With the rapid advancement of intelligent transportation systems and the increasing adoption of connected and autonomous vehicles, edge computing has become a crucial enabler for real-time vehicular applications. However, the highly dynamic and uncertain nature of vehicular environments presents significant challenges to conventional deterministic resource management methods. This paper proposes a novel distributed resource collaborative optimization framework based on probabilistic inference to address service quality management issues in dynamic vehicular edge computing scenarios. The framework innovatively employs Bayesian learning to replace traditional deterministic prediction models, enabling intelligent task orchestration through probabilistic representations of vehicular resource availability. By integrating a joint communication and computation resource allocation mechanism, the approach markedly reduces task processing latency while ensuring service reliability. Extensive experiments on real-world road scenario datasets demonstrate that, compared to mainstream edge computing solutions, the proposed method achieves significant improvements in latency performance (26.1% reduction), resource utilization efficiency (89.3% utilization rate), and system stability (96.8% task success rate).