An Improved YOLOv8 Based Smart Road Stud Detection Method

Smart road studs are widely used for road safety and traffic data collection. Their accurate and reliable detection, and integration into the perception and control modules of connected and autonomous vehicles (CAVs), enhances road boundary detection, vehicle localization, and driving safety. However, real-time, accurate and reliable detection of the small-sized smart road studs is challenging for fast moving CAVs, especially in harsh environments. To address the challenges, we first build a real-world smart road stud dataset, and then propose and validate a lightweight and efficient smart road stud detection model based on the you only look once 8th version (YOLOv8). We then introduce a novel downsampling module (DownS) combining the average pooling and the max pooling to reduce the number of parameters and minimize information loss during downsampling. Furthermore, we replace the loss function with Normalized Wasserstein Distance (NWD) loss to reduce sensitivity to location deviations in small target detection. Finally, we deploy a real-time smart road stud detection system on an experimental vehicle to validate the feasibility and effectiveness of the proposed algorithm. The experimental results demonstrate that the proposed algorithm significantly enhances the accuracy and efficiency of smart road stud detection, increasing the mean average precision by 9.58% and reducing the number of parameters by 13.71 %. Our dataset is available at: https://github.com/wky-xidian/smart-road-stud-dataset.