使用JavaCV进行并发视频帧处理

时间:2017-07-19 14:34:01

标签: java multithreading javacv

我有一个使用JavaCV逐帧过滤视频的代码。代码如下所示

try (FFmpegFrameGrabber grabber = createGrabber()) {
    grabber.start();
    try (FFmpegFrameRecorder recorder = createRecorder(grabber)) {
        recorder.start();
        Frame frame;
        while ((frame = grabber.grab()) != null) {
            Frame editedFrame = filterFrame(frame); //This takes a long time.
            recorder.setTimestamp(grabber.getTimestamp());
            recorder.record(editedFrame);
        }
    }
}

由于行Frame editedFrame = filterFrame(frame);需要相当长的时间,是否可以利用多线程,因此整个过程可以更快?我正在考虑使用ExecutorServiceLinkedBlockingQueue之类的东西一次处理几个帧,然后根据时间戳记录回来。

1 个答案:

答案 0 :(得分:0)

这是pattern producer

Producer类用于执行长任务(filterFrame):

import java.util.concurrent.Callable;

public class Producer implements Callable<Frame> {

    private final Frame frame;

    public Producer(Frame frame) {
        this.frame = frame;
    }

    @Override
    public Frame call() throws Exception {
        return filterFrame(frame);
    }
}

消费者类用于商店框架:

import java.util.Map;
import java.util.concurrent.BlockingQueue;
import java.util.concurrent.Future;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.TimeUnit;

public class Consumer implements Runnable {

    private final BlockingQueue<Map.Entry<Long, Future<Frame>>> queue;
    public Boolean continueProducing = Boolean.TRUE;
    private final FFmpegFrameRecorder recorder;

    public Consumer(FFmpegFrameRecorder recorder,
                    BlockingQueue<Map.Entry<Long, Future<Frame>>> queue) {
        this.recorder = recorder;
        this.queue = queue;
    }

    @Override
    public void run() {
        try {
            Map.Entry<Long, Future<Frame>> entry = this.queue.poll(2, TimeUnit.SECONDS);
            Frame editedFrame = entry.getValue().get();
            recorder.setTimestamp(entry.getKey());
            recorder.record(editedFrame);
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
    }
}

管理任务和关闭threadPool的主要部分

try (FFmpegFrameGrabber grabber = createGrabber()) {
            grabber.start();
            try (FFmpegFrameRecorder recorder = createRecorder(grabber)) {
                recorder.start();
                Frame frame;
                ExecutorService threadPool = Executors.newWorkStealingPool();
                BlockingQueue<Map.Entry<Long, Future<Frame>>> queue = new LinkedBlockingQueue<>();
                threadPool.execute(new Consumer(recorder, queue));
                while ((frame = grabber.grab()) != null) {
                    queue.put(new Map.Entry<Long, Future<Frame>>() {
                        @Override
                        public Long getKey() {
                            return grabber.getTimestamp();
                        }

                        @Override
                        public Future<Frame> getValue() {
                            return threadPool.submit(new Producer(frame)); // Frame editedFrame = filterFrame(frame); //This takes a long time.
                        }

                        @Override
                        public Future<Frame> setValue(Future<Frame> value) {
                            return null;
                        }
                    });
                }
                threadPool.shutdownNow();
            }
        }

注意:这不是复制粘贴解决方案,需要为您的代码进行一些自定义。如果你分享更多信息,我会改变它