Mask R-CNN-based Vehicle Speed Estimation for Enhancing Road Safety and Control

Precise vehicle speed assessment is essential for enhancing road safety and enforcing traffic laws. Traditional speed measuring methods, including radar and inductive loop sensors, often exhibit limitations in terms of scalability and cost efficiency. This paper presents a vehicle speed estimation system based on a Mask Region-based Convolutional Neural Network (Mask R-CNN), which utilizes deep learning (DL) for accurate object identification, segmentation, and motion tracking in real-world traffic conditions. The system analyses video frames to identify cars, monitor their progress, and calculate speeds using calibrated distance measurements and frame rate evaluation. An experimental study utilized a benchmark traffic dataset and authentic highway video footage. The proposed system achieved a mean speed estimation error of 3.1 km/h, significantly improving upon traditional computer vision techniques. The detection accuracy improved by 96%, with a Mean Intersection over Union (mIoU) of 89.4%, providing dependable vehicle segmentation. The system functions in real-time at 18 frames per second (FPS) using an NVIDIA RTX 3080 GPU, making it appropriate for extensive deployment. The findings demonstrate the viability of Mask R-CNN for precise, non-invasive, and economical vehicle speed assessment, thereby enhancing road safety and traffic regulation.