Recently, the institute signed a memorandum of understanding (MoU) with Hyderabad City Police to gain access to video data from the city’s CCTV network. The technology is in the ready-to-be-deployed stage and the institute has already filed for its patent rights.
This is how it works: The solution is partially installed in cameras and partially on the servers of the central police control room. A software is also installed on an embedded card attached to CCTV cameras which helps detect violators (motorcyclists without helmets) by sending out an alert to the central alert database.
“It will be fully automatic along with a web interface to verify the alerts by the operators (traffic police, etc.). From there, it will be connected to the existing RTO website to generate challans (fines) and send a notification to the riders through SMS,” explained Dinesh Singh, one of the research scholars behind this solution.
Number of road accidents due to bike riders without helmets has been alarming. A Delhi Police annual report (released in 2017) revealed that of the total number of fatal accidents in the city in 2016, 35-40 per cent of the deaths were due to riders “not wearing helmets” or “poor quality helmets”.
“We have done significant lab experiments on sparse traffic at the IITH campus as well as dense traffic from Hyderabad city CCTV network. The results are motivating us to develop the complete software to deploy the system for realworld use,” said C Krishna Mohan, associate professor, IITH. Krishna along with research scholars—Singh and C Vishnu— are the trio behind this technology.
This solution can be deployed at the entrances of cities like toll bridges or on the selected checkpoints, especially road intersections.
Next step for the researchers is to get an industry partner on board for making the technology commercially available.
The solution uses convolutional neural network technology (applied to analyse visual images) which primarily mimics the human brain using AI. The proposed system can easily be extended to other kinds of traffic applications, like detection of tripling, zigzag bike driving, traffic violations by other kinds of vehicles, according to Mohan.
“Recent studies show that human surveillance proves ineffective, as the duration of monitoring of videos increases, the errors made by humans also increases,” Mohan said.
The research team has also developed a novel framework for automatic detection of road accidents in the surveillance videos. Mohan said that they have also proposed a framework for ‘Snatch Theft Detection’ in city-wide surveillance.