Design of an Integrated Platform for Drone Target Recognition and Servo Rudder Tracking based on Raspberry Pi 5

Target identification and tracking in security and other contexts are critically needed as drone technology becomes more widely used. Problems with traditional radar detection methods include their high cost, vulnerability to interference from the environment, and complicated device placement. Issues with deep learning-based methods include their high reliance on computer resources and poor transfer efficiency for embedded environment models. In order to accomplish closed-loop feedback from visual perception to motion control, this study suggests an integrated platform for drone target recognition and servo tracking based on the Raspberry Pi 5. This platform uses the nenn framework for quantitative pruning to increase embedded inference speed, optimises the model structure for unmanned aerial vehicle scenarios, and incorporates the YOLOv5s lightweight object detection model; designed to create a hardware closed loop using a Raspberry Pi 5 as the central controller, a PCA9685-driven dual axis servo motor, and an 8megapixel CSI camera; An incremental PID controller is activated to modify the servo angle by determining the offset between the target and the image centre. The NCNN framework is optimised at the hardware level to enhance performance, while the NMS method and image preprocessing are utilised at the software level. According to experiments, the platform offers a low-cost, high real-time, lightweight solution for drone security, achieving a mAP50 detection rate of $93.3 \%$ and a tracking delay of less than 200 ms in complex backgrounds. In the future, data augmentation and model optimisation can increase the robustness of complex environments, making it appropriate for situations like airports.