Lidar sensors are fundamental and crucial devices for self-driving vehicles, due to their capabilities of environmental information acquisition to develop feasible driving areas and motion decisions. Even though lidars provide 3D point clouds for accurate positions, the complex data processing takes arduous efforts for online calculation, including segmentation, filtering, clustering and verification. Furthermore, sensor fusion with multiple lidar sensors (such as Velodyne HDL-32c and Ibeo-Lux 4L) aggravates the computational load, which prevents exact surrounding detection from being applied for planning and control in real time. To address the problem, this paper proposes a systematic filtering algorithm based on occupancy rates of two categories obstacle vehicle detection method for autonomous ground vehicles, considering lidar calibration, ground segmentation, ego-vehicle removal of 3D lidar point clouds. Another adaptive searching (AS) algorithm on density-based spatial clustering of applications with noise (DBSCAN) is proposed to coordinate the characteristics of scanned points with respect to distances to the lidar set-up point. An indoor perception test with a fully-instrumented autonomous hybrid electric vehicle within complicated surroundings was conducted, with 3D point cloud fused data provided by one Velodyne HDL-32c lidar sensor and three Ibeo-Lux 4L scanners, which has verified the effectiveness of the proposed approach for obstacle detection.