In the field of autonomous driving, the importance of the occupancy grid data structure cannot be ignored. The occupancy grid has advantages such as reducing data complexity, improving computational efficiency, and facilitating path planning. By constructing an accurate occupancy grid dataset, researchers can better understand and analyze the distribution of objects in the environment, providing strong support for tasks such as object detection and path planning. This paper proposes a new method for constructing an occupancy dataset, which first constructs dense voxels based on point cloud data, then extracts semantics through two methods, and finally filters the grid based on the visible area to obtain the ground truth of the Occupancy dataset(Named as VA-OCC dataset.). By replacing the existing dataset in the paper with the VA-OCC dataset, better IOU scores and visualization effects can be achieved.