Computer Vision: How Robots Recognise What They See
A camera gives a robot a grid of numbers. Software finds edges, shapes and finally objects in those numbers.
The robot repeats this loop many times a second. Tap a step to pause on it.
To a robot, a photo is just a big grid of brightness and colour numbers. Computer vision looks for patterns in that grid: edges first, then shapes, then whole objects such as a cup or a person.
Modern systems learn these patterns from many labelled example images instead of hand-written rules.
A robot can sort parcels, spot a ripe fruit, read a sign or find the cup it needs to pick up, without a fixed position for everything.
The robot must be told exactly where everything is. Move the cup two centimetres and it misses.
Bad lighting, glare or an unusual angle can fool a vision system that worked fine in the lab.
What does a camera first give the robot?