Citilog brings automatic incident detection to existing PTZ traffic cameras that lack presets, using AI-based scene calibration to detect stopped vehicles, pedestrians, and debris without manual configuration.
Road operators have spent decades expanding video coverage on highways, bridges, tunnels, and other critical corridors. Pan-tilt-zoom, or PTZ, traffic cameras are central to that infrastructure because a single camera provides flexible visibility across large operational areas.
Yet, many of these cameras remain passive monitoring tools, useful only when an operator is actively watching the screen. Consequently, the industry has built far more camera coverage than automated detection coverage. For agencies pursuing Vision Zero goals, closing this gap without requiring a major roadside build-out has become a pressing priority.
The PTZ challenge
The primary obstacle to deploying automatic incident detection (AID) on PTZ traffic cameras involves contextual awareness. Standard computer vision algorithms require a defined digital road surface mask to understand where travelled lanes and shoulders are located, which help reduce false alarms from irrelevant background movement. With a fixed camera, that context remains constant.
PTZ cameras complicate that model because the scene changes whenever the camera pans, tilts, or zooms. The established solution has been preset positions. When a PTZ camera returns to a known view, the system loads the corresponding digital mask and analyses the scene much like a fixed camera. Citilog’s AID has supported preset-based PTZ cameras for more than a decade, including the broader incident detection set and traffic data collection when each preset has been configured.
However, that approach leaves a substantial population of traffic cameras disconnected from automated safety networks. Some PTZ cameras do not support presets. In other deployments, position feedback does not reach the incident detection platform. The operator has video, poles, power, and communications, but the analytics cannot reliably determine where the travelled roadway sits in the current view.
Automatic scene calibration
Citilog addresses that gap through automatic, AI-based calibration. Instead of depending on a predefined mask tied to a known camera position, the system observes live traffic movement, builds an understanding of vehicle trajectories, and infers the travelled roadway from that flow. The result is an automatically generated detection zone. In peak traffic conditions, the learning phase can take just a few minutes.
When the camera is moved to a different view, the system recognises the change and learns the new scene automatically, with no manual intervention. Once the scene is initialised, the system detects stopped vehicles, pedestrians, and debris within the generated zone and automatically reports incidents to the traffic management system. These detections are backed by Citilog’s deep learning models trained on more than 28 years of real-world incident data, which cut false alarms by up to 90% by filtering out shadows, reflections, and weather.
Because this update works with any traffic camera delivering a standard RTSP or ONVIF Profile S stream, agencies can extend AID without adding new cameras, cabling, or roadside infrastructure. For operators, that means earlier awareness of events using existing traffic camera networks that might otherwise depend on manual observation or external reports.
The life-saving potential of early automatic detection was demonstrated last year in the Vuache Tunnel in France, where a heavy goods vehicle stopped in the southbound bore before catching fire. An automatic incident detection alert prompted the control centre to close the tunnel before the fire fully developed, preventing other vehicles from entering and allowing the driver to be safely evacuated.
That lesson is increasingly relevant for traffic operations. The easiest and least costly step in improving roadway safety is adding software intelligence to existing infrastructure, not adding more hardware. As agencies pursue Vision Zero objectives under staffing and budget constraints, converting passive PTZ views into active detection coverage can help close the gap between simply seeing the road and reacting fast enough to save lives.
