Facial Recognition in Python （Dnn）

人脸识别（五）

5.1 前言

       续上一篇博客《人脸识别（四）》，之前的人脸识别都是用Haar特征进行识别，这次换用DNN网络进行人脸识别。

5.2 准备工作

       只是在之前博客的代码上略做改动，因此不必再多说什么了。

       和博客[《人脸识别（三）》](https://blog.csdn.net/lrglgy/article/details/90112977)一样，不过这里需要多下载四个神经网络模型文件，它们分别是tensorflow神经网络模型文件[opencv_face_detector_uint8.pb](https://github.com/Roggu123/Algorithm/blob/master/Practice/CV/FaceDetection/FaceDetect_py/model/opencv_face_detector_uint8.pb)、模型配置文件[opencv_face_detector.pbtxt](https://github.com/Roggu123/Algorithm/blob/master/Practice/CV/FaceDetection/FaceDetect_py/model/opencv_face_detector.pbtxt)，还有Caffe神经网络模型[res10_300x300_ssd_iter_140000_fp16.caffemodel](https://github.com/Roggu123/Algorithm/blob/master/Practice/CV/FaceDetection/FaceDetect_py/model/res10_300x300_ssd_iter_140000_fp16.caffemodel)、对应的模型配置文件[deploy.prototxt](https://github.com/Roggu123/Algorithm/blob/master/Practice/CV/FaceDetection/FaceDetect_py/model/deploy.prototxt)。  
## 5.3 代码  
# -*- coding: utf-8 -*-
# @Author   作者 : BogeyDa！！
# @FileName 文件名 : FaceDetect_Dnn.py
# @Software 创建文件的IDE : PyCharm
# @Blog     博客地址 : https://blog.csdn.net/lrglgy
# @Time     创建时间 : 2019-05-15 11:23
#
# @reference参考 : 博客：同上
#                 代码：[face_detection_opencv_dnn.py](https://github.com/spmallick/learnopencv/blob/master/FaceDetectionComparison/face_detection_opencv_dnn.py)
# @Log      代码说明：利用DNN网络识别人脸，他人Demo
from __future__ import division
import cv2
import time
import sys

def detectFaceOpenCVDnn(net, frame):
    frameOpencvDnn = frame.copy()
    frameHeight = frameOpencvDnn.shape[0]
    frameWidth = frameOpencvDnn.shape[1]
    blob = cv2.dnn.blobFromImage(frameOpencvDnn, 1.0, (300, 300), [104, 117, 123], False, False)

    net.setInput(blob)
    detections = net.forward()
    bboxes = []
    flag = 0
    for i in range(detections.shape[2]):
        confidence = detections[0, 0, i, 2]
        if confidence > conf_threshold:
            x1 = int(detections[0, 0, i, 3] * frameWidth)
            y1 = int(detections[0, 0, i, 4] * frameHeight)
            x2 = int(detections[0, 0, i, 5] * frameWidth)
            y2 = int(detections[0, 0, i, 6] * frameHeight)
            bboxes.append([x1, y1, x2, y2])
            cv2.rectangle(frameOpencvDnn, (x1, y1), (x2, y2), (0, 255, 0), int(round(frameHeight/150)), 8)
            flag = 1
    return frameOpencvDnn, bboxes, flag

if __name__ == "__main__" :

    # OpenCV DNN supports 2 networks.
    # 1. FP16 version of the original caffe implementation ( 5.4 MB )
    # 2. 8 bit Quantized version using Tensorflow ( 2.7 MB )
    # 选择进行人脸识别的网络：1.Tensorflow  2.Caffe
    print("Please choose the net you prefer:")
    print("1.Tensorflow        2.Caffe")
    DNN = input("Input:\n")
    if DNN == '2':
        modelFile = "model/res10_300x300_ssd_iter_140000_fp16.caffemodel"
        configFile = "model/deploy.prototxt"
        net = cv2.dnn.readNetFromCaffe(configFile, modelFile)
    else:
        modelFile = "model/opencv_face_detector_uint8.pb"
        configFile = "model/opencv_face_detector.pbtxt"
        # net = cv2.dnn.readNetFromTensorflow(configFile, modelFile)
        net = cv2.dnn.readNetFromTensorflow(modelFile, configFile)

    conf_threshold = 0.7   #设定神经网络阈值

    # 确定进行人脸识别的文件来源source, source=0表示摄像头，否则为指定路径的文件
    source = 0
    if len(sys.argv) > 1:
        source = sys.argv[1]

    # 定义人脸识别中涉及的参数及操作，
    # cap--被识别文件，frame--视频帧，frame_count--视频帧数，tt_opencvDnn--进行人脸识别的总时间
    # vid_writer--保存视频
    cap = cv2.VideoCapture(source)
    hasFrame, frame = cap.read()
    vid_writer = cv2.VideoWriter('Results/Dnn/output-dnn-{}.avi'.format(str(source).split(".")[0]),cv2.VideoWriter_fourcc('M','J','P','G'), 15, (frame.shape[1],frame.shape[0]))
    frame_count = 0
    tt_opencvDnn = 0
    reg = 0

    # 通过循环读取视频每一帧并进行人脸识别
    while(1):
        hasFrame, frame = cap.read()
        if not hasFrame:
            break
        frame_count += 1
        t = time.time()
        # flag表示是否识别出人脸
        outOpencvDnn, bboxes, flag = detectFaceOpenCVDnn(net,frame)
        if(flag==1):
            reg += 1
        tt_opencvDnn += time.time() - t
        fpsOpencvDnn = frame_count / tt_opencvDnn
        accuracy = reg/frame_count
        label1 = "OpenCV DNN ; FPS : {:.2f}：".format(fpsOpencvDnn)
        label2 = "OpenCV DNN ; Acc : {:.2f}".format(accuracy)
        cv2.putText(outOpencvDnn, label1, (10,50), cv2.FONT_HERSHEY_SIMPLEX, 1.4, (0, 0, 255), 3, cv2.LINE_AA)
        cv2.putText(outOpencvDnn, label2, (10,100), cv2.FONT_HERSHEY_SIMPLEX, 1.4, (0, 0, 255), 3, cv2.LINE_AA)

        cv2.imshow("Face Detection Comparison", outOpencvDnn)

        vid_writer.write(outOpencvDnn)
        if frame_count == 1:
            tt_opencvDnn = 0

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    print("The overall accuracy is:", accuracy)
    cv2.destroyAllWindows()
    vid_writer.release()

5.3 改进之处

       这次使用神经网络进行人脸识别，效果要比利用Harr特征识别要好一些。例如在头部活动较快时，Harr特征很多时候是识别不出人脸的，而DNN基本都能识别出来。
       这次的代码又在视频窗口中添加了识别准确率的显示，可以更好地与其它人脸识别方式进行对比，识别准确率显示如下（FPS是帧率，Acc是准确率）：

5.4 参考

face-detection-opencv-dlib-and-deep-learning-c-python
face_detection_opencv_dnn.py