1. Challenges

Expressway traffic operations today face the triple challenges of high volume, high speed, and high risk. Traditional video monitoring is susceptible to weather and lighting, resulting in low recognition rates for subtle abnormal behaviors and frequent false/missed alarms. Incident handling relies on manual detection and reporting, often missing the optimal response window. Sensing, decision-making, and information dissemination are fragmented, lacking efficient coordination. Massive trajectory data cannot be quickly analyzed, and abnormal behavior mining lags behind—severely constraining the safety management and emergency response capabilities of expressways.

2. Our Solution

This system uses radar-video fusion technology as its core, integrating millimeter-wave radar, high-definition video, edge computing, and intelligent dome cameras to build an integrated “cloud-edge-end” incident linkage system. Radar accurately captures motion information while video synchronously extracts visual features. Spatial-temporal fusion generates continuous trajectories with unique identity. Edge computing enables on-site data fusion and incident assessment, achieving second-level detection and automatic alerts for incidents such as abnormal stop, reverse drive, congestion, and illegal lane change. Dome cameras are triggered for snapshot evidence, and warning information is pushed to VMS, broadcasting devices, etc., forming a full closed loop of “sensing – assessment – alert – handling – archiving”.

2.1 System Composition

• Front-end sensing layer: Millimeter-wave radar and HD video are fused. Radar collects vehicle position, speed, heading angle; fixed cameras extract license plate, vehicle type, and other visual features; dome cameras handle incident evidence capture—forming an all-weather multi-source sensing network.

• Edge computing layer: Built-in intelligent algorithms complete incident assessment within milliseconds, triggering radar-dome linkage for automatic snapshot evidence capture. Structured data is uploaded to the platform in real time, enabling rapid edge response and coordinated in-depth analysis at the center.

• Business application layer: The platform presents macroscopic network situations via digital twin, supports full-process incident closed-loop handling, and triggers alerts through VMS and broadcasting—forming a fast closed loop of “sensing – assessment – evidence – alert – handling – archiving”.

2.2 How it works

Radar continuously tracks lane-level vehicle trajectories, outputting real-time position, speed, heading angle, and other motion data. Video synchronously extracts visual features such as license plate and vehicle type. The edge computing unit performs spatial-temporal fusion to generate continuous trajectories with unique identity. Built-in intelligent algorithms complete incident assessment within milliseconds. Upon detection, the edge unit immediately drives the dome camera to pan, tilt, and zoom based on radar coordinates, capturing HD video and snapshots for evidence. The structured incident package is uploaded to the management platform via a dedicated network. The platform automatically matches pre-defined response plans based on incident severity, pushing alerts to VMS and broadcasting devices, and simultaneously notifying road administration, traffic police, rescue teams, and other relevant departments. After the incident is resolved, it is automatically archived.

2.3 System Functions

• Traffic data sensing: Continuously outputs lane-level trajectory data (position, speed, heading angle, acceleration) for full vehicle tracking and path reconstruction. Provides multi-section statistics: traffic volume, speed, occupancy, headway, etc. Outputs area status data including vehicle count, space occupancy, and distribution coefficient—building a precise data foundation for network monitoring and situation assessment.

• Traffic incident sensing: Based on radar-video fusion, achieves all-weather accurate identification and second-level alerts. Incident types include: abnormal stop, overspeed/underspeed, reverse drive, illegal lane change, congestion, pedestrian intrusion, and non-motor vehicle intrusion.

• Incident linkage handling: Once the radar locks onto an incident, it generates target information and pushes it to the edge computing unit. The edge unit drives the dome camera to automatically pan, tilt, and zoom for HD video and snapshot capture. Radar trajectory and video stream are fused to generate a structured incident package. The package is uploaded via dedicated network to the management platform for automatic categorization and archiving, supporting multi-dimensional retrieval and playback for accident accountability and traffic analysis.

3. Core Values

• All‑weather high‑reliability sensing – Radar unaffected by lighting, rain, snow, or fog, ensuring 7×24 uninterrupted accurate detection—fully compensating for video’s failure in harsh environments.

• Second‑level detection and closed‑loop handling – Incident detection completes within milliseconds; automatic dome camera evidence capture, VMS alerts, and multi‑department coordination form a complete handling loop—greatly improving response efficiency.

• Accurate identification with low false alarms – Deep fusion of radar trajectory and video features effectively filters environmental interference, significantly reducing false and missed alarms.

• Low‑latency edge computing – Incident assessment is offloaded to roadside edge nodes, avoiding data transmission delays and ensuring millisecond‑level response.

• Closed‑loop evidence chain with full traceability – Incident processes are automatically linked with radar trajectory replay and HD video recordings, forming a complete evidence chain to support post‑event accountability and traffic analysis.