Estimating depth from a single low-altitude aerial image captured by an Unmanned Aerial System (UAS) has become a recent research focus. This method has a wide range of applications in 3D modeling, digital terrain models, and target detection. Traditional 3D reconstruction requires multiple images, while UAV depth estimation can complete the task with just one image, thus having higher efficiency and lower cost. This study aims to use deep learning to estimate depth from a single UAS low-altitude remote sensing image. We propose a novel global and local mixed multi-scale feature enhancement network for monocular depth estimation in low-altitude remote sensing scenes, which exchanges information between feature maps of different scales during the forward process through convolutional operations while maintaining the maximum scale feature map. At the same time, we propose a Global Scene Attention (GSA) module in the decoder part of the depth network, which can better focus on object edges, distinguish foreground and background in the UAV field of view, and ultimately demonstrate excellent performance. Finally, we design several loss functions for the low-altitude remote sensing field to constrain the network to reach its optimal state. We conducted extensive experiments on public dataset UAVid 2020, and the results show that our method outperforms state-of-the-art methods.

Robust traffic object detection has significant meanings for applications like intelligent transportation and surveillance. However, detection in the open world faces various adverse factors like occlusion, bad weather, and low light. Current endeavors focus on proposing better learnable modules for better feature representation, but their generalization ability is limited by the quality and diversity of training data. In this paper, we propose an IA-Det to achieve robust detection at various challenging transportation scenes by fully exploring semantics and suppressing the noises. First, an Iterative Attention based on t -mixture probability model is proposed to collect rich semantics at the localization branch. Unlike earlier attention methods, the IA module inherits the robust characteristic of t -mixture, and asymmetrically redistributes the attention to features based on their similarity to semantic centers. Therefore, out-of-distribution features can receive greater opportunities for clustering into their semantic neighbors rather than being suppressed as noises. For the classification branch, an effective and efficient Classification Responding Block is designed to activate classification-aware semantics from the collected attention map. Comprehensive experiments on normal, rainy, foggy, and low light fusion datasets show that the proposed IA-Det achieves state-of-the-art performance with an impressive generalization ability.

UAV-based target positioning methods are in great demand in fields, such as national defense and urban management. In previous studies, the localization accuracy of UAVs in complex scenes was difficult to be guaranteed. Target positioning methods need to improve the accuracy with guaranteed computational speed. The purpose of this study is to improve the accuracy of target localization while using only UAV information. With the introduction of depth estimation methods that perform well, the localization errors caused by complex terrain can be effectively reduced. In this study, a new target position system is developed. The system has these features: real-time target detection and monocular depth estimation based on video streams. The performance of the system is tested through several target localization experiments in complex scenes, and the results proved that the system can accomplish the expected goals with guaranteed localization accuracy and computational speed.