摘要
随着自动驾驶技术和点云处理技术的不断发展,如何对点云数据中的点的相关特征进行分析就成为关键问题。针对点云特征分析问题,很多的文献中对点云特征值采取手动输入的方法。然而,该类方法无法适应实际环境中复杂多变的情况。为了解决该问题,该文提出了一种新的基于深度学习网络的三维点云检测框架。该框架以深度学习作为点云分析工具,并引入三维视觉机理,在点云分析时网络会通过已有数据集的训练自动寻找需要注意的点云区域,从而极大地减少了计算成本。在测试时,采用自动驾驶点云数据集对网络进行应用。实验结果表明,与其他三维点云物体检测方法相比,基于深度学习网络的三维点云检测方法能够在对硬件环境要求较低的情况下实现良好的运行效果。
With the continuous development of automatic driving technology and point cloud processing technology,how to analyze the relevant characteristics of points in point cloud data has become an essential issue.For the problem of point cloud feature analysis,the hand-crafted method for features of point cloud is adopted in many literatures,which cannot adapt to the complex and changeable situation in the actual environment.In order to solve this problem,we propose a new 3D point cloud detection framework based on deep learning network.The framework uses deep learning as a tool for point cloud analysis,and introduces three-dimensional visual mechanism.During point cloud analysis,the network will automatically find the point cloud fields which need attention through the training of existing datasets,which reduces the computational cost greatly.During the test,we use the point cloud dataset of automatic driving applied to the network.The experimental results show that compared with other 3D point cloud object detection methods,the proposed 3D point cloud detection method based on deep learning network can achieve good performance under the condition of low hardware requirements.
作者
高焱
熊风光
GAO Yan;XIONG Feng-guang(School of Big Data,North University of China,Taiyuan 030051,China)
出处
《计算机技术与发展》
2022年第3期76-83,共8页
Computer Technology and Development
基金
国家重点研发计划(2018YFB2101504)
山西省重点研发计划(201803D121081,201903D121147)
山西省自然科学基金(201901D111150)
山西省回国留学人员科研资助项目(2020-113)。
关键词
点云
检测
三维点云检测框架
三维视觉机理
自动驾驶点云数据集
point cloud
detection
3D point cloud detection framework
three-dimensional visual mechanism
dataset of automatic driving