He told the NVIDIA AI Conference in Sydney about how the use of several Jetson TX2-based edge computing devices and LPWAN networks can be used to monitor in real time the flow of vehicles and pedestrians in a network.
The conference is the premier event on artificial intelligence and deep learning, and showcases the latest breakthroughs from universities, start-ups and major enterprises in a wide range of fields such as smart cities, autonomous machines, virtual reality and more.
“Each device in the monitored network processes the live feed from its own camera,” he told the conference.
“Each frame is analysed by an object detector (YOLO v3) to extract the pedestrians and vehicles in the frame.
“Then the detections passed to a tracker algorithm (based on a Kalman filter) to determine their trajectories.
“Once a frame has been processed, it is discarded and only aggregated indicators are sent over the LPWAN network to a dashboard (number of detections, types, trajectories), limiting the privacy issues and the bandwidth requirements.
“This solution is a key component of a project aiming to better understand and predict the pedestrian and vehicles flows around the Liverpool CBD in order to ease congestion, provide better transport options and improve health and safety.”
Liverpool City Council is teaming up with UOW and IT integration company, Meshed, to measure pedestrian and vehicle movements around the CBD based on CCTV images, and data from smart phones.
The 13-month project started in March and was made possible by a Smart Cities and Suburbs grant for $120,000 from the Federal Government and matched dollar-for-dollar by Liverpool City Council.
SMART researchers Dr Nicolas Verstaevel and Dr Johan Barthélemy are developing an algorithm that will allow the city’s existing CCTV cameras – as well as additional cameras – to recognise and count pedestrians and vehicles.
Dr Barthélemy is an early career researcher at SMART and the leader of the Digital Living Lab & SMART IoT Hub where he develops sensors for various applications such as smart cameras, connected beer kegs, water level sensors and low cost gas sensors.
In addition he is also focusing on the development of new tools and frameworks to simulate large scale agent-based micro-simulation.
You can watch Dr Johan Barthélemy’s presentation here