Empirical Study of the Performance of Object Detection Methods on Road Marking Dataset
Nguyen D. Vo, Vy Le, Thien C. Lai, Khanh B. T. Duong, Yen Tran · 2022 RIVF International Conference on Computing and Communication Technologies (RIVF) · 2022
Due to the rising need for research into vehicle automation, many traffic-related datasets for the problem of object detection in traffic images were introduced. Specifically, road marking detection is one of the most essential tasks involved in vehicle automation. In this research, we looked into the problem of road markings detection and conducted experiments, evaluation and comparisons of the performance of different state-of-the-art object detection methods on the CeyMo Road Marking dataset, a recently published benchmark that poses many challenges. Experiment results show that recent state-of-the-art object detectors obtained very competitive performance on the benchmark. Particularly, YOLOv7 performed remarkably better compared to other methods, with [email protected], [email protected] and mAP scores at 90.7%, 76.2% and 69.5%, respectively, making it the current best method on the dataset. The substantial accuracy and time advantage of YOLOv7 also demonstrate the rapid improvements of recent research into object detection.