Detecting Objects from Space: An Evaluation of Deep-Learning Modern Approaches

• DOI: 10.3390/electronics9040583
• Publication Date: 2020-03-31T17:27:19Z
• Authors: Khang Nguyen; Nhut T. Huynh; Phat C. Nguyen; Khanh-Duy Nguyen; Nguyen D. Vo; Tam V. Nguyen
• Keywords: Drone; Boosting; Aerial image; Robustness

Unmanned aircraft systems or drones enable us to record capture many scenes from the bird’s-eye view and they have been fast deployed a wide range of practical domains, i.e., agriculture, aerial photography, delivery surveillance. Object detection task is one core steps in understanding videos collected drones. However, this very challenging due unconstrained viewpoints low resolution captured videos. While deep-learning modern object detectors recently achieved great success general benchmarks, PASCAL-VOC MS-COCO, robustness these on images by not well studied. In paper, we present an evaluation state-of-the-art including Faster R-CNN (Faster Regional CNN), RFCN (Region-based Fully Convolutional Networks), SNIPER (Scale Normalization for Image Pyramids with Efficient Resampling), Single-Shot Detector (SSD), YOLO (You Only Look Once), RetinaNet, CenterNet We conduct experiments VisDrone2019 dataset which contains 96 39,988 annotated frames provide insights into efficient images.