Road-Assisted Cooperative Model Training and Inference for Perception in Intelligent Networked Vehicular Systems
- Publication Type:
- Thesis
- Issue Date:
- 2025
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
The perception module in an autonomous driving (AD) system strives to accurately represent the surrounding environment. However, ensuring accurate and reliable perception in dynamic traffic scenarios remains a significant challenge, such as model uncertainty, perception uncertainty and wireless communication constraints between connected autonomous vehicles (CAVs) and intelligent roadside units (IRSUs). To address these challenges, this thesis develops a series of methods for robust perception in intelligent networked vehicular systems, focusing on road-assisted cooperative model training and inference. To mitigate the model uncertainty, a road-assisted cooperative training framework is introduced, along with a road supervised data annotation algorithm. Additionally, a roadside sensor placement algorithm is developed to facilitate optimal knowledge sharing between CAVs and IRSUs. To reduce the communication latency among CAVs and IRSUs, a network topology optimization algorithm is devised to minimize latency under varying network conditions. For improving perception robustness, this thesis incorporates diverse V2V communication channel models into the cooperative inference system, providing a systematic analysis of performance degradation due to communication impairments. Building upon this analysis, a joint weighting and denoising framework is developed to correct both CAV-level and pixel-level feature distortions, thereby enhancing the resilience of shared intermediate features in cooperative perception.
Please use this identifier to cite or link to this item: