python+mediapip 实现AI姿态检测健身姿态检测追踪项目
python+mediapip 实现AI姿态检测健身姿态检测追踪项目
最近研究mediapipe 这个东东，感觉有点意思，有点上瘾。如果实现了姿态检测，那么我们可以用这些姿态检测的坐标做一下项目了，比如说，如何检测健身举哑铃的动作检测，虽然功能十分简单，但是要用Python 去实现一个动作的检测，在代码层次来讲还是很繁琐的。 下面讲解一下如何使用python+opencv+mediapipe实现姿态检测，并对举哑铃这个动作进行识别。
要实现上面所说的功能，需要实现以下步骤，下面我们一步一步的实现下面的步骤，以完成整个的功能。
- Install and Import Dependencies
- Make Detections
- Determining Joints
- Calculate Angles
- Curl Counter
1. Install and Import Dependencies
首先是运行环境的检测， 要看下你的mediapipe 与opencv-python 依赖是成功安装，并且摄像头能成功的采集到你的那张帅脸，如果这些都没有问题，那么就可以进行下面的操作了。
pip install mediapipe opencv-python
import cv2
import mediapipe as mp
import numpy as np
mp_drawing = mp.solutions.drawing_utils
mp_pose = mp.solutions.pose
# VIDEO FEED
cap = cv2.VideoCapture(0)
while cap.isOpened():
ret, frame = cap.read()
cv2.imshow('Mediapipe Feed', frame)
if cv2.waitKey(10) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
2. Make Detections
cap = cv2.VideoCapture(0)
## Setup mediapipe instance
## 开启我们的姿态检测进程函数，这里有两个指标检测置信度和跟踪置信度，作用是控制模型检测的准确度和灵敏度
with mp_pose.Pose(min_detection_confidence=0.5, min_tracking_confidence=0.5) as pose:
while cap.isOpened():
ret, frame = cap.read()
# Recolor image to RGB
## 摄像投的数据都是以BGR的形式，但是模型处理要以RGB,所以这里的颜色空间要进行转换
image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
image.flags.writeable = False
# Make detection
results = pose.process(image)
# Recolor back to BGR
image.flags.writeable = True
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
# Render detections
##然后进行画点和连线的操作了
mp_drawing.draw_landmarks(image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS,
mp_drawing.DrawingSpec(color=(245,117,66), thickness=2, circle_radius=2),
mp_drawing.DrawingSpec(color=(245,66,230), thickness=2, circle_radius=2)
)
cv2.imshow('Mediapipe Feed', image)
if cv2.waitKey(10) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
代码的功能我基本上已经在注释中阐述明白了， 完成了上面的步骤，那么基本上就能够检测到你的姿态的模型了。
3.Determining Joints
下图是姿态检测模型中的33个关节点，我们会获取所需要的关节点的坐标，然后进行算法的计算，以达到我们所需的功能需求。下面我会获取我们的其中的关节坐标，然后计算肘部关节的角度，
# Extract landmarks
try:
landmarks = results.pose_landmarks.landmark
print(landmarks)
except:
pass
我们可以添加两行代码将我们每一个坐标点打出来看一下
可以看到，每个关节的坐标点都可以获取到。
另外，还可以用以下代码验证有多少个关节点，以及是哪一个关节点。
print(len(landmarks))
33
for lndmrk in mp_pose.PoseLandmark:
print(lndmrk)
PoseLandmark.NOSE
PoseLandmark.LEFT_EYE_INNER
PoseLandmark.LEFT_EYE
PoseLandmark.LEFT_EYE_OUTER
PoseLandmark.RIGHT_EYE_INNER
PoseLandmark.RIGHT_EYE
PoseLandmark.RIGHT_EYE_OUTER
PoseLandmark.LEFT_EAR
PoseLandmark.RIGHT_EAR
PoseLandmark.MOUTH_LEFT
PoseLandmark.MOUTH_RIGHT
PoseLandmark.LEFT_SHOULDER
PoseLandmark.RIGHT_SHOULDER
PoseLandmark.LEFT_ELBOW
PoseLandmark.RIGHT_ELBOW
PoseLandmark.LEFT_WRIST
PoseLandmark.RIGHT_WRIST
PoseLandmark.LEFT_PINKY
PoseLandmark.RIGHT_PINKY
PoseLandmark.LEFT_INDEX
PoseLandmark.RIGHT_INDEX
PoseLandmark.LEFT_THUMB
PoseLandmark.RIGHT_THUMB
PoseLandmark.LEFT_HIP
PoseLandmark.RIGHT_HIP
PoseLandmark.LEFT_KNEE
PoseLandmark.RIGHT_KNEE
PoseLandmark.LEFT_ANKLE
PoseLandmark.RIGHT_ANKLE
PoseLandmark.LEFT_HEEL
PoseLandmark.RIGHT_HEEL
PoseLandmark.LEFT_FOOT_INDEX
PoseLandmark.RIGHT_FOOT_INDEX
landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].visibility
landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value]
landmarks[mp_pose.PoseLandmark.LEFT_WRIST.value]
4. Calculate Angles
接下来 我们获取到了坐标，那么就需要用坐标计算出我们胳膊的角度了，用来判断是否有举哑铃这个动作。
def calculate_angle(a,b,c):
a = np.array(a) # First
b = np.array(b) # Mid
c = np.array(c) # End
radians = np.arctan2(c[1]-b[1], c[0]-b[0]) - np.arctan2(a[1]-b[1], a[0]-b[0])
angle = np.abs(radians*180.0/np.pi)
if angle >180.0:
angle = 360-angle
return angle
上面，就是用于计算坐标的Fun， 角度范围0-180。
如何获取我们所需要的左边呢?
shoulder = [landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].x,landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].y]
elbow = [landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value].x,landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value].y]
wrist = [landmarks[mp_pose.PoseLandmark.LEFT_WRIST.value].x,landmarks[mp_pose.PoseLandmark.LEFT_WRIST.value].y]
获取函数所需的坐标点。
然后看整体代码
cap = cv2.VideoCapture(0)
## Setup mediapipe instance
with mp_pose.Pose(min_detection_confidence=0.5, min_tracking_confidence=0.5) as pose:
while cap.isOpened():
ret, frame = cap.read()
# Recolor image to RGB
image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
image.flags.writeable = False
# Make detection
results = pose.process(image)
# Recolor back to BGR
image.flags.writeable = True
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
# Extract landmarks
try:
landmarks = results.pose_landmarks.landmark
# Get coordinates
shoulder = [landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].x,landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].y]
elbow = [landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value].x,landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value].y]
wrist = [landmarks[mp_pose.PoseLandmark.LEFT_WRIST.value].x,landmarks[mp_pose.PoseLandmark.LEFT_WRIST.value].y]
# Calculate angle
angle = calculate_angle(shoulder, elbow, wrist)
# Visualize angle
cv2.putText(image, str(angle),
tuple(np.multiply(elbow, [640, 480]).astype(int)),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 2, cv2.LINE_AA
)
except:
pass
# Render detections
mp_drawing.draw_landmarks(image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS,
mp_drawing.DrawingSpec(color=(245,117,66), thickness=2, circle_radius=2),
mp_drawing.DrawingSpec(color=(245,66,230), thickness=2, circle_radius=2)
)
cv2.imshow('Mediapipe Feed', image)
if cv2.waitKey(10) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
角度基本上也没有问题了。
5.Curl Counter
那么就是最后一步了，为了实现举哑铃这个动作的检测，实现一个计数器的功能是最基本的。
cap = cv2.VideoCapture(0)
# Curl counter variables
counter = 0
stage = None
## Setup mediapipe instance
with mp_pose.Pose(min_detection_confidence=0.5, min_tracking_confidence=0.5) as pose:
while cap.isOpened():
ret, frame = cap.read()
# Recolor image to RGB
image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
image.flags.writeable = False
# Make detection
results = pose.process(image)
# Recolor back to BGR
image.flags.writeable = True
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
# Extract landmarks
try:
landmarks = results.pose_landmarks.landmark
# Get coordinates
shoulder = [landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].x,landmarks[mp_pose.PoseLandmark.LEFT_SHOULDER.value].y]
elbow = [landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value].x,landmarks[mp_pose.PoseLandmark.LEFT_ELBOW.value].y]
wrist = [landmarks[mp_pose.PoseLandmark.LEFT_WRIST.value].x,landmarks[mp_pose.PoseLandmark.LEFT_WRIST.value].y]
# Calculate angle
angle = calculate_angle(shoulder, elbow, wrist)
# Visualize angle
cv2.putText(image, str(angle),
tuple(np.multiply(elbow, [640, 480]).astype(int)),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 2, cv2.LINE_AA
)
# Curl counter logic
if angle > 160:
stage = "down"
if angle < 30 and stage =='down':
stage="up"
counter +=1
print(counter)
except:
pass
# Render curl counter
# Setup status box
cv2.rectangle(image, (0,0), (225,73), (245,117,16), -1)
# Rep data
cv2.putText(image, 'REPS', (15,12),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0,0,0), 1, cv2.LINE_AA)
cv2.putText(image, str(counter),
(10,60),
cv2.FONT_HERSHEY_SIMPLEX, 2, (255,255,255), 2, cv2.LINE_AA)
# Stage data
cv2.putText(image, 'STAGE', (65,12),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0,0,0), 1, cv2.LINE_AA)
cv2.putText(image, stage,
(60,60),
cv2.FONT_HERSHEY_SIMPLEX, 2, (255,255,255), 2, cv2.LINE_AA)
# Render detections
mp_drawing.draw_landmarks(image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS,
mp_drawing.DrawingSpec(color=(245,117,66), thickness=2, circle_radius=2),
mp_drawing.DrawingSpec(color=(245,66,230), thickness=2, circle_radius=2)
)
cv2.imshow('Mediapipe Feed', image)
if cv2.waitKey(10) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
感兴趣的朋友赶紧去试试吧，不停奥利给，看看计数多少个奥利给！