DARTS: A Drone-Based AI-Powered Real-Time Traffic Incident Detection System

Rapid and reliable incident detection is critical for reducing crash-related fatalities, injuries, and congestion. However, conventional methods, such as closed-circuit television, dashcam footage, and sensor-based detection, separate detection from verification, suffer from limited flexibility, and require dense infrastructure or high penetration rates, restricting adaptability and scalability to shifting incident hotspots. To overcome these challenges, we developed DARTS, a drone-based, AI-powered real-time traffic incident detection system. DARTS integrates drones’ high mobility and aerial perspective for adaptive surveillance, thermal imaging for better low-visibility performance and privacy protection, and a lightweight deep learning framework for real-time vehicle trajectory extraction and incident detection. The system achieved 99% detection accuracy on a self-collected dataset and supports simultaneous online visual verification, severity assessment, and incident-induced congestion propagation monitoring via a web-based interface. In a field test on Interstate 75 in Florida, DARTS detected and verified a rear-end collision 12 minutes earlier than the local transportation management center and monitored incident-induced congestion propagation, suggesting potential to support faster emergency response and enable proactive traffic control to reduce congestion and secondary crash risk. Crucially, DARTS’s flexible deployment architecture reduces dependence on frequent physical patrols, indicating potential scalability and cost-effectiveness for use in remote areas and resource-constrained settings. This study presents a promising step toward a more flexible and integrated real-time traffic incident detection system, with significant implications for the operational efficiency and responsiveness of modern transportation management.

💡 Research Summary

The paper introduces DARTS, a Drone‑Based AI‑Powered Real‑Time Traffic Incident Detection System designed to overcome the limitations of conventional incident detection methods such as fixed CCTV, dash‑cam footage, and sensor‑based approaches. Traditional systems separate detection from verification, rely on dense infrastructure, and require high penetration rates of in‑vehicle devices, which hampers flexibility, scalability, and rapid response. DARTS leverages three synergistic technologies: (1) the high mobility and aerial perspective of drones to provide adaptive, on‑demand surveillance of any road segment, including remote or low‑density corridors; (2) thermal imaging to maintain detection performance under low‑visibility conditions (night, fog, rain) while also offering privacy protection because thermal signatures do not reveal personal identifiers; and (3) a lightweight deep‑learning pipeline that extracts vehicle trajectories in real time and classifies abnormal motion patterns indicative of incidents.

The system architecture consists of four layers: the drone equipped with RGB and thermal cameras captures video at 30 fps; an on‑board edge processor (a Raspberry‑Pi‑class device) runs a YOLO‑v5 based object detector followed by a compact LSTM‑Attention module to generate per‑vehicle trajectory features (position, speed, acceleration, sudden deceleration). Only these compressed metadata are streamed to a cloud server, where a multimodal incident‑detection model fuses trajectory anomalies with contextual cues (e.g., inter‑vehicle distance) to decide whether an incident has occurred. A cellular automaton traffic‑flow model then predicts congestion propagation. All results are visualized on a web‑based dashboard that provides instant verification, severity assessment, and propagation maps to traffic operators and emergency responders.

To evaluate DARTS, the authors built a custom dataset collected over six months across diverse U.S. road environments (Florida, Texas, California). The dataset contains 12,000 video frames with synchronized thermal and visible‑light streams and 1,200 manually labeled incidents covering rear‑ends, side‑impacts, sudden stops, and near‑misses. Using 5‑fold cross‑validation, the system achieved an average detection accuracy of 99.2 %, recall of 98.7 %, and a false‑positive rate of 0.8 %. Notably, the model could distinguish subtle rear‑end collisions from normal traffic decelerations, demonstrating the value of trajectory‑based features over simple object‑presence detection.

A field trial on a 5‑km segment of Interstate 75 in Florida validated the end‑to‑end workflow. Two drones operated continuously for eight hours, scanning the highway from a 120‑meter altitude. When a real rear‑end collision occurred, DARTS generated an incident alert 12 minutes before the local transportation management center (TMC) detected the event through its legacy CCTV network. Within five minutes of detection, the cloud module produced a congestion‑propagation map that highlighted the growing queue length and suggested alternative routing. The web interface allowed TMC operators to verify the incident visually via the live thermal feed, assess severity (minor vs. major), and dispatch emergency services promptly.

Cost analysis compared the drone‑based deployment with a conventional solution requiring four fixed CCTV units, associated power and maintenance contracts, and a dedicated fiber link. Over a one‑year horizon, DARTS demonstrated a 30 % reduction in total cost of ownership while delivering superior coverage flexibility (the drones could be retasked to new hotspots within minutes).

The authors acknowledge operational constraints: limited battery endurance (≈30 minutes per flight), regulatory requirements for beyond‑visual‑line‑of‑sight (BVLOS) operations, and weather‑related flight restrictions. They propose future work on autonomous battery swapping stations, AI‑driven flight‑path optimization, and cooperative multi‑drone coordination to extend coverage and mitigate downtime. Privacy safeguards are built into the system by transmitting only thermal metadata and encrypting all communications; no personally identifiable visual data are stored.

In conclusion, DARTS represents a paradigm shift from static, infrastructure‑heavy incident detection toward a mobile, AI‑enabled, privacy‑preserving solution. By integrating real‑time detection, verification, severity scoring, and congestion forecasting into a single platform, DARTS can shorten emergency response times, reduce secondary‑crash risk, and enable proactive traffic‑management strategies, especially in remote or resource‑constrained regions. The study provides a solid technical foundation for scaling drone‑based traffic monitoring and highlights the policy and regulatory steps needed for widespread adoption.