Monocular 3D perception tasks, such as object detection and depth estimation, play a pivotal role in autonomous driving and traffic management systems. These tasks enable vehicles and roadside systems to perceive three-dimensional information of the environment using only a single camera input, thereby offering a cost-effective and scalable solution. However, while on-vehicle perception methods have been extensively studied, roadside BEV (Bird’s Eye View) perception algorithms remain relatively underexplored. To enhance the capabilities of roadside BEV, we propose a plugin named Multi-Scale Hybrid Attention (MSHA) and integrate it into BEVSpread. MSHA captures global context and models long-range dependencies by concatenating original and enhanced features along the channel dimension to achieve multi-scale recognition. It dynamically adjusts channel and spatial weights to reinforce key features and suppress redundant information. Additionally, channel shuffling facilitates cross-channel information exchange, enhancing the richness and diversity of features, thereby significantly improving the performance of downstream visual tasks. Extensive quantitative experiments on the DAIR-V2X and Rope3D datasets demonstrate the superior performance of our approach, making it a promising solution for real-world traffic scenarios.