Validation of Speed Limit Sign Recognition System in Virtual Environments: A Simulation-Based Approach

This paper presents a comprehensive approach to validate a speed limit sign recognition system, which is one of the Advanced Driving Assistance Systems in the virtual environment. Its necessity and advantages are emphasized for enhancing automotive safety and efficiency. The recognition of speed limit signs is highlighted as crucial to improving the functionality of driver assistance systems and the development of autonomous vehicles. However, the testing of these recognition systems using actual vehicles can be identified as entailing latent risks in addition to being time-consuming and financially demanding. To address these challenges, a simulation-based validation method is proposed, eliminating the hazards and reducing both the time and financial costs associated with real-world testing. Most of the research has used simulation techniques to test or enhance the performance of the algorithm that comprises the system, but usually neglecting the reliability of the simulation-based validation themselves. This paper focuses on validating the system in simulation and the reliability of the virtual environments through experiments. Finally, the effectiveness of simulation-based validation is elicited. The recognition system used in this paper is based on the You Only Look Once (YOLO) algorithm, renowned for object detection tasks. A diverse set of virtual data to mimic a wide range of real-world scenarios has been used to test the system. This paper presents a detailed comparison between the outcomes derived from tests conducted with real data and those obtained from virtual environment simulations. The results suggest that simulation-based validation can be a possible method for assessing speed limit sign recognition systems, with performance closely matching that in real-world conditions.