Smart tracking system learns to recognise faces
1 min read
A young designer is working on a project to enable machines to 'see' as part of his PhD at the University of Surrey.
![](/media/1yidae1p/predator1.jpg?width=1002&height=564&bgcolor=White&v=1d9d7651e9525a0)
According to Zdenek Kalal, the Predator system can learn individual facial features, then recognise and tag the person – even from a photograph. The result is a real time tracking that improves over time.
Kalal aims to create self improving real time visions that can simultaneously track, learn and detect an unknown object in a video stream. His Tracking-Learning-Detection (TLD) system – also known as Predator makes minimal assumptions about an object, the scene or the camera's motion and requires only initialisation by a bounding box while operating in real time.
He has also developed a learning method that optimally combines boosting with bootstrapping and enables efficiently process of large training data sets. Using the method he built a real time multiview face detector.
His PhD advisors were Krystian Mikolajczyk and Jiri Matas, while the external PhD examiner was Andrew Fitzgibbon from Microsoft Research Cambridge.