Development of a Deep Learning-Based Self-Calibrating Radar Speed Measurement System

Radar speedometers, crucial in vehicular speed measurement, encounter challenges in maintaining accuracy over extended periods. This study introduces a self-calibrating radar speed measurement system employing a convolutional neural network (CNN) to enhance data accuracy and reliability. We present an economical neural network-based speed measurement system that utilizes digital image processing and YOLO v7 object detection neural network for target vehicle extraction. Additionally, a dedicated CNN for speed calculation is designed and trained using a custom dataset. The network's parameters are optimized through data ablation experiments. Experimental validation confirms the system's capability to measure vehicle speed with high accuracy and stability. The developed CNN for speed calculation effectively identifies potential radar errors and excludes anomalous data, thus facilitating the radar's automatic calibration. This feature ensures long-term precision of the speed measurement equipment, significantly improving the reliability and sustainability of speed measurement operations.