基于深度学习算法的三维激光雷达主动成像目标检测

3D Lidar active imaging target detection based on deep learning algorithm

为了提升目标检测效果,设计基于深度学习算法的三维激光雷达主动成像目标检测方法。利用深度学习算法的生成对抗网络,剔除三维激光雷达主动成像过程中的杂波干扰,得到无杂波干扰的目标三维图像;设计生成模型与对抗模型的损失函数,利用端到端深度神经网络的点云编码层,在无杂波干扰的目标三维图像内,提取目标三维图像特征,并输入目标检测层内;通过目标检测层输出目标检测候选框,利用非极大值结合混合置信度,确定最终目标检测框,完成三维激光雷达主动成像目标检测。实验结果表明:该方法可有效剔除杂波干扰,得到无杂波干扰的目标三维图像;该方法可有效完成三维激光雷达主动成像目标检测,且在不同目标运动模糊长度像素时,该方法目标检测的交并比均高于阈值,说明该方法的目标检测精度较高。

In order to improve the effect of target detection,a 3D LiDAR active imaging target detection method based on deep learning algorithm is designed.By using the generative adversarial network of deep learning algorithm,the clutter interference in the active imaging process of 3D LiDAR is eliminated,and the three-dimensional target im-age without clutter interference is obtained.The loss function of the generative model and the countermeasure model is designed.The point cloud coding layer of the end-to-end deep neural network is used to extract the 3D image features of the target without clutter interference and input them into the target detection layer.The target detection candidate frame is output through the target detection layer,and the final target detection frame is determined by the combination of non-maximum value and mixed confidence,and the 3D LiDAR active imaging target detection is completed.The experimental results show that the proposed method can effectively eliminate clutter interference and obtain three-di-mensional target images without clutter interference.The proposed method can effectively complete the target detection in 3D LiDAR active imaging,and the intersection ratio of the proposed method is higher than the threshold value for different target motion blur length pixels,indicating that the proposed method has high target detection accuracy.