What is computer vision?

We live in a world that wants to make everything smarter, making everything smarter also means reconsidering the place of the human/employee in many cases. Does it look like we are living in a science fiction movie? That is possible, but hopefully with a better ending than in most cases when the world is taken over by robots.😅

Making certain processes smarter also has some advantages, human mistakes may become a thing of the past, we can make predictions, switch much faster within certain processes and so much more. Is the future that computers will take over many human tasks? The question is rather when do we start.

A hot topic is now artificial intelligence, computer vision and machine learning. You have probably heard the radio commercial from Aion bank that uses artificial intelligence to help you manage your money and reduce costs. It is the first financial institution to take a different approach and to be very innovative. Would you like to know more about how artificial intelligence, computer vision and machine learning actually works? Read more under the picture.

What is Computer Vision?

Computer vision falls under artificial intelligence, which is the all-encompassing term. Specifically, it means that our computers, like you and I all have one or more, take over human tasks. The computer is, as it were, able to identify objects and on the basis of that it makes a decision and carries out the action.

Did you know that unlocking your smartphone with your face works via computer vision? That's why our smartphone no longer recognizes our face when we wear a mouth mask. The features and facets of the image, our face, that is already in the memory do not match the face that is shown, there is a nose, mouth and chin missing that defines the edges.

Tip to have your face with mouth mask recognized by your mobile phone.😉

Image for post

What is the connection between computer vision and artificial intelligence?

Computer vision is part of artificial intelligence. It is concerned with giving 'vision' to computers, understanding an image in other words. The aim of computer vision is to teach computers how to identify, classify and categorise the visual world. Just like we do with our eyes.

How does computer vision work?

We look at the world and we all see images. Computers don't see images but only numbers, binary numbers. An image is made up of pixels, each of which has a numerical value. Computer vision converts each pixel that makes up an image into binary data (combination of zeros and ones).

Training a computer vision system

A computer vision system should be powered just like a child. A computer vision system is not fed with peas and carrots but with data, a lot of data. Without prior training, the computer vision system cannot know what certain objects look like, what their characteristics are and how they are constructed. That is why thousands or millions of images of specific objects are presented to the system. In this way, the system can gradually learn the difference between a bird and an airplane.

We need to feed the system with sufficient photographic material so that it can understand the exact nuances at pixel level, which then define the individual components of the larger image. The system uses that information to form an idea of what a bird, for example, is. In turn, the system can predict that anything with two legs and wings should be labeled a bird. As explained earlier, computer vision does not see the two legs and wings of the bird, but looks for patterns in the binary values of the pixels. Just like in the game 'search the same or memory' it searches for the same binary combinations.

Image for post

Repeat!

At some point the output has reached an acceptable and accurate level. However, repetition remains important to make the system smarter.

With many repetitions, the system has an idea of what a photo of a bird, for example, should contain. If you show the system a new image and want to know if there is a bird on that image, then the system compares each pixel of the image with all those other images of birds with which it has been trained. If the input meets a minimum threshold of comparable pixels then the artificial intelligence explains that it is a bird.

In practice

A few weeks ago there was a lot of commotion on the beach in Blankeberge. Amok makers caused a brawl between themselves and the police. Mayor Lippens of the other seaside resort, Knokke-Heist, reacted immediately. He repelled all day-tourists who wanted to visit our beautiful North Sea.

A few weeks before this event we installed some smart cameras at the rescue post in Knokke-Heist. The smart cameras keep an eye on the crowds and count how many people are now on the beach. The privacy of all people in bikinis, swimming trunks or shorts is 100% guaranteed! We don't transmit the images, they remain safe on the camera itself. What is transmitted is the result, so the number of people on the beach. The images remain in the safe environment of the camera.

After the incident in Blankeberge and after keeping out the day tourists in Knokke-Heist we could observe a nice drop in population on the beach.

Image for post

How does this work technically?

A camera, connected to Ethernet, takes a picture at that x-time. Using an algorithm, it recognizes people and counts them. This result is sent to Azure. In terms of hardware, we use the NVIDIA Jetson computer. This is TPU (Tensor processing unit). Thanks to its strong graphics processor, it is ideal for artificial intelligence. This TPU runs our IoT edge with Python.

The challenge!

A camera that recognizes and counts people sounds very easy. If you know, apart from the current situation with corona, that our beaches are always crowded and there are a lot of people on one square meter, it soon becomes less easy for the camera to recognize every person.

But we did it! Our detection of the people on the beach is extremely good. So we easily recognize 100 people in the picture over a distance of about 100 meters!

Image for post

Aptus to the rescue!

Do you have your own challenge for us, or are you simply mega interested in this technology? Let us know!