Detection of Vehicle Lights and Classification of Its State

The lights of a vehicle are an essential part of it. It not only helps to drive but also helps in controlling the vehicle. In this paper, a novel method is presented for the detection of vehicle headlights and taillights. This paper also classifies the state of the vehicle lights. Vehicles that are near to the camera are considered for the experiment and the size of the vehicle light had a significant impact on the detection accuracy. Indian Driving Dataset (IDD) was used for the training and testing of the model. IDD consists of the rural and urban roads of Hyderabad and Bengaluru City. It covers various light conditions like day and night and scenarios like a smaller number of vehicles and a greater number of vehicles for detection. Some data preprocessing was also performed before using the dataset for the experiment. The model first detects the vehicle light and then it also classifies the state of the vehicle light. For vehicle light detection, some of the widely used object detection models such as YOLO, Faster RCNN, and MobileNet are used. For the classification of the state of the vehicle light, a color thresholding approach was used to develop the system. The system achieved a mean average precision(mAP) of 91.1%. It classifies it as Headlight on, Taillight on, or light off. The model performance was also evaluated and tested on IDD videos and it was very promising.