Cross-view geo-localization aims to match images of the same target from different platforms, e.g ., drone and satellite. It is a challenging task due to the changing appearance of targets and environmental content from different views. Most methods focus on obtaining more comprehensive information through feature map segmentation, while inevitably destroying the image structure, and are sensitive to the shifting and scale of the target in the query. To address the above issues, we introduce simple yet effective part-based representation learning, shifting-dense partition learning (SDPL). We propose a dense partition strategy (DPS), dividing the image into multiple parts to explore contextual information while explicitly maintaining the global structure. To handle scenarios with non-centered targets, we further propose the shifting-fusion strategy, which generates multiple sets of parts in parallel based on various segmentation centers, and then adaptively fuses all features to integrate their anti-offset ability. Extensive experiments show that SDPL is robust to position shifting, and performs competitively on two prevailing benchmarks, University-1652 and SUES-200. In addition, SDPL shows satisfactory compatibility with a variety of backbone networks ( e.g. , ResNet and Swin). https://github.com/C-water/SDPL_release.
15 December 2025 in IEEE Transactions on Geoscience and Remote Sensing
This paper investigates the challenges of UAV-view geo-localization under extreme environmental changes, where significant cross-domain style differences can lead to degradation of model performance. Existing methods primarily focus on mitigating domain shift issues caused by environmental factors but generally overlook the direct interference of environmental noise. We argue that mitigating environmental noise is equally critical for extracting discriminative cross-view features and introduce a pixel-to-feature co-enhancement network (P2FCN). P2FCN comprises a style-noise dual suppression module (SNDS) and a part-based multi-dimensional feature learning strategy (PMDFL). Specifically, the SNDS module mitigates the stylistic discrepancies between cross-environment images through pixel-level dynamic adjustment, while reducing the introduction of environmental noise to a certain degree. PMDFL improves the representation generalization by purifying discriminative information within partitioned channel-and spatial-wise feature subspaces. Extensive experiments on two widely used benchmarks, i.e., University-1652 and SUES-200, demonstrate that the proposed method achieves state-of-the-art performance in multiple environments compared to existing methods.
28 December 2023 in IEEE Transactions on Instrumentation and Measurement
Visual relocalization plays a crucial role in unmanned systems, which automatically estimates the camera pose in a known scene with the input visual information. However, existing visual relocalization methods encounter challenges in relocating images with significant viewpoint variations, and they often lack the ability to self-identify the precision of the relocated camera pose, leading to low accuracy and recall. In this study, we propose a novel visual relocalization method that leverages a hierarchical image fern encoding and matching technique to accurately relocalize under large viewpoint differences. Furthermore, we introduce a self-supervised pose optimization module that self-identify the precision of the relocated camera poses and refines the false or inaccurate poses via a looped correction strategy. To identify the precision without ground truth, we translate the problem of precision identification into the classification of precision range and design a multi-SVM module to classify the precision range of a relocated camera pose by the output values of the Iterative Closest Point (ICP). We evaluate our proposed method on two widely used datasets, 7-Scenes and 12-Scenes, and show that our method substantially reduces false positives in camera relocalization. In particular, our method achieves more than a 20% improvement in recall with 1cm/1° accuracy compared to the state-of-the-art methods.
2 July 2024 in IEEE Transactions on Instrumentation and Measurement
Real-time 3D reconstruction of indoor scenes is important in computer vision and graphics. Existing solutions mainly use a handheld RGB or RGB-D camera to perform the task. However, these solutions do not consider information other than visual features to estimate camera poses and require intensive labor to record the scene. In this paper, we design and develop a motor-driven rotating acquisition device with three synchronized RGB-D cameras to automate the acquisition and propose a novel method to assist the camera pose estimation by the angle encoders of the motor. Firstly, benifited from the design choice of the device, camera pose estimation can be restricted in a 1-D space, and motor and visual information are combined together to achieve high accuracy. Furthermore, motor-based loop detection and global optimization are proposed to refine the camera poses and the scene geometry. Extensive evaluations are performed on both virtual and real datasets to demonstrate that the proposed indoor scene reconstruction system outperforms existing solutions regarding both reconstruction efficiency and accuracy.