Ideas for computer vision start-up

20th July 2017

So, you are interested in pursuing "deep tech" and want to know some mundane problems out there that you want to tackle with your startup. Below are some problems that can be solved with computer vision technology. Some are hard and some are pretty easy to solve.

By the way, check out some of my OpenCV with Golang tutorials as well after reading this article.

Time to retire the barcode? Maybe not, but add computer vision to help out.

Imagine, instead of relying on a human to print and stick barcodes on ... let say fruits, use computer vision technology to look at the fruits and tell you what it is. For an example, I pick up an orange and the computer vision scanner will be able to tell me that it is an orange from the US or South Africa or even better if I pick up the wrong type and it is tangerine instead of orange, the computer vision system will alert me. I always have trouble recognizing the type of fish that my mom asks me to buy from the market. Instead of taking photo of the target fish to buy and verify with my mom via WhatsApp, it would be better if I have an application that can help me out. If the marketplace has this kind of technology, yeah... it will be helpful to consumers like me. Maybe it will even increase sales and reinforce customer loyalty.

Smart shopping cart.

Ok, I walk around the supermarket with this smart shopping cart. It will be able to tell me how much the total weight and how much the items in the cart cost whenever I put some items that I want to purchase in the cart.

Crowd or computer vision aided purchase

I take a picture of an item that I like to know more and want to know where to buy. Upload the photo to an app or the web and the crowd will be able to tell me what it is and where I can purchase the item. If crowd wisdom is not accurate, maybe a computer vision software will be able to recognize the item and pull useful data out from its database.

Daily health scanner or fashion recommender

We all get naked before taking shower or bath. Why not build a computer vision system that will record the naked whole body each time before taking shower and alert us if there's any anomaly. Such as an extra mole or missing some hair. Even better, recommend us on what to wear after shower base on the clothes available in our wardrobe and a hash of our emotion, our location, the weather forecast and the state of the economy.

小胖车

Stock or inventory management

Instead of closing your store for few days to do inventory checking. Why not employ computer vision technology to do it real time and without closing the store. Whenever a customer takes an item off the shelf, the computer vision will be able to take note that an item has been deducted and log the changes into the database in real time. If the item in stock count hits re-order level, the system will automatically dispatch purchase order to the supplier for restocking. Nothing new about this, the technology is already out there and the current model uses weight instead of computer vision. Not that hard to implement, just need to train a classifier to recognize the target item and then use the classifier database to count how many items left on the shelf. For example, in this tutorial on how to count the number of faces or vehicles in a photo, it demonstrates how easy it is to count the number of faces or vehicles with a given classifier database. You just need to build a classifier database to detect the number of a target item.

Scan public area for known criminals or terrorist suspects.

Yup, this is what computer vision good at. Scan the public areas and cross reference faces with a database of known criminals. Apprehend the criminals when the opportunity arises. However, at the moment this kind of system can be easily defeated by "noise". The criminals just need to wear a mask or a t-shirt with a lot of faces on it. If your computer vision startup can silence the noise and have high accuracy rate, your system will be in demand by security agencies.

Christian Gertenbach

Scan for copyright infringement.

I got a t-shirt designer friend, Chow Hon Lam a.k.a FlyingMouse ( https://flyingmouse365.com ). He is a brilliant designer and every now and then he or his friends will stumble upon thieves that stole his design and resell the design without paying him royalty fee. Some of those thiefs are reputable fashion house as well. Perhaps you can create a computer vision software or crawler that will crawl the web to scan for copyright infringement and alert the owner. A startup is already working on this idea. See my blog post about Binded.

Want to know more what computer vision can do? Watch The Cirle Movie to explore some good ideas and ... some really creepy ideas.