Szenenwechsel

Mit der Szenenwechselanalyse lassen sich Szenenwechsel in einem Video erkennen.

In dem folgenden Beispiel für Python ist gut zu sehen, wie Video Intelligence Streaming genutzt werden kann, um Szenenwechseln in einem Video zu identifizieren.

Java

import com.google.api.gax.rpc.BidiStream;
import com.google.cloud.videointelligence.v1p3beta1.StreamingAnnotateVideoRequest;
import com.google.cloud.videointelligence.v1p3beta1.StreamingAnnotateVideoResponse;
import com.google.cloud.videointelligence.v1p3beta1.StreamingFeature;
import com.google.cloud.videointelligence.v1p3beta1.StreamingLabelDetectionConfig;
import com.google.cloud.videointelligence.v1p3beta1.StreamingVideoAnnotationResults;
import com.google.cloud.videointelligence.v1p3beta1.StreamingVideoConfig;
import com.google.cloud.videointelligence.v1p3beta1.StreamingVideoIntelligenceServiceClient;
import com.google.cloud.videointelligence.v1p3beta1.VideoSegment;
import com.google.protobuf.ByteString;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.util.Arrays;

class StreamingShotChangeDetection {

  // Perform streaming video detection for shot changes
  static void streamingShotChangeDetection(String filePath) {
    // String filePath = "path_to_your_video_file";

    try (StreamingVideoIntelligenceServiceClient client =
        StreamingVideoIntelligenceServiceClient.create()) {

      Path path = Paths.get(filePath);
      byte[] data = Files.readAllBytes(path);
      // Set the chunk size to 5MB (recommended less than 10MB).
      int chunkSize = 5 * 1024 * 1024;
      int numChunks = (int) Math.ceil((double) data.length / chunkSize);

      StreamingLabelDetectionConfig labelConfig = StreamingLabelDetectionConfig.newBuilder()
          .setStationaryCamera(false)
          .build();

      StreamingVideoConfig streamingVideoConfig = StreamingVideoConfig.newBuilder()
          .setFeature(StreamingFeature.STREAMING_SHOT_CHANGE_DETECTION)
          .setLabelDetectionConfig(labelConfig)
          .build();

      BidiStream<StreamingAnnotateVideoRequest, StreamingAnnotateVideoResponse> call =
          client.streamingAnnotateVideoCallable().call();

      // The first request must **only** contain the audio configuration:
      call.send(
          StreamingAnnotateVideoRequest.newBuilder()
              .setVideoConfig(streamingVideoConfig)
              .build());

      // Subsequent requests must **only** contain the audio data.
      // Send the requests in chunks
      for (int i = 0; i < numChunks; i++) {
        call.send(
            StreamingAnnotateVideoRequest.newBuilder()
                .setInputContent(ByteString.copyFrom(
                    Arrays.copyOfRange(data, i * chunkSize, i * chunkSize + chunkSize)))
                .build());
      }

      // Tell the service you are done sending data
      call.closeSend();

      for (StreamingAnnotateVideoResponse response : call) {
        StreamingVideoAnnotationResults annotationResults = response.getAnnotationResults();

        for (VideoSegment segment : annotationResults.getShotAnnotationsList()) {
          double startTimeOffset = segment.getStartTimeOffset().getSeconds()
              + segment.getStartTimeOffset().getNanos() / 1e9;
          double endTimeOffset = segment.getEndTimeOffset().getSeconds()
              + segment.getEndTimeOffset().getNanos() / 1e9;

          System.out.format("Shot: %fs to %fs\n", startTimeOffset, endTimeOffset);
        }
      }
    } catch (IOException e) {
      e.printStackTrace();
    }
  }
}

Node.js

/**
 * TODO(developer): Uncomment these variables before running the sample.
 */
// const path = 'Local file to analyze, e.g. ./my-file.mp4';
const {
  StreamingVideoIntelligenceServiceClient,
} = require('@google-cloud/video-intelligence').v1p3beta1;
const fs = require('fs');

// Instantiates a client
const client = new StreamingVideoIntelligenceServiceClient();
// Streaming configuration
const configRequest = {
  videoConfig: {
    feature: 'STREAMING_SHOT_CHANGE_DETECTION',
  },
};
const readStream = fs.createReadStream(path, {
  highWaterMark: 5 * 1024 * 1024, //chunk size set to 5MB (recommended less than 10MB)
  encoding: 'base64',
});
//Load file content
const chunks = [];
readStream
  .on('data', chunk => {
    const request = {
      inputContent: chunk.toString(),
    };
    chunks.push(request);
  })
  .on('close', function() {
    // configRequest should be the first in the stream of requests
    stream.write(configRequest);
    for (let i = 0; i < chunks.length; i++) {
      stream.write(chunks[i]);
    }
    stream.end();
  });

const stream = client.streamingAnnotateVideo().on('data', response => {
  //Gets annotations for video
  const annotations = response.annotationResults;
  const shotChanges = annotations.shotAnnotations;
  console.log(JSON.stringify(shotChanges));
  if (shotChanges.length === 1) {
    console.log(`The entire video is one shot.`);
  }
  shotChanges.forEach(shot => {
    console.log(
      ` Shot: ${shot.startTimeOffset.seconds || 0}` +
        `.${(shot.startTimeOffset.nanos / 1e6).toFixed(0)}s to ${shot
          .endTimeOffset.seconds || 0}` +
        `.${(shot.endTimeOffset.nanos / 1e6).toFixed(0)}s`
    );
  });
});

Python

from google.cloud import videointelligence_v1p3beta1 as videointelligence

# path = 'path_to_file'

client = videointelligence.StreamingVideoIntelligenceServiceClient()

# Set streaming config.
config = videointelligence.types.StreamingVideoConfig(
    feature=(videointelligence.enums.StreamingFeature.
             STREAMING_SHOT_CHANGE_DETECTION))

# config_request should be the first in the stream of requests.
config_request = videointelligence.types.StreamingAnnotateVideoRequest(
    video_config=config)

# Set the chunk size to 5MB (recommended less than 10MB).
chunk_size = 5 * 1024 * 1024

# Load file content.
stream = []
with io.open(path, 'rb') as video_file:
    while True:
        data = video_file.read(chunk_size)
        if not data:
            break
        stream.append(data)

def stream_generator():
    yield config_request
    for chunk in stream:
        yield videointelligence.types.StreamingAnnotateVideoRequest(
            input_content=chunk)

requests = stream_generator()

# streaming_annotate_video returns a generator.
responses = client.streaming_annotate_video(requests)

# Each response corresponds to about 1 second of video.
for response in responses:
    # Check for errors.
    if response.error.message:
        print(response.error.message)
        break

    for annotation in response.annotation_results.shot_annotations:
        start = (annotation.start_time_offset.seconds +
                 annotation.start_time_offset.nanos / 1e9)
        end = (annotation.end_time_offset.seconds +
               annotation.end_time_offset.nanos / 1e9)

        print('Shot: {}s to {}s'.format(start, end))