the tank.

Automatic video saving if things gets agitated. And alert if any of these fails.

The product could be easily done with cheap processor, network connection, sensors and controllers, oriented for remote control and have solid foundation to be sold to traveling peoples. So you can look at the video, press feed, the feeder start to operate and you see your fish on target. You know every company has a k-car.

Once you user base start to grow, you start to expand development of the capability of your client devices and use remote centralized AI with the video feed of all your customers who said yes to eula without reading then use it to train your AI to give the best advice.

Once your AI masterminded all aspects of fish keeping and can politely give advice without being turned off. And guide you breeding any species it has in catalog...

It's Your devices, your AI server... Time for monthly subscriptions for advanced breeding support.

 

[seangee]

I think there would likely be more interest in an inexpensive monitoring system for basic parameters such as pH, ammonia, nitrate, TDS (possibly), temperature and lighting.
Behavioural monitoring would require masses of data as there is so much variance between species and even combinations of species. This is further complicated by environmental triggers. Temperature changes are the obvious one where a relatively small change can trigger spawning which in turn completely changes the behaviour. But the same changes are also triggered in many species by changes in barometric pressure and/or water quality.

 

[GaryE]

I'm very much into fish behaviour, and this makes me skeptical about your project. One of the mistakes made by aquarists is generalization. Fish do this, fish do that. In reality, the word 'fish' gets us into trouble because the meaning is so superficial. There can be patterns of behaviour within a family or within a Genus, but interpreting what a species does is still an art in many cases, and is a total mystery in others.
Then you look at estimates of how many aquatic species are kept in aquariums...
We all know the stories of what happened when (some) humans grasped we weren't on a planet the universe revolved around. I suspect that over the next decades we'll see the start of what happens when we (we being people like me, outside of scientific circles) realize we aren't the only intelligent life on Earth, and that many different types of intelligences surround us. I don't believe that one has sunk in yet.
What we believed fish behaviour was 50 years ago could have been monitored by AI, but where we are now and where we may be going in our understanding could make this an enormously complex project, even if it just strives to monitor the basic behaviours.
Diversity is going to make aspects of your project enormously complex.
AI stands to help unravel the mysteries, but that's a whole other project.

Now, crusty old fishkeepers - and observation on aquarist behaviour... I know of some reef keepers who have one single aquarium supported and monitored by technology that is beyond that of 1980. They would be your target - the engineers of the hobby. Locally, I speak with 6 such aquarists, even though I am a freshwater fish guy. One of them has a cam on his tank, but he also has crippling levels of anxiety and is a control freak. In the freshwater scene here, I know people who have put up cams, and lost interest with the low resolution images.
Monitoring systems are expensive, and outside of the reef tank world, this tends not to be a rich person's hobby.
It has another aspect that makes me skeptical - as people gain experience in the hobby and if they develop a 'popular science' approach, they expand. Multiple tanks, multiple monitoring systems? I know very few aquarists 'advanced enough' in their interests to want a remote monitoring system who haven't already leapt into a multi-aquarium set up.
We are also renowned cheapskates. Many of us are broke, and many of us have the North American belief system that devalues learning and spending on interests. We have some age related dropouts who hate technology and brag they have no smart phone - that's not far off a description of me as tech isn't central to my interests. I use my phone a lot, but not nearly to the level of several of the members here. I know it's not about me, but I think I'm kind of representative of a lot of your potential market as it stands now.
I'd see this working as a marketable tool if it dealt with hard data (water parameters) and recognized the paralytic diversity of fish behaviour; if it could survive with the tiny market of single tank reef fish engineers; and if we had a large influx of young tech savvy aquarists coming into the domain. I'm afraid I don't see that.

 

[Rusty_Shackleford]

Something you may want to include in the AI programming. Recognizing breeding/mating behavior.
Not to discourage you but I've been thinking of your potential market and it's pretty small. Maybe some hard core breeders or reefers and the hobbyists that just want everything to do with tech. But the average aquarist probably not.

 

[Uberhoust]

It would be pretty easy to get a neural network to identify the fish, I have thought of that myself a few times. Getting the network, or AI to assess behavior or determine when to feed might take a lot more work. Additionally, now you have to have a computer running full time to process all the video frames, this would likely use more watts than the heater. Finally, you don't need AI to monitor the sump or provide a webcam view of your fishtank, this can be done with less expensive sensors and a controller, the reef guys probably already have something for this.

On the other hand it would be cool to have a system monitor your tank and give you information, (with video clips?) about how your fish are interacting, but with so many variations in people's fish tanks I think the training would have to be done by the user, at least initially, or you would have to get the users to id certain behaviors from their tanks onto videos collect from their tanks and send them in to train the AI. Training AI can be a time consuming and expensive task

My back of the napkin issues I came up with include:
1. How to mount the camera so it has a clear view, keeping reflections out can be difficult, the best idea I came up with was to create a shroud at one end of the tank and permanently mount the camera there.
2. Fish tanks often have many hidden areas, so the camera might not see the fish jump, you might want to consider fish counting to indicate when a fish goes missing.
3. Some fish should be easy to identify as an individual, but others might not have enough visual differences to be uniquely identified.
4. From my experience, you would need a reasonable computer with a GPU, seems like a bit of expense for many people. (based on neural nets).

 

[Fishkeeper123]

Thanks for clicking on my thread!

I’m an engineering student at Brown University, trying to gauge interest in the use of a generative artificial intelligence for our legacy aquarium monitoring systems (mainly Apex). Specifically, I am wondering if fellow hobbyists would find it useful to have a personal "aquarium AI agent" that uses a behavior-recognition camera to track fish activity and provide insights on stress events, aggression, and feeding patterns.

I think the addition of a "smart" camera could be useful for the following applications:

- Optimized feeding: With an automatic feeder, pattern recognition software can detect when fish begin to slow their feeding response and automatically stop the feeder — ensuring everyone eats and preventing excess food from sinking to the bottom.

- Aggression detection: If you suspect aggressive behavior in your tank, the smart system can isolate and record those incidents for playback, helping you identify the bullies.

- Jump alerts: If a fish jumps out while you're not watching, the camera automatically captures and saves the footage so you can review what happened.

- Remote sump monitoring: If you receive a low water-level alert while at work, you can tap to open the sump cam and visually inspect your sump from anywhere.

- Enjoyment while away: Sometimes, you just want to watch your beautiful tank remotely — and a smart cam provides more tailored insights than a standard pet camera.

Feel free to share your thoughts by replying to the thread.

I've also included a 1-2 min (anonymous) google survey to better understand our opinions on AI for improving tank monitoring. I'll be sure to share the results!

Click here for the 1-2 min Anonymous Survey

I think it’s a nice idea. I currently have a camera on the tank so I can see it when I’m away. It’s pretty good but not amazing. Im not sure if there’s a way but seeing fish clearly in the dark would be nice. This might be a really common thing my £15 camera doesn’t have but it would be good to add to the design if you can. There’s a lot of things you are considering for this camera and some of them such as the aggression indicator is cool and would be helpful so I know how much aggression goes on in my tank. The automatic feeder thing is also interesting and would be beneficial if people went on holiday. However, the other 2 wouldn’t have any effect as I don’t have a sump and I have a lid on the tank and I’m not sure that loads have sumps(unless for really large tanks) and lidless tanks(in the UK I’ve not seen a lidless decent size tank before, maybe not looking hard enough tho).

 

[Sir Guppy]

Thanks for clicking on my thread!

I’m an engineering student at Brown University, trying to gauge interest in the use of a generative artificial intelligence for our legacy aquarium monitoring systems (mainly Apex). Specifically, I am wondering if fellow hobbyists would find it useful to have a personal "aquarium AI agent" that uses a behavior-recognition camera to track fish activity and provide insights on stress events, aggression, and feeding patterns.

I think the addition of a "smart" camera could be useful for the following applications:

- Optimized feeding: With an automatic feeder, pattern recognition software can detect when fish begin to slow their feeding response and automatically stop the feeder — ensuring everyone eats and preventing excess food from sinking to the bottom.

- Aggression detection: If you suspect aggressive behavior in your tank, the smart system can isolate and record those incidents for playback, helping you identify the bullies.

- Jump alerts: If a fish jumps out while you're not watching, the camera automatically captures and saves the footage so you can review what happened.

- Remote sump monitoring: If you receive a low water-level alert while at work, you can tap to open the sump cam and visually inspect your sump from anywhere.

- Enjoyment while away: Sometimes, you just want to watch your beautiful tank remotely — and a smart cam provides more tailored insights than a standard pet camera.

Feel free to share your thoughts by replying to the thread.

I've also included a 1-2 min (anonymous) google survey to better understand our opinions on AI for improving tank monitoring. I'll be sure to share the results!

Click here for the 1-2 min Anonymous Survey

This could come in handy to spot some external diseases