Analysis of Factors Influencing Traffic Sign Recognition Based on Multi-Sensor Perception Data

When traffic signs are blocked or faded, the ability of automated vehicles (AVs) to perceive and recognize them is lost or reduced. Previous research has improved and optimized the recognition algorithm but has not systematically analyzed the environmental and design factors that can lead to the AV’s failure to perceive and recognize traffic signs. This study uses the multiple sensors in the equipped vehicles of the Tongji University Road and Traffic Holographic Data Acquisition System to perceive and recognize traffic signs in six crash hotspots in Shanghai, China. Failure to recognize traffic signs was analyzed considering the influence of four sets of factors: traffic sign design, vehicle characteristics, environmental factors, and road design. A feature importance analysis was conducted utilizing the machine learning algorithm model CatBoost. Shapley additive explanation values were used to analyze the positive and negative relationships as well as the contribution rates of each influencing factor. The results revealed that the major factors influencing sign perception and recognition were positioning of traffic sign, lighting, blockage, cleanliness, fading, and reflectivity. Cleanliness had the most significant impact on the perception and recognition of traffic signs compared with the other influencing factors, and it also showed significant influencing interaction with the other factors. Fading was the second most important factor. The results of this study can guide designers of traffic signs to promote safety in mixed traffic environments that include both AVs and human-driven vehicles.