python opencv调用摄像头并追踪移动物体,浅析Python+OpenCV使用摄像头追踪人脸面部血液变化实现脉搏评估...
使用摄像头追踪人脸由于血液流动引起的面部色素的微小变化实现实时脉搏评估。
效果如下(演示视频):
由于这是通过比较面部色素的变化评估脉搏所以光线、人体移动、不同角度、不同电脑摄像头等因素均会影响评估效果,实验原理是面部色素对比,识别效果存在一定误差,各位小伙伴且当娱乐,代码如下:
import cv2
import numpy as np
import dlib
import time
from scipy import signal
# Constants
WINDOW_TITLE = 'Pulse Observer'
BUFFER_MAX_SIZE = 500 # Number of recent ROI average values to store
MAX_VALUES_TO_GRAPH = 50 # Number of recent ROI average values to show in the pulse graph
MIN_HZ = 0.83 # 50 BPM - minimum allowed heart rate
MAX_HZ = 3.33 # 200 BPM - maximum allowed heart rate
MIN_FRAMES = 100 # Minimum number of frames required before heart rate is computed. Higher values are slower, but
# more accurate.
DEBUG_MODE = False
# Creates the specified Butterworth filter and applies it.
def butterworth_filter(data, low, high, sample_rate, order=5):
nyquist_rate = sample_rate * 0.5
low /= nyquist_rate
high /= nyquist_rate
b, a = signal.butter(order, [low, high], btype='band')
return signal.lfilter(b, a, data)
# Gets the region of interest for the forehead.
def get_forehead_roi(face_points):
# Store the points in a Numpy array so we can easily get the min and max for x and y via slicing
points = np.zeros((len(face_points.parts()), 2))
for i, part in enumerate(face_points.parts()):
points[i] = (part.x, part.y)
min_x = int(points[21, 0])
min_y = int(min(points[21, 1], points[22, 1]))
max_x = int(points[22, 0])
max_y = int(max(points[21, 1], points[22, 1]))
left = min_x
right = max_x
top = min_y - (max_x - min_x)
bottom = max_y * 0.98
return int(left), int(right), int(top), int(bottom)
# Gets the region of interest for the nose.
def get_nose_roi(face_points):
points = np.zeros((len(face_points.parts()), 2))
for i, part in enumerate(face_points.parts()):
points[i] = (part.x, part.y)
# Nose and cheeks
min_x = int(points[36, 0])
min_y = int(points[28, 1])
max_x = int(points[45, 0])
max_y = int(points[33, 1])
left = min_x
right = max_x
top = min_y + (min_y * 0.02)
bottom = max_y + (max_y * 0.02)
return int(left), int(right), int(top), int(bottom)
# Gets region of interest that includes forehead, eyes, and nose.
# Note: Combination of forehead and nose performs better. This is probably because this ROI includes eyes,
# and eye blinking adds noise.
def get_full_roi(face_points):
points = np.zeros((len(face_points.parts()), 2))
for i, part in enumerate(face_points.parts()):
points[i] = (part.x, part.y)
# Only keep the points that correspond to the internal features of the face (e.g. mouth, nose, eyes, brows).
# The points outlining the jaw are discarded.
min_x = int(np.min(points[17:47, 0]))
min_y = int(np.min(points[17:47, 1]))
max_x = int(np.max(points[17:47, 0]))
max_y = int(np.max(points[17:47, 1]))
center_x = min_x + (max_x - min_x) / 2
left = min_x + int((center_x - min_x) * 0.15)
right = max_x - int((max_x - center_x) * 0.15)
top = int(min_y * 0.88)
bottom = max_y
return int(left), int(right), int(top), int(bottom)
def sliding_window_demean(signal_values, num_windows):
window_size = int(round(len(signal_values) / num_windows))
demeaned = np.zeros(signal_values.shape)
for i in range(0, len(signal_values), window_size):
if i + window_size > len(signal_values):
window_size = len(signal_values) - i
curr_slice = signal_values[i: i + window_size]
if DEBUG_MODE and curr_slice.size == 0:
print ('Empty Slice: size={0}, i={1}, window_size={2}'.format(signal_values.size, i, window_size))
print (curr_slice)
demeaned[i:i + window_size] = curr_slice - np.mean(curr_slice)
return demeaned
# Averages the green values for two arrays of pixels
def get_avg(roi1, roi2):
roi1_green = roi1[:, :, 1]
roi2_green = roi2[:, :, 1]
avg = (np.mean(roi1_green) + np.mean(roi2_green)) / 2.0
return avg
# Returns maximum absolute value from a list
def get_max_abs(lst):
return max(max(lst), -min(lst))
# Draws the heart rate graph in the GUI window.
def draw_graph(signal_values, graph_width, graph_height):
graph = np.zeros((graph_height, graph_width, 3), np.uint8)
scale_factor_x = float(graph_width) / MAX_VALUES_TO_GRAPH
# Automatically rescale vertically based on the value with largest absolute value
max_abs = get_max_abs(signal_values)
scale_factor_y = (float(graph_height) / 2.0) / max_abs
midpoint_y = graph_height / 2
for i in range(0, len(signal_values) - 1):
curr_x = int(i * scale_factor_x)
curr_y = int(midpoint_y + signal_values[i] * scale_factor_y)
next_x = int((i + 1) * scale_factor_x)
next_y = int(midpoint_y + signal_values[i + 1] * scale_factor_y)
cv2.line(graph, (curr_x, curr_y), (next_x, next_y), color=(0, 255, 0), thickness=1)
return graph
# Draws the heart rate text (BPM) in the GUI window.
def draw_bpm(bpm_str, bpm_width, bpm_height):
bpm_display = np.zeros((bpm_height, bpm_width, 3), np.uint8)
bpm_text_size, bpm_text_base = cv2.getTextSize(bpm_str, fontFace=cv2.FONT_HERSHEY_DUPLEX, fontScale=2.7,
thickness=2)
bpm_text_x = int((bpm_width - bpm_text_size[0]) / 2)
bpm_text_y = int(bpm_height / 2 + bpm_text_base)
cv2.putText(bpm_display, bpm_str, (bpm_text_x, bpm_text_y), fontFace=cv2.FONT_HERSHEY_DUPLEX,
fontScale=2.7, color=(0, 255, 0), thickness=2)
bpm_label_size, bpm_label_base = cv2.getTextSize('BPM', fontFace=cv2.FONT_HERSHEY_DUPLEX, fontScale=0.6,
thickness=1)
bpm_label_x = int((bpm_width - bpm_label_size[0]) / 2)
bpm_label_y = int(bpm_height - bpm_label_size[1] * 2)
cv2.putText(bpm_display, 'BPM', (bpm_label_x, bpm_label_y),
fontFace=cv2.FONT_HERSHEY_DUPLEX, fontScale=0.6, color=(0, 255, 0), thickness=1)
return bpm_display
# Draws the current frames per second in the GUI window.
def draw_fps(frame, fps):
cv2.rectangle(frame, (0, 0), (100, 30), color=(0, 0, 0), thickness=-1)
cv2.putText(frame, 'FPS: ' + str(round(fps, 2)), (5, 20), fontFace=cv2.FONT_HERSHEY_PLAIN,
fontScale=1, color=(0, 255, 0))
return frame
# Draw text in the graph area
def draw_graph_text(text, color, graph_width, graph_height):
graph = np.zeros((graph_height, graph_width, 3), np.uint8)
text_size, text_base = cv2.getTextSize(text, fontFace=cv2.FONT_HERSHEY_DUPLEX, fontScale=1, thickness=1)
text_x = int((graph_width - text_size[0]) / 2)
text_y = int((graph_height / 2 + text_base))
cv2.putText(graph, text, (text_x, text_y), fontFace=cv2.FONT_HERSHEY_DUPLEX, fontScale=1, color=color,
thickness=1)
return graph
# Calculate the pulse in beats per minute (BPM)
def compute_bpm(filtered_values, fps, buffer_size, last_bpm):
# Compute FFT
fft = np.abs(np.fft.rfft(filtered_values))
# Generate list of frequencies that correspond to the FFT values
freqs = fps / buffer_size * np.arange(buffer_size / 2 + 1)
# Filter out any peaks in the FFT that are not within our range of [MIN_HZ, MAX_HZ]
# because they correspond to impossible BPM values.
while True:
max_idx = fft.argmax()
bps = freqs[max_idx]
if bps < MIN_HZ or bps > MAX_HZ:
if DEBUG_MODE:
print ('BPM of {0} was discarded.'.format(bps * 60.0))
fft[max_idx] = 0
else:
bpm = bps * 60.0
break
# It's impossible for the heart rate to change more than 10% between samples,
# so use a weighted average to smooth the BPM with the last BPM.
if last_bpm > 0:
bpm = (last_bpm * 0.9) + (bpm * 0.1)
return bpm
def filter_signal_data(values, fps):
# Ensure that array doesn't have infinite or NaN values
values = np.array(values)
np.nan_to_num(values, copy=False)
# Smooth the signal by detrending and demeaning
detrended = signal.detrend(values, type='linear')
demeaned = sliding_window_demean(detrended, 15)
# Filter signal with Butterworth bandpass filter
filtered = butterworth_filter(demeaned, MIN_HZ, MAX_HZ, fps, order=5)
return filtered
# Get the average value for the regions of interest. Will also draw a green rectangle around
# the regions of interest, if requested.
def get_roi_avg(frame, view, face_points, draw_rect=True):
# Get the regions of interest.
fh_left, fh_right, fh_top, fh_bottom = get_forehead_roi(face_points)
nose_left, nose_right, nose_top, nose_bottom = get_nose_roi(face_points)
# Draw green rectangles around our regions of interest (ROI)
if draw_rect:
cv2.rectangle(view, (fh_left, fh_top), (fh_right, fh_bottom), color=(0, 255, 0), thickness=2)
cv2.rectangle(view, (nose_left, nose_top), (nose_right, nose_bottom), color=(0, 255, 0), thickness=2)
# Slice out the regions of interest (ROI) and average them
fh_roi = frame[fh_top:fh_bottom, fh_left:fh_right]
nose_roi = frame[nose_top:nose_bottom, nose_left:nose_right]
return get_avg(fh_roi, nose_roi)
# Main function.
def run_pulse_observer(detector, predictor, webcam, window):
roi_avg_values = []
graph_values = []
times = []
last_bpm = 0
graph_height = 200
graph_width = 0
bpm_display_width = 0
# cv2.getWindowProperty() returns -1 when window is closed by user.
while cv2.getWindowProperty(window, 0) == 0:
ret_val, frame = webcam.read()
# ret_val == False if unable to read from webcam
if not ret_val:
print ("ERROR: Unable to read from webcam. Was the webcam disconnected? Exiting.")
shut_down(webcam)
# Make copy of frame before we draw on it. We'll display the copy in the GUI.
# The original frame will be used to compute heart rate.
view = np.array(frame)
# Heart rate graph gets 75% of window width. BPM gets 25%.
if graph_width == 0:
graph_width = int(view.shape[1] * 0.75)
if DEBUG_MODE:
print ('Graph width = {0}'.format(graph_width))
if bpm_display_width == 0:
bpm_display_width = view.shape[1] - graph_width
# Detect face using dlib
faces = detector(frame, 0)
if len(faces) == 1:
face_points = predictor(frame, faces[0])
roi_avg = get_roi_avg(frame, view, face_points, draw_rect=True)
roi_avg_values.append(roi_avg)
times.append(time.time())
# Buffer is full, so pop the value off the top to get rid of it
if len(times) > BUFFER_MAX_SIZE:
roi_avg_values.pop(0)
times.pop(0)
curr_buffer_size = len(times)
# Don't try to compute pulse until we have at least the min. number of frames
if curr_buffer_size > MIN_FRAMES:
# Compute relevant times
time_elapsed = times[-1] - times[0]
fps = curr_buffer_size / time_elapsed # frames per second
# Clean up the signal data
filtered = filter_signal_data(roi_avg_values, fps)
graph_values.append(filtered[-1])
if len(graph_values) > MAX_VALUES_TO_GRAPH:
graph_values.pop(0)
# Draw the pulse graph
graph = draw_graph(graph_values, graph_width, graph_height)
# Compute and display the BPM
bpm = compute_bpm(filtered, fps, curr_buffer_size, last_bpm)
bpm_display = draw_bpm(str(int(round(bpm))), bpm_display_width, graph_height)
last_bpm = bpm
# Display the FPS
if DEBUG_MODE:
view = draw_fps(view, fps)
else:
# If there's not enough data to compute HR, show an empty graph with loading text and
# the BPM placeholder
pct = int(round(float(curr_buffer_size) / MIN_FRAMES * 100.0))
loading_text = 'Computing pulse: ' + str(pct) + '%'
graph = draw_graph_text(loading_text, (0, 255, 0), graph_width, graph_height)
bpm_display = draw_bpm('--', bpm_display_width, graph_height)
else:
# No faces detected, so we must clear the lists of values and timestamps. Otherwise there will be a gap
# in timestamps when a face is detected again.
del roi_avg_values[:]
del times[:]
graph = draw_graph_text('No face detected', (0, 0, 255), graph_width, graph_height)
bpm_display = draw_bpm('--', bpm_display_width, graph_height)
graph = np.hstack((graph, bpm_display))
view = np.vstack((view, graph))
cv2.imshow(window, view)
key = cv2.waitKey(1)
# Exit if user presses the escape key
if key == 27:
shut_down(webcam)
# Clean up
def shut_down(webcam):
webcam.release()
cv2.destroyAllWindows()
exit(0)
def main():
detector = dlib.get_frontal_face_detector()
# Predictor pre-trained model can be downloaded from:
# http://sourceforge.net/projects/dclib/files/dlib/v18.10/shape_predictor_68_face_landmarks.dat.bz2
try:
predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')
except RuntimeError as e:
print ('ERROR: \'shape_predictor_68_face_landmarks.dat\' was not found in current directory. ' \
'Download it from http://sourceforge.net/projects/dclib/files/dlib/v18.10/shape_predictor_68_face_landmarks.dat.bz2')
return
webcam = cv2.VideoCapture(0)
if not webcam.isOpened():
print ('ERROR: Unable to open webcam. Verify that webcam is connected and try again. Exiting.')
webcam.release()
return
cv2.namedWindow(WINDOW_TITLE)
run_pulse_observer(detector, predictor, webcam, WINDOW_TITLE)
# run_pulse_observer() returns when the user has closed the window. Time to shut down.
shut_down(webcam)
if __name__ == '__main__':
main()
总结
以上所述是小编给大家介绍的浅析Python+OpenCV使用摄像头追踪人脸面部血液变化实现脉搏评估,希望对大家有所帮助,如果大家有任何疑问请给我留言,小编会及时回复大家的。在此也非常感谢大家对脚本之家网站的支持!
如果你觉得本文对你有帮助,欢迎转载,烦请注明出处,谢谢!
python opencv调用摄像头并追踪移动物体,浅析Python+OpenCV使用摄像头追踪人脸面部血液变化实现脉搏评估...相关推荐
- Python人脸面部血液变化脉搏评估系统
目录 1.项目简介 2.系统适用行业和用途 3.系统的技术特点和功能 3.1.ROI选取及定位
- python opencv调用cuda_Win10使用VS2019从源码编译OpenCV 4.4 + CUDA 11.0 + Cudnn 8.0 + python3
本文主要介绍Win10使用VS2019从源码编译OpenCV 4.4,并使用opencv_contrib支持CUDA 11.0 + Cudnn 8.0,以及对python3的支持. 1 首先准备安装环 ...
- python怎么调用另一个py文件的变量,Python中py文件引用另一个py文件变量的方法
Python中py文件引用另一个py文件变量的方法 最近自己初学Python,在编程是遇到一个问题就是,怎样在一个py文件中使用另一个py文件中变量,问题如下: demo1代码 import requ ...
- python无法调用安装的包_如何使用python命令和安装包进行安装Django框架
在python语言的框架中,Django框架是一种Web应用框架,这个框架的版本需要跟python的版本搭配上.其中,Django框架可以使用命令安装,也可以使用下载压缩包进行安装.下面利用实例说明D ...
- linux中python如何调用matlab的数据_?如何在Python中加入多个数据帧?
全文共2235字,预计学习时长7分钟 来源:Pexels 初学Python编程的人,面临的是各种未知的挑战. 下面是一个几乎让所有更有抱负的数据科学家都感到意外的场景: 你正在处理一个从多个源收集数据 ...
- python是否基于c语言_通过实例浅析Python对比C语言的编程思想差异
我一直使用 Python,用它处理各种数据科学项目. Python 以易用闻名.有编码经验者学习数天就能上手(或有效使用它). 听起来很不错,不过,如果你既用 Python,同时也是用其他语言,比如说 ...
- python中对象和变量的关系_浅析python中的类变量和对象变量
刚学python,学到了有关于类和对象的地方.对一个概念有点模糊,后来通过实践编码找到一定规律 在python中 class test(object): id=2 name='tt' list=['t ...
- python如何调用图片-python调用图片
广告关闭 腾讯云11.11云上盛惠 ,精选热门产品助力上云,云服务器首年88元起,买的越多返的越多,最高返5000元! python本身也有识别图片转文字的框架,但是相比调用接口,识别的精度就略显不行 ...
- labelme 语义分割数据集_labelme自定义语义分割数据集 Python文件调用labelme_json_to_da...
labelme自定义语义分割数据集 Python文件调用labelme_json_to_da labelme自定义语义分割数据集 Python文件调用labelme_json_to_dataset 批 ...
最新文章
- Cognitive Inference:认知推理下的常识知识库资源、常识推理测试评估与中文实践项目索引...
- Hibernate之检索方式
- SharePoint2013自定义登录页面
- 第三届广东省强网杯网络安全大赛WEB题writeup
- bigdecimal不保留小数_金钱要使用BigDecimal数据类型(使用double的已经被公司开除了)...
- MyEclipse8 GA 下载地址 注册码 优化指南
- 百度地图标点点击变色_《和平精英》版本爆料第三弹:雪地洞穴开启!组队标点功能升级~...
- Dom4j操作XML
- 计算机系统基础:CPU相关知识笔记
- AIX5.3安装oracle10g
- 分布式数据库架构及企业实践--基于Mycat中间件pdf
- Word、Excel、PPT使用技巧与实战方法大全(2013完整版) 免积分下载
- 【机器视觉】图像预处理
- 计算机视觉的研究热点 卷积神经网络,计算机视觉领域研究热点与前沿分析
- 如何在Mac上释放内存?Mac清除RAM教程
- 尚德机构退费难的问题
- 河南科技学院计算机对口分数线,2010年河南科技学院对口招生分数线是多少?...
- Zabbix使用SMTP发送邮件报警并且制定报警内容
- Runtime.getRuntime().availableProcessors()
- [香橙派]win10使用简短的cmd命令实现ssh快速登录