1. 基于MiPi相机的人脸识别
在没有看过我写的MiPi相机相关作品之前,先看这篇文章:基于旭日派的Ros系统小车的再开发——使用python脚本调用MIPI相机(学习笔记)-CSDN博客
其他相关作品:
基于旭日派的Ros系统小车的再开发——使用python脚本调用深度相机(学习笔记)-CSDN博客
1.1 Jupyter Notenook代码:
没有jupyter环境的,首先还是一样去找方法去安装,C站很多教程教你怎么安装jupyter环境,这里不多做赘述。直接先上代码:
@ 人脸检测器(默认):haarcascade_frontalface_default.xml
@ 人脸检测器(侧视):haarcascade_profileface.xml
这两个文件统一说一下,私信我或者去我主页找付费下载即可。后面的代码均是基于这两个xml文件来进行的。
###下面的代码jupyter内应该是n个模块,这里为方便起见,我直接放一块了,统一说明,中间用回车空三行的就是不同的模块,注意使用时去区分。###
# 中文开关,默认为英文 Chinese switch. The default value is English
g_ENABLE_CHINESE = False
Name_widgets = {
'Start': ("Start", "开始"),
'Close_Camera': ("Close_Camera", "关闭摄像头")
}
# 显示摄像头组件 Camera display widgets
import ipywidgets.widgets as widgets
from IPython.display import display
import cv2
import time
import numpy as np
import threading
#bgr8转jpeg格式
def bgr8_to_jpeg(value, quality=75):
return bytes(cv2.imencode('.jpg', value)[1])
from SunriseRobotLib import Mipi_Camera
img_width=320
img_height=240
g_camera = Mipi_Camera(img_width, img_height)
if g_camera.isOpened():
print("Open Camera OK")
else:
print("Fail To Open Camera")
g_stop_program = False
g_start_function = False
g_car_runtime = -1
# image_widget = widgets.Image(format='jpeg', width=320, height=240)
# 摄像头显示控件 Camera display widgets
DISPLAY_WIDTH = img_width
DISPLAY_HEIGHT = img_height
image_widget = widgets.Image(format='jpeg', width=DISPLAY_WIDTH, height=DISPLAY_HEIGHT)
# 开始 Start
Button_Start = widgets.Button(
description=Name_widgets['Start'][g_ENABLE_CHINESE],
button_style='success', # 'success', 'info', 'warning', 'danger' or ''
tooltip='Description',
icon='uncheck' )
# 关闭摄像头 Close_Camera
Button_Close_Camera = widgets.Button(
value=False,
description=Name_widgets['Close_Camera'][g_ENABLE_CHINESE],
button_style='danger', # 'success', 'info', 'warning', 'danger' or ''
tooltip='Description',
icon='uncheck' )
# 按键按下事件处理 Key press event processing
def on_button_clicked(b):
global g_stop_program
b.icon = 'check'
with output:
print("Button clicked:", b.description)
if b.description == Name_widgets['Close_Camera'][g_ENABLE_CHINESE]:
# 停止线程,释放摄像头 Stop the thread and release the camera
Button_Start.icon = 'uncheck'
g_stop_program = True
time.sleep(.1)
g_camera.release()
b.icon = 'uncheck'
# 功能按钮按下事件处理 Function button press event handling
def on_button_function_clicked(b):
global g_start_function
if b.description == Name_widgets['Start'][g_ENABLE_CHINESE]:
if b.icon == 'uncheck':
b.icon = 'check'
g_start_function = True
else:
b.icon = 'uncheck'
g_start_function = False
# time.sleep(.1)
with output:
print("Button clicked:", b.description, g_start_function)
# 关联按键事件回调 Button event callbacks
Button_Close_Camera.on_click(on_button_clicked)
Button_Start.on_click(on_button_function_clicked)
face_haar = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
# face_haar = cv2.CascadeClassifier("haarcascade_profileface.xml")
def face_detect(image):
gray_img = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
faces = face_haar.detectMultiScale(gray_img, 1.1, 3)
for face_x,face_y,face_w,face_h in faces:
return "1", (face_x,face_y,face_w,face_h)
return None, (0,0,0,0)
face_haar = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
# face_haar = cv2.CascadeClassifier("haarcascade_profileface.xml")
def face_detect(image):
gray_img = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
faces = face_haar.detectMultiScale(gray_img, 1.1, 3)
for face_x,face_y,face_w,face_h in faces:
return "1", (face_x,face_y,face_w,face_h)
return None, (0,0,0,0)
# 启动摄像头显示任务 Start the camera display task
thread1 = threading.Thread(target=task_processing)
thread1.setDaemon(True)
thread1.start()
# 创建一个横向的盒子容器,以便将图像小部件相邻放置
# create a horizontal box container to place the image widget next to eachother
image_container = widgets.HBox([image_widget])
button_group = widgets.HBox([Button_Start, Button_Close_Camera])
output = widgets.Output()
box_display = widgets.VBox([image_container, button_group, output])
display(box_display)
这里jupyter notenook代码相比于后面的几个代码来说有一个很好的一点就是,jupyter这里做到了多线程去进行识别,点击弹出的start窗口实时显示的图像才会进行人脸识别,关闭状态下,仅仅是做一个图像采集的作用,这样做的好处就是避免了资源损耗,避免机器长时间进行识别而导致机器寿命下降等等一系列原因。
1.2 Python代码:
import cv2
import time
import ipywidgets.widgets as widgets
from IPython.display import display
from SunriseRobotLib import Mipi_Camera
# 加载人脸识别模型
face_haar = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
width = 320
height = 240
g_camera = Mipi_Camera(width, height, debug=True)
if g_camera.isOpened():
print("Open Camera OK!")
else:
print("Fail To Open Camera!")
def bgr8_to_jpeg(value, quality=75):
return bytes(cv2.imencode('.jpg', value)[1])
image_widget = widgets.Image(format='jpeg', width=320, height=240)
#display(image_widget)
try:
m_fps = 0
t_start = time.time()
saved_count = 1 # 统计已经保存的文件数量
while g_camera.isOpened():
ret, frame = g_camera.read()
if not ret:
print("Camera Read Fail!")
break
# 将帧转为灰度图像
gray_img = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# 检测人脸位置
faces = face_haar.detectMultiScale(gray_img, 1.1, 3)
# 遍历检测到的人脸
for (face_x, face_y, face_w, face_h) in faces:
# 绘制矩形框显示人脸位置
cv2.rectangle(frame, (face_x, face_y), (face_x+face_w, face_y+face_h),
(0, 255, 255), 1)
m_fps = m_fps + 1
fps = m_fps / (time.time() - t_start)
if (time.time() - t_start) >= 2:
m_fps = fps
t_start = time.time() - 1
text = "FPS: " + str(int(fps))
cv2.putText(frame, text, (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 255), 1)
image_widget.value = bgr8_to_jpeg(frame)
# 检测是否按下 s 键
k = cv2.waitKey(1)
if k == ord('s'):
# 获取当前的日期时间戳
timestamp = time.strftime("%Y%m%d_%H%M%S", time.localtime())
# 生成指定格式的文件名
filename = f"{saved_count:04d}_{timestamp}.jpg"
# 保存图像至本地文件夹
cv2.imwrite(filename, frame)
# 提示用户已经成功保存当前图像,并更新已保存的文件数量
print(f"Saved {filename} successfully!")
saved_count += 1
cv2.imshow('Frame', frame)
# 等待键盘输入并判断是否关闭窗口
if cv2.waitKey(1) & 0xFF == ord(' '):
break
except Exception as e:
print("Exception: ", e)
这里的python代码写的也很简单,并没有做复杂的多线程化处理,比较python脚本嘛,一切从简即可,所以相较于上面的jupyter代码不足之处就是没有做优化处理,所以执行之后可能会出现掉帧或者卡顿的情况,(可能出现的原因是基于机器的配置而定的);对比jupyter代码好处就是使用python脚本会很方便,易于修改和直接。不过个人觉得python和jupyter各有特色,并没有站队的意思。
2. 基于系统相机的人脸识别
1. 系统相机:
import cv2
import time
import ipywidgets.widgets as widgets
from IPython.display import display
# 设置窗口大小
window_width = 640
window_height = 480
# 加载人脸识别模型
face_haar = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')
# 使用电脑摄像头
g_camera = cv2.VideoCapture(0)
# 设置摄像头的宽度和高度
width = 320
height = 240
g_camera.set(3, width)
g_camera.set(4, height)
# 创建窗口并设置大小
cv2.namedWindow('Frame', cv2.WINDOW_NORMAL)
cv2.resizeWindow('Frame', window_width, window_height)
def bgr8_to_jpeg(value, quality=75):
return bytes(cv2.imencode('.jpg', value)[1])
image_widget = widgets.Image(format='jpeg', width=320, height=240)
display(image_widget)
try:
m_fps = 0
t_start = time.time()
while True:
ret, frame = g_camera.read()
if not ret:
print("Camera Read Fail!")
break
# 将帧转为灰度图像
gray_img = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# 检测人脸位置
faces = face_haar.detectMultiScale(gray_img, 1.1, 3)
# 遍历检测到的人脸
for (face_x, face_y, face_w, face_h) in faces:
# 绘制矩形框显示人脸位置
cv2.rectangle(frame, (face_x, face_y), (face_x+face_w, face_y+face_h), (0, 255, 255), 1)
m_fps = m_fps + 1
fps = m_fps / (time.time() - t_start)
if (time.time() - t_start) >= 2:
m_fps = fps
t_start = time.time() - 1
text = "FPS: " + str(int(fps))
cv2.putText(frame, text, (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 255), 1)
image_widget.value = bgr8_to_jpeg(frame)
cv2.imshow('Frame', frame)
# 等待键盘输入并判断是否关闭窗口
if cv2.waitKey(1) & 0xFF == ord(' '):
break
except Exception as e:
print("Exception: ", e)
# 释放摄像头对象
g_camera.release()
# 关闭所有的窗口
cv2.destroyAllWindows()
这里g_camera = cv2.VideoCapture(0)表示调用默认系统的摄像头,其他的代码属性写法和MiPi相机的很像,都可以实时采集识别人脸,也可以按下键盘的‘s’键进行保存,按下键盘的空格键进行退出程序。