With the potential of reshaping the future of mobility, connected and autonomous vehicles (CAVs) offer a potential opportunity to transform the world with significant social, industrial, and economic benefits. One major challenge of CAVs is the driving safety while meeting the occlusions. To resolve this problem, the researches on various types of sensor technologies and inference models have been advanced in recent years. However, the sensory data collected by an individual vehicle is insufficient to offer occlusion-aware autonomous driving. Alternatively, as a promising solution, we propose a perception task-oriented information sharing (PTOIS) network, in which each CAVs is able to achieve occlusion-free environmental awareness based on the perception data sharing via vehicle-to-everything (V2X) communications. Moreover, a game-theoretical computing resource allocation strategy is designed in the PTOIS framework to provide distributed CAVs with real-time and high-reliability perception data fusion. Extensive numerical results demonstrate that the real-time and high-reliability performance of the proposed resource allocation strategy, and the flexibility of PTOIS in the ability of adapting to diverse driving situations.