Design and Implementation of an Annotation-Driven Drone Autonomy Tool Using YOLOv8–V11 Architectures for Real-Time Object Detection and Distance Estimation

Autonomous UAV perception under Size–Weight–Power (SWaP) constraints is hindered by architectural fragmentation, where annotation, detection, and ranging modules are optimized independently, leaving cross-module synergies underexploited. This research introduces a unified framework that eliminates fragmentation through coordinated, cross-stage optimization for embedded deployment. An adaptive annotation engine combining Datumaro sparse encodings with foundation-model distillation reduces memory usage by 92.3% and labeling effort by 68.7%, while maintaining segmentation quality (mIoU = 0.891 ± 0.023). A mission-aware detection pipeline—featuring YOLO architecture benchmarking (v8–v11) and TensorRT INT8 quantization with kernel fusion—achieves 84.7% mAP@0.5 at 30.6 FPS on Jetson Orin NX (75 W), representing a 10.3-point gain over modular baselines. A hybrid ranging policy adaptively switches between Vincenty geodesics (GNSS-enabled, 2.8 m RMSE) and calibrated monocular geometry (GNSS-denied, 0.48 m RMSE), reducing drift by 67.3%. Replicated factorial analysis on 16,266 scene-disjoint aerial images reveals statistically significant cross-module interactions (δ12= 0.23, δ13= 0.31, δ23= 0.18;p< 0.001), yielding a 29.3% end-to-end performance amplification over additive optimization. The framework generalizes robustly, maintaining 94.2% cross-dataset retention on VisDrone/COCO/UAV benchmarks and achieving 99.84% collision-free operation across 847 autonomous sorties, establishing a reproducible pathway toward SWaP-constrained UAV autonomy.