The paper proposes and considers a method for improving the analog FPV video stream, which is generated and transmitted via a 5.8 GHz communication channel, for further real-time object recognition using the YOLOv9 neural network. The proposed approach is based on a multi-stage preprocessing pipeline. It includes noise reduction, interference compensation, brightness and contrast correction, deinterlacing, frame stabilization, motion blur compensation, resolution enhancement, and color normalization to improve the quality and informativeness of digitized FPV frames. Experiments were conducted using the BetaFPV Meteor 75 Pro FPV drone and the C03 FPV camera. The results obtained demonstrated an improvement in image quality, an increase in the level of detection confidence, and an increase in the number of successfully detected objects. In particular, the average contrast value increases by 45–50%, the detail and clarity of object boundaries increase, and YOLOv9's confidence in the correctness of detection increases by 23%. This confirms the effectiveness of the proposed method for the development of modern technologies in robotic systems based on computer vision in conditions of analog video signal transmission.

Terms—computer vision, FPV drone, modern technologies, object detection, preprocessing pipeline, robotic systems, video enhancement, YOLOv9