Remote Traffic Light Detection and Recognition Based on Deep Learning

Abstract

As we all know, the signal indicating system of traffic signal lamp plays an obvious role in effectively controlling crossing, ensuring traffic order, guiding traffic flow, improving road capacity and urban traffic stride. Crossing an intersection according to the traffic light is the basic principle of traffic behavior. In real life, the driver may be tired driving or dangerous state of using mobile phone while driving the vehicle, or have the intention to catch the yellow light, so the illegal behavior of crossing the intersection without following the traffic light is still relatively common (especially in China and other developing countries and poor areas). Traffic accidents occur frequently, and intersection is one of the places where traffic accidents occur frequently. At the present stage, this paper mainly studies traffic light detection and recognition based on YOLOv5 model and YOLOv5+DeepSort. The trained model can be used in vehicle system or intelligent recognition field, and it is hoped that when the vehicle approaches or passes through the intersection, it can provide the necessary information for the driver, and can be used for auxiliary driving. Effectively reduce the number of traffic accidents at intersections.