深度学习之人物年龄预测

1.年龄检测

论文地址：《Age and Gender Classification using Convolutional Neural Networks》
论文作者提出了一个简单的类似AlexNet的网络结构，该网络总共学习了8个年龄段：

0-2
4-6
8-12
15-20
25-32
38-43
48-53
60-100

注意：这些年龄段不是连续的
首先，要搞懂检测年龄适合用回归还是用分类来做
举个栗子：

1. 回归

2. 分类

年龄预测是基于面部外观，有的人保养的好，显得年轻，实际年龄与测得会有差别。在不结合其它有关信息作推断情况下，网络模型很难预测到实际的真实年龄。若看作是回归问题，模型很难预测到图像中年龄的一个准确值，而看作是分类问题，预测一个年龄段相对来说模型更容易训练，比回归产生更高的准确性。

2.思路方法

自动识别年龄步骤：

1. 检测出输入图像或视频中的人脸
2. 提取面部感兴趣区域（ROI）
3. 用年龄检测器预测人物的年龄
4. 返回结果

对于检测人脸的分类器：

分类器	优缺点
Haar级联	速度快，嵌入式设备上运行，但准确性低
HOG +线性SVM	相比Haar级联精确，但速度慢，对遮挡，面部角度变化时检测效果不好
深度学习检测器	相比以上两者效果最佳，但需消耗更多计算资源

3.代码实现

环境：

win10
pycharm
anaconda3
python3.7
opencv4.2.0

对于OpenCV尽量用最新版本，可参考这篇仅一个命令行简单快速安装：https://blog.csdn.net/y459541195/article/details/104851892

文件结构：

3.1 单张图像检测代码

import numpy as np
import cv2"""
#图片年龄预测
执行：
python test_age.py """
# 检测年龄段
AGE_LIST = ["(0-2)","(4-6)","(8-12)","(15-20)","(25-32)","(38-43)","(48-53)","(60-100)"]# 人脸检测模型路径
prototxtPathF ="./models/face_detector/face_deploy.prototxt"
weightsPathF = "./models/face_detector/res10_300x300_ssd_iter_140000.caffemodel"
# 加载人脸模型
faceNet = cv2.dnn.readNet(prototxtPathF,weightsPathF)# 年龄检测模型
prototxtPathA ="./models/age_detector/age_deploy.prototxt"
weightsPathA = "./models/age_detector/age_net.caffemodel"
#加载模型
ageNet = cv2.dnn.readNet(prototxtPathA,weightsPathA)#获取图像
image = cv2.imread("./input/test01.jpg")
src = image.copy()
(h,w)= image.shape[:2]# 构造blob
blob = cv2.dnn.blobFromImage(image,1.0,(300,300),(104,177,123))
# 送入网络计算
faceNet.setInput(blob)
detect = faceNet.forward()
# 检测
for i in range(0,detect.shape[2]):confidence = detect[0,0,i,2]# 过滤掉小的置信度,计算坐标,提取面部roi,构造面部blob特征if confidence > 0.5:box = detect[0,0,i,3:7]*np.array([w,h,w,h])(startX,startY,endX,endY) = box.astype("int")face = image[startY:endY,startX:endX]faceBlob = cv2.dnn.blobFromImage(face, 1.0, (227, 227),(78.4263377603, 87.7689143744, 114.895847746),swapRB=False)# 预测年龄ageNet.setInput(faceBlob)predictions = ageNet.forward()i = predictions[0].argmax()age = AGE_LIST[i]ageConfidence = predictions[0][i]#显示打印text = "age{}:{:.2f}%".format(age,ageConfidence*100)print(text)#绘制显示框y = startY - 10 if startY - 10 > 10 else startY + 10cv2.rectangle(image, (startX, startY), (endX, endY),(0, 0, 255), 2)cv2.putText(image, text, (startX, y),cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 0, 255), 2)cv2.imshow("Result",image)
cv2.waitKey(0)

3.2 视频流检测代码

import numpy as np
import cv2
import imutils"""
#视频流年龄预测
执行：
python test_video_age.py
"""
def detect_age(frame,faceNet,ageNet,minConfidence=0.5):# 检测年龄段AGE_LIST = ["(0-2)","(4-6)","(8-12)","(15-20)","(25-32)","(38-43)","(48-53)","(60-100)"]#定义空列表存放结果results = [](h,w)= frame.shape[:2]# 构造blobblob = cv2.dnn.blobFromImage(frame,1.0,(300,300),(104,177,123))# 送入网络计算faceNet.setInput(blob)detect = faceNet.forward()# 检测for i in range(0,detect.shape[2]):confidence = detect[0,0,i,2]# 过滤掉小的置信度,计算坐标,提取面部roi,if confidence > minConfidence:box = detect[0,0,i,3:7]*np.array([w,h,w,h])(startX,startY,endX,endY) = box.astype("int")face = frame[startY:endY,startX:endX]# 过滤干扰if face.shape[0]<20 or face.shape[1]<20:continue# 构造面部blob特faceBlob = cv2.dnn.blobFromImage(face, 1.0, (227, 227),(78.4263377603, 87.7689143744, 114.895847746),swapRB=False)# 预测年龄ageNet.setInput(faceBlob)predictions = ageNet.forward()i = predictions[0].argmax()age = AGE_LIST[i]ageConfidence = predictions[0][i]# 构造字典存放结果dicts = {"location":(startX,startY,endX,endY),"age":(age,ageConfidence)}results.append(dicts)return results# 人脸检测模型路径
prototxtPathF ="./models/face_detector/face_deploy.prototxt"
weightsPathF = "./models/face_detector/res10_300x300_ssd_iter_140000.caffemodel"
# 加载人脸模型
faceNet = cv2.dnn.readNet(prototxtPathF,weightsPathF)# 年龄检测模型
prototxtPathA ="./models/age_detector/age_deploy.prototxt"
weightsPathA = "./models/age_detector/age_net.caffemodel"
#加载模型
ageNet = cv2.dnn.readNet(prototxtPathA,weightsPathA)#获取视频图像
videoPath = "./input/test1.mp4"
vs = cv2.VideoCapture(videoPath)#处理视频流
while True:(grabbed,frame) = vs.read()# 判断是否结束if not grabbed:print("无视频读取...")breakframe = imutils.resize(frame,width=720)#调用上面函数计算results = detect_age(frame,faceNet,ageNet,minConfidence=0.5)for  i in results:#显示信息text = "age{}:{:.2f}%".format(i["age"][0], i["age"][1] * 100)(startX,startY,endX,endY) = i["location"]# 绘制显示框y = startY - 10 if startY - 10 > 10 else startY + 10cv2.rectangle(frame, (startX, startY), (endX, endY),(0, 0, 255), 2)cv2.putText(frame, text, (startX, y),cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 0, 255), 2)# 显示cv2.imshow("Result", frame)key = cv2.waitKey(1) & 0xFF# 按q键退出循环if key == ord("q"):break
cv2.destroyAllWindows()
vs.release()

4.测试结果

4.1 单张图像测试

虚拟环境下命令行输入：

python test_age.py

效果一：

效果二：

效果三：

4.2 视频流检测

深度学习之年龄检测

视频地址： https://www.bilibili.com/video/BV16g4y187iQ/

Reference：
1.https://talhassner.github.io/home/publication/2015_CVPR
2.https://github.com/dpressel/rude-carnie
3.https://github.com/GilLevi/AgeGenderDeepLearning