Robust Vehicle Speed Estimation Based on Vision Sensor Using YOLOv5 and DeepSORT

For driving safety purposes, the policymakers make regulations on the speed limit on the highway because vehicles high speed may cause an accident. Vision-based vehicle speed estimation requires limited human resources and minimizes human error. Therefore, the authors of this research propose a vehicle speed estimation pipeline using You Only Look Once version 5 (YOLOv5) for vehicle detection, tracking vehicles using Simple Online and Realtime Tracking with a Deep Association Metric (DeepSORT), and speed measurement process. The dataset built for this work is taken in Gadjah Mada University using a camera in bird view orientation. The detection precision and recall of the of YOLOv5l pre-trained model used are both about 100% and has fps 10.762. The speed value reaches a Root Mean Square Error (RMSE) of 3.926 and a Mean Absolute Error (MAE) of 3.155. The RMSE obtained is compared with previous research and has the most significant value