一种基于EdgeBoard的智能车系统设计与实现  

Design and implementation of intelligent vehicle system based on EdgeBoard

作　　者：曹月花 李辉 CAO Yuehua;LI Hui(Hangzhou Dianzi University Information Engineering College,Hangzhou 310000,China)

机构地区：[1]杭州电子科技大学信息工程学院,浙江杭州310000

出　　处：《现代电子技术》2022年第18期166-170,共5页Modern Electronics Technique

基　　金：教育部2020年度第一批产学合作协同育人项目(202002009030)。

摘　　要：文中提出一种基于EdgeBoard的智能车系统,主要研究内容包括智能车车模的搭建、深度学习模型的训练、智能车控制等。首先,在百度AI Studio平台上部署飞桨深度学习框架,以计算卡EdgeBoard为主处理器,板载ATmega2560内核的WBOT控制器为下位机,CMOS高分辨率摄像头为视觉模块,闭环编码电机和智能舵机为动力装置,运用超声波、磁敏等各类传感器并使用CNC铝板搭建车模结构,从而构建一套完整的智能车模型;其次,通过深度学习训练模型,实现道路数据信息采集和数据的预处理,构建深度学习框架对数据集进行训练;再应用智能车的控制算法实现训练完成的模型调用、获取摄像头拍摄的数据、EdgeBoard对拍摄到的道路信息和任务信息的处理、EdgeBoard主处理器与WBOT下位机的通信、WBOT命令的接收以及控制指令的发送等功能;最后,通过实验对该智能系统的有效性进行验证。结果表明:所设计的智能车可以在设定的赛道上实现自主寻迹、定点停车、物料搬运、任务识别等功能;相比较于传统的智能车,文中装载深度学习模型的智能车寻迹效率更快,识别率高,对车道限制少,具有较强的鲁棒性和抗干扰能力,可以应用于智能交通系统中。A intelligent vehicle system based on EdgeBoard is designed,which mainly includes the intelligent car model building,deep learning model training and intelligent car control. The deep learning framework of flying oars is deployed on Baidu AI studio platform,the computer card EdgeBoard is taken as the main processor,the WBOT controller with the ATmega2560 core on board is taken as the lower computer,the high-resolution camera CMOS is used as the vision module,the closed-loop coding motor and the smart steering gear are used as the power devices,and various sensors such as ultrasonic and magnetic sensors are used to build the car model structure by means of CNC aluminum panels,so that a complete set of intelligent car model is constructed. The collection of road data information and road data preprocessing are realized by means of the deep learning training model,and the deep learning framework are constructed to train the data set. The control algorithm of intelligent car is used to realize the tasks of model call,data acquisition from camera,road information and task information processing by EdgeBoard,communication between EdgeBoard main processor and WBOT lower computer,WBOT command receiving and control command sending. The effectiveness of the intelligent system is verified by experiments. The results show that the designed intelligent vehicle can realize the functions of autonomous tracking,fixed-point parking,material handling and task recognition on the set track. In comparison with the traditional intelligent vehicle,the intelligent vehicle with deep learning model has faster tracking efficiency,accurate recognition rate,less lane restrictions,strong robustness and anti-interference ability,and can be applied to the intelligent transportation system.

关 键 词：智能车 百度飞桨 深度学习 控制算法 EdgeBoard WBOT控制器 人工智能 自主寻迹 

分 类 号：TN915-34[电子电信—通信与信息系统]