Abstract: Aiming at the problems of high memory requirements, calibration alignment and noise interference in the construction of global space occupancy grid map based on heterogeneous distance sensors, a direct gridding method is proposed for constructing the fused occupancy grid map. Firstly, a ring buffer is used to implement the sliding update of the grid map, reducing the number of grids to lower memory consumption. Meanwhile, the perception and control error model is adopted to derive the map update error formula during high-speed flight, the number of sliding grids is established, and the relationship among grid resolution, range scaling factor and flight speed is constructed. Secondly, the scale transformation factors of imaging radar are inversely solved based on the grid resolution and radar detection model, allowing for direct mapping and alignment of LiDAR and imaging radar point clouds to a unified grid space. Finally, a strategy for updating the "probability topographic map" is proposed, and anti-interference ability and robustness of the grid map in complex environments are enhanced by weighting the occupancy probability and introducing update coefficient. Results show that the sliding update strategy significantly reduces both the memory usage and construction time of the grid map; meanwhile, the range of construction time decreases by 16 ms, reducing computational fluctuations; in the smoke interference scene in the NTU4DRadLM open-source dataset based on LiDAR and imaging radar, the direct gridding method reduces the latency and enhances the resistance to interference while constructing a multi-source heterogeneous fusion-based grid map. In the actual large-scale complex terrain task, the memory occupancy remains stable within the range of 1.5~2 GB, thereby meeting the resource constraints of the embedded computing unit in UAV (unmanned aerial vehicle).