EVERY year, millions of pilgrims flood to Mecca in Saudi Arabia to perform the fifth pillar of Islam – the Hajj. Sadly, some go missing or die amid the faithful throngs, and in a gathering of such size, finding or identifying those people is difficult.
It might be about to get easier, thanks to a face recognition system being developed at Umm Al-Qura University in Mecca. Using photographs taken of every pilgrim that enters the country, the system will help the authorities scour CCTV images for the faces of the missing, or help them identify the deceased.
The system is part of a broader project called Crowdsensing – funded by – which aims to predict and manage the flow of pilgrims using a network of 1500 CCTV cameras in and around Mecca’s Grand Mosque (Masjid al-Haram). The large crowds have led to several disasters in recent years, including a stampede that killed 362 pilgrims in 2006.
Advertisement
The face-recognition software works by first analysing all of the faces in a database and highlighting features that vary most from face to face – these are the most useful for searches. When an unknown face is submitted for identification, its features are then extracted and compared to the database. For people presumed missing in the crowd, the software will scan CCTV images in search of a match.
So far, the system has only been tested on small groups of photos, including a collection of six images for each of 100 people who made the pilgrimage last year. When trained using two of the six images of each person, the system correctly identified another of the images, taken from a different angle, 93 per cent of the time.
Lead researcher says the completed system will eventually hold images of each person who enters Saudi Arabia for pilgrimage every year – roughly 2 million people. As well as allowing the authorities to use it to match entry photos to CCTV images to look for missing people, the plan is to allow concerned family members to upload an image of a missing relative through a web browser.
Running automatic face detection on a database containing so many people will be a world first, says Mark Hansen, a computer vision researcher at the University of the West of England in Bristol. However, he points to some unanswered questions, such as how the software will analyse images if subjects’ faces are obscured or their appearance changes. Men traditionally shave their heads during the Hajj, for example, which could throw the system off.
