As intelligent vehicles technology advances, the need for high-precision and high-real-time 3D environmental perception has become a focal point of research. However, methods based on the cartesian coordinate system often suffer from low space occupancy ratios and poor adaptability. Consequently, this paper proposes a Light Detection and Ranging (LiDAR)-based 3D perception method utilizing polar coordinate voxels aimed at enhancing the space occupancy rate and adaptability, thereby improving 3D perception performance. Initially, LIDAR point cloud data is voxelized using polar coordinates to achieve voxels with high space occupancy and adaptability. Subsequently, the Graph Convolutional Networks is employed to extract features from the point clouds within the voxels as well as from the polar coordinate voxels themselves. Finally, an efficient 2D backbone network and a CenterHeader detection head are utilized to determine object categories and 3D perception outcomes. Experimental results on the nuScenes dataset indicate that the proposed model achieves a 59.0% NDS and 57.3% mAP score, with an inference speed of 18.5Hz, setting a new State Of The Art (SOTA) benchmark compared to rival models. The model's perception outcomes remain stable and effective in challenging complex scenarios within the nuScenes dataset and real vehicle datasets, further demonstrating its excellent robustness and generalizability.