Urban Multi-object Recognition under Lidar Point Cloud Scale Filtering

During laser radar data collection, point cloud data distortion occurs due to factors such as carrier motion and environmental conditions, which interferes with target recognition, leading to target position deviation and inaccurate identification. To address this, a method for multi-object recognition in urban planning areas based on laser radar point cloud scale filtering is proposed. After correction and compensation of the laser radar point cloud data, the target area is effectively segmented into non-ground points and ground points using an elevation difference threshold. The rBRIEF algorithm is introduced to establish feature descriptors for non-ground points, which are then input into an support vectormachine (SVM) prediction function. Through feature classification and discrimination, multi-object recognition in urban planning areas is achieved. The results show that the proposed method reduces point cloud distortion, making the corrected point clouds more closely approximate real-world scenarios and improving recognition accuracy. Significant improvements are observed in error metrics such as root mean square error (RMSE), with a pixel accuracy of 93.0 % and a mean intersection over union (mIoU) of 81.0 %. Illustrative comparisons demonstrate that the method accurately recognizes multiple object categories, avoiding misidentification and omission, thus providing reliable data for urban planning.