IntermediateComputer VisionFull notebook

Human Pose Estimation using OpenCV DNN and OpenPose

Notebook-based pose estimation tutorial using OpenCV DNN, OpenPose BODY_PARTS and POSE_PAIRS, image inference and video/webcam workflow.

Watch YouTube Video Download Notebook / Source Code

This download currently contains the uploaded notebook PoseEstimation_during_Video.ipynb. Replace it later with your final cleaned code ZIP if required.

Human Pose Estimation using OpenCV DNN and OpenPose

Overview

This tutorial is based on your uploaded PoseEstimation_during_Video.ipynb notebook. It explains how to load the TensorFlow OpenPose graph, define body parts and pose pairs, run pose estimation on an image, then extend the same logic to video and webcam streams. All notebook code cells and output images are included as editable tutorial material.

Editing tip: This page is controlled from data/tutorials.ts. Change the title, text, images, Python code, YouTube link, and download link there.

What you will learn

Load the OpenPose TensorFlow graph in OpenCV DNN
Define BODY_PARTS and POSE_PAIRS
Estimate body keypoints on an image
Draw skeleton connections
Adapt the image pipeline to video or webcam input

Suggested tutorial flow

01

OpenCV DNN setup

Load the TensorFlow graph and define image input size and threshold.

02

Pose dictionary setup

Define body parts and skeleton connections.

03

Image pose estimation

Run forward pass and draw detected body joints and lines.

04

Video demo

Process video frames with the same pose estimation function.

05

Webcam demo

Use webcam input for realtime pose estimation.

Text explanation template

Notebook-based tutorial material

This tutorial is based on your uploaded PoseEstimation_during_Video.ipynb notebook. It explains how to load the TensorFlow OpenPose graph, define body parts and pose pairs, run pose estimation on an image, then extend the same logic to video and webcam streams. All notebook code cells and output images are included as editable tutorial material.

How this page is organized

The page shows the full notebook workflow as tutorial sections. Each code cell is rendered as a separate Python code block. If the notebook cell produced an image, graph, screenshot or display output, it is shown under the matching code block and also in the results gallery.

How to edit this later

Open data/tutorials.ts and edit the object with this tutorial slug. Replace resultImages with your final screenshots, update codeBlocks with cleaned code, change videoUrl to the exact YouTube link, and replace the download URL with your final source-code ZIP or GitHub link.

Images, screenshots and results

Add your result screenshots, graphs, model outputs, detected objects, confusion matrices, or interface snapshots here.

human-pose-estimation-opencv-openpose notebook output cell 5 — Notebook output from cell 5: plt.imshow(img) ##BGR

human-pose-estimation-opencv-openpose notebook output cell 6 — Notebook output from cell 6: plt.imshow(cv.cvtColor(img, cv.COLOR_BGR2RGB))

human-pose-estimation-opencv-openpose notebook output cell 9 — Notebook output from cell 9: plt.imshow(cv.cvtColor(estimated_image, cv.COLOR_BGR2RGB))

Python code explanation

Use one or more code blocks. After each block, explain what the code does in simple teaching language.

Notebook cell 0: import cv2 as cv ## pip install opencv-python

pythonexample

import cv2 as cv  ## pip install opencv-python
import matplotlib.pyplot as plt

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 1: net = cv.dnn.readNetFromTensorflow("graph_opt.pb") ## weights

pythonexample

net = cv.dnn.readNetFromTensorflow("graph_opt.pb") ## weights

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 2: inWidth = 368

pythonexample

inWidth = 368
inHeight = 368
thr = 0.2

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 3: BODY_PARTS = { "Nose": 0, "Neck": 1, "RShoulder": 2, "RElbow": 3, "RWrist": 4,

pythonexample

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"] ]

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 4: img = cv.imread("pose.png")

pythonexample

img = cv.imread("pose.png")

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 5: plt.imshow(img) ##BGR

pythonexample

plt.imshow(img) ##BGR

Notebook output

<matplotlib.image.AxesImage at 0x18475abd088>

human-pose-estimation-opencv-openpose notebook output cell 5 — Notebook output from cell 5: plt.imshow(img) ##BGR

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 6: plt.imshow(cv.cvtColor(img, cv.COLOR_BGR2RGB))

pythonexample

plt.imshow(cv.cvtColor(img, cv.COLOR_BGR2RGB))

Notebook output

<matplotlib.image.AxesImage at 0x18476ffa888>

human-pose-estimation-opencv-openpose notebook output cell 6 — Notebook output from cell 6: plt.imshow(cv.cvtColor(img, cv.COLOR_BGR2RGB))

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 7: #255/2 = 127.5

pythonexample

#255/2 = 127.5
def pose_estimation(frame):
    frameWidth = frame.shape[1]
    frameHeight = frame.shape[0]
    net.setInput(cv.dnn.blobFromImage(frame, 1.0, (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 = []## list
    
    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 > thr 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))
    return frame

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 8: estimated_image = pose_estimation(img)

pythonexample

estimated_image = pose_estimation(img)

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 9: plt.imshow(cv.cvtColor(estimated_image, cv.COLOR_BGR2RGB))

pythonexample

plt.imshow(cv.cvtColor(estimated_image, cv.COLOR_BGR2RGB))

Notebook output

<matplotlib.image.AxesImage at 0x1847707ad88>

human-pose-estimation-opencv-openpose notebook output cell 9 — Notebook output from cell 9: plt.imshow(cv.cvtColor(estimated_image, cv.COLOR_BGR2RGB))

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 11: cap = cv.VideoCapture('Record_2020_10_13_23_32_51_93.mp4')

pythonexample

cap = cv.VideoCapture('Record_2020_10_13_23_32_51_93.mp4')
cap.set(3,800)
cap.set(4,800)

if not cap.isOpened():
    cap = cv.VideoCapture(0)
if not cap.isOpened():
    raise IOError("Cannot open video")

while cv.waitKey(1) < 0: 
    hasFrame,frame = cap.read()
    if not hasFrame:
        cv.waitKey()
        break
        
    frameWidth = frame.shape[1]
    frameHeight = frame.shape[0]
    net.setInput(cv.dnn.blobFromImage(frame, 1.0, (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 = []## list
    
    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 > thr 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.imshow('Pose estimation Tutorial ', frame)

Notebook output

KeyboardInterrupt: 
[1;31m---------------------------------------------------------------------------
[1;31mKeyboardInterrupt                         Traceback (most recent call last)
[1;32m<ipython-input-12-b7829fd437bc> in <module>[1;34m
     11     hasFrame[1;33m,frame [1;33m= cap[1;33m.read[1;33m([1;33m)[1;33m[1;33m
     12     [1;32mif [1;32mnot hasFrame[1;33m:[1;33m[1;33m
[1;32m---> 13[1;33m         cv[1;33m.waitKey[1;33m([1;33m)[1;33m[1;33m
     14         [1;32mbreak[1;33m[1;33m
     15 [1;33m

[1;31mKeyboardInterrupt:

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Notebook cell 18: cap = cv.VideoCapture(1)

pythonexample

cap = cv.VideoCapture(1)

cap.set(cv.CAP_PROP_FPS, 10)
cap.set(3,800)
cap.set(4,800)

if not cap.isOpened():
    cap = cv.VideoCapture(0)
if not cap.isOpened():
    raise IOError("Cannot open webcame")

while cv.waitKey(1) < 0: 
    hasFrame,frame = cap.read()
    if not hasFrame:
        cv.waitKey()
        break
        
    frameWidth = frame.shape[1]
    frameHeight = frame.shape[0]
    net.setInput(cv.dnn.blobFromImage(frame, 1.0, (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 = []## list
    
    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 > thr 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.imshow('Pose estimation Tutorial ', frame)

Notebook output

KeyboardInterrupt: 
[1;31m---------------------------------------------------------------------------
[1;31mKeyboardInterrupt                         Traceback (most recent call last)
[1;32mC:\ProgramData\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py in run_code[1;34m(self, code_obj, result, async_)
   3325                 [1;32melse[1;33m:[1;33m[1;33m
[1;32m-> 3326[1;33m                     exec[1;33m(code_obj[1;33m, self[1;33m.user_global_ns[1;33m, self[1;33m.user_ns[1;33m)[1;33m[1;33m
   3327             [1;32mfinally[1;33m:[1;33m[1;33m

[1;32m<ipython-input-13-19aa5f334ae0> in <module>[1;34m
     34         [1;31m# could be detected this way.[1;33m[1;33m[1;33m
[1;32m---> 35[1;33m         _[1;33m, conf[1;33m, _[1;33m, point [1;33m= cv[1;33m.minMaxLoc[1;33m(heatMap[1;33m)[1;33m[1;33m
     36         x [1;33m= [1;33m(frameWidth [1;33m* point[1;33m[[1;36m0[1;33m][1;33m) [1;33m/ out[1;33m.shape[1;33m[[1;36m3[1;33m][1;33m[1;33m

[1;31mKeyboardInterrupt: 

During handling of the above exception, another exception occurred:

[1;31mKeyboardInterrupt                         Traceback (most recent call last)
[1;32mC:\ProgramData\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py in run_code[1;34m(self, code_obj, result, async_)
   3328                 [1;31m# Reset our crash handler in place[1;33m[1;33m[1;33m
[1;32m-> 3329[1;33m                 sys[1;33m.excepthook [1;33m= old_excepthook[1;33m[1;33m
   3330         [1;32mexcept SystemExit [1;32mas e[1;33m:[1;33m[1;33m

This code block is extracted from the uploaded Jupyter notebook. Replace or rewrite the explanation later with your final teaching notes for this step.

Download source code

Use this area to provide the complete source code, GitHub repository, Google Drive ZIP, notebook, dataset, or premium download link.

Download Notebook / Source Code

Related Tutorials

Keep learning

Image Recognition using MobileNetV2

Deep Learning45 min

Image Recognition using MobileNetV2

Actual notebook-based tutorial for loading images in Python, preprocessing for MobileNetV2 and decoding ImageNet predictions.

Read tutorial Watch video

Facial Emotion Recognition using DeepFace and OpenCV

Computer Vision50 min

Facial Emotion Recognition using DeepFace and OpenCV

Actual notebook-based tutorial for detecting faces, analyzing emotions with DeepFace and drawing emotion labels using OpenCV.

Read tutorial Watch video

Driver Identification Using Deep Learning

Research Project45 min

Driver Identification Using Deep Learning

Driver identification pipeline using deep learning for recognizing driver identity from image/video input.

Read tutorial Watch video