Home People Research 3DGS Papers News Contact

Research Projects

⚫ Cross-Driver Domain Generalization for Improved Drowsiness Recognition Based on EEG Signals

This paper proposes a domain-generalized EEG drowsiness detection framework using hybrid feature mapping and self-attention mechanisms. The method achieves 81.34% accuracy, surpassing state-of-art by 1.97%, with frontal lobe delta/theta/alpha bands showing optimal performance.

Method innovation Performance metrics Key findings

Paper

⚫ 3DGS Real-time Light and Shadow Rendering Effects

This paper proposes an alternating interaction fusion method that mutually enhances LiDAR and image BEV features through local attention mechanisms, avoiding unilateral reliance or parallel redundancy. The approach outperforms existing fusion methods on nuScenes dataset while maintaining computational efficiency.

3D object detection Camera-LiDAR fusion Autonomous Vehicles

Paper

⚫ High-Fidelity Static Scene Reconstruction via Dense Point Cloud and Gaussian Splatting Rendering

Novel pipeline combines MVS point clouds and Gaussian Splatting for static scene reconstruction, integrating dynamic removal, depth-attention refinement, and uncertainty modeling to achieve photorealistic rendering.

MVS point clouds Gaussian Splatting Photorealistic rendering

GitHub

Paper

⚫ Editable Static Scene Construction with Realistic Environmental Effects and Component Libraries

Develops an editable static scene system with realistic weather effects (rain/snow/fog), component editing tools, and multi-source databases (high-precision/AI-generated/modular assets) for flexible scene construction.

Scene editing plugin Environmental effects simulation Component database

GitHub

Paper

⚫ Dynamic Scene Reconstruction and Localization System for Driving Environments

This work develops a dynamic scene reconstruction pipeline for driving videos, featuring motion segmentation, object-level reconstruction, and real-time localization. The system handles complex dynamic elements while maintaining scene context.

Dynamic reconstruction Motion segmentation Scene localization

GitHub

Paper

⚫ Dynamic Scene Reconstruction and Novel View Synthesis for Driving Scenes

This work enables dynamic scene reconstruction and novel view synthesis for driving scenarios, featuring rigid (vehicles) and non-rigid (pedestrians) object segmentation with dedicated rendering pipelines.

Dynamic reconstruction Novel view synthesis Object-aware rendering

GitHub

Paper