Strategies for Enhancing Static Object Detection Range in Autonomous Vehicles Using Monocular Camera Vision

The performance of object detection models for autonomous driving is increasingly advancing. However, to recognize static traffic targets such as traffic light recognition and traffic sign recognition, the detectable distance in images plays a crucial role in the planning strategies and performance of recognition algorithms for autonomous driving systems. This study analyzes the detection performance and detection range of traffic lights by a monocular camera on an autonomous driving vehicle. The detection model used is the YOLOv7-x model, with the original images segmented to 1x, 1/2x, and 1/3x of the model input size. The distances are obtained through high-precision maps and vehicle-mounted GPS/RTK. The results of this study indicate the scalability of the detection distance of a monocular camera’s object detection in an autonomous driving environment using a single model. It further demonstrates a proportional relationship between the detection distance and the model input size. It confirms the scalability of detection distances with cameras of different fields of view using a single model.