Think about the way Facebook recognizes your friends’ faces when you upload a photo. To do it, computers had to learn to see like us, to understand and interpret what’s going on in a picture or video. This is called computer vision—a part of the larger fields of machine learning and artificial intelligence (AI).

Let’s explore how computer vision might assist humanitarian organizations (it already does), some of the issues of bringing computer vision to middle- and low-income countries, and how ICT can help overcome those.

Are you interested in how AI can be used for further international development and research? Then you’ll be delighted to know that this blog post is part of our ongoing series “AI x ICT” where we explore how AI and ICT together can move humanitarian efforts forward.

A few examples

Getting back to computer vision, when looking around today’s world it becomes clear that computers can already “see” for a while, and this has spawned many uses—not only “recognizing friends”. Here are a few of these uses:

  • Google Drive makes your photos searchable with text queries—you can search for “car” and it will actually find all your photos that contain a car
  • Tesla uses computer vision for its Autopilot feature, so that its cars can get you from A to B without you needing to do all that much—though I hope you’re still paying attention to the road!
  • Remember that funny app that drew bunny ears on your head while using the selfie camera? Yup, computer vision.
  • Alibaba, the Chinese e-commerce giant uses “smart mirrors” to suggest fashion styles to you.

Humanitarian Computer Vision

So, computer vision is alive and well … in the commercial sector. But what about the humanitarian world? Is computer-vision of any use there?

Absolutely, humanitarian projects bring computer vision to bear in many ways:

  • PlantVillage uses computer vision to look at plant leaves to detect diseases, teaching farmers how to spot this themselves and what to do about it. Read more about PlantVillage on the FAO website.
  • The Humanitarian OpenStreet Map (HOT) uses computer vision to identify for example houses in satellite images. (We talked about this as an example of humanitarian AI before here.)
  • Wadwhani AI applies computer vision to e-healthcare in various ways. For example, it helps diagnose skin conditions such as monkeypox, scabies, eczema, or psoriasis.

Challenges in Low-Income Countries

Applying computer vision, or any high-tech solution to low- or middle-income countries brings challenges. I’m sure you’re familiar with them, so I’ll keep this brief with a few examples:

  • Computer vision needs an image or a video—that means a camera. While smart phones combine cameras with an internet connection they’re not exactly available everywhere—but they set the bar for any computer vision solution.
  • Data. Affordable internet connectivity is not a given everywhere. In the AI world, this is especially a problem for computer vision, since images and even more so videos eat lots of data.
  • Assuming a farmer has a smartphone, and they have the data to spare to send a picture … where do they send it? Upload it to Dropbox? (And then?) Is there a custom app?

What are ways that we can work with what is there?

Computer Vision Meets ICT

Assuming the existence of smartphones with occasional internet, how can we address the issue of connectivity and communication?

One option is to keep things locally. (Also called “edge computing”.) The PlantVillage project above went this route, developing custom apps that can do the computer vision part locally. This option is awesome—where it is feasible.

Downsides include costly custom app development for different platforms, the update problem, and you obviously can’t use large AI models (the “brains” that you feed thousands of documents, videos or images to).

That’s where messaging apps come into play. Many people in low-income countries use apps like WhatsApp already. They are easy to use, and people are familiar with them.

Imagine the farmer popping up WhatsApp, messaging a chatbot, and selecting “identify disease from photo” from a menu. The farmer then sends a picture they took earlier that day and sends it over to the chat bot in WhatsApp. Within seconds the reply comes in, identifying the disease, and recommending steps to take.

Using messaging apps like WhatsApp to facilitate the communication gives you an off-the-shelf solution to connect computer vision prowess with the people who need it.

The AI Gap

This “off-the-shelf” part is important. Not just for the people that are being helped with computer vision, but also for the humanitarian organization in the middle of it.

After all, AI magic is great, but if only Microsoft and a handful of giant mega corps can afford to use it then it’s really not all that impactful.

Only when even a small, local NGO on a budget can leverage AI’s magic to help its constituents will AI live up to its potential—and transform the lives of everyone.