Long Zhuang

Accurate extrinsic calibration between solid-state LiDARs and cameras is essential for reliable multimodal perception, but targetless calibration remains challenging because the two sensors provide heterogeneous measurements and observe scene boundaries from different viewpoints. This paper presents a targetless solid-state LiDAR-camera calibration framework based on depth-domain matching and visibility-aware foreground boundary refinement. To reduce the modality gap between the LiDAR and camera modalities, a monocular depth estimator is used to transform RGB images into depth-like representations. Spherical projection constructs a LiDAR front-view map without relying on manually specified initial extrinsics, while preserving the integrity of the LiDAR data. Rather than treating the resulting SuperGlue matches as physically correct correspondences, we analyze the mismatch mechanism near foreground occlusion boundaries and refine the initial matches by constructing a foreground boundary confidence map from the LiDAR range map. The refined 2D-3D correspondences are then used in a two-stage estimation procedure, including RANSAC-PnP initialization and weighted nonlinear reprojection-error minimization. Experiments on the MIAS-LCEC dataset, together with qualitative evaluation on ACSC and a self-collected campus dataset, show that the proposed method improves calibration accuracy and robustness compared with representative targetless and recent calibration methods.

In recent years, significant progress has been made in place recognition (PR) methods using single camera or 4-D millimeter-wave radar. However, cameras exhibit limited robustness in complex weather and lighting conditions, while 4-D radar lacks rich semantic information. To address these issues, this study proposes a novel radar-camera fusion model for PR, termed RCPRNet. First, traditional multimodal feature fusion methods struggle to capture the hierarchical and representational differences between 4-D radar and cameras, often introducing redundancy and noise. To overcome this limitation, this article introduces the adaptive Gaussian fusion module (AGFM), which dynamically adjusts modality-specific feature weights through probabilistic modeling, thereby enabling more robust and effective feature fusion. Second, to address the challenge of feature enhancement under the severe sparsity of 4-D radar point clouds, we propose the sparse local-global feature enhancement module (SLGFEM). This module performs multihead attention operations only on sparse subsets composed of local key neighborhoods and randomly sampled global points, significantly improving recognition accuracy while reducing computational cost. Finally, two novel 4-D radar PR datasets are constructed, SNAIL4DPR and SEU4DPR, with the latter collected in-house. RCPRNet achieves state-of-the-art (SOTA) performance on both datasets.

Extrinsic calibration of 4D millimeter-wave radar and camera is a prerequisite for multi-sensor fusion in complex environments. Online calibration algorithms can dynamically correct jitter errors caused by vibrations and have increasingly become the predominant approach. However, online calibration for 4D radar-camera systems is constrained by the sparsity of radar point clouds, making it challenging to establish feature correspondences between radar and camera. This paper proposes a novel 4D radar-camera online calibration algorithm, RCOC Distill, which leverages cross-modal knowledge distillation (CMKD) to enhance 4D radar representations with LiDAR derived features, thereby improving calibration accuracy. A Local Correlation Layer (LCL) is introduced to compute feature similarity within a local window via convolution operations, reducing computational burden while maintaining local consistency. Additionally, a Radar-LiDAR Cross-Modal Alignment (RLCMA) module is proposed to further enhance radar feature density, mitigating inefficiencies in knowledge transfer from Li DAR to 4D radar. The proposed method is evaluated on two datasets (NTU4DRaDLM and Dual-Radar). Compared to state-of the-art models, RCOC-Distill achieves improvements of 38.73% and 23.48% in translation and rotation errors, respectively, on the Dual-Radar dataset. These results validate the superiority of the proposed method. Our code and dataset are publicly available at https://github.com/Zhuanglong2/RCOC-Distill.

LiDAR SLAM technology plays a critical role in applications such as 3D mapping and autonomous navigation. However, most existing SLAM systems are built on the assumption of a static environment, which is frequently violated in real-world scenarios. This results in significant map noise and incorrect data associations caused by dynamic elements. In addition, while loop closure modules help mitigate cumulative drift in LiDAR odometry, their performance often degrades severely in the presence of viewpoint changes and geometrically degenerate environments. To address these challenges, a semantic-aided LiDAR odometry method is introduced, which first leverages semantic and geometric information to extract different types of feature points, and then estimates odometry under semantic consistency constraints. Furthermore, a semantic-enhanced loop closure network is designed, which incorporates a Semantic-Spatial Fusion (SSF) module to integrate loop closure and semantic features, and a Correspondence Confidence Evaluation (CCE) module to filter out unreliable point correspondences for registration. These two components are then integrated into a complete semantic SLAM system, enabling the construction of globally consistent semantic maps. Experiments on the SemanticKITTI dataset and a customized campus scenario validate the robustness of the proposed loop closure network in both loop closure detection and global registration. When integrated with the semantic-enhanced odometry and loop closure, the resulting system achieves higher mapping accuracy than several existing SLAM methods and produces globally consistent semantic maps.

Modern millimeter-wave (mmWave) radar heat map-based object detection techniques have shown significant potential. However, radar heat maps still pose challenges for fine-grained object differentiation, and fusing camera maps with radar heat maps remains challenging. This paper proposes a multi-sensor radar object segmentation (ROS) method that fuses radar range-angle (RA) heat maps with camera RGB maps, leveraging the advantages of camera sensors for extracting semantic information and mmWave radar for extracting object position information. A method for mapping RGB to the RA coordinate system (RMTR) is proposed, eliminating the need for additional training branches and sensor calibration. A Transformer is utilized to predict the range and angle matrices of each object separately from a global perspective. These matrices then undergo an element-wise multiplication (Hadamard product) operation to generate the final pseudo-RA map. For the fusion method, this paper introduces the cross-attention fusion module (CAF), which uses the cross-attention mechanism to achieve more efficient fusion by treating the pseudo-RA feature map as the Query and the RA feature map as the Key and Value. Extensive experiments show that the proposed method achieves state-of-the-art (SOTA) performance on the CRUW and CARRADA datasets, with fewer parameters and reduced computational complexity. Our code for RC-ROSNet is available at https://github.com/Zhuanglong2/RCROSNet.

The online calibration of 4D millimeter-wave radar and camera is crucial for advancing perception and SLAM technologies in complex environments. It eliminates the reliance on manual labeling, offering real-time and convenience. However, the sparse nature of 4D radar point clouds presents challenges in establishing correspondences with camera images. This letter proposes an online 4D radar-camera online calibration method (4DRC-OC) that utilizes unified depth map representations for auxiliary training, ensuring feature alignment and modal unification between the two sensors. Due to the limited useful information within sparse depth maps, 4DRC-OC uses dynamic convolution to adaptively capture detailed features. Furthermore, this letter designs a correlation module based on channel-wise fusion (CMCF) that computes correlations between error depth maps and RGB-derived depth maps, thereby enhancing features to facilitate extrinsic parameter regression. Experimental results on the Dual-Radar dataset validate the superiority of the proposed approach in extrinsic calibration.

The rapid growth of commercial unmanned aerial vehicles (UAVs) has led to more incidents. These include illegal activities by malicious actors. Such activities pose significant threats to public facilities and people’s safety. Therefore, distinguishing UAVs from other targets is a critical measure in preventing potential hazards. This paper proposes an enhanced feature extraction and global modeling model for UAV recognition, named CT-RURnet. The model directly learns from radar echo data, firstly using the convolutional neural networks module to quickly extract local features, and then using the Transformer module to establish global connections among these local features. The proposed model primarily focuses on learning the micro-Doppler features produced by targets, which facilitates UAV recognition. The used datasets include both simulated data and the Real Doppler RAD-DAR (RDRD) dataset. The proposed network achieves accuracy rates of 98.7% (10 SNR), 97.9% (5 SNR), and 86.95% (0 SNR) on the simulated dataset under varying SNR conditions and 97.14% on the real dataset. Compared to other baseline models, the CT-RURnet consistently delivers superior results. Ablation experiments are conducted to verify the contribution of each module to the overall network performance.

A novel microstrip filtering antenna with slot coupling feed is presented in this work. An asymmetric interdigital coupling structure is used for the feed to excite the patch antenna through gap-fed coupling. Introducing a U-shaped slot on the patch surface modifies the current path to attain different resonant modes. The asymmetric coupled fingers in the low-frequency band generate a radiation null of −64 dBi, while additional resonances introduced in the high band broaden the bandwidth from 4.9 to 5.3 GHz. A horizontally shorted microstrip branch produces another null point of −28 dBi between the bands, enabling steeper roll-off and further improvement in frequency selectivity. The proposed filtering antenna provides a flexible filter response without requiring extra filtering circuits, with appreciable peak gains (6.1 dBi and 7.1 dBi) and stable radiation characteristics. This makes it suitable for WLAN (Wireless Local Area Network) and ISM (Industrial Scientific Medical) systems applications.