Skip to content

Commit

Permalink
Project import generated by Copybara.
Browse files Browse the repository at this point in the history 
GitOrigin-RevId: 5b4c149782c086ebf9ef390195fb260ad0103217
  • Loading branch information
MediaPipe Team authored and chuoling committed Feb 27, 2021
1 parent 350fbb2 commit a92cff7
Show file tree
Hide file tree
Showing 2 changed files with 147 additions and 121 deletions.
126 changes: 5 additions & 121 deletions docs/solutions/pose.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,8 @@
layout: default
title: Pose
parent: Solutions
has_children: true
has_toc: false
nav_order: 5
---

Expand All @@ -21,10 +23,9 @@ nav_order: 5
## Overview

Human pose estimation from video plays a critical role in various applications
such as
[quantifying physical exercises](#pose-classification-and-repetition-counting),
sign language recognition, and full-body gesture control. For example, it can
form the basis for yoga, dance, and fitness applications. It can also enable the
such as [quantifying physical exercises](./pose_classification.md), sign
language recognition, and full-body gesture control. For example, it can form
the basis for yoga, dance, and fitness applications. It can also enable the
overlay of digital content and information on top of the physical world in
augmented reality.

Expand Down Expand Up @@ -387,121 +388,6 @@ on how to build MediaPipe examples.
* Target:
[`mediapipe/examples/desktop/upper_body_pose_tracking:upper_body_pose_tracking_gpu`](https://github.com/google/mediapipe/tree/master/mediapipe/examples/desktop/upper_body_pose_tracking/BUILD)

## Pose Classification and Repetition Counting

One of the applications
[BlazePose](https://ai.googleblog.com/2020/08/on-device-real-time-body-pose-tracking.html)
can enable is fitness. More specifically - pose classification and repetition
counting. In this section we'll provide basic guidance on building a custom pose
classifier with the help of a
[Colab](https://drive.google.com/file/d/19txHpN8exWhstO6WVkfmYYVC6uug_oVR/view?usp=sharing)
and wrap it in a simple
[fitness app](https://mediapipe.page.link/mlkit-pose-classification-demo-app)
powered by [ML Kit](https://developers.google.com/ml-kit). Push-ups and squats
are used for demonstration purposes as the most common exercises.

![pose_classification_pushups_and_squats.gif](../images/mobile/pose_classification_pushups_and_squats.gif) |
:--------------------------------------------------------------------------------------------------------: |
*Fig 4. Pose classification and repetition counting with MediaPipe Pose.* |

We picked the
[k-nearest neighbors algorithm](https://en.wikipedia.org/wiki/K-nearest_neighbors_algorithm)
(k-NN) as the classifier. It's simple and easy to start with. The algorithm
determines the object's class based on the closest samples in the training set.
To build it, one needs to:

* Collect image samples of the target exercises and run pose prediction on
them,
* Convert obtained pose landmarks to a representation suitable for the k-NN
classifier and form a training set,
* Perform the classification itself followed by repetition counting.

### Training Set

To build a good classifier appropriate samples should be collected for the
training set: about a few hundred samples for each terminal state of each
exercise (e.g., "up" and "down" positions for push-ups). It's important that
collected samples cover different camera angles, environment conditions, body
shapes, and exercise variations.

![pose_classification_pushups_un_and_down_samples.jpg](../images/mobile/pose_classification_pushups_un_and_down_samples.jpg) |
:--------------------------------------------------------------------------------------------------------------------------: |
*Fig 5. Two terminal states of push-ups.* |

To transform samples into a k-NN classifier training set, either
[basic](https://drive.google.com/file/d/1z4IM8kG6ipHN6keadjD-F6vMiIIgViKK/view?usp=sharing)
or
[extended](https://drive.google.com/file/d/19txHpN8exWhstO6WVkfmYYVC6uug_oVR/view?usp=sharing)
Colab could be used. They both use the
[Python Solution API](#python-solution-api) to run the BlazePose models on given
images and dump predicted pose landmarks to a CSV file. Additionally, the
extended Colab provides useful tools to find outliers (e.g., wrongly predicted
poses) and underrepresented classes (e.g., not covering all camera angles) by
classifying each sample against the entire training set. After that, you'll be
able to test the classifier on an arbitrary video right in the Colab.

### Classification

Code of the classifier is available both in the
[extended](https://drive.google.com/file/d/19txHpN8exWhstO6WVkfmYYVC6uug_oVR/view?usp=sharing)
Colab and in the
[ML Kit demo app](https://mediapipe.page.link/mlkit-pose-classification-demo-app).
Please refer to them for details of the approach described below.

The k-NN algorithm used for pose classification requires a feature vector
representation of each sample and a metric to compute the distance between two
such vectors to find the nearest pose samples to a target one.

To convert pose landmarks to a feature vector, we use pairwise distances between
predefined lists of pose joints, such as distances between wrist and shoulder,
ankle and hip, and two wrists. Since the algorithm relies on distances, all
poses are normalized to have the same torso size and vertical torso orientation
before the conversion.

![pose_classification_pairwise_distances.png](../images/mobile/pose_classification_pairwise_distances.png) |
:--------------------------------------------------------------------------------------------------------: |
*Fig 6. Main pairwise distances used for the pose feature vector.* |

To get a better classification result, k-NN search is invoked twice with
different distance metrics:

* First, to filter out samples that are almost the same as the target one but
have only a few different values in the feature vector (which means
differently bent joints and thus other pose class), minimum per-coordinate
distance is used as distance metric,
* Then average per-coordinate distance is used to find the nearest pose
cluster among those from the first search.

Finally, we apply
[exponential moving average](https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average)
(EMA) smoothing to level any noise from pose prediction or classification. To do
that, we search not only for the nearest pose cluster, but we calculate a
probability for each of them and use it for smoothing over time.

### Repetition Counter

To count the repetitions, the algorithm monitors the probability of a target
pose class. Let's take push-ups with its "up" and "down" terminal states:

* When the probability of the "down" pose class passes a certain threshold for
the first time, the algorithm marks that the "down" pose class is entered.
* Once the probability drops below the threshold, the algorithm marks that the
"down" pose class has been exited and increases the counter.

To avoid cases when the probability fluctuates around the threshold (e.g., when
the user pauses between "up" and "down" states) causing phantom counts, the
threshold used to detect when the state is exited is actually slightly lower
than the one used to detect when the state is entered. It creates an interval
where the pose class and the counter can't be changed.

### Future Work

We are actively working on improving BlazePose GHUM 3D's Z prediction. It will
allow us to use joint angles in the feature vectors, which are more natural and
easier to configure (although distances can still be useful to detect touches
between body parts) and to perform rotation normalization of poses and reduce
the number of camera angles required for accurate k-NN classification.

## Resources

* Google AI Blog:
Expand All @@ -512,5 +398,3 @@ the number of camera angles required for accurate k-NN classification.
* [Models and model cards](./models.md#pose)
* [Web demo](https://code.mediapipe.dev/codepen/pose)
* [Python Colab](https://mediapipe.page.link/pose_py_colab)
* [Pose Classification Colab (Basic)](https://mediapipe.page.link/pose_classification_basic)
* [Pose Classification Colab (Extended)](https://mediapipe.page.link/pose_classification_extended)
142 changes: 142 additions & 0 deletions docs/solutions/pose_classification.md
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,142 @@
---
layout: default
title: Pose Classification
parent: Pose
grand_parent: Solutions
nav_order: 1
---

# Pose Classification
{: .no_toc }

<details close markdown="block">
<summary>
Table of contents
</summary>
{: .text-delta }
1. TOC
{:toc}
</details>
---

## Overview

One of the applications
[BlazePose](https://ai.googleblog.com/2020/08/on-device-real-time-body-pose-tracking.html)
can enable is fitness. More specifically - pose classification and repetition
counting. In this section we'll provide basic guidance on building a custom pose
classifier with the help of [Colabs](#colabs) and wrap it in a simple
[fitness app](https://mediapipe.page.link/mlkit-pose-classification-demo-app)
powered by [ML Kit](https://developers.google.com/ml-kit). Push-ups and squats
are used for demonstration purposes as the most common exercises.

![pose_classification_pushups_and_squats.gif](../images/mobile/pose_classification_pushups_and_squats.gif) |
:--------------------------------------------------------------------------------------------------------: |
*Fig 1. Pose classification and repetition counting with MediaPipe Pose.* |

We picked the
[k-nearest neighbors algorithm](https://en.wikipedia.org/wiki/K-nearest_neighbors_algorithm)
(k-NN) as the classifier. It's simple and easy to start with. The algorithm
determines the object's class based on the closest samples in the training set.

**To build it, one needs to:**

1. Collect image samples of the target exercises and run pose prediction on
them,
2. Convert obtained pose landmarks to a representation suitable for the k-NN
classifier and form a training set using these [Colabs](#colabs),
3. Perform the classification itself followed by repetition counting (e.g., in
the
[ML Kit demo app](https://mediapipe.page.link/mlkit-pose-classification-demo-app)).

## Training Set

To build a good classifier appropriate samples should be collected for the
training set: about a few hundred samples for each terminal state of each
exercise (e.g., "up" and "down" positions for push-ups). It's important that
collected samples cover different camera angles, environment conditions, body
shapes, and exercise variations.

![pose_classification_pushups_un_and_down_samples.jpg](../images/mobile/pose_classification_pushups_un_and_down_samples.jpg) |
:--------------------------------------------------------------------------------------------------------------------------: |
*Fig 2. Two terminal states of push-ups.* |

To transform samples into a k-NN classifier training set, both
[`Pose Classification Colab (Basic)`] and
[`Pose Classification Colab (Extended)`] could be used. They use the
[Python Solution API](./pose.md#python-solution-api) to run the BlazePose models
on given images and dump predicted pose landmarks to a CSV file. Additionally,
the [`Pose Classification Colab (Extended)`] provides useful tools to find
outliers (e.g., wrongly predicted poses) and underrepresented classes (e.g., not
covering all camera angles) by classifying each sample against the entire
training set. After that, you'll be able to test the classifier on an arbitrary
video right in the Colab.

## Classification

Code of the classifier is available both in the
[`Pose Classification Colab (Extended)`] and in the
[ML Kit demo app](https://mediapipe.page.link/mlkit-pose-classification-demo-app).
Please refer to them for details of the approach described below.

The k-NN algorithm used for pose classification requires a feature vector
representation of each sample and a metric to compute the distance between two
such vectors to find the nearest pose samples to a target one.

To convert pose landmarks to a feature vector, we use pairwise distances between
predefined lists of pose joints, such as distances between wrist and shoulder,
ankle and hip, and two wrists. Since the algorithm relies on distances, all
poses are normalized to have the same torso size and vertical torso orientation
before the conversion.

![pose_classification_pairwise_distances.png](../images/mobile/pose_classification_pairwise_distances.png) |
:--------------------------------------------------------------------------------------------------------: |
*Fig 3. Main pairwise distances used for the pose feature vector.* |

To get a better classification result, k-NN search is invoked twice with
different distance metrics:

* First, to filter out samples that are almost the same as the target one but
have only a few different values in the feature vector (which means
differently bent joints and thus other pose class), minimum per-coordinate
distance is used as distance metric,
* Then average per-coordinate distance is used to find the nearest pose
cluster among those from the first search.

Finally, we apply
[exponential moving average](https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average)
(EMA) smoothing to level any noise from pose prediction or classification. To do
that, we search not only for the nearest pose cluster, but we calculate a
probability for each of them and use it for smoothing over time.

## Repetition Counting

To count the repetitions, the algorithm monitors the probability of a target
pose class. Let's take push-ups with its "up" and "down" terminal states:

* When the probability of the "down" pose class passes a certain threshold for
the first time, the algorithm marks that the "down" pose class is entered.
* Once the probability drops below the threshold, the algorithm marks that the
"down" pose class has been exited and increases the counter.

To avoid cases when the probability fluctuates around the threshold (e.g., when
the user pauses between "up" and "down" states) causing phantom counts, the
threshold used to detect when the state is exited is actually slightly lower
than the one used to detect when the state is entered. It creates an interval
where the pose class and the counter can't be changed.

## Future Work

We are actively working on improving BlazePose GHUM 3D's Z prediction. It will
allow us to use joint angles in the feature vectors, which are more natural and
easier to configure (although distances can still be useful to detect touches
between body parts) and to perform rotation normalization of poses and reduce
the number of camera angles required for accurate k-NN classification.

## Colabs

* [`Pose Classification Colab (Basic)`]
* [`Pose Classification Colab (Extended)`]

[`Pose Classification Colab (Basic)`]: https://mediapipe.page.link/pose_classification_basic
[`Pose Classification Colab (Extended)`]: https://mediapipe.page.link/pose_classification_extended

0 comments on commit a92cff7

Please sign in to comment.