Autonomous vehicles (AVs) rely on LiDAR and camera sensors to perceive their environment. However, adverse weather conditions, such as rain, snow, and fog, negatively affect these sensors, reducing their reliability by introducing unwanted noise. Effective denoising of multimodal sensor data is crucial for safe and reliable AV operation in such circumstances. Existing denoising methods primarily focus on unimodal approaches, addressing noise in individual modalities without fully leveraging the complementary nature of LiDAR and camera data. To enhance multimodal perception in adverse weather, we propose a novel Adaptive Cross-Modal Denoising (ACMD) framework, which leverages modality-specific self-denoising encoders, followed by an Adaptive Bridge Controller (ABC) to evaluate residual noise and guide the direction of cross-modal denoising. Following this, the Cross-Modal Denoising (CMD) module is introduced, which selectively refines the noisier modality using semantic guidance from the cleaner modality. Synthetic noise was added to both sensors’ data during training to simulate real-world noisy conditions. Experiments on the WeatherKITTI dataset show that ACMD surpasses traditional unimodal denoising methods (Restormer, PathNet, BM3D, PointCleanNet) by 28.2% in PSNR and 33.3% in CD, and outperforms state-of-the-art fusion models by 16.2% in JDE. The ACMD framework enhances AV reliability in adverse weather conditions, supporting safe autonomous driving.

In dense crowds and complex electromagnetic environments of metro stations, UWB-based seamless payment suffers from limited positioning accuracy and insufficient stability. A promising solution is to incorporate the vision modality, thereby enhancing localization robustness through cross-modal trajectory alignment. Nevertheless, high similarity among passenger trajectories, modality imbalance between vision and UWB, and UWB drift in crowded conditions collectively pose substantial challenges to trajectory alignment in metro stations. To address these issues, this paper proposes a multi-modal trajectory progressive alignment algorithm under modality imbalance. Specifically, a progressive alignment mechanism is introduced, which leverages the alignment probabilities from previous time steps to exploit the temporal continuity of trajectories, thereby gradually increasing confidence in alignments while mitigating the uncertainty of individual matches. In addition, contrastive learning with the InfoNCE loss is employed to enhance the model’s ability to learn from scarce but critical positive samples and to ensure stable matching on the UWB modality. Experimental results demonstrate that the proposed method consistently outperforms baseline approaches in both off-peak and peak periods, with its matching error rate reduced by 68% compared to the baseline methods during peak periods.

License plate recognition is a critical perception module in intelligent transportation systems, yet severe occlusion remains a fundamental challenge in unconstrained traffic scenes because the missing visual evidence is physically absent rather than merely degraded. Although multi-view observations provide complementary cues for recovering occluded characters, existing fusion strategies usually operate in the spatial domain and therefore suffer from feature misalignment, especially when Vision Transformers rely on local patch-wise positional encoding. To address this issue, we propose Wavelet-Enhanced Transformer, a domain-transformed multi-view recognition framework for occluded license plates. The central idea is to shift feature fusion from the spatial domain to the wavelet domain, where discrete wavelet transform decomposes encoded features into low-frequency structural components and high-frequency detail components. This transformation alleviates spatial semantic ambiguity caused by viewpoint variation and positional mismatch, enabling more robust cross-view feature aggregation. On top of this representation, we design an order-agnostic memory fusion mechanism to progressively accumulate complementary evidence from multiple partial observations, and introduce a discriminator-based stopping module to adaptively terminate unnecessary iterations for efficient deployment. Experiments on the CBLPRD-330k dataset and severe occlusion stress tests demonstrate the superiority of the proposed method. Under 20–50% character occlusion, our model maintains over 95% character accuracy and consistently outperforms mainstream baselines such as LPRNet, TrOCR, and PP-OCRv5 in character accuracy, length accuracy, and full-match rate (FMR). These results indicate that wavelet-domain feature fusion provides an effective and robust solution for highly occluded license plate recognition in complex traffic environments.