点击下方卡片，关注“机器视觉与AI深度学习”
视觉/图像重磅干货，第一时间送达!
Raspberry Pi 4（建议 2GB 或更高） Coral USB 加速器 适用于 Raspberry Pi 的兼容 USB-C 电源 microSD 卡（16GB 或更高） Raspberry Pi 操作系统（Raspbian 版本：11（bullseye）— 64） Raspberry Pi 相机模块 v3
在您的电脑上下载并安装 Raspberry Pi Imager。 将 microSD 插入读卡器并打开 Raspberry Pi Imager 我选择 Raspberry PI 4 作为设备，选择 Raspberry Pi OS (Bullseye) 作为基础镜像。
选择存储并转到下一步。
单击编辑设置。输入您想要为您的操作系统设置的用户名和密码。
另外在服务中启用 ssh 并进行密码验证
sudo apt-get updatesudo apt-get -y upgradesudo apt-get -y install libjpeg-dev libtiff5-dev libjasper-dev libpng12-devsudo apt-get -y install libavcodec-dev libavformat-dev libswscale-dev libv4l-devsudo apt-get -y install libxvidcore-dev libx264-devsudo apt-get -y install qt4-dev-toolssudo apt-get -y install libatlas-base-dev
pip install opencv-python然后我们安装 Tensorflow Lite
pip install tflite-runtime我们现在将安装 Coral 加速器。请记住，在安装加速器之前，请将其从 Raspberry Pi 上拔下。
echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.listcurl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -sudo apt-get updateInstall the libedgetpu library by issuing:
通过发出以下命令安装 libedgetpu 库：
sudo apt-get install libedgetpu1-stdfrom picamera2 import Picamera2camera = Picamera2()camera.configure(picam2.create_preview_configuration(main={"format": 'XRGB8888', "size": (640, 480)}))camera.start()image = camera.capture_array()cv2.imwrite('test.jpg', image)
import cv2import PoseNet.engine.utils as utilsfrom PoseNet.engine.pose_engine import PoseEnginefrom datetime import datetime# Model Path_MODEL_PATH = "PoseNet/model/posenet_resnet_50_416_288_16_quant_edgetpu_decoder.tflite"# Frame shape_FRAME_WEIGHT, _FRAME_HEIGHT = 1024, 768# Threshold of the accuracy_THERESHOLD = 0.50def detect_pose(callback_function, quit_on_key=True):# Initating Interpreterengine = PoseEngine(_MODEL_PATH)# Initiating camera instancecamera = utils.init_camera(_FRAME_WEIGHT, _FRAME_HEIGHT)# Initialize frame rate calculationframe_rate_calc = 1freq = cv2.getTickFrequency()# Define the codec and create VideoWriter objectfourcc = cv2.VideoWriter_fourcc('F','M','P','4')video_name= f"./PoseNet/captured_video/{datetime.today().strftime('%Y%m%d%H%M%S')}.avi"# FPS for recording video. Setting 6 as the real fps is 6 to 7fps = 6.0# Video Recorder instanceout = cv2.VideoWriter(video_name,fourcc, fps, (_FRAME_WEIGHT, _FRAME_HEIGHT))while True:# Grab frame from video streamimage = camera.capture_array()# Start timer (for calculating frame rate)t1 = cv2.getTickCount()# Getting the input details the model expect_, src_height, src_width, _ = engine.get_input_tensor_shape()# The main Magic is happening hereposes, _ = engine.DetectPosesInImage(image)# Draw the lines in the keypointsoutput_image = utils.draw_keypoints_from_keypoints(poses, image, _THERESHOLD, src_width, src_height)# Converting the image from 640x480x4 to 640x480x3output_image=cv2.cvtColor(output_image, cv2.COLOR_BGRA2BGR)out.write(output_image)# flipping the image for displayoutput_image = cv2.flip(output_image, 1)# Draw framerate in corner of frame# Calculate frameratet2 = cv2.getTickCount()time1 = (t2-t1)/freqframe_rate_calc= 1/time1# Adding the FPS Text to the outputcv2.putText(output_image,'FPS: {0:.2f}'.format(frame_rate_calc),(30,50),cv2.FONT_HERSHEY_SIMPLEX,1,(255,255,0),2,cv2.LINE_AA)# Call back function calling here with the imagecallback_function(output_image)# Key to quite displayif cv2.waitKey(1) == ord('q') and quit_on_key:break# Clean upout.release()cv2.destroyAllWindows()
