Search Results (87)

The extraction and matching of interest points are fundamental to many geometric computer vision tasks. Traditionally, matching is performed by assigning descriptors to interest points and identifying correspondences based on descriptor similarity. This work introduces a technique whereby interest points are inherently associated during detection, eliminating the need for computing, storing, transmitting, or matching descriptors. Although the matching accuracy is marginally lower than that of conventional approaches, our method completely eliminates the need for descriptors, leading to a drastic reduction in memory usage for localization systems. We assess its effectiveness by comparing it against both classical handcrafted methods and modern learned approaches. Full article

Low-cost RGB intraoral cameras are accessible alternatives to intraoral scanners; however, generating panoramic images is challenging due to narrow fields of view, textureless surfaces, and specular highlights. This study proposes a robust stitching framework and identifies optimal acquisition parameters to overcome these limitations. All experiments were conducted exclusively on a mandibular dental phantom model. Geometric consistency was further validated using repeated physical measurements of mandibular arch dimensions as ground-truth references. We employed a deep learning-based approach using SuperPoint and SuperGlue to extract and match features in texture-poor environments, enhanced by a central-reference stitching strategy to minimize cumulative drift errors. To validate the feasibility in a controlled setting, we conducted experiments on dental phantoms varying working distances (1.5–3.0 cm) and overlap ratios. The proposed method detected approximately 19–20 times more valid inliers than SIFT, significantly improving matching stability. Experimental results indicated that a working distance of 2.5 cm offers the optimal balance between stitching success rate and image detail for handheld operation, while a 1/3 overlap ratio yielded superior geometric integrity. This system demonstrates that robust 2D dental mapping is achievable with consumer-grade sensors when combined with advanced deep feature matching and optimized acquisition protocols. Full article

To address the application requirements of Visual Simultaneous Localization and Mapping (VSLAM) in intracavitary environments and the scarcity of gold-standard datasets for deep learning methods, this study proposes a hybrid attention mechanism combined with U-Net for vascular branch point extraction in endoluminal images (SuperVessel). The network is initialized via transfer learning with pre-trained SuperRetina model parameters and integrated with a vascular feature detection and matching method based on dual branch fusion and structure enhancement, generating a pseudo-gold-standard vascular branch point dataset. The framework employs a dual-decoder architecture, incorporates a dynamic up-sampling module (CBAM-Dysample) to refine local vessel features through hybrid attention mechanisms, designs a Dice-Det loss function weighted by branching features to prioritize vessel junctions, and introduces a dynamically weighted Triplet-Des loss function optimized for descriptor discrimination. Experiments on the Vivo test set demonstrate that the proposed method achieves an average Area Under Curve (AUC) of 0.760, with mean feature points, accuracy, and repeatability scores of 42,795, 0.5294, and 0.46, respectively. Compared to SuperRetina, the method maintains matching stability while exhibiting superior repeatability, feature point density, and robustness in low-texture/deformation scenarios. Ablation studies confirm the CBAM-Dysample module’s efficacy in enhancing feature expression and convergence speed, offering a robust solution for intracavitary SLAM systems. Full article

Visual SLAM systems face significant performance degradation under dynamic lighting conditions, where traditional feature extraction methods suffer from reduced keypoint detection and unstable matching. This paper presents LDFE-SLAM, a novel visual SLAM framework that addresses illumination challenges through a Light-Aware Deep Front-End (LDFE) architecture. Our key insight is that low-light degradation in SLAM is fundamentally a geometric feature distribution problem rather than merely a visibility issue. The proposed system integrates three synergistic components: (1) an illumination-adaptive enhancement module based on EnlightenGAN with geometric consistency loss that restores gradient structures for downstream feature extraction, (2) SuperPoint-based deep feature detection that provides illumination-invariant keypoints, and (3) LightGlue attention-based matching that filters enhancement-induced noise while maintaining geometric consistency. Through systematic evaluation of five method configurations (M1–M5), we demonstrate that enhancement, deep features, and learned matching must be co-designed rather than independently optimized. Experiments on EuRoC and TUM sequences under synthetic illumination degradation show that LDFE-SLAM maintains stable localization accuracy (∼1.2 m ATE) across all brightness levels, while baseline methods degrade significantly (up to 3.7 m). Our method operates normally down to severe lighting conditions (30% ambient brightness and 20–50 lux—equivalent to underground parking or night-time streetlight illumination), representing a 4–6× lower illumination threshold compared to ORB-SLAM3 (200–300 lux minimum). Under severe (25% brightness) conditions, our method achieves a 62% tracking success rate, compared to 12% for ORB-SLAM3, with keypoint detection remaining above the critical 100-point threshold, even under extreme degradation. Full article

In Search and Rescue (SAR) missions, UAVs must be capable of detecting small objects from complex and noise-prone maritime images. Existing small object detection methods typically rely on super-resolution techniques or complex structural designs, which often demand significant computational resources and fail to meet the real-time requirements for small mobile devices in SAR tasks. To address this challenge, we propose SAQ-YOLO, an efficient small object detection model based on the YOLO framework. We design a Small Object Auxiliary Query branch, which uses deep semantic information to guide the fusion of shallow features, thereby improving small object capture efficiency. Additionally, SAQ-YOLO incorporates a series of lightweight channel, spatial, and group (large kernel) gated attention mechanisms to suppress background clutter in complex maritime environments, enhancing feature extraction at a low computational cost. Experiments on the SeaDronesSee dataset demonstrate that, compared to YOLOv11s, SAQ-YOLO reduces the number of parameters by approximately 70% while increasing mAP@50 by 2.1 percentage points. Compared to YOLOv11n, SAQ-YOLO improves mAP@50 by 8.7 percentage points. When deployed on embedded platforms, SAQ-YOLO achieves an inference latency of only 35 milliseconds per frame, meeting the real-time requirements of maritime SAR applications. These results suggest that SAQ-YOLO provides an efficient and deployable solution for UAV SAR operations in vast and highly dynamic marine environments. Future work will focus on enhancing the robustness of the detection model. Full article

To address the issue of decreased localization accuracy and robustness in existing visual SLAM systems caused by imprecise identification of dynamic regions in complex dynamic scenes—leading to dynamic interference or reduction in valid static feature points, this paper proposes a dynamic visual SLAM method integrating instance-level motion classification, temporally adaptive super-pixel segmentation, and optical flow propagation. The system first employs an instance-level motion classifier combining residual flow estimation and a YOLOv8-seg instance segmentation model to distinguish moving objects. Then, temporally adaptive super-pixel segmentation algorithm SLIC (TA-SLIC) is applied to achieve fine-grained dynamic region partitioning. Subsequently, a proposed dynamic region missed-detection correction mechanism based on optical flow propagation (OFP) is used to refine the missed-detection mask, enabling accurate identification and capture of motion regions containing non-rigid local object movements, undefined moving objects, and low-dynamic objects. Finally, dynamic feature points are removed, and valid static features are utilized for pose estimation. The localization accuracy of the visual SLAM system is validated using two widely adopted datasets, TUM and BONN. Experimental results demonstrate that the proposed method effectively suppresses interference from dynamic objects (particularly non-rigid local motions) and significantly enhances both localization accuracy and system robustness in dynamic environments. Full article

To address the problem of degraded positioning accuracy in traditional visual–inertial navigation systems (VINS) due to interference from moving objects in dynamic scenarios, this paper proposes an improved algorithm based on the VINS-Fusion framework, which resolves this issue through a synergistic combination of multi-scale feature optimization and real-time dynamic feature elimination. First, at the feature extraction front-end, the SuperPoint encoder structure is reconstructed. By integrating dual-branch multi-scale feature fusion and 1 × 1 convolutional channel compression, it simultaneously captures shallow texture details and deep semantic information, enhances the discriminative ability of static background features, and reduces mis-elimination near dynamic–static boundaries. Second, in the dynamic processing module, the ASORT (Adaptive Simple Online and Realtime Tracking) algorithm is designed. This algorithm combines an object detection network, adaptive Kalman filter-based trajectory prediction, and a Hungarian algorithm-based matching mechanism to identify moving objects in images in real time, filter out their associated dynamic feature points from the optimized feature point set, and ensure that only reliable static features are input to the backend optimization, thereby minimizing pose estimation errors caused by dynamic interference. Experiments on the KITTI dataset demonstrate that, compared with the original VINS-Fusion algorithm, the proposed method achieves an average improvement of approximately 14.8% in absolute trajectory accuracy, with an average single-frame processing time of 23.9 milliseconds. This validates that the proposed approach provides an efficient and robust solution for visual–inertial navigation in highly dynamic environments. Full article

Mine rescue robots operate in extremely adverse subterranean environments, where the acquired video data are frequently affected by severe jitter and motion distortion. Such instability leads to the loss of critical visual information, thereby reducing the reliability of rescue decision-making. To address this issue, a dual-channel visual stabilization framework based on the SuperPoint network is proposed, extending the traditional ORB descriptor framework. Here, dual-channel refers to two configurable and mutually exclusive feature extraction paths—an ORB-based path and a SuperPoint-based path—that can be flexibly switched according to scene conditions and computational requirements, rather than operating simultaneously on the same frame. The subsequent stabilization pipeline remains unified and consistent across both modes. The method employs an optimized detector head that integrates deep feature extraction, non-maximum suppression, and boundary filtering to enable precise estimation of inter-frame motion. When combined with smoothing filters, the approach effectively attenuates vibrations induced by irregular terrain and dynamic operational conditions. Experimental evaluations conducted across diverse scenarios demonstrate that the proposed algorithm achieves an average improvement of 27.91% in Peak Signal-to-Noise Ratio (PSNR), a 55.04% reduction in Mean Squared Error (MSE), and more than a twofold increase in the Structural Similarity Index (SSIM) relative to pre-stabilized sequences. Moreover, runtime analysis indicates that the algorithm can operate in near-real-time, supporting its practical deployment on embedded mine rescue robot platforms.These results verify the algorithm’s robustness and applicability in environments requiring high visual stability and image fidelity, providing a reliable foundation for enhanced visual perception and autonomous decision-making in complex disaster scenarios. Full article

Underwater object detection is critical for marine resource utilization, ecological monitoring, and maritime security, yet it remains constrained by optical degradation, high energy consumption, and vulnerability to adversarial perturbations. To address these challenges, this study proposes an Adaptive Vision Transformer (A-ViT)-based detection framework. At the hardware level, a systematic power-modeling and endurance-estimation scheme ensures feasibility across shallow- and deep-water missions. Through the super-resolution reconstruction based on the Hybrid Attention Transformer (HAT) and the staged enhancement with the Deep Initialization and Deep Inception and Channel-wise Attention Module (DICAM), the image quality was significantly improved. Specifically, the Peak Signal-to-Noise Ratio (PSNR) increased by 74.8%, and the Structural Similarity Index (SSIM) improved by 375.8%. Furthermore, the Underwater Image Quality Measure (UIQM) rose from 3.00 to 3.85, while the Underwater Color Image Quality Evaluation (UCIQE) increased from 0.550 to 0.673, demonstrating substantial enhancement in both visual fidelity and color consistency. Detection accuracy is further enhanced by an improved YOLOv11-Coordinate Attention–High-order Spatial Feature Pyramid Network (YOLOv11-CA_HSFPN), which attains a mean Average Precision at Intersection over Union 0.5 (mAP@0.5) of 56.2%, exceeding the baseline YOLOv11 by 1.5 percentage points while maintaining 10.5 ms latency. The proposed A-ViT + ROI reduces inference latency by 27.3% and memory usage by 74.6% when integrated with YOLOv11-CA_HSFPN and achieves up to 48.9% latency reduction and 80.0% VRAM savings in other detectors. An additional Image-stage Attack QuickCheck (IAQ) defense module reduces adversarial-attack-induced latency growth by 33–40%, effectively preventing computational overload. Full article

The integration of SLAM with Gaussian splatting presents a significant challenge: achieving compatibility between real-time performance and high-quality rendering. This paper introduces a novel SLAM system named SFGS-SLAM (SuperFeats Gaussian Splatting SLAM), restructured from tracking to mapping, to address this issue. A new keypoint detection network is designed and characterized by fewer parameters than existing networks such as SuperFeats, resulting in faster processing speeds. This keypoint detection network is augmented with a global factor graph incorporating the GICP (Generalized Iterative Closest Point) odometry, reprojection-error factors and loop-closure constraints to minimize drift. It is integrated with the Gaussian splatting as the mapping part. By leveraging the reprojection error, the proposed system further reduces odometry error and improves rendering quality without compromising speed. It is worth noting that SFGS-SLAM is primarily designed for static indoor environments and does not explicitly model or suppress dynamic disturbances. Comprehensive experiments were conducted on various datasets to evaluate the performance of our system. Extensive experiments on indoor and synthetic datasets show that SFGS-SLAM achieves accuracy comparable to state-of-the-art SLAM while running in real time. SuperFeats reduces matching latency by over 50%, and joint optimization significantly improves global consistency. Our results demonstrate the practicality of combining lightweight feature matching with dense Gaussian mapping, highlighting trade-offs between speed and accuracy. Full article

With the increasing demand for implementing deep-learning models on devices on resource-constrained devices, the development of power-efficient neural networks has become imperative. This paper introduces HADQ-Net, a novel framework for optimizing deep convolutional neural networks (CNNs) through Quantization-Aware Training (QAT). By compressing 32-bit floating-point (FP32) precision weights and activation values to lower bit-widths, HADQ-Net significantly reduces memory footprint and computational complexity while maintaining high accuracy. We propose adaptive quantization limits based on the statistical properties of each layer or channel, coupled with normalization techniques, to enhance quantization efficiency and accuracy. The framework includes algorithms for QAT, quantized convolution, and quantized inference, enabling efficient deployment of deep CNN models on edge devices. Extensive experiments across tasks such as super-resolution, classification, object detection, and semantic segmentation demonstrate the trade-offs between accuracy, model size, and computational efficiency under various quantization levels. Our results highlight the superiority of QAT over post-training quantization methods and underscore the impact of quantization types on model performance. HADQ-Net achieves significant reductions in memory footprint, computational complexity, and energy consumption, making it ideal for resource-constrained environments without sacrificing performance. Full article

Methane (CH₄) is known as the most potent greenhouse gas in the short term. With the growing urgency of mitigating climate change and monitoring CH₄ emissions, many emerging satellite systems have been launched in the past decade to observe CH₄ and other greenhouse gases from space. These satellites are either capable of pinpointing and quantifying super emitters or deriving regional emissions with a more frequent revisit time. This study aims to reconcile emissions estimated from point source satellites and those from regional mapping satellites, and to investigate the potential of integrating point-based quantification and regional-based quantification techniques. To do that, we quantified CH₄ emissions from the Permian Basin separately by applying the divergence method to the TROPOMI Level-2 data product, as well as an event-based approach using CH₄ plumes quantified by Carbon Mapper systems. The resulting annual CH₄ emissions estimates from the Permian Basin in 2024 are 1.83 ± 0.96 Tg and 1.26 [0.78, 2.02] Tg for divergence and event-based methods, respectively. The divergence-based emissions estimate shows a more comprehensive spatial distribution of emissions across the Permian Basin, whereas the event-based approach highlights the grid cells with the short-duration super-emitters. The emissions from grids with detectable emissions under both methods show strong agreement (R² ≈ 0.642). After substituting the overlap cells’ values from divergence-based emissions estimation with those from event-based estimation, the combined emissions estimate is 2.68 [1.88, 3.54] Tg, which is reconciled with Permian Basin emissions estimates from previous studies. We found that CH₄ emissions from the Permian Basin gradually reduced over the past five years. Furthermore, this case study indicates the potential for integrating estimations from both methods to generate a more comprehensive regional emissions estimate. Full article

Contact-force perception is a critical component of safe robotic grasping. With the rapid advances in embodied intelligence technology, humanoid robots have enhanced their multimodal perception capabilities. Conventional force sensors face limitations, such as complex spatial arrangements, installation challenges at multiple nodes, and potential interference with robotic flexibility. Consequently, these conventional sensors are unsuitable for biomimetic robot requirements in object perception, natural interaction, and agile movement. Therefore, this study proposes a sensorless external force detection method that integrates SuperPoint-Scale Invariant Feature Transform (SIFT) feature extraction with finite element analysis to address force perception challenges. A visual analysis method based on the SuperPoint-SIFT feature fusion algorithm was implemented to reconstruct a three-dimensional displacement field of the target object. Subsequently, the displacement field was mapped to the contact force distribution using finite element modeling. Experimental results demonstrate a mean force estimation error of 7.60% (isotropic) and 8.15% (anisotropic), with RMSE < 8%, validated by flexible pressure sensors. To enhance the model’s reliability, a dual-channel video comparison framework was developed. By analyzing the consistency of the deformation patterns and mechanical responses between the actual compression and finite element simulation video keyframes, the proposed approach provides a novel solution for real-time force perception in robotic interactions. The proposed solution is suitable for applications such as precision assembly and medical robotics, where sensorless force feedback is crucial. Full article

Artificial intelligence (AI) is revolutionarily shaping the entire landscape of medicine and particularly the privileged field of radiology, since it produces a significant amount of data, namely, images. Currently, AI implementation in radiology is continuously increasing, from automating image analysis to enhancing workflow management, and specifically, pediatric neuroradiology is emerging as an expanding frontier. Pediatric neuroradiology presents unique opportunities and challenges since neonates’ and small children’s brains are continuously developing, with age-specific changes in terms of anatomy, physiology, and disease presentation. By enhancing diagnostic accuracy, reducing reporting times, and enabling earlier intervention, AI has the potential to significantly impact clinical practice and patients’ quality of life and outcomes. For instance, AI reduces MRI and CT scanner time by employing advanced deep learning (DL) algorithms to accelerate image acquisition through compressed sensing and undersampling, and to enhance image reconstruction by denoising and super-resolving low-quality datasets, thereby producing diagnostic-quality images with significantly fewer data points and in a shorter timeframe. Furthermore, as healthcare systems become increasingly burdened by rising demands and limited radiology workforce capacity, AI offers a practical solution to support clinical decision-making, particularly in institutions where pediatric neuroradiology is limited. For example, the MELD (Multicenter Epilepsy Lesion Detection) algorithm is specifically designed to help radiologists find focal cortical dysplasias (FCDs), which are a common cause of drug-resistant epilepsy. It works by analyzing a patient’s MRI scan and comparing a wide range of features—such as cortical thickness and folding patterns—to a large database of scans from both healthy individuals and epilepsy patients. By identifying subtle deviations from normal brain anatomy, the MELD graph algorithm can highlight potential lesions that are often missed by the human eye, which is a critical step in identifying patients who could benefit from life-changing epilepsy surgery. On the other hand, the integration of AI into pediatric neuroradiology faces technical and ethical challenges, such as data scarcity and ethical and legal restrictions on pediatric data sharing, that complicate the development of robust and generalizable AI models. Moreover, many radiologists remain sceptical of AI’s interpretability and reliability, and there are also important medico-legal questions around responsibility and liability when AI systems are involved in clinical decision-making. Future promising perspectives to overcome these concerns are represented by federated learning and collaborative research and AI development, which require technological innovation and multidisciplinary collaboration between neuroradiologists, data scientists, ethicists, and pediatricians. The paper aims to address: (1) current applications of AI in pediatric neuroradiology; (2) current challenges and ethical considerations related to AI implementation in pediatric neuroradiology; and (3) future opportunities in the clinical and educational pediatric neuroradiology field. AI in pediatric neuroradiology is not meant to replace neuroradiologists, but to amplify human intellect and extend our capacity to diagnose, prognosticate, and treat with unprecedented precision and speed. Full article

In traditional centralized steel plant production monitoring systems, there are two major problems. On the one hand, the limited shooting angles of cameras make it impossible to capture comprehensive information. On the other hand, using multiple cameras to display monitoring screens separately on a large screen leads to clutter and easy omission of key information. To address the above-mentioned issues, this paper proposes an image stitching technique based on an improved LightGlue algorithm. First of all, this paper employs the SuperPoint (Self-Supervised Interest Point Detection and Description) algorithm as the feature extraction algorithm. The experimental results show that this algorithm outperforms traditional algorithms both in terms of feature extraction speed and extraction accuracy. Then, the LightGlue (Local Feature Matching at Light Speed) algorithm is selected as the feature matching algorithm, and it is optimized and improved by combining it with the Agglomerative Clustering (AGG) algorithm. The experimental results indicate that this improvement effectively increases the speed of feature matching. Compared with the original LightGlue algorithm, the matching efficiency is improved by 26.2%. Finally, aiming at the problems of parallax and ghosting existing in the image fusion process, this paper proposes a pixel dynamic adaptive fusion strategy. A local homography matrix strategy is proposed in the geometric alignment stage, and a pixel difference fusion strategy is proposed in the pixel fusion stage. The experimental results show that this improvement successfully solves the problems of parallax and ghosting and achieves a good image stitching effect. Full article