With advancements in artificial intelligence, Bird’s Eye View (BEV)-based map segmentation methods have found extensive applications across various domains. These approaches establish BEV feature spaces and leverage the powerful learning capabilities of neural networks to achieve effective segmentation. Sensor fusion in BEV spaces has demonstrated its practicality for tasks such as 3D detection and map segmentation. However, existing methods face challenges, including inaccuracies in camera-based BEV estimation and limited perception of distant areas due to the sparse nature of Light Detection And Ranging (LiDAR) points. In this paper, we propose an attention-based fusion/decoder mechanism that cross-enhances LiDAR features and camera information. This method improves the learning of depth estimation in the camera branch and induces accurate localization of dense camera features in the BEV space. It also facilitates effective BEV fusion between spatially synchronized features, addressing the limitations of current methods and advancing the precision of BEV-based map segmentation.