A Novel Traffic Signal System for Self-Driving

Current traffic light recognition systems are primarily designed for human interpretation, making autonomous vehicles susceptible to lighting interference. This paper proposes the novel traffic signal system (NTSS) to address this issue. Comparative experiments between the traditional traffic signal (TTS) and NTSS using the resnet18 model were conducted across various scenarios, including normal, light interference, and dawn conditions. The results showed that NTSS outperformed TTS in all scenarios, particularly at dawn, where NTSS achieved an accuracy of 0.95 compared to TTS's 0.90. Under light interference, NTSS maintained higher accuracy (0.85) compared to TTS (0.60). Deploying NTSS on a model car confirmed its practical applicability and potential for real-world use. These findings suggest that NTSS significantly improves traffic signal recognition.