• Solutions


May 10, 2023

Incident detection, automated

Traffic Technology International | May 10, 2023

Al can extend the ability of staff to monitor the roadways, but deep learning is essential to make the technology even smarter and to further improve traffic safety.

Forward-thinking cities around the world are using Al in traffic management to improve safety on roads and reduce congestion. Adaptive traffic control is one such example of Al that has been in use for decades, analyzing data from traffic sensors and adjusting traffic signal timings in real-time. More recently, predictive traffic analytics enable transportation agencies to anticipate traffic conditions and provide updates to drivers and operators via traffic management systems. The biggest challenge with most Al technologies is that they cannot match the keen eye and cognition of human traffic professionals, leading to false positives. Indeed, an unreliable Al algorithm is the equivalent of the boy who cried wolf, distracting from actual roadway calamities. Meanwhile most transportation agencies cannot realistically staff enough operators to actively watch all major arteries for incidents and respond in real-time to emergencies or congestion. That is where deep learning comes in. Here’s how it works: video-based traffic technologies powered by Al, such as Citilog’s Applied Deep Learning (ADL) for incident management, are fed thousands of actual examples of traffic incidents accumulated over more than 25 years, such as wrong way driving, stopped vehicles, smoke in tunnels, pedestrians and bicyclists, debris, slow vehicles, and congestion. This database of incidents trains the Al to be more effective and efficient at recognizing authentic events. It is the equivalent of sending your Al off to university for a degree in literal

The factor by which Citilog’s Applied Deep Learning (ADL) reduces false positive traffic incident alerts

  1. Citilog’s Applied Deep Learning (ADL) solution can run a specifically developed neural network targeted at eliminating false positive detections 2. The New York State Bridge Authority (NYSBA) uses Citilog’s ADL solution

street smarts. The result is a dramatic reduction in false positive alerts caused by environmental factors such as shadows, snow, rain, and other weather conditions, improving accuracy by a factor of 10. The technology can even identify wrong way drivers with cameras monitoring bidirectional roadways, and differentiate between bicyclists, pedestrians and motorcycles including within tunnels. Stopped vehicles can be classified as a car, van, truck, or motorcycle so operators quickly know from


alerts what type of vehicle to report to first responders without needing to analyze the video recording. The Maryland Transportation Authority in the US is using ADL today to proactively respond to incidents in mere seconds. Meanwhile, the New York State Bridge Authority is in the process of upgrading their incident detection system with Citilog ADL to improve operational efficiency and deploy emergency responders even faster. Ultimately, human intervention is needed for critical steps of the process such as verifying incidents, alerting the emergency responders, and the emergency responders themselves. Yet, Al enriched by deep learning can make staff responsibilities easier, faster, and more effective. When it comes to traffic accidents, faster response times can mean lives saved. Al offers immense benefits to the transportation industry, but Al needs deep learning to identify and minimize false positives. Deep learning has the potential to revolutionize the field of traffic safety and traffic management. Now that the technology is no longer theoretical, all it takes is for transportation agencies to incorporate the solution into their operations.