Computer Vision: Applications, Challenges and the Future

Hello students, welcome to Kwickprep. Your phone unlocks when it sees your face. But why does it sometimes fail in a dark room? Today we will see where computer vision is used and why seeing is hard for a machine. We will also cover privacy and the future.

Let us quickly recall the meaning. Computer vision is the domain of AI that helps computers understand images and videos. A computer does not see a picture like we do. It sees a grid of tiny dots called pixels, and each pixel is a number. A trained model learns patterns in these numbers, so it can recognise objects, faces and text.

Before applications, let us name four common tasks. Classification answers, what is in this image, for example a cat or a dog. Object detection answers, what is here and where is it, like finding every car on a road. Segmentation marks exactly which pixels belong to an object, like the outline of a tumour in a scan. Recognition answers, who or which one is it, as in face unlock.

Our first application area is healthcare. Vision models study X-rays, CT scans and MRI scans, and mark areas that look unusual, such as signs of tuberculosis in a chest X-ray. In India, AI tools screen photos of the eye to find damage caused by diabetes, which helps where eye doctors are few. Remember, the model only helps the doctor; the final decision stays with a trained doctor.

Next, retail, which means selling goods in shops and online. Some stores use cameras at self-checkout counters that recognise products, so you do not need to scan every barcode. Cameras on shelves notice when a shelf is empty or an item is in the wrong place. Shopping apps offer visual search, where you click a photo of a shoe and the app finds similar shoes.

Now agriculture, which matters a lot in India. A farmer can click a photo of a sick leaf, and a phone app suggests which disease or pest it may be. Drones fly over large fields and their cameras show which parts of the crop are weak or dry. In mandis and factories, machines use cameras to sort fruit and grain by size, colour and damage.

Our fourth area is security. At many Indian airports, DigiYatra lets passengers enter by face recognition instead of showing a boarding pass again and again. Traffic cameras read vehicle number plates automatically, which helps with e-challans and parking systems. Smart CCTV systems can raise an alert when they notice unusual activity, such as a person entering a closed area at night.

Pause the video and think about this. A face unlock works perfectly at noon in a bright garden. Now the same person tries the same phone at night, in a room with no light. Will it still work, and why? Keep your answer in mind, because the next slide explains it.

So what makes seeing hard for a machine? Lighting changes every pixel number, so a dark room makes the same face look new. That is why our night unlock may fail. Viewpoint or angle changes the shape, so a face from the side looks very different from the front. Occlusion means part of the object is hidden, like a mask covering half a face. A cluttered background, like a busy market, makes it hard to separate the object from everything around it.

The third big challenge is bias. Bias means the model gives better results for some groups of people and worse results for others. It usually happens because the training images do not represent everyone, for example mostly light-skinned faces or mostly city roads. Studies have found some face systems make more mistakes on darker-skinned women, and that is unfair. The fix is to collect diverse data and test the model separately on every group before using it.

Now, privacy, which means your right to control information about yourself. Cameras can capture and identify faces in public places without asking the person. When many cameras are linked, a system can track where someone goes all day. If a database of faces is stolen, you cannot change your face the way you change a password. In India, the Digital Personal Data Protection Act of 2023 treats such data as personal data, so it needs a clear purpose and consent.

So how should computer vision be used responsibly? First, ask for consent and clearly tell people why their images are collected. Second, collect only the images that are needed, store them safely and delete them when the purpose is over. Third, keep a human in charge of serious decisions, such as arresting someone or refusing treatment.

Finally, where is computer vision heading? Edge AI means the model runs on the phone or camera itself. This is faster, and images stay private because they are not sent to a server. Multimodal AI models can look at a photo and answer questions about it in words. Self-driving vehicles and robots are learning to understand whole scenes, not just single objects. AR glasses and 3D vision will let devices understand rooms and places around us.

Let us recap what we learnt today. Computer vision is used in healthcare, retail, agriculture and security. It struggles with lighting, angle, hidden parts and biased training data. Privacy matters because faces can be captured and tracked, and face data cannot be reset. The future brings edge AI, models that see and talk, smarter robots and AR.