基于深度学习的高分辨率三维激光雷达成像目标检测方法  

High-resolution 3D lidar imaging target detection method based on deep learning

作　　者：苗长芬[1] MIAO Changfen(School of computer and information engineering,Xinxiang University,Xinxiang Henan 453003,China)

机构地区：[1]新乡学院计算机与信息工程学院,河南新乡453003

出　　处：《激光杂志》2024年第6期233-237,共5页Laser Journal

基　　金：河南省自然科学基金项目(No.202300320436)。

摘　　要：高分辨率三维激光雷达成像目标检测中受到双目视觉耦合干扰导致检测精度不高,提出基于深度学习的高分辨率三维激光雷达成像目标检测方法。采用视觉传感器与惯性传感器实现对高分辨率三维激光雷达图像采集,通过激光点云特征提取和动态场景图像融合的方法实现对高分辨率的三维激光雷达合成,采用补偿滤波方法实现对三维激光雷达成像目标检测过程中的点云畸变矫正处理,通过局部滑动窗口的图特征匹配和深度学习模型,建立激光雷达成像目标检测的深度特征匹配模型,根据特征匹配结果和激光雷达帧图匹配结果,实现对高分辨率三维激光雷达成像目标检测。仿真测试结果表明,采用该方法进行目标检测的激光雷达成像在噪声干扰下,峰值信噪比最低为41 dB,检测精度达到0.96。The detection accuracy is not high due to the interference of binocular vision coupling in high-resolution 3D lidar imaging target detection.A method of high-resolution 3D lidar imaging target detection based on deep learning is proposed.The vision sensor and inertial sensor are used to collect high-resolution 3D lidar images,and the method of laser point cloud feature extraction and dynamic scene image fusion is used to synthesize high-resolution 3D lidar.The compensation filtering method is used to correct the point cloud distortion in the process of 3D lidar imaging target detection.The depth feature matching model of lidar imaging target detection is established through the map feature matching and depth learning model of local sliding window,and the high-resolution 3D lidar imaging target detection is realized according to the feature matching results and the lidar frame matching results.The simulation test results show that the laser radar imaging using this method for target detection has a minimum peak signal-to-noise ratio of 41 dB under noise interference,and the detection accuracy reaches 0.96.

关 键 词：深度学习 高分辨率 三维激光雷达成像 目标检测 

分 类 号：TN232[电子电信—物理电子学]