Video Traffic Detection

Definition and Technical Principles

Video traffic detection is a non-intrusive monitoring method based on computer vision technology, utilising roadside-mounted cameras to capture traffic footage and automatically analyse and extract traffic parameters via advanced algorithms. Unlike conventional detection systems such as inductive loops that require slot-cutting into road pavements during installation, video detection relies on optical cameras to capture visual feeds, applying image processing algorithms to extract crucial data including traffic volume, vehicle speed, and lane occupancy. Today, video sensors account for up to 60% of all roadside data collection hardware, establishing themselves as the mainstream technology for traffic flow monitoring.

Core Detection Models

The current mainstream technical framework for video traffic detection relies heavily on deep learning object detection models. The YOLO (You Only Look Once) family is among the most widely deployed real-time object detection algorithms. As the latest iteration in the series, YOLOv8 delivers significant gains in both accuracy and processing speed, enabling rapid vehicle detection and classification directly from live video streams. Academic research highlights that YOLOv8-based systems can seamlessly perform vehicle detection, speed estimation, tracking, and attribute recognition (including vehicle colour and size profiling). In expressway incident detection scenarios, pairing YOLOv8 with multi-object tracking algorithms like Deep SORT allows real-time calculation of critical safety metrics—such as time-to-collision, time-to-occupancy, and crash-avoidance deceleration—for risk assessment.

Another prominent framework is Faster R-CNN, which generates candidate regions via Region Proposal Networks (RPN) before performing classification and bounding-box regression, offering superior localisation precision in vehicle detection and classification tasks. For expressway traffic incident monitoring, algorithmic models combining YOLOv5s and DeepSORT have likewise demonstrated verified operational capabilities in tracking traffic volume and vehicle speeds.

Detection Capabilities and Application Scenarios

The functional capabilities of video traffic detection systems span several operational dimensions. In traffic parameter acquisition, these systems facilitate vehicle detection, classification, volume counting, and speed estimation. Field trials utilising roadside video streams demonstrate that YOLOv8 paired with the ByteTrack tracking algorithm achieved a mean absolute error of just 5.33 km/h and a marginal bias of +0.67 km/h in speed estimation. In expressway operations and management, video analytics technology underpins automated number plate recognition (ANPR), traffic volume monitoring, incident detection, and road environment analysis. Drone-mounted video analytics software deployed by traffic police can identify nine distinct traffic offences—including illegal parking, lane straddling, and emergency lane misuse—alongside six types of traffic incidents such as congestion, roadworks, and pedestrians on the carriageway, delivering an abnormal stationary vehicle detection accuracy of at least 70%.

Non-Intrusive Advantages and Limitations

As a non-intrusive technology, the primary advantage of video detection lies in its comparatively lower installation and maintenance overheads. Compared to inductive loop sensors that necessitate destructive pavement cutting and are vulnerable to road surface degradation, video detection achieves continuous, wide-area, multi-lane surveillance without compromising the road structure. Nonetheless, video-based systems face distinct physical limitations: detection accuracy can be markedly degraded by adverse lighting, severe weather changes, and vehicle occlusions. Furthermore, conventional background modelling techniques, such as frame differencing and Gaussian Mixture Models (GMM), remain prone to missed detections in slow-moving traffic or visually cluttered scenes. Current research is focusing on bolstering system robustness through advanced techniques like spatio-temporal video analytics to effectively tackle the challenges posed by adverse weather and complex driving environments.