You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
CameraView/docs/_docs/frame-processing.md

8.0 KiB

layout title description order disqus
page Frame Processing Process each frame in real time 6 1

We support frame processors that will receive data from the camera preview stream. This is a useful feature with a wide range of applications. For example, the frames can be sent to a face detector, a QR code detector, the Firebase Machine Learning Kit, or any other frame consumer.

cameraView.addFrameProcessor(new FrameProcessor() {
    @Override
    @WorkerThread
    public void process(@NonNull Frame frame) {
        long time = frame.getTime();
        Size size = frame.getSize();
        int format = frame.getFormat();
        int userRotation = frame.getRotationToUser();
        int viewRotation = frame.getRotationToView();
        if (frame.getDataClass() == byte[].class) {
            byte[] data = frame.getData();
            // Process byte array...
        } else if (frame.getDataClass() == Image.class) {
            Image data = frame.getData();
            // Process android.media.Image...
        }
    }
}

For your convenience, the FrameProcessor method is run in a background thread so you can do your job in a synchronous fashion. Once the process method returns, internally we will re-use the Frame instance and apply new data to it. So:

  • you can do your job synchronously in the process() method. This is recommended.
  • if you must hold the Frame instance longer, use frame = frame.freeze() to get a frozen instance that will not be affected. This is discouraged because it requires copying the whole array. Also, starting from v2.5.0, this is not allowed when Camera2 is used.

Process synchronously

Processing synchronously, for the duration of the process() method, is the recommended way of using processors, because it solves different issues:

  • avoids the need of calling frame = frame.freeze() which is a very expensive operation
  • the engine will automatically drop frames if the process() method is busy, so you'll only receive frames that you can handle
  • we have already allocated background threads for you, so there's no need to create another

Some frame consumers might have a built-in asynchronous behavior. But you can still block the process() thread until the consumer has returned.

@Override
@WorkerThread
public void process(@NonNull Frame frame) {
    
    // EXAMPLE 1:
    // Firebase and Google APIs will often return a Task.
    // You can use Tasks.await() to complete the task on the current thread.
    // Read: https://developers.google.com/android/guides/tasks#blocking
    try {
        result = Tasks.await(firebaseDetector.detectInImage(firebaseImage));
    catch (Exception e) {
        // Firebase task failed.
    }
    
    
    // EXAMPLE 2:
    // For other async consumers, you can use, for example, a CountDownLatch.
    
    // Step 1: create the latch.
    final CountDownLatch latch = new CountDownLatch(1);
    
    // Step 2: launch async processing here...
    // When processing completes or fails, call latch.countDown();
    
    // Step 3: after launching, block the current thread.
    latch.await();
}

Frame Data

Starting from v2.5.0, the type of data offered by frame.getData() depends on the camera engine that created this frame:

  • The Camera1 engine will offer byte[] arrays
  • The Camera2 engine will offer android.media.Image objects

You can check this at runtime by inspecting the data class using frame.getDataClass().

Frame Size

The Camera2 engine offers the option to set size constraints for the incoming frames.

cameraView.setFrameProcessingMaxWidth(maxWidth);
cameraView.setFrameProcessingMaxHeight(maxWidth);

With other engines, these API have no effect.

Frame Format

The Camera2 engine offers the option to set the frame format as one of the ImageFormat constants. The default is ImageFormat.YUV_420_888.

cameraView.setFrameProcessingFormat(ImageFormat.YUV_420_888);
cameraView.setFrameProcessingFormat(ImageFormat.YUV_422_888);

With the Camera1 engine, the incoming format will always be ImageFormat.NV21. You can check which formats are available for use through CameraOptions.getSupportedFrameProcessingFormats().

Advanced: Thread Control

Starting from v2.5.1, you can control the number of background threads that are allocated for frame processing work. This should further push you into perform processing actions synchronously and can be useful if processing is very slow with respect to the preview frame rate, in order to avoid dropping too many frames.

You can change the number of threads by calling setFrameProcessingExecutors(). Whenever you do, we recommend that you also change the frame processing pool size to a compatible value. The frame processing pool size is roughly the number of Frame instances that can exist at any given moment. We recommend that this value is set to the number of executors plus 1. For example:

  • Single threaded (default):
cameraView.setFrameProcessingExecutors(1);
cameraView.setFrameProcessingPoolSize(2);
  • Two threads:
cameraView.setFrameProcessingExecutors(2);
cameraView.setFrameProcessingPoolSize(3);

XML Attributes

<com.otaliastudios.cameraview.CameraView
    app:cameraFrameProcessingMaxWidth="640"
    app:cameraFrameProcessingMaxHeight="640"
    app:cameraFrameProcessingFormat="0x23"
    app:cameraFrameProcessingPoolSize="2"
    app:cameraFrameProcessingExecutors="1"/>
Frame API Type Description
camera.addFrameProcessor(FrameProcessor) - Register a FrameProcessor.
camera.removeFrameProcessor(FrameProcessor) - Removes a FrameProcessor.
camera.clearFrameProcessors() - Removes all FrameProcessors.
camera.setFrameProcessingMaxWidth(int) - Sets the max width for incoming frames.
camera.setFrameProcessingMaxHeight(int) - Sets the max height for incoming frames.
camera.getFrameProcessingMaxWidth() int Returns the max width for incoming frames.
camera.getFrameProcessingMaxHeight() int Returns the max height for incoming frames.
camera.setFrameProcessingFormat(int) - Sets the desired format for incoming frames. Should be one of the ImageFormat constants.
camera.getFrameProcessingFormat() - Returns the format for incoming frames. One of the ImageFormat constants.
camera.setFrameProcessingPoolSize(int) - Sets the frame pool size, roughly the number of Frames that can exist at any given moment. Defaults to 2, which fits all use cases unless you change the executors.
camera.getFrameProcessingPoolSize() - Returns the frame pool size.
camera.setFrameProcessingExecutors(int) - Sets the processing thread size. Defaults to 1, but can be increased if your processing is slow and you are dropping too many frames. This should always be tuned together with the frame pool size.
camera.getFrameProcessingExecutors() - Returns the processing thread size.
frame.getDataClass() Class<T> The class of the data returned by getData(). Either byte[] or android.media.Image.
frame.getData() T The current preview frame, in its original orientation.
frame.getTime() long The preview timestamp, in System.currentTimeMillis() reference.
frame.getRotationToUser() int The rotation that should be applied to the byte array in order to see what the user sees. Can be useful in the processing phase.
frame.getRotationToView() int The rotation that should be applied to the byte array in order to match the View / Activity orientation. Can be useful in the drawing / rendering phase.
frame.getSize() Size The frame size, before any rotation is applied, to access data.
frame.getFormat() int The frame ImageFormat. Defaults to ImageFormat.NV21 for Camera1 and ImageFormat.YUV_420_888 for Camera2.
frame.freeze() Frame Clones this frame and makes it immutable. Can be expensive because requires copying the byte array.
frame.release() - Disposes the content of this frame. Should be used on frozen frames to release memory.