Does your robot know when to cross the road?

This paper presents a novel approach to enhance a robot’s proficiency in correctly identifying Pedestrian Traffic Light (PTL) signs in adverse conditions. The pivotal aspect of this endeavor lies in the accurate recognition and interpretation of the pedestrian traffic light signals. By discerning signals such as the red pedestrian icon, countdown timer, or absence of display, the robot can make informed decisions to abstain from crossing. Conversely, upon detecting the green pedestrian icon, the robot can safely traverse the road. The primary objective of this study is to improve a model capable of processing images taken from the perspective of pedestrians, providing precise outputs corresponding to the exhibited PTL sign and improve its accuracy in unfavourable conditions. Considering the global variability in PTL designs, our focus is tailored to the specific street conditions prevalent in Japan. By enhancing the data the model is currently trained on, this research strives to enhance the robotic system’s adeptness in recognizing and responding to PTL indications within Japanese crosswalks.