基于深度学习的智能辅助驾驶系统设计  

Design of Intelligent Auxiliary Driving System Based on Deep Learning

作　　者：林付春 张荣芬 刘宇红 LIN Fuchun;ZHANG Rongfen;LIU Yuhong(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025

出　　处：《贵州大学学报（自然科学版）》2018年第1期73-77,共5页Journal of Guizhou University:Natural Sciences

基　　金：贵州省科技计划项目(黔科合平台人才[2016]5707)

摘　　要：本文提出一种基于机器视觉和深度学习的智能辅助驾驶系统。软件系统主要在阿里云服务器上运行,利用云端服务器实现对深度学习卷积神经网络的训练和对几种马路上常见障碍物的检测识别,并采用双目视觉,根据视觉差,计算出障碍物与车辆的距离。硬件系统采用嵌入式ARM9作为中央控制单元,搭载Linux系统,协调4G无线通信模块、摄像头采集模块和智能语音播报模块进行运作。实现前端图像的采集压缩、与云服务器的通信以及将障碍物种类和距离播报的功能。本系统使用前端硬件在马路上采集障碍物图像发送到服务器的caffe框架上,在训练好的卷积神经网络上进行测试,结果表明该系统能实时准确地将车辆周围信息以语音的方式播报给驾驶者。A scheme design of intelligent auxiliary driving system based on machine vision and deep learning was proposed.The software system is mainly running on Ali cloud server,using of cloud servers to achieve the training of deep learning convolution neural network and several common obstacles detection and identification on the road,while using binocular vision,according to the visual difference,calculate the distance between the obstacle and the vehicle.Hardware system uses embedded ARM9 as the central control unit,equipped with Linux system,coordination 4G wireless communication module,camera acquisition module and intelligent voice broadcast module,to achieve the front-end image acquisition and compression,communication with the cloud server and implement the obstacle type and distance broadcast function.The system uses front-end hardware to collect obstacle images on the road and send to the server’s caffe frame of the convolution neural network pre-trained to carry out the test.The results show that the system can be real-time and tell accurate information on the vehicle around the way to the driver.

关 键 词：辅助驾驶 机器视觉 深度学习 卷积神经网络 CORTEX-A9 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]