基于openCV和paddlehub分别实现视频流的人脸识别项目以及简易跟踪功能
一、文章说明
本次文章主要目的记录一下三月份学习内容并且保存记载自己的学习成果。
二、项目内容
1、基于openCV框架实现视频流的简单人脸识别
import cv2
import numpy as np
if __name__ == '__main__':
video = cv2.VideoCapture('D:\openCVProject\student1.mp4') # 视频 频率30帧/s
face_detector = cv2.CascadeClassifier('D:/openCVProject/data/haarcascades/haarcascade_frontalface_default.xml')
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out = cv2.VideoWriter('output5min2.mp4', 0x7634706d, 30.0, (1920, 1080))#视频输出地址
print("开始读取视频")
while True:
retval, image = video.read()
if retval == False:
print("视频播放完成")
break
# 把图像转换为灰度图像，减少图片信息
# gray = cv2.cvtColor(image,code=cv2.COLOR_BGR2GRAY)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
print("视频")
faces = face_detector.detectMultiScale(gray) # 检测灰度图片
print(faces)
for x, y, w, h in faces:
# x，y是坐标，w为宽度，h为高度，color为矩形的颜色，thickness为线宽
cv2.rectangle(image, pt1=(x, y), pt2=(x + w, y + h), color=[0, 0, 255], thickness=2)
# 等待用户输入
key = cv2.waitKey(1)
out.write(image)
cv2.imshow('video', image)
if key == ord('q'):
print("退出视频")
break
# out.release()
video.release() # 释放内存
cv2.destroyAllWindows()
演示视频：https://www.bilibili.com/video/BV17v4y1J78a/?vd_source=2fa09a0655d11f6cfae7287ba7e9df31
2、基于PaddleHub框架实现视频流的简单人脸识别以及简单跟踪功能
import paddlehub as hub
import cv2
import math
import random
import string
from PIL import Image, ImageDraw, ImageFont
import copy
import numpy as np
mask_detector = hub.Module(name="pyramidbox_lite_server_mask")
temp_mark = 0
length = 0
temp_sum = 0
glob_score = 0
self_score = 0
# # 定义字符串长度和数量
# str_len = 3
# str_num = 55
#
# # 生成随机字符串列表
# str_list = [''.join(random.sample(string.ascii_letters + string.digits, str_len)) for i in range(str_num)]
list_name = ["林劲松", '韦思琪', '官远航', '曾宏睿', '肖梓宏', '陈棋濠', '陈沛琳', '袁子杰', '李佳泽', '吴雨墨',
'林嘉宇', '李艺嘉', '赖弘煊', '左晟宇', '刘飞', '许孟辰', '吴瑞晨', '刘雅涵', '邹玉婷', '廖子涵',
'章子怡', '彭予棠', '芦芸', '陈彦泽', '张文成', '谭玉婷', '康悦函', '李懿轩', '龚垲锋', '李佳鑫',
'彭志煊', '施美奂', '薛雅雯', '胡俊宇', '夏桂清', '叶璟琰', '郑晨曦', '郭凯泽', '施美轮', '邓浩宇',
'郑阳洋', '吴少尧', '黄靖然', '苏奕凡', '赵钰彤', '杜林东', '韦思琪', '王珅瑜', '吴佳豪', '许睿浩'
'许梓晨',
'詹泽云', '苏奕凡', '官远航', '吴瑞晨', '官远航', '李诺依', '官远航', '郭铠泽', '芦芸',
'许家诚', '薛雅雯', '桑统治', '范梓雯', '刘雅涵', '刘锦龙', '官远航', '刘飞', '官远航', '官远航',
'官远航',
'官远航', '袁子杰', '陈棋濠', '袁子杰', '官远航', '廖凌姗', '官远航', '官远航', '官远航', '官远航',
'林嘉宇''蒋 艳', '陈世玲', '官远航', '颜资琪', '官远航', '施美奂', '许孟辰', '黄增鑫', '卢 芸', '李艺嘉',
'官远航', '官远航', '官远航', '官远航', '官远航', '兰雨嘉', '李艺嘉', '陈子毅', '官远航', '官远航',
'官远航', '陈怀刚', '官远航', '吴瑞晨', '沈佩圻', '李诺依', '王若熙', '郭铠泽', '芦芸', '许家诚',
'薛雅雯', '桑统治', '官远航', '刘雅涵', '刘锦龙', '官远航', '康悦涵', '范俊文', '官远航', '官远航',
'官远航', '廖子涵', '陈棋濠', '袁子杰', '邹 蒙', '廖凌姗', '黄玄烨', '蔡金余', '王静娴', '官远航',
'林嘉宇''蒋 艳', '陈世玲', '章子怡', '颜资琪', '官远航', '施美奂', '吴宛儒', '黄增鑫', '卢 芸', '李艺嘉',
'官远航', '官远航', '官远航', '官远航', '官远航', '兰雨嘉', '官远航', '陈子毅', '官远航', '林嘉宇',
'詹泽云', '陈怀刚', '林亿森', '吴瑞晨', '沈佩圻', '李诺依', '王若熙', '郭铠泽', '赖韦铭', '官远航',
'薛雅雯', '桑统治', '官远航', '刘雅涵', '刘锦龙', '陈棋濠', '陈棋濠', '范俊文', '陈棋濠', '官远航',
'官远航', '官远航', '陈棋濠', '蔡麒乐', '邹 蒙', '廖凌姗', '官远航', '蔡金余', '王静娴', '林嘉宇',
'兰昊益''蒋 艳', '陈世玲', '章子怡', '颜资琪', '官远航', '苏鸿圣', '吴宛儒', '黄增鑫', '陈彦泽', '许孟辰',
'康致轩', '康悦涵', '官远航', '官远航', '官远航', '兰雨嘉', '陈柳丹', '陈子毅', '黄宗灿', '林嘉宇',
'詹泽云', '陈怀刚', '林亿森', '吴瑞晨', '沈佩圻', '李诺依', '王若熙', '郭铠泽', '官远航', '许家诚',
'薛雅雯', '桑统治', '辛添淇', '刘雅涵', '刘锦龙', '官远航', '潘栩若', '范俊文', '林嘉宇', '官远航',
'袁子杰', '陈柏霖', '周 婷', '蔡麒乐', '邹 蒙', '廖凌姗', '黄玄烨', '蔡金余', '王静娴', '王媛媛',
'兰昊益' '蒋 艳', '陈世玲', '章子怡', '颜资琪', '林泽楷', '苏鸿圣', '吴宛儒', '黄增鑫', '卢 芸', '楼雨馨',
'康致轩', '康悦涵', '官远航', '连欣怡', '官远航', '兰雨嘉', '陈柳丹', '陈子毅', '黄宗灿', '官远航',
'詹泽云', '陈怀刚', '林亿森', '吴瑞晨', '沈佩圻', '李诺依', '王若熙', '郭铠泽', '许孟辰', '许家诚',
'薛雅雯', '桑统治', '辛添淇', '刘雅涵', '刘锦龙', '黄馨怡', '潘栩若', '范俊文', '丁诗诗', '荣伟宏',
'王亿宸', '陈柏霖', '袁子杰', '蔡麒乐', '邹 蒙', '廖凌姗', '黄玄烨', '蔡金余', '王静娴', '王媛媛',
'兰昊益' '蒋 艳', '陈世玲', '章子怡', '颜资琪', '林泽楷', '苏鸿圣', '吴宛儒', '黄增鑫', '卢 芸', '楼雨馨',
'康致轩', '王芷筠', '王可晗', '连欣怡', '官远航', '兰雨嘉', '陈柳丹', '陈子毅', '黄宗灿', '李仁杰',
'詹泽云', '陈怀刚', '林亿森', '吴瑞晨', '沈佩圻', '李诺依', '王若熙', '郭铠泽', '赖韦铭', '许家诚',
'薛雅雯', '桑统治', '辛添淇', '刘雅涵', '刘锦龙', '黄馨怡', '潘栩若', '范俊文', '丁诗诗', '荣伟宏',
'王亿宸', '陈柏霖', '周 婷', '蔡麒乐', '邹 蒙', '廖凌姗', '黄玄烨', '蔡金余', '王静娴', '王媛媛',
'兰昊益' '蒋 艳', '陈世玲', '章子怡', '颜资琪', '林泽楷', '苏鸿圣', '吴宛儒', '黄增鑫', '卢 芸', '楼雨馨',
'康致轩', '王芷筠', '王可晗', '连欣怡', '官远航', '兰雨嘉', '陈柳丹', '陈子毅', '黄宗灿', '李仁杰',
'詹泽云', '陈怀刚', '林亿森', '吴瑞晨', '沈佩圻', '李诺依', '王若熙', '郭铠泽', '赖韦铭', '许家诚',
'薛雅雯', '桑统治', '辛添淇', '刘雅涵', '刘锦龙', '黄馨怡', '潘栩若', '范俊文', '丁诗诗', '荣伟宏',
'王亿宸', '陈柏霖', '周 婷', '蔡麒乐', '邹 蒙', '廖凌姗', '黄玄烨', '蔡金余', '王静娴', '王媛媛',
'兰昊益' '蒋 艳', '陈世玲', '章子怡', '颜资琪', '林泽楷', '苏鸿圣', '吴宛儒', '黄增鑫', '卢 芸', '楼雨馨',
'康致轩', '王芷筠', '王可晗', '连欣怡', '官远航', '兰雨嘉', '陈柳丹', '陈子毅', '黄宗灿', '李仁杰',
'金 鑫']
name_ndarray = np.array(list_name)
# str_list = [chr(i) for i in range(97, 123)]
class Point1:
def __init__(self):
self.x = 0
self.y = 0
self.z = 0
self.score = 0
lis1 = [] # 结构体数组
class Point2:
def __init__(self2):
self2.xx = 0
self2.yy = 0
self2.zz = 0
lis2 = [] # 结构体数组
# 封装函数
def cv2AddChineseText(img, text, position, textColor=(0, 255, 0), textSize=3):
if (isinstance(img, np.ndarray)): # 判断是否OpenCV图片类型
print("openCV success")
img = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
# 创建一个可以在给定图像上绘图的对象
draw = ImageDraw.Draw(img)
# 字体的格式
fontStyle = ImageFont.truetype(
"simsun.ttc", textSize, encoding="utf-8")
# 绘制文本
draw.text(position, text, textColor, font=fontStyle)
# 转换回OpenCV格式
return cv2.cvtColor(np.asarray(img), cv2.COLOR_RGB2BGR)
def mask_detecion(img):
global temp_sum
temp_sum = temp_sum + 1
input_dict = {"data": [img]}
result = mask_detector.face_detection(data=input_dict)
count = len(result[0]['data'])
if count < 1:
# print('There is no face detected!')
pass
else:
# labels = [] # 存储每个人脸标签
global glob_score
global self_score
global temp_mark
global temp_count
global length
global Point1
global Point2
temp_conf = 0
if temp_mark == 0:
faces_coordinates = []
for i in range(0, count):
lis1.append(Point1()) # 添加一个结构体
# print(result[0]['data'][i])
# label = result[0]['data'][i].get('label')
score = float(result[0]['data'][i].get('confidence'))
print("置信率")
print(score)
if score > 0.8:
temp_conf = temp_conf + 1
# if temp_sum % 60 == 0:
x1 = int(result[0]['data'][i].get('left'))
y1 = int(result[0]['data'][i].get('top'))
x2 = int(result[0]['data'][i].get('right'))
y2 = int(result[0]['data'][i].get('bottom'))
# faces_coordinates.append([x1, y1, x2, y2])
print("标记点的坐标\n")
print(x1)
print(y1)
lis1[i].x = x1
lis1[i].y = y1
lis1[i].z = temp_count
lis1[i].score = score
label = label = name_ndarray[lis1[i].z]
print(label)
self_score = round(lis1[i].score * 100, 2)
selfscore = str(self_score)
print("每个人脸的置信率")
print(selfscore)
cv2.rectangle(img, (x1, y1), (x2, y2), (255, 200, 0), 2)
cv2.putText(img, selfscore, (x1 + 45, y1), 0, 0.7, (0, 255, 200), 1)
# img1 = cv2AddChineseText(img, label, (x1, y1), (0, 255, 0), 3)
img = cv2AddChineseText(img, label, (x1, y1 - 15), (0, 255, 200), 10)
temp_count = temp_count + 1
temp_mark = temp_mark + 1
# if temp_sum % 60 == 0:
k = temp_conf / count
glob_score = round(k * 100, 2)
globscore = str(glob_score)
cv2.putText(img, globscore, (30, 30), 0, 0.8, (0, 255, 200), 3)
img = cv2AddChineseText(img, '课堂人数：51', (10, 60), (0, 255, 200), 25)
img = cv2AddChineseText(img, '课堂气氛：', (10, 30), (0, 255, 200), 25)
length = temp_count
print("每一帧有几张人脸图像\n")
print(count)
print(temp_count)
else:
print("每一帧有几张人脸图像\n")
print(count)
list_len = length
for i in range(0, count):
# if i > length-1:
# lis1.append(Point1())
# label = result[0]['data'][i].get('label')
score = float(result[0]['data'][i].get('confidence'))
print("置信率")
print(score)
if score > 0.8:
temp_conf = temp_conf + 1
x1 = int(result[0]['data'][i].get('left'))
y1 = int(result[0]['data'][i].get('top'))
x2 = int(result[0]['data'][i].get('right'))
y2 = int(result[0]['data'][i].get('bottom'))
# faces_coordinates.append([x1, y1, x2, y2])
print("标记点的坐标")
print(x1)
print(y1)
min_dis = float('inf')
num = 0
for j in range(0, list_len - 1):
distance = math.sqrt((lis1[j].x - x1) ** 2 + (lis1[j].y - y1) ** 2)
if distance < min_dis:
min_dis = distance
if distance < 20:
# lis1[i].z = lis1[j].z
lis1[j].x = x1
lis1[j].y = y1
num = 1
b = lis1[j].z
if temp_sum % 60 == 0:
lis1[j].score = score
label = name_ndarray[b]
self_score = round(lis1[j].score * 100, 2)
print(label)
break
print("本次检测坐标与上一帧坐标之间的距离最小为：")
print(distance)
print("\n")
if num == 0:
lis1.append(Point1()) # 添加一个结构体
temp_count = temp_count + 1
lis1[length - 1].x = x1
lis1[length - 1].y = y1
lis1[length - 1].z = temp_count
lis1[length - 1].score = score
# label = str(lis1[length - 1].z)
label = name_ndarray[lis1[length - 1].z]
self_score = round(lis1[length - 1].score * 100, 2)
print("label failed")
print(label)
# label = str(lis1[i].z)
print("输出中文")
# percentscore = '{:.2%}'.format(score)
# sumscore = str(percentscore)
# cv2.rectangle(img, (x1, y1), (x2, y2), (255, 200, 0), 2)
selfscore = str(self_score)
# self_score = round()
cv2.putText(img, selfscore, (x1 + 45, y1), 0, 0.5, (0, 255, 0), 1)
img = cv2AddChineseText(img, label, (x1, y1 - 15), (0, 255, 0), 15)
cv2.rectangle(img, (x1, y1), (x2, y2), (255, 200, 0), 2)
# cv2.putText(img, label, (x1, y1), 0, 0.6, (0, 255, 0), 2)
print("输出边框")
length = temp_count
print("一共用了多少个名字")
print(length)
# k = round(temp_conf / count, 4)
# percent = '{:.2%}'.format(k)
# sum = str(percent)
if temp_sum % 60 == 0:
k = temp_conf / count
glob_score = round(k * 100, 2)
globscore = str(glob_score)
# focus = 'Student concentration'
# cv2.putText(img, focus, (10, 20), 0, 1, (0, 255, 200), 4)
cv2.putText(img, globscore, (120, 50), 0, 0.8, (0, 255, 200), 1)
img = cv2AddChineseText(img, '课堂人数：51 人', (10, 60), (0, 255, 200), 25)
img = cv2AddChineseText(img, '课堂气氛：', (10, 30), (0, 255, 200), 25)
return img
if __name__ == '__main__':
cap = cv2.VideoCapture('D:\openCVProject\student30fps10min.mp4') # 视频文件检测
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out = cv2.VideoWriter('outputstudent10mintest3.mp4', 0x7634706d, 30.0, (1920, 1080))
temp_count = 1 # 出现的第几张图片
length = 0
if (cap.isOpened()): # 视频打开成功
while (True):
ret, frame = cap.read() # 读取一帧
# gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
result = mask_detecion(frame)
print("***\n")
cv2.imshow('mask_detection', result)
out.write(result)
if cv2.waitKey(1) & 0xFF == 27: # 按下Esc键退出
break
else:
print('open video failed!')
cap.release()
cv2.destroyAllWindows()
演示视频：https://www.bilibili.com/video/BV1Yg4y1u7X9/?vd_source=2fa09a0655d11f6cfae7287ba7e9df31
通过比较上下两帧位相似地点的左边距离来判断是否是同一张人脸图像，只能实现简单的人脸跟踪效果。
9504
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
