Automatic recognition of emergency vehicles in images from video recorders

In this article, a system for automatic recognition, including detection and classification, of emergency vehicles based on images from video recorders is proposed. The system utilizes YOLO v8 artificial neural network. Recognition accuracy tests were conducted on a specially prepared database of frames from real recordings. It includes marked and unmarked police vehicles (which are very difficult to distinguish and typically not included in automatic recognition systems), fire, ambulance, military vehicles and other reference images. The experiments concerned the recognition of vehicle types and the state of emergency lighting. The highest value of total (for all classes) F1 measure is over 86%. In selected classes the F1 reaches 91%, while precision is 94% and sensitivity is 89%. They can be considered as satisfactory results, what indicates the possibility of using the presented system in practice.