Citilog harnesses AI deep-learning

First published at ITS World Congress - October 23, 2019

Jean Marie Guyon of Citilog

False alarms have long been an issue with highway monitoring systems that incorporate automatic incident detection; now Citilog has launched what it is calling CT-ADL which uses applied AI to reduce false positives by as much as 90%.

CT-ADL (Citilog Applied Deep Learning) is powered by the company’s purpose-built image recognition neural network and can differentiate between actual accidents and occurrences that frequently generate false alarms. Citilog is a subsidiary of Sweden’s Axis Communications, whose highway monitoring systems have been collecting data for many years.

According to Jean-Marie Guyon, Citilog’s VP sales and marketing, a wide range of common occurrences – shadows, reflections, heavy rain – can be frequently incorrectly flagged by current automatic incident detection (AID) systems.

“AID systems have been around for decades, and false alarms have been an issue since they first appeared,” he says.

“We decided the solution to this was deep learning, a subset of AI. With many hours of highway monitoring video footage collected by Axis, we had access to huge amounts of data.”

Over the past two years, Citilog used this data to train its neural network to differentiate between real incidents and accidents, and false alarms.

“Our approach was to apply AI and deep learning to overcoming an ongoing issue for highways authorities, and we’ve shown we can reduce the occurrence of false alarms by 90%,” says Guyon.

Citilog is now looking at other applications for the concept.

“There is also potential to apply CT-ADL to additional fields, such as identifying incidents relating to pedestrians, bicycles and other elements in the smart cities environment,” he says.

Booth 467