Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
97 tokens/sec
GPT-4o
53 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
5 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments (2201.09201v2)

Published 23 Jan 2022 in cs.CV and cs.AI

Abstract: Unmanned Aerial Vehicles (UAVs) rely on satellite systems for stable positioning. However, due to limited satellite coverage or communication disruptions, UAVs may lose signals from satellite-based positioning systems. In such situations, vision-based techniques can serve as an alternative, ensuring the self-positioning capability of UAVs. However, most of the existing datasets are developed for the geo-localization tasks of the objects identified by UAVs, rather than the self-positioning task of UAVs. Furthermore, the current UAV datasets use discrete sampling on synthetic data, such as Google Maps, thereby neglecting the crucial aspects of dense sampling and the uncertainties commonly experienced in real-world scenarios. To address these issues, this paper presents a new dataset, DenseUAV, which is the first publicly available dataset designed for the UAV self-positioning task. DenseUAV adopts dense sampling on UAV images obtained in low-altitude urban settings. In total, over 27K UAV-view and satellite-view images of 14 university campuses are collected and annotated, establishing a new benchmark. In terms of model development, we first verify the superiority of Transformers over CNNs in this task. Then, we incorporate metric learning into representation learning to enhance the discriminative capacity of the model and to lessen the modality discrepancy. Besides, to facilitate joint learning from both perspectives, we propose a mutually supervised learning approach. Last, we enhance the Recall@K metric and introduce a new measurement, SDM@K, to evaluate the performance of a trained model from both the retrieval and localization perspectives simultaneously. As a result, the proposed baseline method achieves a remarkable Recall@1 score of 83.05% and an SDM@1 score of 86.24% on DenseUAV. The dataset and code will be made publicly available on https://github.com/Dmmm1997/DenseUAV.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (5)
  1. Ming Dai (9 papers)
  2. Enhui Zheng (5 papers)
  3. Zhenhua Feng (27 papers)
  4. Jiedong Zhuang (9 papers)
  5. Wankou Yang (43 papers)
Citations (19)

Summary

  • The paper demonstrates that transformer-based models outperform CNNs in urban UAV self-positioning tasks by leveraging global contextual information.
  • It introduces DenseUAV, a novel dataset with dense UAV-view images that addresses data gaps for effective cross-view geo-localization.
  • The research proposes metric learning and a new SDM@K evaluation metric to enhance model discrimination and accurately assess positioning performance.

Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments

The research paper "Vision-Based UAV Self-Positioning in Low-Altitude Urban Environments" addresses the challenges faced by Unmanned Aerial Vehicles (UAVs) when satellite-based positioning systems, such as GPS, are unavailable. This includes coverage issues or signal disruptions in urban settings. To solve this issue, the authors propose a vision-based method providing an alternative for UAV self-localization.

Contributions of the Paper

The main contributions of this work include:

  1. DenseUAV Dataset: The authors present DenseUAV, the first publicly available dataset specifically designed for UAV self-positioning. DenseUAV distinguishes itself by capturing real UAV-view images in dense sampling patterns, an innovation that addresses data insufficiencies in existing datasets tailored for cross-view geo-localization.
  2. Transformer Superiority: The paper emphasizes the effectiveness of Transformers over CNNs in UAV self-positioning tasks. Experiments demonstrate that Transformers, due to their superior ability to understand global contextual information, are well-suited for this task.
  3. Metric and Mutual Learning: The integration of metric learning enhances the discriminative capability of models by addressing modality discrepancies. Additionally, a mutually supervised learning approach is introduced to facilitate joint learning from multi-modal datasets, enhancing the alignment of UAV and satellite image features.
  4. New Evaluation Metric - SDM@K: The Recall@K metric, traditionally used in image retrieval tasks, is complemented by a new metric, SDM@K. This metric evaluates both retrieval and localization simultaneously, which provides a more nuanced assessment of a model's performance in UAV self-positioning.

Experimental Analysis

The authors conduct extensive experiments which include:

  • Data Augmentation Impact: Four types of data augmentation methods are evaluated, namely random rotation, random affine transformation, random brightness, and random erasing. These aim to increase data diversity in terms of lighting conditions, spatial information, and viewpoint discrepancies.
  • Backbone Network Comparison: The results underscore the significant advantage Transformer architectures have over CNNs for this application, with ViT-S emerging as a favorable architecture when considering both performance and computational efficiency.
  • Prediction Head Variations: Experiments explore different pooling and chunking strategies for the prediction head. Block-based methods like FSRA and LPN demonstrate improved performance, suggesting the importance of precise feature alignment in high-density datasets.
  • Data Source Variation: An analysis of drone flight altitudes, scales, and temporal offsets of satellite images reveals that multi-scale and multi-time node datasets enhance the robustness and accuracy of UAV self-positioning models.

Implications and Future Work

This paper has notable implications for urban UAV navigation systems where GPS reliability is challenged. The dataset, DenseUAV, not only sets a new benchmark but also prompts further research into UAV applications in urban environments. The proposed evaluation metric, SDM@K, allows future studies to more accurately gauge model performance in practical settings.

For future research, extending the dataset's scope beyond urban settings to incorporate different terrain conditions and height distributions could further validate and refine UAV self-positioning technologies. Additionally, exploring perspective transformation or generative adversarial network (GAN)-based approaches could mitigate the modality disparities between UAV and satellite images more effectively.

In summary, this paper offers a substantial advancement in UAV self-localization methods by introducing a novel dataset and framework, potentially impacting a wide array of UAV applications in low-altitude urban environments.