How a computer sees
A camera follows a ball across a cricket ground. What does it do with each picture?
- Image acquisition: a camera takes the picture.
- Pre-processing: noise, the stray marks from a camera, and light are tidied.
- Feature extraction: features, such as edges and shapes, are found.
- Object categorisation: the picture is named, like cow or bus.
- Object identification, also called detection: each object is found and a box marks it.
- Object tracking: one object is followed through video.
It tidies each picture, finds the ball's edges, then follows the ball.
A feature is a clue in a picture. An edge is a line where the colour changes sharply.
A vision system usually tidies the picture first. Then it finds features, and names, locates or follows objects.
Locating an object means finding where it is.
Steps 4 to 6 are jobs that use the features. A system may do only one of them.
Pre-processing matters. Noise can look like a real edge. So the computer may find an edge that is not there.
A video is many pictures, called frames. Tracking follows an object from frame to frame.
Phones use these steps to unlock with a face. Doctors use them to study scans, such as X-rays. Farm cameras can use them to check crop health.
Switch the steps on. Watch what the edge step finds.
Loading lab…