Smart Cities: Reducing Congestion with Deep Learning

In India, manual traffic management has become highly impractical due to rapid urbanization. Additionally, central monitoring systems are facing scalability issues as they process increasing amounts of data received from hundreds of traffic cameras. Deep learning experts at Siemens Corporate Technology are developing an intelligent traffic management solution to tackle these problems.

 

by Tomonica Chandran

The Central Silk Board Junction in Bengaluru, India is one of countless intersections in the city that is regularly blocked by huge traffic jams. However, such problems might soon become a thing of the past. A team of deep learning and machine intelligence experts at Siemens Corporate Technology is developing a traffic management solution that fully automates traffic control and monitoring. A prototype is currently being tested at the Electronic City campus in Bengaluru in collaboration with the Electronics City Township Authority (ELCITA). Rama NS, CEO of ELCITA, is confident the solution will significantly improve traffic management in Bengaluru. “It will provide us with traffic information that we don’t currently have, and help improve our management of commuter traffic,” says Rama.

Automatic Traffic Light Control in Real Time

The solution developed by Corporate Technology captures video streams from several cameras installed around ELCITA and processes them using deep learning techniques based on artificial intelligence. As the processing takes place at intersections themselves, this avoids the delays experienced with previous solutions in which video streams from hundreds of traffic cameras are sent to a central cloud environment to be processed and monitored. With the new approach, typical traffic management tasks such as vehicle detection, traffic density estimation and control of traffic lights can be automated for real-time performance. Where necessary, however, the solution can also be used to process video streams in a central cloud environment – to obtain a complete overview of the traffic situation in the city, for example.

Pedestrians, motorcycles or cars – as the Siemens' solution evaluates the camera images, it also classifies the depicted objects in real time.

Autonomous Driving on chaotic City Streets

David Borst from the MindSphere Application Center at Siemens Mobility has taken a keen interest in the Bengaluru tests. “The better we understand traffic patterns, the better we’ll be able to manage them,” he explains.

 

Vinay Sudhakaran, a Senior Key Expert on infield learning at Siemens Corporate Technology in India, can even envisage applications beyond traffic management. For instance, he points out that the ability to identify different vehicle types can be useful in urban planning. “Deep learning techniques can also be employed to detect accidents and automatically notify police and ambulance services. The technology could also capture visual evidence of traffic violations, extract vehicle numbers and automatically generate traffic tickets.”  Moreover, he adds that developments in vehicle detection technology can bring autonomous navigation closer to reality, even on chaotic city streets.

Green Corridor

ELCITA field trials are currently in progress with promising results revealing realistic and consistent figures for traffic density. “We are now ready to move on to a simulation environment where we want to find out how artificial intelligence can manage traffic signals at multiple intersections in order to achieve a green corridor effect,” says Vinay. “However, we will need to optimize our deep learning models in order to do this. We are looking to the Indian Institute of Science to support us, and a collaboration agreement has already been signed.”

2018-04-13

Tomonica Chandran

Picture credits: from top: 1. Picture gettyimages

Subscribe to our Newsletter

Stay up to date at all times: everything you need to know about electrification, automation, and digitalization.