基于交叉自注意力机制的LiDAR点云三维目标检测

针对基于深度学习的激光雷达(light detection and ranging, LiDAR)点云三维(3D)目标检测对小目标的检测精度较低和噪声干扰问题，提出一种基于交叉自注意力机制的3D点云目标检测方法CSA-RCNN (cross self-attention region convolutional neural network)。利用交叉自注意力(cross self-attention, CSA)同时学习点云的坐标和特征，并设计多尺度融合(multi-scale fusion, MF)模块自适应捕捉各层级多尺度特征。此外，还设计重叠采样策略对感兴趣目标区域选择性地重采样以获得更多前景点，有效降低了噪声采样。在广泛使用的KITTI数据集上进行算法性能测试，结果表明，本文方法对行人等小目标的检测精度有较大提升，平均精度均值相比PointRCNN等4种经典算法均获得提升，显著提高3D点云目标的检测性能。

LiDAR point cloud 3D object detection based on cross self-attention mechanism

Aiming at the low detection precision of small objects and noise interference in light detection and ranging (LiDAR) point cloud 3D object detection based on deep learning,a 3D point cloud object detection method CSA-RCNN(cross self-attention region cnn)based on cross self-attention mechanism was proposed.The cross self-attention was used to learn the coordinates and features of the point cloud simultaneously,and a multi-scale fusion (MF) module was designed to adaptively capture multi-scale features at each level.In addition,an overlapping sampling strategy was designed to selectively resample the target region of interest to obtain more foreground points,effectively reducing noise sampling.The algorithm performance test was carried out on the widely used KITTI dataset.The results show that the detection precision of the method in this paper for small objects such as pedestrians is greatly improved,and the average precision mean value is increased compared with four classical algorithms such as PointRCNN,which significantly improves the performance of 3D point cloud object detection.