OpenCV实现人体姿态估计(人体关键点检测）OpenPose

OpenPose人体姿态识别项目是美国卡耐基梅隆大学（CMU）基于卷积神经网络和监督学习并以Caffe为框架开发的开源库。可以实现人体动作、面部表情、手指运动等姿态估计。适用于单人和多人，具有极好的鲁棒性。是世界上首个基于深度学习的实时多人二维姿态估计应用，基于它的实例如雨后春笋般涌现。

其理论基础来自Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields ，是CVPR 2017的一篇论文，作者是来自CMU感知计算实验室的曹哲（http://people.eecs.berkeley.edu/~zhecao/\#top）,Tomas Simon,Shih-En Wei,Yaser Sheikh 。

人体姿态估计技术在体育健身、动作采集、3D试衣、舆情监测等领域具有广阔的应用前景，人们更加熟悉的应用就是抖音尬舞机。

OpenPose的效果并不怎么好，强烈推荐《2D Pose人体关键点检测(Python/Android /C++ Demo) 》2D Pose人体关键点实时检测(Python/Android /C++ Demo)_pan_jinquan的博客-CSDN博客，提供了C++推理代码和Android Demo

人体关键点检测需要用到人体检测，请查看鄙人另一篇博客：2D Pose人体关键点实时检测(Python/Android /C++ Demo)_pan_jinquan的博客-CSDN博客

OpenPose项目Github链接：https://github.com/CMU-Perceptual-Computing-Lab/openpose

OpenCV实现的Demo链接//github.com/PanJinquan/opencv-learning-tutorials/blob/master/opencv_dnn_pro/openpose-opencv/openpose_for_image_test.py）

1、实现原理

输入一幅图像，经过卷积网络提取特征，得到一组特征图，然后分成两个岔路，分别使用 CNN网络提取Part Confidence Maps 和 Part Affinity Fields；

得到这两个信息后，我们使用图论中的 Bipartite Matching（偶匹配）求出Part Association，将同一个人的关节点连接起来，由于PAF自身的矢量性，使得生成的偶匹配很正确，最终合并为一个人的整体骨架；
最后基于PAFs求Multi-Person Parsing—>把Multi-person parsing问题转换成graphs问题—>Hungarian Algorithm(匈牙利算法)
（匈牙利算法是部图匹配最常见的算法，该算法的核心就是寻找增广路径，它是一种用增广路径求二分图最大匹配的算法。)

2、实现神经网络

watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM4MTA2OTIz_size_16_color_FFFFFF_t_70

阶段一：VGGNet的前10层用于为输入图像创建特征映射。

阶段二：使用2分支多阶段CNN，其中第一分支预测身体部位位置（例如肘部，膝部等）的一组2D置信度图（S）。如下图所示，给出关键点的置信度图和亲和力图 - 左肩。

watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM4MTA2OTIz_size_16_color_FFFFFF_t_70 1

第二分支预测一组部分亲和度的2D矢量场（L），其编码部分之间的关联度。如下图所示，显示颈部和左肩之间的部分亲和力。

watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM4MTA2OTIz_size_16_color_FFFFFF_t_70 2

阶段三：通过贪心推理解析置信度和亲和力图，对图像中的所有人生成2D关键点。

3.OpenCV-OpenPose实现推理代码

# -*-coding: utf-8 -*-
"""
    @Project: python-learning-notes
    @File   : openpose_for_image_test.py
    @Author : panjq
    @E-mail : pan_jinquan@163.com
    @Date   : 2019-07-29 21:50:17
"""
import cv2 as cv
import os
import glob
BODY_PARTS = {"Nose": 0, "Neck": 1, "RShoulder": 2, "RElbow": 3, "RWrist": 4,
              "LShoulder": 5, "LElbow": 6, "LWrist": 7, "RHip": 8, "RKnee": 9,
              "RAnkle": 10, "LHip": 11, "LKnee": 12, "LAnkle": 13, "REye": 14,
              "LEye": 15, "REar": 16, "LEar": 17, "Background": 18}
POSE_PAIRS = [["Neck", "RShoulder"], ["Neck", "LShoulder"], ["RShoulder", "RElbow"],
              ["RElbow", "RWrist"], ["LShoulder", "LElbow"], ["LElbow", "LWrist"],
              ["Neck", "RHip"], ["RHip", "RKnee"], ["RKnee", "RAnkle"], ["Neck", "LHip"],
              ["LHip", "LKnee"], ["LKnee", "LAnkle"], ["Neck", "Nose"], ["Nose", "REye"],
              ["REye", "REar"], ["Nose", "LEye"], ["LEye", "LEar"]]
def detect_key_point(model_path, image_path, out_dir, inWidth=368, inHeight=368, threshhold=0.2):
    net = cv.dnn.readNetFromTensorflow(model_path)
    frame = cv.imread(image_path)
    frameWidth = frame.shape[1]
    frameHeight = frame.shape[0]
    scalefactor = 2.0
    net.setInput(
        cv.dnn.blobFromImage(frame, scalefactor, (inWidth, inHeight), (127.5, 127.5, 127.5), swapRB=True, crop=False))
    out = net.forward()
    out = out[:, :19, :, :]  # MobileNet output [1, 57, -1, -1], we only need the first 19 elements
    assert (len(BODY_PARTS) == out.shape[1])
    points = []
    for i in range(len(BODY_PARTS)):
        # Slice heatmap of corresponging body's part.
        heatMap = out[0, i, :, :]
        # Originally, we try to find all the local maximums. To simplify a sample
        # we just find a global one. However only a single pose at the same time
        # could be detected this way.
        _, conf, _, point = cv.minMaxLoc(heatMap)
        x = (frameWidth * point[0]) / out.shape[3]
        y = (frameHeight * point[1]) / out.shape[2]
        # Add a point if it's confidence is higher than threshold.
        points.append((int(x), int(y)) if conf > threshhold else None)
    for pair in POSE_PAIRS:
        partFrom = pair[0]
        partTo = pair[1]
        assert (partFrom in BODY_PARTS)
        assert (partTo in BODY_PARTS)
        idFrom = BODY_PARTS[partFrom]
        idTo = BODY_PARTS[partTo]
        if points[idFrom] and points[idTo]:
            cv.line(frame, points[idFrom], points[idTo], (0, 255, 0), 3)
            cv.ellipse(frame, points[idFrom], (3, 3), 0, 0, 360, (0, 0, 255), cv.FILLED)
            cv.ellipse(frame, points[idTo], (3, 3), 0, 0, 360, (0, 0, 255), cv.FILLED)
    t, _ = net.getPerfProfile()
    freq = cv.getTickFrequency() / 1000
    cv.putText(frame, '%.2fms' % (t / freq), (10, 20), cv.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0))
    cv.imwrite(os.path.join(out_dir, os.path.basename(image_path)), frame)
    cv.imshow('OpenPose using OpenCV', frame)
    cv.waitKey(0)
def detect_image_list_key_point(image_dir, out_dir, inWidth=480, inHeight=480, threshhold=0.3):
    image_list = glob.glob(image_dir)
    for image_path in image_list:
        detect_key_point(image_path, out_dir, inWidth, inHeight, threshhold)
if __name__ == "__main__":
    model_path = "pb/graph_opt.pb"
    # image_path = "body/*.jpg"
    out_dir = "result"
    # detect_image_list_key_point(image_path,out_dir)
    image_path = "./test.jpg"
    detect_key_point(model_path, image_path, out_dir, inWidth=368, inHeight=368, threshhold=0.05)