4-树莓派+英特尔神经棒2代部署人工智能应用

安装配置树莓派和英特尔神经棒二代

张子豪 2019-11-14

修改者：白宸羽

材料准备

树莓派3B+主板
Intel movidius 神经计算棒2代

顺便提一句，英特尔神经棒二代的英文名字叫做Intel® Neural Compute Stick 2

树莓派配件：购买指南见下方链接

https://www.bilibili.com/video/av47594798

https://zhuanlan.zhihu.com/p/47257546

配置树莓派系统

下载镜像、烧录镜像、配置树莓派

检查

uname -m

输出armv7l说明正常。

换源

sudo nano /etc/apt/sources.list

用#注释掉原文件内容，用以下内容取代：

deb http://mirrors.tuna.tsinghua.edu.cn/raspbian/raspbian/ buster main contrib non-free rpi

deb-src http://mirrors.tuna.tsinghua.edu.cn/raspbian/raspbian/ buster main contrib non-free rpi

保存，退出。

sudo nano /etc/apt/sources.list.d/raspi.list

用#注释掉原文件内容，用以下内容取代：

deb http://mirror.tuna.tsinghua.edu.cn/raspberrypi/ buster main ui

deb-src http://mirror.tuna.tsinghua.edu.cn/raspberrypi/ buster main ui

保存，退出。
使用sudo apt-get update命令更新软件源列表。

安装cmake

sudo apt install cmake

下载OpenVINO toolkit for Raspbian安装包

官网介绍：https://docs.openvinotoolkit.org/latest/_docs_install_guides_installing_openvino_raspbian.html

The OpenVINO™ toolkit quickly deploys applications and solutions that emulate human vision. Based on Convolutional Neural Networks (CNN), the toolkit extends computer vision (CV) workloads across Intel® hardware, maximizing performance. The OpenVINO toolkit includes the Intel® Deep Learning Deployment Toolkit (Intel® DLDT).

The OpenVINO™ toolkit for Raspbian* OS includes the Inference Engine and the MYRIAD plugins. You can use it with the Intel® Movidius™ Neural Compute Stick (Intel® NCS) or the Intel® Neural Compute Stick 2 plugged in one of USB ports.

用自己电脑点击下方链接下载2019年10月25日发布的OpenVINO最新版本，再用filezilla传到树莓派的Downloads文件夹下

https://download.01.org/opencv/2019/openvinotoolkit/R3/l_openvino_toolkit_runtime_raspbian_p_2019.3.334.tgz

cd ~/Downloads/

tar -xf l_openvino_toolkit_runtime_raspbian_p_2019.3.334.tgz

给解压之后的文件重命名为inference_engine_vpu_arm

source inference_engine_vpu_arm/bin/setupvars.sh

sh inference_engine_vpu_arm/install_dependencies/install_NCS_udev_rules.sh

sed -i "s|<INSTALLDIR>|$(pwd)/inference_engine_vpu_arm|" inference_engine_vpu_arm/bin/setupvars.sh

设置环境变量

sudo nano /home/pi/.bashrc

在最后一行增加以下内容

source /home/pi/Downloads/inference_engine_vpu_arm/bin/setupvars.sh

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-MWrMQL8b-1617724188910)(C:\Users\40743\AppData\Roaming\Typora\typora-user-images\1573722567418.png)]

重开一个新的命令行窗口，登陆后提示[setupvars.sh] OpenVINO environment initialized则说明配置成功。

为英特尔神经棒2代配置USB规则

sudo usermod -a -G users "$(whoami)"

sh inference_engine_vpu_arm/install_dependencies/install_NCS_udev_rules.sh

提示：

Updating udev rules...
Udev rules have been successfully installed.

则说明安装成功了。

把英特尔神经棒插在树莓派的USB口上

编译

cd inference_engine_vpu_arm/deployment_tools/inference_engine/samples

mkdir build && cd build

cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_FLAGS="-march=armv7-a"

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-OUGhQB6q-1617724188912)(C:\Users\40743\AppData\Roaming\Typora\typora-user-images\1573722696477.png)]

make -j2 object_detection_sample_ssd

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-DuwfGO3B-1617724188914)(C:\Users\40743\AppData\Roaming\Typora\typora-user-images\1573723423978.png)]

下载神经棒模型的模型文件和权重文件

wget --no-check-certificate https://download.01.org/openvinotoolkit/2018_R4/open_model_zoo/face-detection-adas-0001/FP16/face-detection-adas-0001.bin

wget --no-check-certificate https://download.01.org/openvinotoolkit/2018_R4/open_model_zoo/face-detection-adas-0001/FP16/face-detection-adas-0001.xml

人脸检测

<path_to_image>换成你的人脸图片的绝对路径

pi@raspberrypi:~/Downloads/inference_engine_vpu_arm/deployment_tools/inference_engine/samples/build $ ./armv7l/Release/object_detection_sample_ssd -m face-detection-adas-0001.xml -d MYRIAD -i /home/pi/Downloads/face/1.jpg

Python中调用OpenCV进行人脸检测

import cv2 as cv

# 同济子豪兄 20191114
# 参考：https://blog.csdn.net/qqqzmy/article/details/85213414
print('开始人脸检测')
# 载入模型和权重文件
net = cv.dnn.readNet('face-detection-adas-0001.xml', 'face-detection-adas-0001.bin')

# Specify target device 
net.setPreferableTarget(cv.dnn.DNN_TARGET_MYRIAD)

# 读入图片，将图片路径更换为自己的人脸图片
frame = cv.imread('./6.jpg')
# Prepare input blob and perform an inference 
blob = cv.dnn.blobFromImage(frame, size=(672, 384), ddepth=cv.CV_8U)
net.setInput(blob) 
out = net.forward()
# Draw detected faces on the frame 
for detection in out.reshape(-1, 7): 
    confidence = float(detection[2])
    
    # xmin,ymin-框的左上角点坐标
    xmin = int(detection[3] * frame.shape[1]) 
    ymin = int(detection[4] * frame.shape[0])

    # xmax,ymax-框的右下角点坐标
    xmax = int(detection[5] * frame.shape[1]) 
    ymax = int(detection[6] * frame.shape[0])
    if confidence > 0.5:
        cv.rectangle(frame, (xmin, ymin), (xmax, ymax), color=(0, 255, 0))
        
# 存储检测后画框的人脸图片
cv.imwrite('./out.png', frame)
print('人脸检测完成,已保存检测结果图像')

配置树莓派摄像头

sudo apt-get install python-opencv

sudo apt-get install fswebcam

插上树莓派摄像头

配置树莓派摄像头

sudo nano /etc/modules

在这个文件末尾添加一行

bcm2835-v4l2

重启树莓派

输入命令

vcgencmd get_camera

如果得到下面的结果，则证明摄像头连接成功

raspistill -o image.jpg

调用摄像头拍一张照片，命名为image.jpg，存储在/pi/home路径，也就是桌面左上角资源管理器一打开显示的那个路径。如果能看到摄像头上红灯亮，目录里面有照片，则进一步说明摄像头配置正确。

ls /dev

右下角的video0即为摄像头。

fswebcam 10  test.jpg 

执行后会 延时10帧 拍摄 （给个准备时间）产生 一张 名称为 test 的图片，存储在/home/pi目录中。

配置展示窗口

打开VNC Viewer

在树莓派命令行中运行这条命令

export DISPLAY=:0.0

树莓派摄像头实时人脸检测

# coding=utf-8
# face-detection-camera.py
import cv2 as cv
import numpy as np

# 同济子豪兄 20191114
# 参考：https://blog.csdn.net/c20081052/article/details/89945575

print('开始人脸摄像头实时检测')
# 载入模型文件和权重文件
net = cv.dnn.readNet('face-detection-adas-0001.xml', 'face-detection-adas-0001.bin') 
# Specify target device 
net.setPreferableTarget(cv.dnn.DNN_TARGET_MYRIAD)

# 从摄像头中读取图像帧
cap=cv.VideoCapture(0)
while(1):
    # 获取一帧图像
    ret,frame=cap.read()
    # Prepare input blob and perform an inference 
    frame=cv.resize(frame,(480,320),interpolation=cv.INTER_CUBIC)
    blob = cv.dnn.blobFromImage(frame, size=(672, 384), ddepth=cv.CV_8U) 
    net.setInput(blob) 
    out = net.forward()
    
    # 绘制人脸框
    for detection in out.reshape(-1, 7): 
        confidence = float(detection[2])
        # 获取左上角图片的坐标
        xmin = int(detection[3] * frame.shape[1]) 
        ymin = int(detection[4] * frame.shape[0])
        # 获取右下角图片的坐标
        xmax = int(detection[5] * frame.shape[1]) 
        ymax = int(detection[6] * frame.shape[0])
        if confidence > 0.5:
            cv.rectangle(frame, (xmin, ymin), (xmax, ymax), color=(0, 255, 0))
    # 展示图像
    cv.imshow("capture",frame)
    if cv.waitKey(1)&0xFF==ord('q'):
        # 每1毫秒监听一次键盘的动作，按q键结束，并保存图片
        cv.imwrite('out.png', frame)
        print("save one image done!")
        break
# 关闭摄像头及显示窗口
cap.release()
cv.destroyAllWindows()

print('人脸摄像头实时检测完成')

运行这个脚本文件

python3 face-detection-camera.py

pture",frame)
if cv.waitKey(1)&0xFF==ord(‘q’):
# 每1毫秒监听一次键盘的动作，按q键结束，并保存图片
cv.imwrite(‘out.png’, frame)
print(“save one image done!”)
break

关闭摄像头及显示窗口

cap.release()
cv.destroyAllWindows()

print(‘人脸摄像头实时检测完成’)

运行这个脚本文件

```shell
python3 face-detection-camera.py