Gesture recognition guide for Android

The MediaPipe Gesture Recognizer task lets you recognize hand gestures in real time, and provides the recognized hand gesture results and hand landmarks of the detected hands. These instructions show you how to use the Gesture Recognizer with Android apps. The code sample described in these instructions is available on GitHub.

You can see this task in action by viewing the web demo. For more information about the capabilities, models, and configuration options of this task, see the Overview.

Code example

The MediaPipe Tasks example code is a simple implementation of a Gesture Recognizer app for Android. The example uses the camera on a physical Android device to continuously detect hand gestures, and can also use images and videos from the device gallery to statically detect gestures.

You can use the app as a starting point for your own Android app, or refer to it when modifying an existing app. The Gesture Recognizer example code is hosted on GitHub.

Download the code

The following instructions show you how to create a local copy of the example code using the git command line tool.

To download the example code:

1. Clone the git repository using the following command:

   git clone https://github.com/google-ai-edge/mediapipe-samples
   

2. Optionally, configure your git instance to use sparse checkout, so you have only the files for the Gesture Recognizer example app:

   cd mediapipe
   git sparse-checkout init --cone
   git sparse-checkout set examples/gesture_recognizer/android
   

After creating a local version of the example code, you can import the project into Android Studio and run the app. For instructions, see the Setup Guide for Android.

Key components

The following files contain the crucial code for this hand gesture recognition example application:

• GestureRecognizerHelper.kt - Initializes the gesture recognizer and handles the model and delegate selection.
• MainActivity.kt - Implements the application, including calling GestureRecognizerHelper and GestureRecognizerResultsAdapter.
• GestureRecognizerResultsAdapter.kt - Handles and formats the results.

Setup

This section describes key steps for setting up your development environment and code projects specifically to use Gesture Recognizer. For general information on setting up your development environment for using MediaPipe tasks, including platform version requirements, see the Setup guide for Android.

Dependencies

The Gesture Recognizer task uses the com.google.mediapipe:tasks-vision library. Add this dependency to the build.gradle file of your Android app:

dependencies {
    implementation 'com.google.mediapipe:tasks-vision:latest.release'
}

Model

The MediaPipe Gesture Recognizer task requires a trained model bundle that is compatible with this task. For more information on available trained models for Gesture Recognizer, see the task overview Models section.

Select and download the model, and store it within your project directory:

<dev-project-root>/src/main/assets

Specify the path of the model within the ModelAssetPath parameter. In the example code, the model is defined in the GestureRecognizerHelper.kt file:

baseOptionBuilder.setModelAssetPath(MP_RECOGNIZER_TASK)

Create the task

The MediaPipe Gesture Recognizer task uses the createFromOptions() function to set up the task. The createFromOptions() function accepts values for the configuration options. For more information on configuration options, see Configuration options.

The Gesture Recognizer supports 3 input data types: still images, video files, and live video streams. You need to specify the running mode corresponding to your input data type when creating the task. Choose the tab corresponding to your input data type to see how to create the task and run inference.

val baseOptionsBuilder = BaseOptions.builder().setModelAssetPath(MP_RECOGNIZER_TASK)
val baseOptions = baseOptionBuilder.build()

val optionsBuilder =
    GestureRecognizer.GestureRecognizerOptions.builder()
        .setBaseOptions(baseOptions)
        .setMinHandDetectionConfidence(minHandDetectionConfidence)
        .setMinTrackingConfidence(minHandTrackingConfidence)
        .setMinHandPresenceConfidence(minHandPresenceConfidence)
        .setRunningMode(RunningMode.IMAGE)

val options = optionsBuilder.build()
gestureRecognizer =
    GestureRecognizer.createFromOptions(context, options)
    

val baseOptionsBuilder = BaseOptions.builder().setModelAssetPath(MP_RECOGNIZER_TASK)
val baseOptions = baseOptionBuilder.build()

val optionsBuilder =
    GestureRecognizer.GestureRecognizerOptions.builder()
        .setBaseOptions(baseOptions)
        .setMinHandDetectionConfidence(minHandDetectionConfidence)
        .setMinTrackingConfidence(minHandTrackingConfidence)
        .setMinHandPresenceConfidence(minHandPresenceConfidence)
        .setRunningMode(RunningMode.VIDEO)

val options = optionsBuilder.build()
gestureRecognizer =
    GestureRecognizer.createFromOptions(context, options)
    

val baseOptionsBuilder = BaseOptions.builder().setModelAssetPath(MP_RECOGNIZER_TASK)
val baseOptions = baseOptionBuilder.build()

val optionsBuilder =
    GestureRecognizer.GestureRecognizerOptions.builder()
        .setBaseOptions(baseOptions)
        .setMinHandDetectionConfidence(minHandDetectionConfidence)
        .setMinTrackingConfidence(minHandTrackingConfidence)
        .setMinHandPresenceConfidence(minHandPresenceConfidence)
        .setResultListener(this::returnLivestreamResult)
        .setErrorListener(this::returnLivestreamError)
        .setRunningMode(RunningMode.LIVE_STREAM)

val options = optionsBuilder.build()
gestureRecognizer =
    GestureRecognizer.createFromOptions(context, options)
    

The Gesture Recognizer example code implementation allows the user to switch between processing modes. The approach makes the task creation code more complicated and may not be appropriate for your use case. You can see this code in the setupGestureRecognizer() function in the GestureRecognizerHelper.kt file.

Configuration options

This task has the following configuration options for Android apps:

Prepare data

Gesture Recognizer works with images, video file and live stream video. The task handles the data input preprocessing, including resizing, rotation and value normalization.

The following code demonstrates how to hand off data for processing. Theses samples include details on how to handle data from images, video files, and live video streams.

import com.google.mediapipe.framework.image.BitmapImageBuilder
import com.google.mediapipe.framework.image.MPImage

// Convert the input Bitmap object to an MPImage object to run inference
val mpImage = BitmapImageBuilder(image).build()
    

import com.google.mediapipe.framework.image.BitmapImageBuilder
import com.google.mediapipe.framework.image.MPImage

val argb8888Frame =
    if (frame.config == Bitmap.Config.ARGB_8888) frame
    else frame.copy(Bitmap.Config.ARGB_8888, false)

// Convert the input Bitmap object to an MPImage object to run inference
val mpImage = BitmapImageBuilder(argb8888Frame).build()
    

import com.google.mediapipe.framework.image.BitmapImageBuilder
import com.google.mediapipe.framework.image.MPImage

// Convert the input Bitmap object to an MPImage object to run inference
val mpImage = BitmapImageBuilder(rotatedBitmap).build()
    

In the Gesture Recognizer example code, the data preparation is handled in the GestureRecognizerHelper.kt file.

Run the task

The Gesture Recognizer uses the recognize, recognizeForVideo, and recognizeAsync functions to trigger inferences. For gesture recognition, this involves preprocessing input data, detecting hands in the image, detecting hand landmarks, and recognizing hand gesture from the landmarks.

The following code demonstrates how to execute the processing with the task model. These samples include details on how to handle data from images, video files, and live video streams.

val result = gestureRecognizer?.recognize(mpImage)
    

val timestampMs = i * inferenceIntervalMs

gestureRecognizer?.recognizeForVideo(mpImage, timestampMs)
    ?.let { recognizerResult ->
        resultList.add(recognizerResult)
    }
    

val mpImage = BitmapImageBuilder(rotatedBitmap).build()
val frameTime = SystemClock.uptimeMillis()

gestureRecognizer?.recognizeAsync(mpImage, frameTime)
    

Note the following:

• When running in the video mode or the live stream mode, you must also provide the timestamp of the input frame to the Gesture Recognizer task.
• When running in the image or the video mode, the Gesture Recognizer task will block the current thread until it finishes processing the input image or frame. To avoid blocking the user interface, execute the processing in a background thread.
• When running in the live stream mode, the Gesture Recognizer task doesn’t block the current thread but returns immediately. It will invoke its result listener with the recognition result every time it has finished processing an input frame. If the recognition function is called when the Gesture Recognizer task is busy processing another frame, the task will ignore the new input frame.

In the Gesture Recognizer example code, the recognize, recognizeForVideo, and recognizeAsync functions are defined in the GestureRecognizerHelper.kt file.

Handle and display results

The Gesture Recognizer generates a gesture detection result object for each recognition run. The result object contains hand landmarks in image coordinates, hand landmarks in world coordinates, handedness(left/right hand), and hand gestures categories of the detected hands.

The following shows an example of the output data from this task:

The resulted GestureRecognizerResult contains four components, and each component is an array, where each element contains the detected result of a single detected hand.

• Handedness

  Handedness represents whether the detected hands are left or right hands.

• Gestures

  The recognized gesture categories of the detected hands.

• Landmarks

  There are 21 hand landmarks, each composed of x, y and z coordinates. The x and y coordinates are normalized to [0.0, 1.0] by the image width and height, respectively. The z coordinate represents the landmark depth, with the depth at the wrist being the origin. The smaller the value, the closer the landmark is to the camera. The magnitude of z uses roughly the same scale as x.

• World Landmarks

  The 21 hand landmarks are also presented in world coordinates. Each landmark is composed of x, y, and z, representing real-world 3D coordinates in meters with the origin at the hand’s geometric center.

GestureRecognizerResult:
  Handedness:
    Categories #0:
      index        : 0
      score        : 0.98396
      categoryName : Left
  Gestures:
    Categories #0:
      score        : 0.76893
      categoryName : Thumb_Up
  Landmarks:
    Landmark #0:
      x            : 0.638852
      y            : 0.671197
      z            : -3.41E-7
    Landmark #1:
      x            : 0.634599
      y            : 0.536441
      z            : -0.06984
    ... (21 landmarks for a hand)
  WorldLandmarks:
    Landmark #0:
      x            : 0.067485
      y            : 0.031084
      z            : 0.055223
    Landmark #1:
      x            : 0.063209
      y            : -0.00382
      z            : 0.020920
    ... (21 world landmarks for a hand)

The following images shows a visualization of the task output:

In the Gesture Recognizer example code, the GestureRecognizerResultsAdapter class in the GestureRecognizerResultsAdapter.kt file handles the results.