Enhanced Vehicle Speed Detection Considering Lane Recognition Using Drone Videos in California
Amirali Ataee Naeini 1 , Ashkan Teymouri 2 , Ghazaleh Jafarsalehi 2 , and H. Michael Zhang 2 · 2025
Read Enhanced Vehicle Speed Detection Considering Lane Recognition Using Drone Videos in California by Amirali Ataee Naeini 1 , Ashkan Teymouri 2 , Ghazaleh Jafarsalehi 2 , and H. Michael Zhang 2 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-11239.txt.
Introduction
Amirali Naeini and co-authors improve in 2025 vehicle speed checks from drone video with lane awareness, tested in California. Per-lane speed data matter for carpool lane checks and for trucks, which face distinct limits. Past drone tools often missed lanes and used weak class sets.
A fine-tuned detector trained on near eight hundred top-down images spots each vehicle, its lane, and its class as car or heavy vehicle. Drone footage from Northern California tests the system across height, view span, and speed. Best runs show mean speed error under one mile per hour. The system aims at real road checks with clear lane detail.
Catalog record
| CREATOR | Amirali Ataee Naeini 1 , Ashkan Teymouri 2 , Ghazaleh Jafarsalehi 2 , and H. Michael Zhang 2 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-11239
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Enhanced Vehicle Speed Detection Considering Lane Recognition Using Drone Videos in California Amirali Ataee Naeini 1 , Ashkan Teymouri 2 , Ghazaleh Jafarsalehi 2 , and H. Michael Zhang 2 Abstract The increase in vehicle numbers in California, driven by inadequate transportation systems and sparse speed cameras, necessitates effective vehicle speed detection. Detecting vehicle speeds per lane is critical for monitoring High-Occupancy Vehicle (HOV) lane speeds, distinguishing between cars and heavy vehicles with differing speed limits, and enforcing lane restrictions for heavy vehicles. While prior works utilized YOLO (You Only Look Once) for vehicle speed detection, they often lacked accuracy, failed to identify vehicle lanes, and offered limited or less practical classification categories. This study introduces a fine-tuned YOLOv11 model, trained on almost 800 bird’s-eye view images, to enhance vehicle speed detection accuracy which is much higher compare to the previous works. The proposed system identifies the lane for each vehicle and classifies vehicles into two categories: cars and heavy vehicles. Designed to meet the specific requirements of traffic monitoring and regulation, the model also evaluates the effects of factors such as drone height, distance of Region of Interest (ROI), and vehicle speed on detection accuracy and speed measurement. Drone footage collected from Northern California was used to assess the proposed system. The fine-tuned YOLOv11 achieved its best performance with a mean absolute error (MAE) of 0.97 mph and mean squared error (MSE) of 0.94 mph 2 \text{mph}^{2} , demonstrating its efficacy in addressing challenges in vehicle speed detection and classification. Keywords: Speed Detection, YOLO, Lane Detection, Drone, Fine-tune, Heavy Vehicle, California I Introduction Traffic congestion and speed monitoring have become increasingly critical challenges in urban planning and transportation management, particularly in high-traffic regions like California. According to the Texas A&M Transportation Institute, California cities rank among the most congested in the United States, with drivers in Los Angeles experiencing an average of 62 hours of delay per year due to congestion [ 1 ] . With an ever-growing number of vehicles on the road and an inadequacy of fixed-speed enforcement systems, ensuring compliance with traffic regulations remains a formidable task. Speeding is a major factor in road safety, accounting for nearly one-third of all traffic-related fatalities nationwide [ 2 ] . Effective speed detection and classification of vehicles per lane are essential for enforcing traffic laws, improving road safety, and optimizing traffic flow. Traditional speed detection systems, such as fixed-speed cameras and radar-based devices, suffer from several limitations, including limited coverage, high installation costs, and the inability to accurately distinguish between different types of vehicles. Fixed-speed cameras provide static monitoring and lack flexibility, whereas radar-based solutions require expensive infrastructure and frequent maintenance. Drones, on the other hand, offer an efficient and cost-effective alternative for large-scale traffic surveillance, providing high-resolution imagery for real-time speed monitoring [ 3 ] . More recently, deep learning-based object detection methods, particularly YOLO (You Only Look Once), have been employed for vehicle detection and speed estimation due to their efficiency and real-time processing capabilities [ 4 ] . However, many of these models lack the necessary accuracy for real-world applications, struggle detect lanes, and fail to differentiate between cars and heavy vehicles [ 5 ] , which is crucial for enforcing regulations specific to high-occupancy vehicle (HOV) lanes and truck restrictions. To address these challenges, this study introduces an enhanced vehicle speed detection system utilizing a fine-tuned YOLOv11 model trained on nearly 800 bird’s-eye view images. Unlike traditional ground-based surveillance methods, drone-based monitoring provides a broader coverage area and more flexible deployment, making it well-suited for real-time traffic surveillance applications [ 3 ] . We employed drones to record bird’s-eye view videos of freeways and roads in Northern California, ensuring comprehensive and dynamic vehicle tracking. Our dataset consists of two hours of recorded data, which was analyzed to evaluate and validate the effectiveness of our proposed algorithm, ensuring its applicability to real-world traffic conditions. The proposed approach offers significant improvements over prior works by addressing three key challenges: (1) improving speed detection accuracy, (2) distinguishing between cars and heavy vehicles, and (3) incorporating lane recognition to associate detected vehicles with specific lanes. Accurate vehicle classification is critical for transportation enforcement, as heavy vehicles are often subject to stricter speed regulations and lane restrictions [ 5 ] . By optimizing YOLOv11 for speed estimation and lane detection, our model achieves a 5% increase in speed detection accuracy compared to previous implementations. Additionally, we investigate how factors such as drone altitude, region of interest (ROI) placement, and vehicle speed affect detection performance, offering insights that can further refine drone-based traffic monitoring techniques. The results demonstrate the effectiveness of the proposed system in real-time traffic monitoring, automated enforcement, and data-driven decision-making for transportation agencies. By leveraging deep learning-based vehicle detection in combination with drone surveillance, this approach enhances speed enforcement efficiency and traffic rule compliance [ 3 ] . Our work not only improves speed detection accuracy but also provides a robust framework for intelligent traffic analysis, paving the way for scalable, cost-effective enforcement mechanisms that can be adapted to different urban environments. II Related Work Accurate vehicle speed detection has become a cornerstone of modern traffic monitoring, especially with the rise of drone-based video acquisition and the YOLO family of object detection models. This section revisits the current research landscape across three domains: speed estimation techniques, YOLO-based detection models, and integrated lane recognition and classification systems. It concludes with a synthesis of research gaps that our study addresses. II-A Vision-Based Vehicle Speed Estimation Early attempts at vision-based speed detection focused on geometric and temporal analysis using monocular cameras and homography transformations. Pornpanomchai and Kongkittisan [ 6 ] laid a foundational framework for speed detection via frame differentiation between a vehicle’s entry and exit across a reference field. Koyuncu and Koyuncu [ 7 ] used a similar principle but integrated discrete timestamp measurements across frames, improving accuracy for low-speed vehicles. Cheng et al. [ 8 ] proposed a real-time detection model with KNN-based pixel motion analysis, whereas Kamil et al. [ 9 ] introduced a pipeline with deep homography transformation and YOLOv8 for precise image rectification and multi-frame velocity estimation. The latter achieved low RMSE (2.37 km/h) and demonstrated high consistency across drone altitudes. Recent advancements have integrated object tracking techniques into video pipelines to enhance accuracy and resilience in complex scenes. Yashina et al. [ 10 ] leveraged Kalman filters and motion history to enhance frame-based estimation and infer vehicle aggressiveness based on instantaneous speed. Ahmed et al. [ 11 ] bridged the gap between stationary and non-stationary drone footage by fusing YOLO-based bounding box tracking with pixel-accurate geographic projection, reducing absolute speed estimation error below 1 km/h. These works underline the shift toward learning-based and homography-aware approaches that enhance classical speed estimation paradigms. II-B YOLO-Based Vehicle Detection and Tracking The YOLO architecture has rapidly evolved to accommodate real-time demands in intelligent traffic systems. Redmon et al. [ 4 ] introduced YOLO as a unified detection framework that enabled single-shot prediction of bounding boxes and class probabilities. Subsequent versions, such as YOLOv3 through YOLOv5, focused on scale-aware training and anchor box optimization. Rahman et al. [ 12 ] applied YOLOv3 to drone-based vehicle speed estimation, systematically varying