Under adverse illumination, matchability scores may distinguish valid key points from ambiguous candidates less reliably. Fixed decision thresholds can then retain unreliable correspondences or discard useful candidates, destabilizing geometric estimation with random sample consensus. IPN-LightGlue is introduced to address this problem by coupling an illumination perception network (IPN) with adaptive matchability modulation. The IPN uses a frozen MobileNetV3-Small backbone to predict a normalized image-level illumination score, which is converted into a non-positive bias on LightGlue’s matchability logits. This design leaves the feature extractor unchanged and adds 74 000 trainable parameters. On the evaluation hardware, an isolated IPN forward pass takes 3.399 ± 0.829 ms per image, and the complete pipeline incurs a runtime overhead of 4.169% to 8.027% relative to LightGlue. Mean matching accuracy at a 3-pixel threshold (MMA@3px) increases from 79.9% to 85.2% on the HPatches illumination subset. For the multi-illumination data set, which isolates illumination-direction changes from viewpoint variation, MMA@3px rises from 85.2% to 89.5%, whereas mean corner error decreases by 16.4%. The method also reduces mean corner error from 59.75 pixels to 7.14 pixels on the HPatches viewpoint subset, despite the IPN being supervised only on radiometric variation. Neither contrast-limited adaptive histogram equalization nor zero-reference deep curve estimation, applied before matching, improves MMA@3px over LightGlue in these evaluations. The findings support illumination-aware matchability modulation as a way to improve correspondence accuracy and geometric stability without modifying the input image.
To address the limited field of view (FoV) of digital cameras in close-range photogrammetry, this study develops a novel, ground-based large-scale rotating photogrammetric system by integrating engineering surveying principles with photogrammetric theory. Results demonstrate that the approximate central projection images generated by this system can expand the coverage of a single camera station by more than five times while maintaining a root-mean square image stitching error less than 0.02 mm. The wide-format images generated by this system exhibit superior application accuracy and reconstruction efficiency in three-dimensional modelling compared with traditional small FoV images. The system reduces the root-mean-square error of control points and check points by 0.165 mm and 0.424 mm, respectively, while also significantly shortening the total reconstruction time. The proposed rotating photography system effectively mitigates the limited FoV of digital cameras and provides a novel technical solution for large-scale ground close-range photogrammetry. The systematic methodology for rotary image acquisition established in this work lays a theoretical and technical foundation for the practical engineering application of such systems.
To improve the accuracy, stability, and deployment adaptability of multi-platform signal positioning within complex indoor environments, an improved Grey Wolf optimization indoor positioning method targeting heterogeneous signal sources including iBeacon and WiFi is proposed. First, a unified signal abstraction interface is designed based on the received signal strength indicator. Combined with normalization processing, Kalman filter fusion, and an entropy-driven noise reduction mechanism, a multi-platform adaptive architecture is constructed. Second, guided factor dynamic adjustment, Lévy flight perturbation, and chaotic initialization strategies are introduced to enhance the global exploration capability and local convergence capability during the search process. Third, combined with fuzzy logic–based signal trust modeling and a multi-node cooperative positioning mechanism, fitness reconstruction and candidate solution fusion are implemented to strengthen anti-interference performance under complex channel conditions. Experimental results show that the proposed method achieves stable performance in open office zones, commercial atriums, underground parking lots, high-dynamic pedestrian areas, and extreme interference environments. Under the hybrid platform configuration, the average positioning error reaches as low as 0.66 m, the standard deviation is 0.10 m, and the maximum error is controlled within 1.26 m. Under strong electromagnetic interference and dense dynamic occlusion, the introduction of trust modeling significantly reduces both error fluctuations and convergence iterations. Computational efficiency tests reveal that the average processing time on the server side is 38.6 ms, and the end-to-end latency on mobile terminals is 72.8 ms, satisfying the demands of real-time indoor positioning tasks. The research results demonstrate that the proposed method possesses comprehensive advantages in positioning accuracy, robustness, cross-scenario adaptability, and engineering deploy ability and can offer methodological support for intelligent indoor navigation, Internet of Things positioning, and smart space perception.
Forest vertical structure is important for forest resource monitoring and biomass estimation. Tomographic imaging and phase histogram (PH) are two typical methods for reconstructing forest vertical scattering profiles. The new spaceborne X-band multi-static synthetic aperture radar system, HT-1, can acquire multi-baseline data in a single pass, providing a common data foundation for applying and evaluating both methods. However, the limited penetration of X-band signals produces weak below-canopy ground responses in forest areas, making it difficult to obtain a reliable terrain reference directly from the reconstructed profiles. Under such conditions, the relative performance of the two methods remains unclear. This study develops a unified inversion and comparative evaluation framework for both methods in terms of vertical structure representation, forest height accuracy, and computational efficiency. The Saihanba area is selected as the test site. ICESat-2 data and an XGBoost model are used to correct the external digital elevation model (DEM), providing a common terrain reference for both methods, with airborne lidar data used for validation. The results show that DEM correction reduces DEM root mean square error (RMSE) from 10.06 to 3.47 m. With the corrected DEM, the forest height RMSEs of Capon and PH are 4.30 and 4.50 m. Profile analysis reveals that Capon can reveal double-layer scattering structures in some sparse forest and forest-edge areas, but such structure occurs only locally and cannot serve as a ground reference across the full scene. After baseline selection, PH achieves accuracy close to Capon with much lower computational cost. These findings indicate that PH can balance accuracy and efficiency when an accurate DEM and a suitable baseline are available, making it a practical solution for rapid large-scale forest height inversion under HT-1 data conditions.
To overcome the limitation that the cloth simulation filter (CSF) cannot adapt to spatial terrain heterogeneity—which leads to reduced filtering accuracy in complex terrains—this paper proposes a terrain-complexity–driven adaptive patch-based CSF method, termed A-CSF. First, an adaptive clustering strategy is developed using terrain-complexity metrics, in which slope and roughness features are used to group point clouds with similar topographic characteristics into unified processing units. Second, a multi-parameter prediction model is established to achieve a nonlinear mapping between terrain features and optimal CSF parameter combinations, thereby enabling dynamic, patch-specific parameter prediction. Finally, a confidence-weighted merging strategy based on the standard deviation of slope is designed to effectively resolve classification conflicts in overlapping regions between patches. Experimental results in the OpenGF large-scale point cloud data set demonstrate that compared to the traditional CSF algorithm, A-CSF improves the Kappa coefficient by 9.1% and reduces total error by 29.0% in a complex mountainous scene (test 3). In a mixed-terrain scene (test 1), it achieves a Kappa improvement of 8.5% and a total error reduction of 31.9%. The proposed method exhibits significantly stronger adaptability and higher filtering accuracy in large-scale complex natural terrain conditions (e.g., mountainous and mixed-terrain scenes). However, its robustness remains limited in scenes dominated by large artificial structures with flat rooftops (e.g., test 2), where the cloth sinking effect leads to increased type II errors.
