Pose Detection

Find 33 body landmarks with CVZone's PoseDetector and measure the angles between them

Pose detection

Pose detection finds the position of a person’s body: their head, shoulders, elbows, hips, knees and feet. It’s used in fitness apps, motion capture and games. CVZone’s PoseDetector uses MediaPipe’s pose model, which finds 33 landmarks on the body.

Some useful landmark numbers:

The code

MediaPipe was installed in the first lesson, so you’re ready to go.

import cv2
from cvzone.PoseModule import PoseDetector

cap = cv2.VideoCapture(0)
detector = PoseDetector(detectionCon=0.5, trackCon=0.5)

while True:
    success, img = cap.read()
    if not success:
        break

    img = detector.findPose(img)
    lmList, bboxInfo = detector.findPosition(img, draw=True, bboxWithHands=False)

    if lmList:
        # Mark the centre of the body
        cv2.circle(img, bboxInfo["center"], 5, (255, 0, 255), cv2.FILLED)

        # Distance from the left shoulder (11) to the left wrist (15)
        length, img, info = detector.findDistance(lmList[11][0:2], lmList[15][0:2],
                                                  img=img, color=(255, 0, 0), scale=10)

        # Angle at the left elbow: shoulder (11), elbow (13), wrist (15)
        angle, img = detector.findAngle(lmList[11][0:2], lmList[13][0:2], lmList[15][0:2],
                                        img=img, color=(0, 0, 255), scale=10)

        # Is the elbow bent to about 50 degrees, give or take 10?
        print(int(angle), detector.angleCheck(myAngle=angle, targetAngle=50, offset=10))

    cv2.imshow("Pose Detection", img)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

How it works

• detector.findPose(img) runs the pose model and draws the skeleton on the image
• detector.findPosition(img) returns lmList, a list of 33 [x, y, z] landmarks, and bboxInfo, the box around the body. lmList is empty when nobody is in view
• lmList[11][0:2] takes just the x, y of landmark 11, the left shoulder
• findDistance() measures the distance between two points in pixels and draws it
• findAngle() measures the angle at the middle point. With shoulder, elbow and wrist, that’s how bent the elbow is
• angleCheck() returns True if the angle is within offset degrees of the target

Which side is left?

Landmark 11 is the person’s own left shoulder. If you’re facing the camera, it appears on the right of the picture.

Going further

• Exercise counter: count a bicep curl each time the elbow angle drops below 50 degrees and then goes back above 150
• Posture check: the angle between ear (7), shoulder (11) and hip (23) changes when you slouch at a desk. Try it from the side
• Robot control: raise your left or right arm to steer a robot

< Previous Next >

You can use the arrows  ← → on your keyboard to navigate between lessons.