openpose人体姿态估计

参考博客：Openpose驾驶员危险驾驶检测（抽烟打电话）

人体姿态识别模型—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。
参考：Github开源人体姿态识别项目OpenPose中文文档
OpenPose项目Github链接：https://github.com/CMU-Perceptual-Computing-Lab/openpose

openpose模型有：手，脸，人体姿态。如图：

CoCo人体18点检测-图片

import cv2
import matplotlib.pyplot as plt
#配置文件
protoFile = './models/pose/coco/pose_deploy_linevec.prototxt'  
weightsfile = './models/pose/coco/pose_iter_440000.caffemodel' 
npoints = 18
POSE_PAIRS = [[1,0],[1,2],[1,5],[2,3],[3,4],[5,6],[6,7],[1,8],[8,9],[9,10],[1,11],[11,12],[12,13],[0,14],[0,15],[14,16],[15,17]]
#加载网络
net = cv2.dnn.readNetFromCaffe(protoFile,weightsfile)
#读取图片
im = cv2.imread('kai.jpg')
#（1440*1080）
im = cv2.cvtColor(im,cv2.COLOR_BGR2RGB)
inHeight = im.shape[0]  #1440
inWidth = im.shape[1]   #1080
netInputsize = (368,368)
#转为inpBlob格式
inpBlob = cv2.dnn.blobFromImage(im,1.0/255,netInputsize,(0,0,0),swapRB=True,crop=False)
#归一化后的图片最为输入传入net网络,然后输出
net.setInput(inpBlob)
output = net.forward()  #1*57*46*46


# plt.figure(figsize=(20,10))
# for i in range(npoints):
#     probMap = output[0,i,:,:]  #把18个output[2][3]输出出来  probMap.shape(46*46)
#     displayMap = cv2.resize(probMap,(inWidth,inHeight),cv2.INTER_LINEAR) #1440*1080
#     plt.subplot(3,6,i+1);plt.axis('on');
#     plt.imshow(displayMap,cmap='jet')
# plt.show()


scaleX = float(inWidth) / output.shape[3]    #float(1080)/64 = 23.47826086956522
scaleY = float(inHeight)/ output.shape[2]    #float(1440)/64 = 31.304347826086957
points = []
threshold = 0.1
for i in range(npoints):
    probMap = output[0,i,:,:]  #shape(46*46)
    minVal,prob,minLoc,point =cv2.minMaxLoc(probMap)
    x = scaleX * point[0]
    y = scaleY * point[1]
    if prob > threshold:
        points.append((int(x),int(y)))
    else:
        points.append(None)
    #points[]最后为18个关键点的坐标
    #[(516, 313), (516, 438), (399, 438), (375, 626), (352, 751), (610, 438), (633, 594), (657, 751), (446, 751),
    # (446, 970), (446, 1158), (563, 782), (540, 1001), (540, 1064), (493, 281), (540, 281), (446, 313), (563, 313)]


imPoints = im.copy()
imSkeleton = im.copy()
for i,p in enumerate(points):
    #enumerate把points的值前面带上索引i
    cv2.circle(imPoints,p,8,(255,255,0),thickness=1,lineType=cv2.FILLED)
    cv2.putText(imPoints,'{}'.format(i),p,cv2.FONT_HERSHEY_SIMPLEX,1,(255,0,0),2,lineType=cv2.LINE_AA)
for pair in POSE_PAIRS:
    partA = pair[0]
    partB = pair[1]
    #if points[partA] and points[partB]:
    cv2.line(imSkeleton,points[partA],points[partB],(255, 255,0),2)
    cv2.circle(imSkeleton, points[partA],8,(255,0,0),thickness=-1, lineType=cv2.FILLED)

plt.figure(figsize=(20,10))
plt.subplot(121)
plt.axis('off');plt.imshow(imPoints)
plt.subplot(122)
plt.axis('off');plt.imshow(imSkeleton)
plt.show()

检测结果：

CoCo人体18点检测-视频

把上述代码改为由opencv捕获摄像头，实时进行人体姿态检测

import cv2
def detect(img):
    protoFile = './models/pose/coco/pose_deploy_linevec.prototxt'
    weightsfile = './models/pose/coco/pose_iter_440000.caffemodel'
    npoints = 18
    POSE_PAIRS = [[1, 0], [1, 2], [1, 5], [2, 3], [3, 4], [5, 6], [6, 7], [1, 8], [8, 9], [9, 10], [1, 11], [11, 12],
                  [12, 13], [0, 14], [0, 15], [14, 16], [15, 17]]
    net = cv2.dnn.readNetFromCaffe(protoFile, weightsfile)
    inHeight = img.shape[0]  # 1440
    inWidth = img.shape[1]  # 1080
    netInputsize = (368, 368)
    inpBlob = cv2.dnn.blobFromImage(img, 1.0 / 255, netInputsize, (0, 0, 0), swapRB=True, crop=False)
    net.setInput(inpBlob)
    output = net.forward()

    scaleX = float(inWidth) / output.shape[3]
    scaleY = float(inHeight) / output.shape[2]
    points = []
    threshold = 0.1
    for i in range(npoints):
        probMap = output[0, i, :, :]  # shape(46*46)
        minVal, prob, minLoc, point = cv2.minMaxLoc(probMap)
        x = scaleX * point[0]
        y = scaleY * point[1]
        if prob > threshold:
            points.append((int(x), int(y)))
        else:
            points.append(None)

    for i, p in enumerate(points):
        # enumerate把points的值前面带上索引i
        cv2.circle(img, p, 8, (255, 255, 0), thickness=1, lineType=cv2.FILLED)
        cv2.putText(img, '{}'.format(i), p, cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2, lineType=cv2.LINE_AA)


cap = cv2.VideoCapture(0)
while True:
    ret, frame = cap.read()
    detect(frame)

    cv2.imshow('frame',frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):  # 按‘q’退出
        break

检测结果：

距离计算

为了检测驾驶员是否存在吸烟和打电话等危险行为，可以计算人体特定关键点之间的距离。
吸烟：右手到鼻子的距离或者左手到鼻子的距离。
打电话：右手到右耳的距离或者左手到左耳的距离。

import math
def __distance(A,B):
    if A is None or B is None:
        return 0
    else:
        return math.sqrt((A[0]-B[0])**2 + (A[1]-B[1])**2)

distance0 =  __distance(points[0],points[4]) #右手到鼻子距离
distance1 =  __distance(points[0],points[7]) #左手到鼻子距离
distance2 =  __distance(points[4],points[16]) #右手到右耳距离
distance3 =  __distance(points[7],points[17]) #左手到左耳距离
print('右手到鼻子距离：',distance0,'左手到鼻子距离：',distance1)
print('右手到右耳距离：',distance2,'左手到左耳距离:',distance3

计算结果

右手到鼻子距离： 467.6964827748868 左手到鼻子距离： 460.135849505339
右手到右耳距离： 447.97321348491363 左手到左耳距离: 447.97321348491363