/Video-Pose-Estimation

Primary LanguagePython

Video-Pose-Estimation

This repository is taken as an inspiration from here with the following modifications:

  1. Used MMPose instead of OpenPose.
  2. Used Human model of MMPose that gives 17 keypoints instead of BODY_25 model of OpenPose.
  3. Implemented modules and easy-to-understand code.

Why MMPose?

  1. Flexibility and Customization: MMPose is an open-source toolbox for pose estimation based on PyTorch. It supports both top-down and bottom-up approaches and provides a wide range of models and algorithms for pose, hand, and whole-body estimation. This flexibility allows developers to choose the most suitable approach for their specific use case.
  2. Advanced Features: MMPose offers advanced features such as 2D keypoint and 3D surface estimation. It also supports different industries including sports analysis, robotics, healthcare, security and surveillance, and entertainment.
  3. Performance and Accuracy: MMPose provides state-of-the-art accuracy and fast inference time, making it suitable for real-time pose estimation and multi-person scenarios. It outperforms some alternatives like OpenPose in terms of accuracy and computational efficiency.
  4. Cost: While OpenPose requires a non-refundable $25,000 USD annual royalty for commercial use, MMPose is freely available for commercial use under the Apache 2.0 license.

Keypoints Comparision HUMAN model vs BODY_25 model.

  1. Left: MMPose
  2. Right: OpenPose Keypoints

Serial No mmpose Keypoints Modified openpose Keypoints
0 Nose Nose
1 Left Eye Neck
2 Right Eye Right Shoulder
3 Left Ear Right Elbow
4 Right Ear Right Wrist
5 Left Shoulder Left Shoulder
6 Right Shoulder Left Elbow
7 Left Elbow Left Wrist
8 Right Elbow MidHip
9 Left Wrist Right Hip
10 Right Wrist Right Knee
11 Left Hip Right Ankle
12 Right Hip Left Hip
13 Left Knee Left Knee
14 Right Knee Left Ankle
15 Left Ankle Right Eye
16 Right Ankle Left Eye
17 Right Ear
18 Left Ear
19 Left Big Toe
20 Left Small Toe
21 Left Heel
22 Right Big Toe
23 Right Small Toe
24 Right Heel
Background

How to set up MMPOSE?

To install mmpose, you need to first set up and install mmaction2. Use this documentation to install mmaction2. You can also directly use the steps mentioned below:

  1. conda create --name mmaction python=3.10 -y
  2. conda activate mmaction
  3. conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
  4. Install the following in sequence: pip install -U openmim mim install mmengine mim install mmcv mim install mmdet mim install mmpose
  5. pip install mmaction2

Pose Classification

Just like the original repository, we are classifying four poses as follows:

  1. Hello
  2. Stop gesture
  3. Sitting pose
  4. Standing pose

For the detailed methodology and rule-based approach description, read this passage.

How to run?

Change the video_path in the code and run the following.

python pose.py