摘要
本文以无人驾驶小车为例,重点介绍了如何将深度学习技术在树莓派上实现。无人驾驶小车可以分为感知和决策两个部分,树莓派对电机的控制实现小车的决策部分;通过摄像头采集的视频对小车进行训练,让其具有感知功能,并通过测试,实现其感知部分。由于树莓派的算力有限,几乎无法搭载经典的网络结构,因此需要设计了几个简单的卷积网络,并对这些网络的性能进行对比,最后选择了一个网络作为实现的网络结构。本文的创新之处在于将深度学习技术落地。由于未见过其他文献有类似的想法实现,本文首先将深度学习技术在树莓派上实现,并成功应用于无人驾驶小车。
This paper taking the autonomous driving model car as an example,focuses on how to implement deep learning on Raspberry Pi.Autonomous driving cars can be divided into two parts,that is,perception and decision.It makes a decision through controlling the motor;the perception is realized by a camera which captures videos for training and testing.Because of the limited computing power,Raspberry Pi can hardly use the classical networks.Therefore,several simple convolutional networks were designed and the performance was compared.Finally,the best performance network was chosen as the final structure.The innovation lies in the implementation of deep learning.Since any similar ideas in other literatures have not been seen,Raspberry Pi based on deep learning was first implemented and successfully applied to an autonomous driving model car.
作者
张希靓
王泓清
姜金华
秦琴
曹建清
文静
ZHANG Xiliang;WANG Hongqing;JIANG Jinhua;QIN Qin;CAO Jianqing;WEN Jing(College of Engineering,Shanghai Polytechnic University,Shanghai 201209,China;School of Computer Science,the University of Birmingham,Birmingham,B152TT,UK;School of Mechatronic Engineering and Automation,Shanghai University,Shanghai 200444,China)
出处
《中国体视学与图像分析》
2019年第3期216-224,共9页
Chinese Journal of Stereology and Image Analysis
基金
2018年上海高校青年教师资助计划(ZZEGD18037)
上海第二工业大学校内涵建设项目汽车电子联合实验中心(A11NH90704)
上海第二工业大学校基金(EGD19XQD07,EGD19XQD06)
上海第二工业大学学科基金(XXKZD1603)
科技部重点研发计划-国际合作项目(2017YFE0118700)
欧盟H2020地平线计划欧盟地平线H2020(First Project,No 734599).
关键词
树莓派
深度学习
无人驾驶小车
Raspberry Pi
deep learning
autonomous driving model car
作者简介
张希靓(1984-),男(汉),辽宁丹东,博士,讲师。研究方向:生成对抗网络,对抗攻击,轨迹跟踪,复杂系统建模;通信作者:姜金华,助理工程师。E-mail:jhjiang@sspu.edu.cn
