A different outlook
Concordia Institute for Information Systems Engineering professor Nizar Bouguila is developing ways to enable computers to recognize and understand what they see in real time. This research is expected to improve video surveillance and security in public places.
“Visually, a human can easily and almost immediately differentiate between objects. When we see a car, we know it is a car,” says the Tunisian-born Bouguila, who’s been at Concordia since 2006. “Computers cannot do this with such ease and speed. So, how can we add this intelligence to machines?”
Existing object recognition software computes a complicated amalgam of thousands or millions of identifiable features, a process that requires valuable time. Bouguila’s software detects a few hundred essential features of any object – shape, form, texture, colour, edges, lines, etc. – and identifies their meaning in combination.
Since only the integral features are detected, the application can be used in real time.
The speed and simplicity of the software holds the potential to significantly improve video surveillance. The flow of people through airports, public streets or buildings could be made significantly quicker as faces or abnormal packages could be scanned and processed almost instantly. This, Bouguila believes, will also make these places safer.
Not limited to security, Bouguila says the application can be used in a myriad of different ways; for everything from detecting a specific license plate in moving traffic to video editing. “The software is able to detect a character or any object, delete it and automatically replace it with another desired object,” he says.
Part of Bouguila’s research won the best vision paper award at the Seventh Canadian Conference on Computer and Robot Vision (CRV) in Ottawa last spring.
The award-winning paper “Unsupervised Feature Selection and Learning for Image Segmentation” was developed in collaboration with Mohand Said Allili from the Université du Québec en Outaouais, Djemel Ziou and Sabri Boutemedjet from the Université de Sherbrooke.
Related links:
• Concordia CIISE
• Computer and Robot Vision conference