Image processing for apts, hotels, airbnbs, wework, etc.

The Big Picture

Designing a new house that leaves half the carbon foot print of a traditional house or costs half as much is hard if not impossible. What is possible is two people sharing the same one. This doesn't sound like it takes any advanced technology does it? Why don't people just share all their things if it's so great?

Well for one thing I might be using my toy right now, but if I'm not I don't want someone to break my toy. The same is true with a lot of other things. There are many successful apps and companies being formed around the concept of sharing things people once just hogged to themselves like Airbnb, Uber, Lyft, Zipcar, Wework etc. They don't just help us share anything, but the most valuable things we own like vehicles and real estate. What happens when someone breaks a thing they borrowed from me? How do you get them to pay for it? How do we know if it was actually broken? Answering these questions will be an important as these sorts of companies continue to grow.

I set out to solve this problem for real estate rentals. My basic idea is take a picture of the room before renting it out, then take a picture after they leave and provide a platform to assess and dispute these damages using technologies like OpenCV and Tensorflow.

The Hardware

So the first question I considered is how to take these pictures. How many pictures do you need. If you are simply photographing one room that's 4 walls, a ceiling, and floor that's at least 6 photos if you can capture each of those in a single picture which is hard to do based on the angle most people hold cameras and probably would be too inconvenient for most people to bother with. So I figured it would be easiest to use a 360 camera that had high enough quality to capture the details of most small rooms. I choose the Ricoh Theta V that has an sdk to easily control with a mobile app.

The next problem was figuring out how to take pictures consistently from the exact same spot and angle which is necessary to be objective and to make image processing with OpenCV the easiest and also make people feel like they have privacy and are not being filmed constantly. I didn't want to drill a hole in my wall just to try mounting my camera at different angle and realized that many potential landlords using this product might feel the same. So I bought a couple of command hooks for the adhesive backs

and 3d printed an attachment to use instead of a hook that would allow you screw the ricoh theta v onto it and consistently attach it and remove it in the same place at the same angle every time without any tools.

The Software

I made an Android app to connect to the camera to take the pictures with a wifi connection so you don't have to press a button on it and can leave the room and photograph it perfectly, then host them online since it may be weeks or months between taking the pictures, then once the second picture has been uploaded run a script that use OpenCV on them to try and identify the differences.

Once the differences have been found, crop those sections and organize a gallery of potential problems and present them to the landlord to review. Have them describe the damage and assign a cost to fix it and then send this itemized list with evidence to the former tenant to dissuade them from challenging anything in small claims court or taking equivalent legal actions.

After collecting data for several months from all the landlords using this app I would be able to train an AI model to classify these damages and how much they should cost using a framework such as Tensorflow. This would streamline the data entry process for landlords by guessing what they would consider a problem and how much they would charge similar to how Letgo attempts to name and price pictures of what you are trying to sell after you only take a single picture. Every state and countries has their own mess of regulations about what can be deducted and for what reasons a machine learning algorithm could be very helpful in navigating these problems.

The Business

I would originally offer this application for free, but once I had collected enough data for a good AI algorithm I would charge a small fee to landlords for each generated list of damages.

Right now these sorts of high quality 360 degree cameras cost at least $300 so that barrier alone would probably stop most people who only rent out a single room in their home on Airbnb from using this product. As these 360 cameras become more common and cheaper that could easily change in the next few years. The first customers I would target would be hotels and apartments that manage hundreds of units that would only need to buy one or two 360 cameras to manage them all.

Most parts of this idea would probably be difficult to protect with intellectual property laws, but training an AI model like this would be where the true value that would be difficult to replicate by competitors. Also as these pictures would be taken months or years apart landlords would feel locked into this specific product and as if there would never be a good time to switch because trying to just keep track of these before pictures for such a long period of time would be a hassle.

This would also collect valuable data for insurance companies and these tech companies offering rental platforms.

This might not be solution that every company that rents property would want or use but real estate is one of the largest industries in the world across every country and culture and if even a small fraction of businesses could be convinced an application like this has value creating it would be a worthwhile business venture.

The rules said I had to make a 3 minute video so here is my 3 minute video which says a lot less than I have already told you.

Thank you for considering my application I hope to have a functioning prototype by the time I see you in July.