Hand landmarks detection guide for iOS

The MediaPipe Hand Landmarker task lets you detect the landmarks of the hands in an image. These instructions show you how to use the Hand Landmarker with iOS apps. The code sample described in these instructions is available on GitHub.

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 basic implementation of a Hand Landmarker app for iOS. The example uses the camera on a physical iOS device to detect hand landmarks in a continuous video stream. The app can also detect hand landmarks in images and videos from the device gallery.

You can use the app as a starting point for your own iOS app, or refer to it when modifying an existing app. The Hand Landmarker 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 Hand Landmarker example app:

    cd mediapipe
    git sparse-checkout init --cone
    git sparse-checkout set examples/hand_landmarker/ios/
    

After creating a local version of the example code, you can install the MediaPipe task library, open the project using Xcode and run the app. For instructions, see the Setup Guide for iOS.

Key components

The following files contain the crucial code for the Hand Landmarker example application:

Setup

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

Dependencies

Hand Landmarker uses the MediaPipeTasksVision library, which must be installed using CocoaPods. The library is compatible with both Swift and Objective-C apps and does not require any additional language-specific setup.

For instructions to install CocoaPods on macOS, refer to the CocoaPods installation guide. For instructions on how to create a Podfile with the necessary pods for your app, refer to Using CocoaPods.

Add the MediaPipeTasksVision pod in the Podfile using the following code:

target 'MyHandLandmarkerApp' do
  use_frameworks!
  pod 'MediaPipeTasksVision'
end

If your app includes unit test targets, refer to the Set Up Guide for iOS for additional information on setting up your Podfile.

Model

The MediaPipe Hand Landmarker task requires a trained model that is compatible with this task. For more information about the available trained models for Hand Landmarker, see the task overview Models section.

Select and download a model, and add it to your project directory using Xcode. For instructions on how to add files to your Xcode project, refer to Managing files and folders in your Xcode project.

Use the BaseOptions.modelAssetPath property to specify the path to the model in your app bundle. For a code example, see the next section.

Create the task

You can create the Hand Landmarker task by calling one of its initializers. The HandLandmarker(options:) initializer accepts values for the configuration options.

If you don't need a Hand Landmarker initialized with customized configuration options, you can use the HandLandmarker(modelPath:) initializer to create an Hand Landmarker with the default options. For more information about configuration options, see Configuration Overview.

The Hand Landmarker task supports 3 input data types: still images, video files and live video streams. By default, HandLandmarker(modelPath:) initializes a task for still images. If you want your task to be initialized to process video files or live video streams, use HandLandmarker(options:) to specify the video or livestream running mode. The livestream mode also requires the additional handLandmarkerLiveStreamDelegate configuration option, which enables the Hand Landmarker to deliver hand landmarker results to the delegate asynchronously.

Choose the tab corresponding to your running mode to see how to create the task and run inference.

Swift

Image

import MediaPipeTasksVision

let modelPath = Bundle.main.path(forResource: "hand_landmarker",
                                      ofType: "task")

let options = HandLandmarkerOptions()
options.baseOptions.modelAssetPath = modelPath
options.runningMode = .image
options.minHandDetectionConfidence = minHandDetectionConfidence
options.minHandPresenceConfidence = minHandPresenceConfidence
options.minTrackingConfidence = minHandTrackingConfidence
options.numHands = numHands

let handLandmarker = try HandLandmarker(options: options)
    

Video

import MediaPipeTasksVision

let modelPath = Bundle.main.path(forResource: "hand_landmarker",
                                      ofType: "task")

let options = HandLandmarkerOptions()
options.baseOptions.modelAssetPath = modelPath
options.runningMode = .video
options.minHandDetectionConfidence = minHandDetectionConfidence
options.minHandPresenceConfidence = minHandPresenceConfidence
options.minTrackingConfidence = minHandTrackingConfidence
options.numHands = numHands

let handLandmarker = try HandLandmarker(options: options)
    

Livestream

import MediaPipeTasksVision

// Class that conforms to the `HandLandmarkerLiveStreamDelegate` protocol and
// implements the method that the hand landmarker calls once it finishes
// performing landmarks detection in each input frame.
class HandLandmarkerResultProcessor: NSObject, HandLandmarkerLiveStreamDelegate {

  func handLandmarker(
    _ handLandmarker: HandLandmarker,
    didFinishDetection result: HandLandmarkerResult?,
    timestampInMilliseconds: Int,
    error: Error?) {

    // Process the hand landmarker result or errors here.

  }
}

let modelPath = Bundle.main.path(
  forResource: "hand_landmarker",
  ofType: "task")

let options = HandLandmarkerOptions()
options.baseOptions.modelAssetPath = modelPath
options.runningMode = .liveStream
options.minHandDetectionConfidence = minHandDetectionConfidence
options.minHandPresenceConfidence = minHandPresenceConfidence
options.minTrackingConfidence = minHandTrackingConfidence
options.numHands = numHands

// Assign an object of the class to the `handLandmarkerLiveStreamDelegate`
// property.
let processor = HandLandmarkerResultProcessor()
options.handLandmarkerLiveStreamDelegate = processor

let handLandmarker = try HandLandmarker(options: options)
    

Objective-C

Image

@import MediaPipeTasksVision;

NSString *modelPath = [[NSBundle mainBundle] pathForResource:@"hand_landmarker"
                                                      ofType:@"task"];

MPPHandLandmarkerOptions *options = [[MPPHandLandmarkerOptions alloc] init];
options.baseOptions.modelAssetPath = modelPath;
options.runningMode = MPPRunningModeImage;
options.minHandDetectionConfidence = minHandDetectionConfidence;
options.minHandPresenceConfidence = minHandPresenceConfidence;
options.minTrackingConfidence = minHandTrackingConfidence;
options.numHands = numHands;

MPPHandLandmarker *handLandmarker =
  [[MPPHandLandmarker alloc] initWithOptions:options error:nil];
    

Video

@import MediaPipeTasksVision;

NSString *modelPath = [[NSBundle mainBundle] pathForResource:@"hand_landmarker"
                                                      ofType:@"task"];

MPPHandLandmarkerOptions *options = [[MPPHandLandmarkerOptions alloc] init];
options.baseOptions.modelAssetPath = modelPath;
options.runningMode = MPPRunningModeVideo;
options.minHandDetectionConfidence = minHandDetectionConfidence;
options.minHandPresenceConfidence = minHandPresenceConfidence;
options.minTrackingConfidence = minHandTrackingConfidence;
options.numHands = numHands;

MPPHandLandmarker *handLandmarker =
  [[MPPHandLandmarker alloc] initWithOptions:options error:nil];
    

Livestream

@import MediaPipeTasksVision;

// Class that conforms to the `MPPHandLandmarkerLiveStreamDelegate` protocol
// and implements the method that the hand landmarker calls once it finishes
// performing landmarks detection in each input frame.

@interface APPHandLandmarkerResultProcessor : NSObject 

@end

@implementation APPHandLandmarkerResultProcessor

-   (void)handLandmarker:(MPPHandLandmarker *)handLandmarker
    didFinishDetectionWithResult:(MPPHandLandmarkerResult *)handLandmarkerResult
         timestampInMilliseconds:(NSInteger)timestampInMilliseconds
                           error:(NSError *)error {

    // Process the hand landmarker result or errors here.

}

@end

NSString *modelPath = [[NSBundle mainBundle] pathForResource:@"hand_landmarker"
                                                      ofType:@"task"];

MPPHandLandmarkerOptions *options = [[MPPHandLandmarkerOptions alloc] init];
options.baseOptions.modelAssetPath = modelPath;
options.runningMode = MPPRunningModeLiveStream;
options.minHandDetectionConfidence = minHandDetectionConfidence;
options.minHandPresenceConfidence = minHandPresenceConfidence;
options.minTrackingConfidence = minHandTrackingConfidence;
options.numHands = numHands;

// Assign an object of the class to the `handLandmarkerLiveStreamDelegate`
// property.
APPHandLandmarkerResultProcessor *processor =
  [APPHandLandmarkerResultProcessor new];
options.handLandmarkerLiveStreamDelegate = processor;

MPPHandLandmarker *handLandmarker =
  [[MPPHandLandmarker alloc] initWithOptions:options error:nil];
    

Configuration options

This task has the following configuration options for iOS apps:

Option Name Description Value Range Default Value
running_mode Sets the running mode for the task. There are three modes:

IMAGE: The mode for single image inputs.

VIDEO: The mode for decoded frames of a video.

LIVE_STREAM: The mode for a livestream of input data, such as from a camera. In this mode, resultListener must be called to set up a listener to receive results asynchronously. In this mode, handLandmarkerLiveStreamDelegate must be set to an instance of a class that implements the HandLandmarkerLiveStreamDelegate to receive the hand landmark detection results asynchronously.
{RunningMode.image, RunningMode.video, RunningMode.liveStream} RunningMode.image
numHands The maximum number of hands detected by the Hand landmark detector. Any integer > 0 1
minHandDetectionConfidence The minimum confidence score for the hand detection to be considered successful in palm detection model. 0.0 - 1.0 0.5
minHandPresenceConfidence The minimum confidence score for the hand presence score in the hand landmark detection model. In Video mode and Live stream mode, if the hand presence confidence score from the hand landmark model is below this threshold, Hand Landmarker triggers the palm detection model. Otherwise, a lightweight hand tracking algorithm determines the location of the hand(s) for subsequent landmark detections. 0.0 - 1.0 0.5
minTrackingConfidence The minimum confidence score for the hand tracking to be considered successful. This is the bounding box IoU threshold between hands in the current frame and the last frame. In Video mode and Stream mode of Hand Landmarker, if the tracking fails, Hand Landmarker triggers hand detection. Otherwise, it skips the hand detection. 0.0 - 1.0 0.5
result_listener Sets the result listener to receive the detection results asynchronously when the hand landmarker is in live stream mode. Only applicable when running mode is set to LIVE_STREAM N/A N/A

When the running mode is set to livestream, the Hand Landmarker requires the additional handLandmarkerLiveStreamDelegate configuration option, which enables the Hand Landmarker to deliver hand landmarks detection results asynchronously. The delegate must implement the handLandmarker(_:didFinishDetection:timestampInMilliseconds:error:) method, which the Hand Landmarker calls after processing the hand landmarks detection results for each frame.

Option name Description Value Range Default Value
handLandmarkerLiveStreamDelegate Enables Hand Landmarker to receive the hand landmark detection results asynchronously in livestream mode. The class whose instance is set to this property must implement the handLandmarker(_:didFinishDetection:timestampInMilliseconds:error:) method. Not applicable Not set

Prepare data

You need to convert the input image or frame to an MPImage object before passing it to the Hand Landmarker. MPImage supports different types of iOS image formats, and can use them in any running mode for inference. For more information about MPImage, refer to the MPImage API

Choose an iOS image format based on your use case and the running mode your application requires.MPImage accepts the UIImage, CVPixelBuffer, and CMSampleBuffer iOS image formats.

UIImage

The UIImage format is well-suited for the following running modes:

  • Images: images from an app bundle, user gallery, or file system formatted as UIImage images can be converted to an MPImage object.

  • Videos: use AVAssetImageGenerator to extract video frames to the CGImage format, then convert them to UIImage images.

Swift

// Load an image on the user's device as an iOS `UIImage` object.

// Convert the `UIImage` object to a MediaPipe's Image object having the default
// orientation `UIImage.Orientation.up`.
let image = try MPImage(uiImage: image)
    

Objective-C

// Load an image on the user's device as an iOS `UIImage` object.

// Convert the `UIImage` object to a MediaPipe's Image object having the default
// orientation `UIImageOrientationUp`.
MPImage *image = [[MPPImage alloc] initWithUIImage:image error:nil];
    

The example initializes an MPImage with the default UIImage.Orientation.Up orientation. You can initialize an MPImage with any of the supported UIImage.Orientation values. Hand Landmarker does not support mirrored orientations like .upMirrored, .downMirrored, .leftMirrored, .rightMirrored.

For more information about UIImage, refer to the UIImage Apple Developer Documentation.

CVPixelBuffer

The CVPixelBuffer format is well-suited for applications that generate frames and use the iOS CoreImage framework for processing.

The CVPixelBuffer format is well-suited for the following running modes:

  • Images: apps that generate CVPixelBuffer images after some processing using iOS's CoreImage framework can be sent to the Hand Landmarker in the image running mode.

  • Videos: video frames can be converted to the CVPixelBuffer format for processing, and then sent to the Hand Landmarker in video mode.

  • livestream: apps using an iOS camera to generate frames may be converted into the CVPixelBuffer format for processing before being sent to the Hand Landmarker in livestream mode.

Swift

// Obtain a CVPixelBuffer.

// Convert the `CVPixelBuffer` object to a MediaPipe's Image object having the default
// orientation `UIImage.Orientation.up`.
let image = try MPImage(pixelBuffer: pixelBuffer)
    

Objective-C

// Obtain a CVPixelBuffer.

// Convert the `CVPixelBuffer` object to a MediaPipe's Image object having the
// default orientation `UIImageOrientationUp`.
MPImage *image = [[MPPImage alloc] initWithUIImage:image error:nil];
    

For more information about CVPixelBuffer, refer to the CVPixelBuffer Apple Developer Documentation.

CMSampleBuffer

The CMSampleBuffer format stores media samples of a uniform media type, and is well-suited for the livestream running mode. Live frames from iOS cameras are asynchronously delivered in the CMSampleBuffer format by iOS AVCaptureVideoDataOutput.

Swift

// Obtain a CMSampleBuffer.

// Convert the `CMSampleBuffer` object to a MediaPipe's Image object having the default
// orientation `UIImage.Orientation.up`.
let image = try MPImage(sampleBuffer: sampleBuffer)
    

Objective-C

// Obtain a `CMSampleBuffer`.

// Convert the `CMSampleBuffer` object to a MediaPipe's Image object having the
// default orientation `UIImageOrientationUp`.
MPImage *image = [[MPPImage alloc] initWithSampleBuffer:sampleBuffer error:nil];
    

For more information about CMSampleBuffer, refer to the CMSampleBuffer Apple Developer Documentation.

Run the task

To run the Hand Landmarker, use the detect() method specific to the assigned running mode:

  • Still image: detect(image:)
  • Video: detect(videoFrame:timestampInMilliseconds:)
  • Livestream: detectAsync(image:timestampInMilliseconds:)

Swift

Image

let result = try handLandmarker.detect(image: image)
    

Video

let result = try handLandmarker.detect(
    videoFrame: image,
    timestampInMilliseconds: timestamp)
    

Livestream

try handLandmarker.detectAsync(
  image: image,
  timestampInMilliseconds: timestamp)
    

Objective-C

Image

MPPHandLandmarkerResult *result =
  [handLandmarker detectInImage:image error:nil];
    

Video

MPPHandLandmarkerResult *result =
  [handLandmarker detectInVideoFrame:image
             timestampInMilliseconds:timestamp
                               error:nil];
    

Livestream

BOOL success =
  [handLandmarker detectAsyncInImage:image
             timestampInMilliseconds:timestamp
                               error:nil];
    

The Hand Landmarker code example shows the implementations of each of these modes in more detail. The example code allows the user to switch between processing modes, which may not be required for your use case.

Note the following:

  • When running in video mode or livestream mode, you must also provide the timestamp of the input frame to the Hand Landmarker task.

  • When running in image or video mode, the Hand Landmarker task blocks the current thread until it finishes processing the input image or frame. To avoid blocking the current thread, execute the processing in a background thread using iOS Dispatch or NSOperation frameworks.

  • When running in livestream mode, the Hand Landmarker task returns immediately and doesn't block the current thread. It invokes the handLandmarker(_:didFinishDetection:timestampInMilliseconds:error:) method with the hand landmarker result after processing each input frame. The Hand Landmarker invokes this method asynchronously on a dedicated serial dispatch queue. For displaying results on the user interface, dispatch the results to the main queue after processing the results. If the detectAsync function is called when the Hand Landmarker task is busy processing another frame, the Hand Landmarker ignores the new input frame.

Handle and display results

Upon running inference, the Hand Landmarker task returns a HandLandmarkerResult which contains hand landmarks in image coordinates, hand landmarks in world coordinates and handedness(left/right hand) of the detected hands.

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

The HandLandmarkerResult output contains three components. Each component is an array, where each element contains the following results for a single detected hand:

  • Handedness

    Handedness represents whether the detected hands are left or right 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.

HandLandmarkerResult:
  Handedness:
    Categories #0:
      index        : 0
      score        : 0.98396
      categoryName : Left
  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 image shows a visualization of the task output: