Vehicle tracking in complex traffic environments remains a challenging task due to frequent target occlusion, high visual similarity among vehicles, and cluttered backgrounds. These factors often result in tracking drift and target misidentification, undermining the reliability of tracking systems in real-world applications. To address these challenges, this paper presents L2G-TransT, a vehicle tracking framework based on local-to-global multi-scale attention. The proposed method hierarchically integrates fine-grained local features, mid-scale spatiotemporal context, and global semantic cues through a combination of window attention, shifted window attention, and global self-attention mechanisms. A dynamic weighting strategy is employed to adaptively fuse multi-scale features based on the target’s visibility and appearance changes, thereby enhancing the model’s adaptability to challenging scenarios. In addition, a multi-head cross-attention module is introduced to improve feature discriminability by suppressing background noise and emphasizing target-relevant regions. Extensive experiments demonstrate that L2G-TransT achieves better tracking performance, improving both accuracy and robustness under occlusion and appearance ambiguity. The codes and datasets are available at https://github.com/chd-via-lab/L2G-TransT.git.

Cross-modality fusion for pavement scenes is essential for effective automated defect detection in large-scale infrastructure inspections. Current fusion methods often overlook modality coupling relationships, limiting the clear representation of subtle defects and effective suppression of background noise. To address these issues, we propose a novel cross-modality feature decoupling fusion framework specifically designed for pavement scenarios. This framework explicitly models both modality-specific distinctions and inter-modal correlations. Our method leverages frequency-driven decoupling information extracted from depth images to guide the fusion process effectively. For improved feature representation in large-scale automation, we introduce a two-stage training strategy. First, decoupling information is acquired independently; then, it is fused at the pixel level. Additionally, we design an Adaptive Fusion Rate Block (AFRB) to dynamically allocate features during the Feature Allocation Fusion (FAF) stage. By incorporating frequency-driven decoupling, our pixel-level fusion significantly preserves structural clarity and subtle defect details, effectively capturing even non-significant defects. Benchmark experiments confirm that our approach achieves superior visual quality and enhanced defect detection performance. These results underline its effectiveness and potential for automated large-scale pavement inspection systems.

A major obstacle for high-level tasks is the unpredictable image degradation. While several architectures proposed to address this, they fail under extreme degradation. Therefore, we introduce a novel cross-modality pipeline called AIRMamba, designed to holistically and robustly restore images degraded due to extreme weather. Specifically, we devise a strategy that utilizes infrared images to create compact high-frequency priors for the restoration process. Meanwhile, we leverage long-range modeling capability of Mamba to achieve both feature extraction and interaction. We emphasize extracting low-frequency representations from the ground truth and achieving this task through a regression-based approach. Consequently, AIRMamba can achieve reliable restoration through large-gap cross-domain guidance. To facilitate this task, we have constructed a cross-modality restoration benchmark, named WeatherInfrared. Our pipeline is simple, robust, and outperforms several state-of-the-art methods in benchmark evaluations.

The infrared image can distinguish the targets from the background based on radiation differences, providing more significant target visibility under dense haze. Fusion of visible haze images with infrared prior representations can generate high-quality fused images for high-level tasks. Consequently, we propose a novel dual-modal fusion network structure that makes full use of infrared prior representations for dehazing. Specifically, we emphasize a Multi-modal Feature Extraction Network (MMFE) to extract deep multi-scale features. Meanwhile, we introduce a Multi-scale Feature Extraction module (MSFE), integrating an Efficient Dual Attenion block (EDAB) to efficiently explore more spatial and marginal information. Additionally, we propose a new feature fusion strategy, which calculates feature fusion weights based on an adaptive multi-head self-attention. Therefore, IPRDehazeNet achieves better dehazing results through dual-modal fusion. Experimental results indicate that IPRDehazeNet outperforms various advanced methods.

The widespread availability of similarity queries over trajectory data has led to numerous real-world applications, such as traffic management and path planning. With the proliferation of trajectory data, data owners often outsource storage and computation tasks to the cloud due to limited computing and storage resources. However, this scenario raises sharp security concerns, where it is critical to ensure both the integrity of query results and privacy during query processing. Furthermore, most existing works assume a single-user setting where all query users share the same key, which may lead to query privacy leakage. Therefore, in this article, we take the first step in studying the issue of multiuser and secure trajectory similarity range query (MSRQ). Specifically, inspired by the M-tree, we propose a secure index based on a distributed two-trapdoor public-key cryptosystem (DT-PKC), called M*-tree, and devise secure protocols to support multiuser query processing. We also carefully design a filtering strategy and verification scheme to ensure fast search and integrity guarantees. Finally, we theoretically analyze the security and complexity and empirically evaluate the performance and feasibility of our proposed approach.

Dynamic searchable symmetric encryption (DSSE) ensures that outsourced data can be searched and updated without compromising data availability. Recent efforts on DSSE have mainly focused on exact keyword retrieval, but considering that misspellings are common and practical, it is necessary to support the functionality of fuzzy keyword search. However, most existing fuzzy keyword search schemes do not consider malicious cloud servers and forward-privacy guarantees. The former may lead to the implementation of a fraction of search operations or forge the results and the latter may reveal the association between the newly updated data and previous search tokens. Therefore, in this article, we investigate the issue of dynamic and verifiable fuzzy keyword search with forward security (DVFKF). Specifically, DVFKF first uses locality sensitive hash to map approximate keywords into the same hash bucket, then links the hash buckets to obtain the bucket strings. Next, to accelerate the performance and support forward privacy, we propose a chain index for each bucket string through counters. Further, to guarantee the integrity of the results, we integrate the Merkle hash tree and chain index to verify the correctness and completeness of the results. Finally, we conduct formal security analysis and empirical evaluations to demonstrate the feasibility and practicality of our scheme on real datasets.

Nearshore multi-object tracking (NMOT) aims to locat and identify nearshore objects. Most approaches accomplish this task using radar and remote-sensing technologies. In contrast, video data can describe the visual appearance of nearshore objects without prior information, such as identity, location, or movement. In this study, we introduce a cross-modality pipeline to address the four major challenges of NMOT. First, we propose introducing a cross-modality bi-attention transformer (CBT) manage the information interaction between RGB and thermal infrared videos effectively. This decoupling and guidance mechanism laid the foundation for our subsequent processes. Next, we integrate the outputs of the backbone with historical frames to extract crucial temporal features. Subsequently, we refine small object detection performance by employing multi-scale feature alignment (MFA). Observations are generated by the transformer decoder. To tackle challenges arising from extensive occlusion and interactions induced by waves in NMOT, we propose guiding modulation (GM), supplemented by low-confidence boxes and multi-point corner momentum (MCM) to facilitate association. Our approach is simple, online, and real-time, showcasing outstanding performance in benchmark evaluations. The open-source implementation of our work is available at https://github.com/Ding-JianGang/Cross-Modality-MOT-in-Nearshore-Environments.