A Feature Pyramid Based Multi-stage Framework for Object Detection in Low-altitude UAV Images

This paper proposes a high-performance framework for accurate multi-stage object detection in low-altitude based UAV images. The proposed system employs cascade style architecture with increasing thresholds achieving detection. makes use of highly efficient Feature Pyramid Networks (FPNs) to detect objects small sizes, and various scales which are the main challenge aerial FPNs aim resolve scale variation problems by combining features multiple levels. experiments have been performed on complex dataset VisDrone has categories classes. FPN-Cascade detector supported slicing data horizontally vertically that resulted an advancement 8% mAP when compared base detector. compare performance standard as well augmented dataset. A concrete methodology about training process, hyperparameter tuning, evaluation methods Cascade RCNN is highlighted. achieves state art 30.04% value