31 Oct 2019 | Blog
Israel-based technology company NoTraffic is using AI to transform intersections from danger zones to intelligent decision makers, cutting time delays and carbon dioxide emissions.
Next week, NoTraffic’s Yoav Valinsky, a computer vision researcher, will be going to GTC DC to discuss how the company is operating at the edge in a presentation called “From Theory to Practice: Computer Vision on Edge Devices for Real-Time Optimization.”
Founded by Tal Kreisler, Or Sela and Uriel Katz, NoTraffic uses AI sensors at intersections to analyze traffic and optimize traffic lights.
This proactive approach contrasts with today’s usual intersection technology, such as inductive-loop traffic detectors. Induction loops are installed underground — making it challenging to upgrade or replace — and work like metal detectors to sense cars.
Those traffic detectors are also constrained by the intersection’s fixed time plan. While they can minimize or maximize a light’s duration, the detectors can’t override a light — even if there are no cars coming in that direction.
Using AI at the edge, NoTraffic’s system reduces the cost of installation and maintenance, and gives intersections the ability to prepare for vehicles rather than just react to them.
NoTraffic starts by installing AI sensor units aimed in every direction of an intersection. The units use the NVIDIA Jetson platform and GPU-accelerated frameworks to fuse machine vision and radar, processing data roughly 15 times a second, according to Valinsky.
The sensor units also integrate connected vehicle capabilities based on dedicated short-range communications (DSRC) and cellular vehicle-to-everything (CV2X). DSRC is a system of wireless communication channels between vehicles and infrastructure. CV2X technology provides communication between vehicles, infrastructure and any related entities.
NoTraffic’s units detects and classifies all road users — including cars, buses, trucks, bicycles and pedestrians — at the edge. The processed data is then streamed to an optimization engine installed in the traffic signal control cabinets that are already present at most intersections.
There, the data is used to optimize and manage traffic lights, both at the individual intersection and across a city grid. By placing compute closer to the point of action, NoTraffic’s edge system saves bandwidth and lowers latency for faster calculations.
NoTraffic securely sends data from each intersection to the cloud for further processing and city-grid optimization. It also presents the information in a dashboard designed for city engineers, who can use it for big data analytics, remote monitoring of intersections and the implementation of new traffic policies.
Analyzing in real time provides capabilities such as collision prediction. NoTraffic’s Director of Business Development Ilan Rozenberg explained that the sensor units calculate the speed, acceleration and direction of vehicles, so they can infer when two cars can’t see each other and will probably collide.
The sensors’ ability to classify vehicles also makes it possible to prioritize certain road users. If a city wanted to prioritize public transportation or pedestrians at the intersections surrounding schools on weekday mornings, city engineers would input that policy in their dashboards. NoTraffic’s system would make the changes autonomously.
The company, which is a member of the NVIDIA Inception startup incubator, is currently focused on the U.S. and conducting pilot projects in several cities and counties across the country. Annually, NoTraffic is reducing delays by an estimated 2,700 hours and preventing 33 tons of carbon emissions per intersection.
The company looks forward to making its technology even smarter — the longer it’s implemented, the better it’ll be able to predict vehicle and pedestrian patterns.
To learn more about the AI powering NoTraffic, register for GTC DC, Nov. 4-6.
