Unmanned aerial vehicle real-time target tracking and autonomous navigation system based on partial quantization instance segmentation

Abstract

This invention discloses a real-time target tracking and autonomous navigation system for unmanned aerial vehicles (UAVs) based on partially quantized instance segmentation. The system employs a cue-based instance segmentation neural network model, which, after partial quantization, enables efficient execution on the UAV's central processing unit (CPU) without relying on a dedicated graphics processing unit (GPU). The method mainly includes the following steps: First, the system receives an initial selection of an object of interest from the user as a cue and generates a reference segmentation mask for the target object; second, based on the selected target, the system generates the segmentation mask in real time, calculates the centroid of the mask, and adjusts the UAV's flight path to align the centroid with the target position in the camera frame, thereby autonomously navigating the UAV. Distance is then maintained by comparing the area of ​​the real-time generated mask with the area of ​​the initial reference mask. This system achieves robust autonomous tracking and navigation of arbitrary targets using a lightweight hardware architecture.

Images