This study introduces a novel two-stage method for the detection of traffic lights and the real-time perception of depth/distance in low-resolution dashcam footage, addressing the critical challenges of accurate traffic light identification under diverse lighting and weather conditions and timely depth/distance perception for enhanced road safety. Our approach combines geometric filtering with traditional image processing for initial localization, followed by precise detection using a custom-trained YOLOv8 model. The methodology is further augmented by employing the DeepSORT algorithm for consistent traffic light tracking and the MiDaS model for depth perception, utilizing real-world dashcam data. This integrated system not only improves the inference speed for traffic light detection but also provides timely alerts to drivers, significantly contributing to navigational safety. Our findings, underscored by a high mean average precision of 0.993 and low inference speed, demonstrate the system’s robustness and its practical applicability in enhancing driver assistance systems.