In this paper, we investigate the paradigm of deep learning techniques to enhance the performance of visual-based SLAM systems, particularly in challenging environments. By leveraging deep feature extraction and matching methods, we propose a robust, versatile hybrid visual SLAM framework, Rover-SLAM, aimed at improving adaptability in adverse conditions, such as dynamic lighting conditions, areas with weak textures, and camera jitter. Building on excellent learning-based algorithms of recent years, we designed from scratch a novel system that uses the same feature extraction and matching approaches for all SLAM tasks. Our system supports multiple modes, including monocular, stereo, monocular-inertial, and stereo-inertial configurations, offering flexibility to address diverse real-world scenarios. Through comprehensive experiments conducted on publicly available datasets and self-collected data, we demonstrate the superior performance of our Rover-SLAM system compared to the SOTA approaches. We also conducted an in-depth analysis of the integration of visual SLAM with deep learning methods quantitatively to provide insights for future research endeavors in this domain. The experimental results showcase the system’s capability of achieving higher localization accuracy and robust tracking performance.

In this paper, we propose an anti-occlusion target association algorithm in an overlapping cross-UAV system. This method focuses on associating targets with the same ID detected by different UAVs. Unlike other approaches that rely solely on images captured by optical cameras, our approach fully leverages data from multiple sensors to extract more reliable and stable association features. Therefore, it still maintains robustness in the face of common challenges, such as unreliable image features and occluded targets. First, we introduced a novel similarity measurement model based on spatial consistency (SC), developed through an analysis of altitude and positional relationships among targets detected by multiple UAVs. This approach leverages data fusion from multiple sensors, including cameras, IMU, and GPS. Simultaneously, we use a tracker to directly transmit association relationships from previous frames based on temporal consistency (TC), and design a decision module to adaptively select more robust SC-based strategy or more efficient TC-based strategy. This significantly improves the efficiency of the algorithm without compromising accuracy. After that, the homography constraint is utilized to optimize the preliminary results and restore the occluded targets. To achieve this, we propose a cost evaluation method to enhance the effectiveness of optimization and restoration by filtering out unreliable homography matrices. Extensive experiments verify the effectiveness and robustness of our proposed method.

While traditional oblique photogrammetry offers high-fidelity accuracy, it suffers from significant processing latency, making it less suitable for time-sensitive applications such as emergency response. Real-time incremental aerial mapping, which prioritizes immediate visualization, faces key challenges including the efficiency–accuracy trade-off, long-range drift, and appearance inconsistency under resource constraints and illumination variation. To address these issues, we propose a novel real-time large-scale orthomosaicking system that integrate learning-based VO(visual odometry) with GNSS(global navigation satellite system) via hierarchical alignment for efficient aerial orthomosaic generation. By combining robust hybrid VO with tightly-coupled optimization of GNSS-derived relative transforms, our method effectively mitigates long-range drift in large-scale environments. Furthermore, we introduce a semantic-aware coarse-to-fine mosaic plane fitting and a dual orthorectification strategy to adaptively balances local accuracy and global geomatric consistency. Finally, the multi-band Laplacian blending is employed to ensure seamless appearance transitions. Experimental results demonstrate that our system outperforms commercial solutions in both speed and consistency, particularly in complex urban structures, texture-sparse regions, and flight boundary areas. These results confirm that our method is able to deliver superior mapping quality while maintaining real-time performance.

Online incremental orthophoto generation with multiple unmanned aerial vehicles (UAVs) remains challenging, as it requires accurate, efficient, and scalable mapping from distributed aerial observations. In this paper, we present a centralized GNSS-assisted multi-UAV 3D Gaussian Splatting SLAM framework for online incremental orthophoto mapping. Each UAV independently performs visual odometry to build local submaps, which are first aligned into a unified global coordinate system using GNSS constraints and further refined via inter-agent visual loop closures for improved cross-agent consistency. To enable scalable and high-quality mapping, we introduce two complementary Gaussian map maintenance modules: plane-guided grid-based collaborative densification, which improves mapping quality and accelerates convergence under multi-UAV conditions, and visibility-aware adaptive pruning, which effectively controls redundancy and memory usage. These components allow efficient joint optimization within a unified Gaussian representation. Experiments on multiple aerial datasets using video-derived image frames captured by consumer-grade UAV cameras demonstrate that the proposed system provides a favorable trade-off between geo-consistency, visual fidelity, and efficiency compared with existing methods. Quantitatively, the proposed method achieves a GCP RMSE of 2.32 m, completes multi-UAV orthophoto generation within 3.3–5.5 min, and reduces the total mapping time by approximately 35–55% compared with the corresponding single-UAV setting, while supporting online tracking and incremental orthophoto updates with bounded latency and memory consumption.