Visual simultaneous localization and mapping (VS-LAM) is fundamental for autonomous perception and navigation. However, existing systems often suffer from degraded localization and reconstruction performance in highly dynamic scenes, weak-texture regions, and illumination variations. To address these challenges, this paper proposes DANI-SLAM, a dynamic-aware neural implicit SLAM framework that integrates learning-based feature representation, dynamic scene understanding, and neural implicit mapping into a unified pipeline for robust pose estimation and reliable scene reconstruction. Specifically, DANI-SLAM employs a dynamic-aware keyframe selection strategy to suppress unreliable observations and stabilize front-end tracking. An uncertainty-aware loop closure mechanism is introduced by incorporating pose covariance into candidate retrieval and re-ranking, improving loop reliability and global consistency. Furthermore, an uncertainty-guided pixel-level neural implicit sub-map fusion strategy is designed to suppress inconsistent observations caused by dynamic objects or occlusions, resulting in more complete and geometrically consistent reconstructions. Extensive experiments on four public datasets, including the TUM RGB-D, Replica, ScanNet, and Bonn RGB-D datasets, demonstrate that DANI-SLAM consistently outperforms representative traditional and learning-based SLAM methods in both localization and reconstruction tasks. In highly dynamic TUM RGB-D sequences, DANI-SLAM achieves up to 48% lower ATE RMSE than baseline methods, while also obtaining superior reconstruction quality in terms of Depth L1, PSNR, SSIM, and LPIPS. These results validate the robustness, generalization, and effectiveness of the proposed system. The source code is available at: https://github.com/DANI-SLAM/DANI-SLAM.
Multi-camera imaging and camera arrays have become ubiquitous in many applications, such as autonomous driving, robot control, or virtual reality, and accurate disparity maps of objects and their environment are essential for reliable operation. Despite recent advances in neural networks, correctly estimating the disparity of flat and textureless objects remains challenging. In particular, we consider a scenario defined by flat, textureless surfaces illuminated by a single fixed light source, resulting in one dominant specular reflection visible on the object surface. Under these conditions, reliable geometric and photometric cues are missing, and the specular reflection often causes mispredictions, especially when using conventional methods that rely on texture information to match corresponding pixels. To address this issue, the novel Specular Reflection Disparity Estimation SRDE algorithm is introduced, which is specifically designed for the constrained scenario of planar, textureless objects and single-source illumination. Unlike conventional stereo matching methods, SRDE ignores texture and instead leverages the geometric properties of specular reflections by incorporating the position information of the reflective region, the light source, and the camera setup. We show that SRDE outperforms existing methods by a notable margin, achieving more than a 52% improvement in End Point Error on synthetic images. Further tests demonstrate superior performance on real-world data. Furthermore, we integrate SRDE into existing neural disparity estimation pipelines by selectively replacing predictions in specular regions without modifying the backbone model. This hybrid strategy enables additional performance gains without requiring network retraining.
This paper proposes a state estimation algorithm designed to effectively suppress process and measurement noise outliers and enhance the precision of state estimation. To accurately model heavy-tailed noise contaminated by outliers, we adopt a generalized multivariate Laplace (GML) distribution. Through a rigorous analysis of its statistical properties, we identify a specific parameterization of the GML distribution (the rate parameter of the exponential distribution in the hierarchical model is determined to be 0.2, departing from the conventional setting of 1) that accurately captures the characteristics of the non-stationary noise. Based on this, a state-space model is formulated within a hierarchical Bayesian framework. This model separately accounts for the uncertain noise covariance matrices and the heavy-tailed factors. Given the non-Gaussian nature and analytical intractability of the joint posterior distribution, we employ a fixed-point variational Bayesian approach to derive variational distributions for both noise parameters and the state vector. The superiority of the algorithm is confirmed through a real-world quadrotor unmanned aerial vehicle tracking experiment and numerical simulations. Results show significant improvements in estimation accuracy and robustness over existing filtering techniques.
In this paper, we investigate the reliable state estimation problem of cyber-physical systems over energy-harvesting decode-and-forward relay channels. To extend the transmission distance and improve the reliability of measurement signal delivery, a decode-and-forward relay is deployed in the sensor-estimator communication network. The relay node is equipped with energy-harvesting technology to collect ambient energy. During measurement signal transmission, successive denial-of-service attacks may cause packet dropouts, which are transformed into bounded stochastic input delays of the measurements. This paper aims to design a remote state estimator that guarantees the augmented dynamics to be exponentially ultimately bounded in the mean-square sense. Sufficient conditions for the existence of the desired state estimator are provided through dynamical stability theory and inequality techniques. The estimator gain is determined by solving a series of matrix inequalities. Finally, the validity of the proposed reliable estimator is verified through two simulation examples.
3D object detection plays a critical role in intelligent robotics and autonomous driving, where accurate and robust perception remains challenging under multi-modal fusion settings. Existing bird’s-eye-view (BEV)-based multi-modal methods still suffer from unreliable camera depth estimation, insufficient adaptation to object scale variations, and temporal misalignment caused by dynamic scenes. To address these challenges, this paper proposes EDAFusion, a unified multi-modal 3D object detection framework with enhanced BEV generation and fusion. Specifically, a LiDAR-guided multi-level depth enhancement (LMDE) strategy is introduced to improve the robustness of camera BEV representation under challenging illumination conditions; a dynamic scale attention fusion (DSAF) module is designed to enhance cross-modal interaction across objects of different scales; and a motion-guided temporal aggregation (MTA) method is developed to improve temporal consistency under object motion and partial occlusion. Experiments on the nuScenes test set demonstrate that EDAFusion achieves 72.1% mAP and 74.5% NDS, outperforming BEVFusion by 1.9% mAP and 1.6% NDS. These results demonstrate the effectiveness of the proposed framework for reliable multi-modal 3D perception in intelligent measurement and sensing systems.
Joint calibration is essential for ensuring the geometric consistency and reliability of multimodal fusion perception. However, existing methods are difficult to apply directly to roadside perception systems deployed at urban intersections because reliable external calibration cues are often unavailable. To address this challenge, we propose a targetless trajectory-based joint calibration framework, termed TBJC. The framework exploits object trajectories observed by heterogeneous sensors and comprises three core modules: trajectory tracking, cross-modal trajectory matching, and extrinsic parameter estimation. By leveraging the strong spatiotemporal correlations and stable geometric consistency embedded in cross-modal traffic trajectories, TBJC establishes trajectory-level correspondences between 2D image observations and 3D LiDAR observations. These correspondences are then used to estimate the camera–LiDAR extrinsic transformation. Building on this framework, we develop TBJC-Alpha as a practical implementation for camera–LiDAR calibration. In TBJC-Alpha, trajectory tracking and cross-modal matching combine learning-based association scoring with global assignment optimization. For extrinsic parameter estimation, Efficient Perspective-n-Point (EPnP) first generates an instantaneous pose estimate, after which Random Sample Consensus (RANSAC) is applied to historical pose estimates to suppress transient outliers and improve estimation stability. TBJC-Alpha is systematically compared with representative targetless calibration methods on the V2X-Seq and TUMTraf datasets. Ablation studies examine the effects of key algorithmic components, while sensitivity analyses evaluate performance under different time-window lengths and trajectory densities. The results demonstrate that TBJC-Alpha achieves competitive performance in accuracy, stability, efficiency, and reliability across the evaluated datasets.
The oil-water two-phase flow measurement is crucial in petrochemical industries. The swirl flowmeter is widely used in the flow rate measurement of oil or water media. However, the phase content of oil-water mixed media is hard to obtain directly from the signals of swirl flowmeters. To address this issue, we proposed a Stochastic Featured Transformer (SFT) to estimate the phase contents of oil-water two-phase flow. The essence of SFT is to estimate the motion parameters of the swirl flowmeter’s piezoelectric diaphragm from the perspective of stochastic processes, and jointly train the Transformer encoder with both the parameters and original signals. In the experiment, we used a swirl flowmeter to test oil-water mixtures with oil contents ranging from 1.71% to 29.08% at a fixed total flow rate of 3.80 m³/h. The signals from piezoelectric sensors were collected and input into the SFT for prediction. The experimental results indicate that the proposed SFT model achieves an MAPE of 4.67% and an R² of 0.9875. This method provides a new approach to the swirl flowmeter signal processing, and also puts forward an efficient solution for oil-water measurement.
Reliable quench detection in superconducting machines is hindered by strong induced voltage interference, and suffers from a high risk and cost of direct validation in actual, high-value superconducting machines. To address this challenge, this paper presents a safe, offline experimental testing system to validate a novel quench detection method based on the terminal-voltage differential of symmetrical HTS coils, where a modular magnetic field emulator is constructed to replicate the machine's time-varying electromagnetic environment experienced by HTS coils. Based on this offline experimental testing system, the novel quench detection method is experimentally evaluated under various operating conditions. Results show that the filtered differential signal can remain below the threshold voltage during normal operation of the HTS coils, whereas a clear and detectable signature appears during controlled HTS-coil quench events. This work provides a high-precision, repeatable, and cost-effective off-machine experimental validation scheme, demonstrating the potential of the proposed quench detection method for superconducting machines.
Accurate measurement of the refractive index profile (RIP) is essential for multicore fibers (MCFs) design and optical performance evaluation. Optical computed tomography is a widely adopted measurement technique. The RIP is usually measured with the objective focused at the center of the fiber. However, the limited depth of field of the objective lens (~2 μm) leads to defocused projection images, which is a major factor contributing to the distortion of the outer cores in the measured RIP. In this paper, for the first time, a measurement method based on multi-focus phase sinograms fusion is proposed to address this issue. Following fusion of the upper-focused, center-focused, and lower-focused sinograms, the RIP of the MCF is reconstructed. Simulations and experiments are performed to validate the proposed method. The results demonstrate that, compared with the commonly used center-focused imaging measurements, the proposed method enables more accurate characterization of the outer-core structures.
The power spectrum estimation (PSE) based on discrete Fourier transform (DFT) is widely adopted for line spectrum detection. The coherent averaged PSE (CAPSE) method can improve the temporal gain by compensating for the phase shifts induced by the fractional frequency, but it suffers from blind compensation across all frequency bins after the time-dimensional DFT (TDFT), which raises the noise fluctuation. Furthermore, while beamforming can provide additional spatial gain for weak line spectrum detection, the whole process is computationally expensive and only enhances the line spectrums from the steered direction. To overcome these limitations, an array-based line spectrum detection framework is proposed. Instead of processing the beam output, the framework operates directly on the spatiotemporal phase spectrum across the uniform line array. Specifically, TDFT is applied to the phase sequence of each hydrophone, and line spectrums are detected by identifying fractional frequency lines, which are straight-line structures formed by strong peaks in the TDFT matrix. The framework inherently exploits temporal coherence and inter-hydrophone phase relationships, yielding detection gains beyond single-hydrophone coherent integration while avoiding the azimuth dependence of conventional beamforming. The simulation shows that, compared to the CAPSE method using a single hydrophone, the proposed method reduces the required signal-to-noise ratio by 7.6 dB, 6.6 dB, and 4.1 dB, when achieving a detection probability of 0.9 and with target azimuth linear variations of 0°, 10°, and 20°, respectively. The measured data also verifies the effectiveness of the method.
Wearable electrocardiogram (ECG) monitoring requires reliable acquisition of a sub-millivolt cardiac biopotential under real-use conditions in which the electrode-skin interface, body motion, mains-coupled common-mode interference, and per-patient variability each contribute distinct sources of measurement uncertainty. The analog front-end (AFE) is the signal-conditioning interface between the electrode-skin system and the digital backend; its architectural choices set the input-referred noise, common-mode rejection ratio (CMRR), input impedance, dynamic range, electrode-offset tolerance, and bandwidth fidelity that bound the achievable ECG signal quality. This paper reviews ECG AFE architectures, auxiliary signal-conditioning circuits, and mixed-signal integration techniques reported between 2010 and 2025, tracing each architectural decision to its measurement-quality consequence. Auxiliary circuits including DC-servo loops, ripple-reduction loops, input-impedance enhancement, driven-right-leg interference suppression, and programmable filtering are assessed by their effect on the measurement chain. More than 100 published designs are compared using input-referred noise, CMRR, input impedance, noise efficiency factor (NEF), power efficiency factor (PEF), dynamic range, electrode-offset tolerance, and validation conditions. The review identifies a saturation in noise-efficiency factor among voltage-domain front-ends, systematic gaps in measurement-validation reporting, and the need for greater consistency in benchmarking electrode-impedance imbalance and motion-artifact tolerance. The paper closes with measurement-oriented design guidelines and a research agenda addressing system-level energy management, electrode-AFE codesign, multimodal physiological measurement, and convergence with electrochemical sensing.
This paper presents a FPGA-based dual-channel measurement system for real-time monitoring of short-circuit signatures in silicon carbide metal oxide field effect transistors (SiC MOSFETs). A differential resistor-capacitor (RC) based voltage slew rate (dv/dt) sensing circuit with clamping-diode buffering measures the turn-on drain–source voltage (VDS) transient and converts it into a clean digital event with nanosecond-scale timing, while an optimized desaturation (DESAT) channel measures the on-state VDS under fault-under-load conditions using a minimized blanking capacitor. Both channels are interfaced to an FPGA, which implements programmable timing windows and decision logic, forming a complete mixed-signal measurement chain from high-dv/dt device terminals to digital fault flags. The transfer characteristics and timing behavior of the dv/dt and DESAT paths are analytically modeled and used to guide component selection for bandwidth, settling time, and noise immunity. Experimental characterization on a 1.2 kV SiC MOSFET demonstrates dv/dt edge detection within 35–40 ns and DESAT-based fault indication within 205–240 ns over a 400–800 V DC bus, with <3% variation in detection delay between –80 °C and 150 °C. The proposed system provides robust discrimination between normal switching, hard-switching fault, and fault-under-load events without false triggering under high dv/dt and voltage spikes. Compared with conventional SC monitoring methods using Rogowski-coil or DESAT solutions, the proposed measurement architecture offers a compact, low-cost, and temperature-robust measurement front end for instrumentation of short-circuit phenomena in wide-bandgap power converters.
To improve the navigation accuracy of inertial measurement units (IMUs), a system-level full-parameter calibration method based on a dual-observation ten-position rotation scheme, which integrates attitude–velocity matching with a Kalman filter algorithm, is proposed in this paper. Compared with the conventional 29-position calibration scheme, the proposed method achieves high-precision estimation of all IMU parameters while reducing the total calibration time from 63 min to 22 min, thereby significantly enhancing calibration efficiency. Firstly, comprehensive error models for the gyroscope and accelerometer are established, including scale factor errors, nonorthogonality, sensor-to-turntable misalignment, gyro and accelerometer biases, lever-arm offsets, and quadratic errors. Secondly, a simplified ten-position rotation scheme is designed to maximize the excitation of each error component under limited test configurations. Finally, Kalman filtering is performed using attitude and velocity matching to estimate the IMU error parameters. System-level simulations and experimental tests verify the effectiveness and applicability of the proposed algorithm. Both simulation and experimental results demonstrate that all IMU error parameters are estimated by the proposed method, and the correctness and practical feasibility of the proposed calibration scheme are verified by the designed tests.
This study presents a low‑power discrete‑time (DT) level‑crossing (LC) ADC for biopotential signal measurement in wearable instrumentation. The design employs a hardware-reuse technique that time-multiplexes a single 7-bit CDAC for both coarse LC tracking and fine SAR quantization, thereby eliminating the need for separate DACs. Two-stage LC detection with programmable second-stage windows improves capture of fine signal variations without hardware overhead. Fabricated in 180 nm BCD, the 0.0896 mm² prototype achieves a peak SNDR of 67.6 dB (10.94-bit ENOB) at 100 Hz and maintains 61.2 dB SNDR (9.87-bit ENOB) at a full 20 kHz bandwidth. Full-bandwidth operation consumes 4.68 μW (59.7 fJ/conv.-step); the efficiency improves at lower frequencies to 27 fJ/conv.-step, and the power drops to 0.79 μW for sparse biological signals, demon-strating its suitability for energy-constrained wearable instrumentation.
Smartphone Bluetooth Low Energy (BLE) application programming interfaces (APIs) often report only channel-aggregated received signal strength (RSS), which mixes measurements from different advertising channels and can distort fingerprint features. This paper proposes a channel-aware RSS processing strategy for BLE fingerprinting. When advertising-channel identifiers are available, the method selects the highest-power channel and applies one-sided outlier suppression in the linear-power domain. When channel identifiers are unavailable, an offline kernel density estimation (KDE)–based separation recovers dominant components and selects the highest-power component, while an online lightweight rule avoids unreliable peak detection for short sample windows. Experiments on multiple public datasets using nearest neighbor (NN) and weighted k-nearest neighbors (WkNN) show consistent improvements in several scenarios, with the largest gains observed when the aggregated RSS distribution is strongly multimodal.
Non-contact vital sign measurement is crucial for improving health monitoring instruments, demonstrating high accuracy while ensuring user convenience and comfort in clinical, home, and remote environments. While camera-based remote photoplethysmography (rPPG) and ultra-wideband (UWB) radar micro-motion sensing offer promising approaches, each faces limitations from environmental factors—cameras from illumination and skin variability, radars from motion artifacts and multipath effects. In this paper, we propose a dynamic deep fusion approach for integrating synchronized RGB and RF signals, combining cross-modal complementary features to reconstruct PPG waveforms under realistic conditions. We also introduce a multimodal instrumentation benchmark dataset—the first combining visual and multi-type UWB radar measurements (one FMCW and two IR-UWB for sensing diversity) with ground-truth PPG— collected from 17 subjects at varying distances (0.5–1.5 m), angles (−45° to +45°), and activities (resting, phone use). The dataset supports future evaluation across varying sensing conditions and subject behaviors. In this study, the proposed fusion framework is evaluated on a subset corresponding to the nominal sensing configuration (0.5 m distance, 0° viewing angle, and resting condition), our fusion system achieves an MAE of 0.1029, an MSE of 0.0152, and a heart-rate estimation error of 2.651 ± 2.430 BPM, surpassing unimodal and static fusion baselines. These results demonstrate improved signal fidelity and noise robustness, supporting applications in advanced instrumentation, telemetry, and pervasive monitoring systems. The dataset and code are available at: https://github.com/NguyenVanKhai2412/Camera-UWB Radars-Fusion-VitalEstimation.git.
In ultrasonic partial discharge detection, sidelobe-induced false sources are a primary cause of localization failure. However, conventional planar arrays typically do not fully exploit element degrees of freedom, are computationally expensive, and are prone to local optima, which limits sidelobe suppression and localization reliability. This paper proposes a meshfree ultrasonic array with a physics-informed neural network and chaos-integrated population-restart (PINN-CIPOP) design framework to suppress false sources and improve PD localization accuracy and robustness. First, a meshfree model of a planar array with a fixed aperture is established, and Delaunay triangulation acceleration algorithm is formulated for the minimum spacing constraint, which expands the two-dimensional freedom of the array elements. Second, tent and logistic chaotic mappings are used for population initialization, and a chaotic restart strategy is incorporated into covariance matrix adaptation evolution strategy algorithm to improve global coverage in multi-peak regions and enhance the ability to escape local optima. Third, the peak sidelobe level function smoothed by the log-sum-exp and Kreisselmeier–Steinhauser functions is embedded into the physics-informed neural network as the physical constraint equation, and the surrogate model is pre-trained to reduce the computational cost of fitness evaluations, thereby improving optimization efficiency without sacrificing search accuracy. Experiments over 100 runs reveal average performance gains of 187.1% and 102.4% respectively for two fitness functions, outperforming state-of-the-art algorithms in sidelobe suppression. Finally, field tests of a fabricated prototype confirm its physical realizability and demonstrate accurate, repeatable localization, markedly outperforming a conventional spiral array.
Electromagnetic tracking (EMT) is a core technology for real-time instrument navigation in minimally invasive surgery. Mainstream EMT systems predominantly rely on the physics-based magnetic dipole (MD) model. While this model offers high efficiency, it suffers from two inherent limitations: (1) Model mismatch in the proximal region due to idealized assumptions; (2) Performance degradation in the distal region caused by rapid signal decay. These limitations necessitate a challenging trade-off between accuracy and positioning range. To resolve this, this paper proposes a novel Spatially-Partitioned Physics-Informed Mixture of Experts (PIMoE) model. The architecture employs independent proximal-region and distal-region experts, trained with a physics-informed regularization strategy, to systematically correct the inherent biases of the physical model. Experimental results demonstrate that, compared with the baseline MD model, the proposed method reduces the global root mean square error (RMSE) from 9.89 mm to 3.34 mm over the evaluated 500×500×500 mm3 workspace, and reduces the proximal-region RMSE from 2.83 mm to 1.38 mm within the evaluated 300 × 300 × 350 mm3 subregion. The results indicate that PIMoE provides a physics-informed modeling strategy for improving the 3D positioning accuracy of custom-developed EMT systems in the evaluated metal-free environment.
