No abstract

No abstract

Unmanned aircraft systems or drones enable us to record or capture many scenes from the bird’s-eye view and they have been fast deployed to a wide range of practical domains, i.e., agriculture, aerial photography, fast delivery and surveillance. Object detection task is one of the core steps in understanding videos collected from the drones. However, this task is very challenging due to the unconstrained viewpoints and low resolution of captured videos. While deep-learning modern object detectors have recently achieved great success in general benchmarks, i.e., PASCAL-VOC and MS-COCO, the robustness of these detectors on aerial images captured by drones is not well studied. In this paper, we present an evaluation of state-of-the-art deep-learning detectors 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, and CenterNet for the object detection in videos captured by drones. We conduct experiments on VisDrone2019 dataset which contains 96 videos with 39,988 annotated frames and provide insights into efficient object detectors for aerial images.

Image processing and object detection in aerial images have to deal with a lot of trouble due to the existence of haze, smoke, dust in the atmosphere. These factors can blur objects and severely decline image quality which might lead to incorrect or missing object detection. To solve this problem, this study shows a method that can reduce the bad effect of haze on object detection in aerial images. A combination of a dehazing method called Feature Fusion Attention Network (FFA-Net) and an object detection method named Probabilistic Anchor Assignment (PAA) was conducted to evaluate two hypotheses: (1) haze was a noisy factor and (2) haze was treated as part of objects. Through extensive experiments, the selective dehazing hypothesis, which was used for truck objects, improved the detection result of car and bus from 19.6% to 21.9% and 0.7% to 4.4%, respectively, on the UAVDT-Benchmark-M dataset. This result showed that our approach was effective.