Efficient vision-based vehicle speed estimation

This paper presents a computationally efficient method for vehicle speed estimation from traffic camera footage. We build upon a previous state-of-the-art method that detects 3D bounding boxes of vehicles using vanishing point geometry and a modified RetinaNet object detector.

We show that replacing RetinaNet with a better performing YOLOv6 architecture results in significant improvements in terms of computational efficiency without compromising vehicle speed estimation accuracy. We evaluate the modified method in several variants in terms of vehicle detection and speed estimation accuracy.

Our extensive evaluation across various hardware platforms, including edge devices, demonstrates significant gains in frames per second (FPS) compared to the prior state-of-the-art, while maintaining comparable or improved speed estimation accuracy. We analyze the trade-off between accuracy and computational cost, showing that smaller models utilizing post-training quantization offer the best balance for real-world deployment.

On the BrnoCompSpeed dataset, our best-performing model achieves better speed estimation accuracy, detection precision and recall than the previous state-of-the-art, while being eight times faster. The transferability of our approach is further confirmed by an evaluation on a different dataset.



DOI: 10.1007/s11554-025-01704-z
‍