Teaching computers to count
Estimating crowd size is hard. It takes humans a lot of work and there are often disagreements about the final count.
Being able to automatically estimate crowd size can have enormous public safety benefits for crowd control. Moreover, the techniques can be adapted to other areas such as counting bacteria in a petri dish or pepperoni on a pizza.
This project aims to teach computers to look at a picture of a crowd of people and almost instantaneously figure out how many people are in the crowd. To do this, we employed deep learning – a type of artificial intelligence – to train a neural network to recognise people in a crowd and highlight each one. The result is a system whose counts are very accurate and whose ability to highlight individual members of a crowd exceeds current state-of-the-art systems.
Theme
Securing our future
Booth
SF25
School
Computer Science
Exhibitor
Avraham Chapman