Target Detection and Positioning of AGV Based on YOLO-F Network and Monocular Ranging

High-precision and low-latency target detection and positioning are critical for autonomous Automated Guided Vehicle (AGV) operations in complex environments. This paper proposes a vision-based target detection and positioning method for detecting and estimating the position and attitude of artificial underwater targets. First, an improved lightweight target detection algorithm YOLO-F is designed, and the structure of the YOLOv8 algorithm is modified by introducing the FEM module, the FFM module, and the SCAM module. At the same time, the number of model parameters is optimized to improve the detection speed and accuracy. Secondly, the artificial targets are colabeled by design, and the position and attitude of the targets are calculated based on the geometric information of the designed markers. In addition, for the problem of monocular ranging error, a visual ranging compensation model based on Gaussian process regression is proposed to optimize the distance estimation by quantifying the nonlinear error confidence interval to improve the positioning accuracy effectively. Experiments verify the method’s effectiveness and robustness, providing a vision solution for AGV autonomous operation.