I think there are use cases that requires a specialized hardware. Such as the one you mentioned: outdoor, no power, multiple cams. And there are use cases that using a phone is more appropriate. Like when you have an unused phone lying around. Or when you are often on the go and want a security camera to carry with easily.
This app lets you turn your unused phone into a security camera. It has AI event detection (person, face, animal) and smart alerts, but the smart functionality runs offline. No footage is sent to our server. Actually we don't have a server. You don't need to sign up or create a profile. Notification is sent through Apple's server (we currently have only iOS app).
We are doing beta program (free). We'd love to hear your feedback, and especially learn about your use case.
We have cameras on driveways, etc where there is no power or ethernet. These cameras (and perhaps a solar powered wifi mesh network, which these could form if they had the right software) are a reasonable solution to that problem:
I’d love to have it take video feeds from something like those. (Perhaps via a standard network protocol.)
Store + ml recognize them locally, and simultaneously client-side encrypt and stream them to an s3-compatible bucket (eg backblaze b2).
Bonus points if the gizmo has a sim card and backup battery so it keeps working if power / internet are cut. (Most of the time it’d upload via residential broadband (wifi/ethernet), but while the house is being broken into, it’s fine if it burns a few GB of cell plan data.)
Make sure the box is compatible with some sort of ssd/nvme drive (no moving parts), and not just hdd.
One model for monetizing it: AGPL or BSL it, and offer some cloud side services for convenience features like managing the encrypted bucket, pruning “boring” videos after 30 days, etc. edit: or, provide a nice (open source, simple) nat hole punching vpn service so phones can see the video in real time. Maybe charge for access to bounce servers and automatic setup.
Thanks for sharing this. I was not aware that Reolink was introducing solar-powered cameras. I like their brand because back in 2017 I was searching for 802.3af PoE IP cams and they were the most recommended. Been running a couple since then and have had good results overall. I hope they are still producing them as good now as they were back then.
Thanks for your suggestion. But are these functionalities not better implemented on a desktop computer? A mobile phone is limited in compute capability but very portable, so it easy to use it for occasional situation like when traveling, or when you only need a security camera for a short time and do not want to invest much.
I would give it a try if there was an Android version. I don't have an unused iPhone, basically because their value in the second hand market makes it worth to sell it insted of keeping it.
> I'm just curious what we would love to use object recognition for
Mix in ML to identifying food for their nutritional content, calorie intake monitor and future projection on your health if you stay on this course.
There is an old adage; You are what you eat.
I'm in the process of doing just this. I've taken pictures of every meal/snack whatever I ingested since 2005. I have over 15,000 images of food. Now that ML and computing power has come into play. I want to create this exact thing. Then correlate with my health. I want to see if there are cause and effect to my eating.
Furthermore, I have my exercise history as well.
I am what you call: Quantify Self.
Another use-case for your ask is cataloging or inventory. Two of the hardest and only human capable things. Imagine not a neatly piled but a large heap of things. How does an AI/robot make sense of it? What is it? How many are there? Is it heavy? Moveable? etc...
Your ask is very large open ended question.
What exactly is your endgame?
> quite convenient using Google Photos too
Googl already has too much of me. The last thing I want is all of my pictures stored, triaged, scanned, archived. I'm pretty darn sure they are using them as a great ML dataset. The images you get off of general searches are junk compared to the goldmine of a personally curated gallery. Big difference.
Sorry for not being clear, I meant 'what do you want your camera to understand?
The reason I ask is because if this is not a pressing need for you, it might not be a pressing need for other people. And even if it is a pressing need for some other people, it is only for a few not for most people. Good luck.
My team has been working on a B2B visual recognition startup, but we think we may make our platform available for anyone. The idea is that anyone can make an image recognition app without coding. The app could be used as an object recognizer or a visual dictionary.
The app has image management and content management features for creators. Then public users can use its search function. If you have a team, you can manage access control within the app.
We are not in the situation where we can make automatic sign-up yet. Though the product is ready, our idea in terms of business is still very early. For now, you all can try our app, just give me an extra step to create an account for you.
App creators own the data and decide who get access or get to use their search model. We don't. But yes, the more recognition models are made, the more benefit public users will get out of it.
