High-precision and robust autonomous localization is an essential technology for uncrewed systems to achieve reliable navigation. Traditional map-based positioning methods can provide high localization accuracy, but as the size of the environment increases, these methods often require significant computational resources and memory, making them unsuitable for resource-constrained platforms. To address this, we propose a lightweight localization method based on sparse maps, designed to reduce computational overhead while maintaining high precision and robustness. Our approach combines light detection and ranging (LiDAR) and inertial measurements by extracting planar features from local regions for scan matching. This method reduces data processing complexity, ensuring both real-time performance and localization accuracy. Unlike traditional methods that match LiDAR data with the global map for every frame, we perform global map matching only at specific intervals, applying pose corrections through a time threshold mechanism, thereby significantly reducing computational burden. Additionally, to improve the accuracy and stability of state estimation, this article integrates LiDAR data at specific time intervals with the results of map matching, along with planar feature matching results extracted from local regions, to achieve precise pose correction. Combined with a nonlinear geometric observer featuring strong convergence properties, this approach enhances localization accuracy and robustness through multisensor fusion. Experimental results on publicly available large-scale datasets and our self-collected datasets demonstrate that our method outperforms existing map-based localization methods by achieving the lowest state update time and reducing computational costs by over 50%, while maintaining robust performance in complex environments. Data will be publicly available on https://github.com/iron2523/sparse-map_localization

Low-cost sensors in smartphones accumulate significant drift during long-term Visual-Inertial Odometry (VIO), while Global Navigation Satellite System (GNSS) signals are frequently blocked or degraded in complex urban environments. To address these issues, we propose Mobile-GVIO, a motion-state-aware GNSS-VIO fusion framework designed for mobile devices to achieve stable 2D global positioning. This framework utilizes GNSS constraints to suppress VIO drift error in open-sky areas, while leveraging the high-frequency smoothness of VIO in indoor and small-scale scenarios. To enhance robustness, we design a lightweight GNSS gating module to filter out low-accuracy and abnormal velocity observations. Furthermore, we analyze short-term trajectory geometry to evaluate continuous turning confidence, dynamically adjusting VIO and GNSS factor weights in the factor graph optimization. Experimental results on public datasets and real-world datasets show that, compared with state-of-the-art algorithms such as ORB-SLAM3, VINS-Fusion, and GICI-LIB, proposed Mobile-GVIO achieves higher positioning accuracy, trajectory consistency, and robustness during turns and stationary states. Finally, to benefit the research community, our fine-grained evaluation datasets are publicly released at https://github.com/SZU-Rob-IPNP-Lab/Mobile-GVIO-Dataset.

Infrared small target detection (IRSTD) plays a critical role in aerospace surveillance and remote sensing systems, yet remains challenging due to low signal-to-noise ratios and complex background interference. While conventional approaches suffer from manual parameter sensitivity, causing high false alarm rates, existing deep learning-based methods exhibit limited performance due to their reliance on spatial features alone, which compromises semantic specificity and localization precision. To overcome these limitations, we propose a wavelet-based cross-domain multistage feature fusion network (WCDMF-Net) that synergizes frequency-domain analysis with spatial-contextual learning. Specifically, we construct a parallel spatial-frequency dual-stream architecture that incorporates Haar wavelet transform (HWT) and inverse discrete wavelet transform (iDWT) for multiscale frequency feature extraction and reconstruction. Afterward, we design an adaptive spectral-spatial fusion block that first enhances features within the spatial and spectral domains and then adaptively fuses these refined representations to leverage the complementary strengths of both domains. We further design a multiscale feature aggregation module (MFAM) to aggregate multiscale output feature maps in the decoding phase. Finally, a multilevel cross-entropy loss function is used to achieve accurate target localization and mask prediction. Extensive experiments on NUDT-SIRST, NUAA-SIRST, and IRSTD-1k datasets demonstrate the effectiveness of our WCDMF-Net, compared with other state-of-the-art (SOTA) methods.

The growing need for reliable localization in feature-scarce environments like tunnels and underground facilities has made multi-sensor fusion a key technology for robust indoor positioning. Traditional RGB-based methods often struggle in low-light conditions due to insufficient visual features. Although less susceptible to darkness, infrared imaging is challenged by sparse image features. To overcome these limitations, this paper proposes a loosely-coupled fusion framework that integrates infrared imagery, Ultra-Wideband (UWB), and Inertial Measurement Unit (IMU) data. The system incorporates a geometry-aware dual-network architecture with a channel-spatial attention mechanism to enhance feature discriminability in low-visibility environments. IMU and UWB subsystems provide essential localization constraints. These data streams are fused using an Error State Kalman Filter (ESKF), a weighting module, and a Rauch-Tung-Striebel (RTS) smoother. This integration of infrared-derived pose environments estimates, UWB ranges, and IMU motion predictions enables accurate and robust positioning.. The proposed system was validated in two environments, including a real tunnel scenario. Experimental results show a 37.6% to 53.1% reduction in positioning error compared to a UWB-only system, demonstrating the proposed method's reliability and stability in complex.

Global navigation satellite system (GNSS) has been widely used in modern delivery services and pedestrian navigation systems, providing meter-level or even centimeterlevel positioning accuracy. However, in urban areas, the dense urban environment induces severe multipath and NLOS propagation errors on the order of tens or even hundreds of meters. Under the condition of multiple faults, fault detection and exclusion are difficult, particularly in the cross-street direction. In this study, various GNSS features are explored in machine learning algorithms to identify the sidewalk, and the GNSS position is determined under the sidewalk constraint. Different machine learning techniques were tested, and then the results show that Convolutional Neural Networks (CNN) delivers the best performance, with a street-side classification accuracy of better than 94%. Both static and dynamic tests were carried out to evaluate the positioning performance with the sidewalk constraint. Results show that the positioning accuracy is improved from tens of meters to less than 5 meters.

Multi-agent collaborative perception significantly improves autonomous driving safety by sharing complementary information to overcome individual limitations owing to occlusions. A primary goal is to navigate the critical trade-off between perception performance and communication bandwidth. However, existing methods struggle to achieve this balance, treating all information equally without considering each agent’s specific situation. To address this issue, this study proposes CoGMoE, a novel collaborative perception method that models the V2V communication as a structured, hierarchical reasoning process. Specifically, CoGMoE provides three distinct advantages: i) it selects a sparse set of semantically salient keypoints from each vehicle, significantly reducing communication overhead while preserving important information; ii) it constructs a hierarchical communication graph that establishes direct alignment links between the corresponding position areas of different vehicles, explicitly separating them from the internal links used for context reasoning; and iii) it uses a graph mixture-of-experts (GraphMoE) architecture governed by multi-round expert deliberation to dynamically assign experts for each link type, achieving superior robustness using iterative feature refinement. Extensive experiments on both simulated and real-world datasets demonstrate that our proposed CoGMoE outperforms state-of-the-art collaborative perception methods in achieving detection accuracy and communication bandwidth trade-off.

Visual localization is a critical component in geographic information systems and location-based services, as it enables precise spatial awareness and navigation. Although current learning-based scene coordinate regression methods have achieved superior accuracy in indoor environments, using the entire or the uniformly sampled feature map for training the scene coordinate regression model often leads to less stable localization performance due to the outliers or low-texture areas, posing challenges in adapting to outdoor environments. To tackle this issue, we propose semantic-enhanced scene coordinate regression network (SEN), which improves visual localization capabilities, particularly in outdoor environments. The SEN integrates an auxiliary semantic feature extraction branch to actively determine the reliable region for scene coordinate regression, filtering out the noninformative pixels. Subsequently, the network creates a randomized buffer to regress the scene coordinates, effectively reducing correlated gradients and achieving efficient optimization. The final regression layer accurately maps 2-D pixel coordinates to their corresponding 3-D scene coordinates, and the pose of the camera is obtained through differentiable RANSAC. Extensive experiments on three representative datasets demonstrate that SEN achieves state-of-the-art performance across various scenes, delivering exceptional performance in both indoor and outdoor environments.

Localization is an essential issue for autonomous underwater vehicles (AUVs). Multi-sensor fusion is an effective way to enhance the positioning performance of AUVs. However, underwater multi-sensor fusion faces challenges such as heterogeneous frequency and dynamic availability of sensors. Traditional filter-based algorithms suffer from low accuracy and robustness when sensors become unavailable. The factor graph optimization (FGO) can enable multi-sensor plug-and-play despite data frequency. Therefore, we present an FGO-based strapdown inertial navigation system (SINS) and long baseline location (LBL) system tightly coupled navigation system (FGO-ILNS). Sensors such as Doppler velocity log (DVL), magnetic compass pilot (MCP), pressure sensor (PS), and global navigation satellite system (GNSS) can be tightly coupled with FGO-ILNS to satisfy different navigation scenarios. In this system, we propose a floating LBL slant range difference factor model tightly coupled with IMU preintegration factor to achieve unification of global position above and below water. Furthermore, to address the issue of sensor measurements not being synchronized with the LBL during fusion, we employ forward-backward IMU preintegration to construct sensor factors such as GNSS and DVL. Moreover, we utilize the marginalization method to reduce the computational load of factor graph optimization. Simulation and public dataset experiments have verified that, compared to filter-based algorithms such as the extended Kalman filter and the federal Kalman filter, as well as the state-of-the-art optimization-based algorithm ORB-SLAM3 and VINS-Fusion, our proposed FGO-ILNS leads in accuracy and robustness.

With the widespread adoption of smartphones and IoT technologies, the demand for location-based services (LBSs) has grown significantly. However, positioning accuracy often fails to meet application requirements due to signal interference from architectural obstructions and other factors. To address the non-line-of-sight (NLOS) errors in ultrawideband (UWB) and the error accumulation in pedestrian dead reckoning (PDR), this article proposes a tightly coupled PDR/UWB indoor positioning method with NLOS adaptive correction. By integrating extended Kalman filtering (EKF), the method eliminates pedestrian height parameters through coplanar base station projection and adaptively corrects NLOS ranging errors using chi-square testing. Experimental validation in both nonoccluded and occluded underground parking environments demonstrates that, compared to standalone UWB positioning, the loosely coupled and tightly coupled algorithms improve positioning accuracy by 41.28% and 44.84% in nonoccluded conditions and by 36.52% and 42.89% in occluded conditions, respectively. The results indicate that the tightly coupled algorithm achieves higher precision and exhibits superior robustness in complex environments, providing an effective solution for high-accuracy indoor positioning.

Tunnel is a vital component of modern transportation system and plays a key role in improving traffic efficiency and reducing travel time. Reliable and continuous vehicle positioning in tunnels is crucial for traffic safety. Conventional vehicle navigation systems heavily rely on the GNSS, which provides accurate positioning in open environments but suffers significant performance degradation in GNSS-denied areas such as tunnels, underground parking lots, and dense urban regions. This study proposes a tunnel vehicle positioning method based on magnetic field signals measured by smartphones. Ferromagnetic materials embedded in tunnel structures generate distinctive magnetic field, which can be captured as sequential features along the driving path. A magnetic fingerprint database is constructed for the tunnel, and Fast-DTW algorithm is employed to match observed magnetic sequences with the database to estimate the vehicle's absolute position. Extensive field experiments were conducted in the Shenzhen-Zhongshan tunnel using multiple smartphone models. Results demonstrate that the proposed system achieves stable and accurate positioning throughout the tunnel without requiring additional infrastructure, offering a practical solution for GNSS-denied tunnel navigation.

The high risks in tunnel construction underscore the critical necessity for intelligent tunnel construction. Unmanned tunnel data collection is vital for intelligent construction; additionally, semantic segmentation aids in understanding the environment. However, complex tunnel terrains are challenging for three-dimensional (3D) laser scanning, and diverse interior structures and nontunnel elements complicate accurate segmentation by subsequent networks. Therefore, this paper proposed a tunnel mobile 3D mapping system (TMMS) for complex terrain in construction tunnels using a quadruped robot and simultaneous localization and mapping (SLAM). Additionally, a deep learning-based semantic segmentation network (TUC-Net) is proposed for analysing 3D point clouds in tunnels under construction. The research presented the neighbourhood feature perception enhancement (NFPE) module to enhance the representation of local features, introduce a self-attention (SA) module and improve the loss function to improve network accuracy. The NFPE module enhances feature aggregation for unstructured objects, and the SA module improves the learning of global features, critical for tunnel point cloud segmentation. The TMMS is used to collect point cloud data from tunnels under construction, leading to the creation of the tunnels under construction point clouds (TUCPC) dataset for training and evaluating the TUC-Net network. Compared to other 3D point cloud semantic segmentation methods, the proposed method demonstrated superior performance, achieving an overall accuracy (OA) of 99.45% and a mean intersection over union (mIoU) of 94.06%, surpassing that of other methods by at least 4.41%. In addition, ablation studies were also performed on the NFPE and SA modules to validate their efficacy.

Reliable indoor localization is crucial for location-based services.Unlike outdoor environments where the Global Navigation Satellite System (GNSS) is prevalent, indoor localization systems employ diverse methods to enhance the accuracy of individual devices. However, these methods face limitations, such as the dependence on pre-existing map data and the necessity of installing anchors. The advancement of the Internet of Things (IoT) and the increasing availability of smart devices have enabled the development of more flexible and dynamic indoor localization solutions. In this article, we propose a novel method to enhance indoor localization through cooperative localization framework. The core concept involves utilizing existing robots as mobile robot anchors to enhance pedestrian localization accuracy through interaction with pedestrians, particularly in environments lacking fixed anchors. We employed a factor graph optimization approach to tightly couple intradevice and interdevice data. This integration dynamically adjusts the inclusion of anchor data based on its quality, thereby minimizing error propagation. The experimental results demonstrate that the localization accuracy of our proposed method better than extend Kalman filter algorithms, emphasizing the potential of mobile IoT devices in indoor localization systems.

A robust back-end module with loop closure detection is crucial for accurate positioning and mapping in LiDAR-based simultaneous localization and mapping (SLAM) systems, particularly in Internet of Things (IoT) environments where multiple devices collaborate. Traditional methods that rely on images or point clouds often fail in environments with similar structures or textures, leading to incorrect loop closures. To address this, we propose a novel LiDAR SLAM system that integrates a front-end odometry module, a loop closure detection module using text semantics and geometric constraints, and a global optimization module. By using cameras on an unmanned ground vehicle (UGV), the system captures text information from the environment, enabling semantic matching to identify potential loops. Geometric constraints help eliminate erroneous loops caused by identical text in different locations. Evaluations on datasets with similarly structured environments, such as indoor parking lots, outdoor campus areas, and mixed indoor-outdoor scenes, show that our method significantly improves loop closure detection accuracy and global precision compared to existing state-of-the-art approaches. Our research can support autonomous IoT systems and multiagent systems that rely on accurate positioning and mapping, with potential applications in embodied intelligence, self-driving cars, and smart cities.

Location-based services (LBS) have become an integral part of daily life and work for the general public. However, achieving widespread and accurate positioning in typical indoor environments remains a significant challenge, particularly in multi-floor indoor parking lots where radio frequency signals like WiFi are often unavailable. Indoor magnetic matching presents a viable solution, but it requires reducing mapping costs through the use of crowdsourced data. To tackle this issue, we propose an innovative method for constructing magnetic maps using crowdsourced vehicle data. Our approach introduces a multi-user joint vehicle dead reckoning technique based on graph optimization, which provides consistent directional estimates of crowdsourced vehicle trajectories. Subsequently, we establish associations between different vehicle trajectories using multi-attribute features of the magnetic field. Building on this foundation, we propose a global trajectory optimization with inequality and equality constraints to achieve precise estimation of crowdsourced vehicle trajectories. Testing with simulated data from two three-floor underground parking lots demonstrates that the proposed method, utilizing only on-board smartphone sensor data, achieves plane and elevation errors of less than 2.75 meters (95%) and 0.59 meters (95%), respectively. Additionally, the magnetic matching positioning error based on crowdsourced magnetic sequence maps is less than 2.29 meters (95%).

Cross-modal localization based on natural language provides a flexible and infrastructure-free alternative to conventional positioning systems, yet existing text-point cloud methods struggle with semantic ambiguity, sparse geometry, and weak cross-modal alignment. We propose QGLoc, a query-guided cross-modal localization framework that integrates text, images, and 3D point clouds for large-scale outdoor environments. An annotation-free 3D semantic map is constructed using SAM and CLIP to enable zero-shot labeling. A coarse-to-fine pipeline retrieves candidate regions via text-cell similarity and refines camera pose through visual-geometric fusion and query-guided multi-head cross-attention. A contrastive objective further enforces discriminative alignment between language queries and scene representations. Experiments on the KITTI360Pose benchmark demonstrate consistent improvements over state-of-the-art methods in both coarse retrieval and fine-grained localization.

The precise localization of patrol robots in petrochemical plants is essential for operational efficiency and safety. Traditional localization methods often struggle in these environments due to dynamic conditions and complex structures. This paper introduces a new localization method that combines map-matching with LiDAR-inertial fusion to tackle these challenges. By integrating high-order inertial odometry with nonlinear geometric observers, the method enhances pose estimation accuracy and real-time performance. Key contributions include the development of a lightweight framework for embedded systems and the integration of advanced algorithms to improve accuracy and efficiency. Experiments in a simulated petrochemical environment show the method’s effectiveness, achieving a positioning error of just 2.6 cm and a pose estimation frequency of 50 Hz. These results demonstrate the method’s potential for reliable navigation and monitoring in complex industrial settings, contributing to safer and more sustainable operations.

Pedestrian tracking in surveillance videos is crucial and challenging for precise personnel management. Due to the limited coverage of a single video, the integration of multiple surveillance videos is necessary in practical applications. In the realm of pedestrian management using multiple surveillance videos, continuous pedestrian tracking is quite important. However, prevailing cross-video pedestrian matching methods mainly rely on the appearance features of pedestrians, resulting in low matching accuracy and poor tracking robustness. To address these shortcomings, this paper presents a cross-video pedestrian tracking algorithm, which introduces spatial information. The proposed algorithm introduces the coordinate features of pedestrians in different videos and a linear weighting strategy focusing on the overlapping view of the tracking process. The experimental results show that, compared to traditional methods, the method in this paper improves the success rate of target pedestrian matching and enhances the robustness of continuous pedestrian tracking. This study provides a viable reference for pedestrian tracking and crowd management in video applications.

Accurate localization in GPS-denied environments has always been a core issue in computer vision and robotics research. In indoor environments, vision-based localization methods are susceptible to changes in lighting conditions, viewing angles, and environmental factors, resulting in localization failures or limited generalization capabilities. In this paper, we propose the TransCNNLoc framework, which consists of an encoding–decoding network designed to learn more robust image features for camera pose estimation. In the image feature encoding stage, CNN and Swin Transformer are integrated to construct the image feature encoding module, enabling the network to fully extract global context and local features from images. In the decoding stage, multi-level image features are decoded through cross-layer connections while computing per-pixel feature weight maps. To enhance the framework’s robustness to dynamic objects, a dynamic object recognition network is introduced to optimize the feature weights. Finally, a multi-level iterative optimization from coarse to fine levels is performed to recover six degrees of freedom camera pose. Experiments were conducted on the publicly available 7scenes dataset as well as a dataset collected under changing lighting conditions and dynamic scenes for accuracy validation and analysis. The experimental results demonstrate that the proposed TransCNNLoc framework exhibits superior adaptability to dynamic scenes and lighting changes. In the context of static environments within publicly available datasets, the localization technique introduced in this study attains a maximal precision of up to 5 centimeters, consistently achieving superior outcomes across a majority of the scenarios. Under the conditions of dynamic scenes and fluctuating illumination, this approach demonstrates an enhanced precision capability, reaching up to 3 centimeters. This represents a substantial refinement from the decimeter scale to a centimeter scale in precision, marking a significant advancement over the existing state-of-the-art (SOTA) algorithms. The open-source repository for the method proposed in this paper can be found at the following URL: github.com/Geelooo/TransCNNloc.

Positioning accuracy can be compromised by the heterogeneity of software and hardware among different intelligent mobile devices. This is due to the fact that the heterogeneity of different devices leads to a significant difference in the received signal strength index of the same Bluetooth access point (AP) captured at the same acquisition point of the device. To address this issue, we propose to use the honey badger algorithm back-propagation neural network (HBA-BPNN) model for calibration. The aim of this study is to calibrate the received signal strength indicator (RSSI) received by Bluetooth sensors of distinct intelligent mobile terminal devices to solve software and hardware heterogeneity issues. Second, this article uses an indoor fingerprint localization algorithm based on an improved generalized regression neural network (GRNN) model and combines it with the calibration algorithm to build a better localization model. Finally, we verified the effectiveness of the HBA-BPNN calibration model for different test intelligent mobile terminal devices and then compared and analyzed the calibration algorithm proposed in this study with different calibration algorithms. The experimental comparative analysis results show that the positioning accuracy can reach 0.84 m by combining the proposed calibration algorithm with the positioning algorithm.

Visual perception technology is an important means to facilitate safe navigation for visually impaired people based on Internet of Things (IoT)-enabled camera sensors. However, due to the rapid development of urban traffic systems, traveling outdoors is becoming increasingly complicated. Visually impaired individuals must implement different types of tasks simultaneously, such as finding roads, avoiding obstacles, and viewing traffic lights, which is challenging for both them and navigation assistance methods. To solve these problems, we propose a multitask visual navigation method for visually impaired individuals using an IoT-based camera. A lightweight neural network is designed, which adopts a multitask learning architecture to perform scene classification and path detection tasks simultaneously. We propose two modules, i.e., an enhanced inverted residuals (EIRs) block and a lightweight vision transformer (ViT) block (LWVIT block), to effectively combine the properties of convolutional neural networks (CNNs) and ViT networks. The two modules allow the network to better learn local features and global representations of images while remaining lightweight. The experimental results show that the proposed method can achieve these tasks simultaneously in a lightweight manner, which is important for IoT-based navigation applications. The accuracy of our method in scene classification reaches 91.7%. The path direction and endpoint detection errors are 6.59 degrees and 0.09, respectively, for blind road and 6.81 degrees and 0.06, respectively, for crosswalk. The number of parameters of our method is 0.993 M, which is smaller than that of the comparison methods. An ablation study further demonstrates the effectiveness of the proposed method.