Question

我目前正在使用AVAssetImageGenerator从视频中提取每一帧，但有时它会连续两次返回我相同的图像（它们没有相同的“帧时间”）。有趣的是它总是发生（在我的测试视频中）每5帧。

Here和here是两张图片（在新标签中分别打开，然后切换标签以查看差异）。

这是我的代码：

//setting up generator & compositor
self.generator = [AVAssetImageGenerator assetImageGeneratorWithAsset:asset];
generator.appliesPreferredTrackTransform = YES;
self.composition = [AVVideoComposition videoCompositionWithPropertiesOfAsset:asset];

NSTimeInterval duration = CMTimeGetSeconds(asset.duration);
NSTimeInterval frameDuration = CMTimeGetSeconds(composition.frameDuration);
CGFloat totalFrames = round(duration/frameDuration);

NSMutableArray * times = [NSMutableArray array];
for (int i=0; i<totalFrames; i++) {
    NSValue * time = [NSValue valueWithCMTime:CMTimeMakeWithSeconds(i*frameDuration, composition.frameDuration.timescale)];
    [times addObject:time];
}

AVAssetImageGeneratorCompletionHandler handler = ^(CMTime requestedTime, CGImageRef im, CMTime actualTime, AVAssetImageGeneratorResult result, NSError *error){
    // If actualTime is not equal to requestedTime image is ignored
    if(CMTimeCompare(actualTime, requestedTime) == 0) {
        if (result == AVAssetImageGeneratorSucceeded) {
            NSLog(@"%.02f     %.02f", CMTimeGetSeconds(requestedTime), CMTimeGetSeconds(actualTime));
            // Each log have differents actualTimes.
            // frame extraction is here...
        }
    }
};

generator.requestedTimeToleranceBefore = kCMTimeZero;
generator.requestedTimeToleranceAfter = kCMTimeZero;
[generator generateCGImagesAsynchronouslyForTimes:times completionHandler:handler];

知道它可能来自哪里？

Answer 1

请参阅AVAssetImageGenerator的以下属性。您应该为两个属性设置kCMTimeZero以获取确切的帧。

/* The actual time of the generated images will be within the range [requestedTime-toleranceBefore, requestedTime+toleranceAfter] and may differ from the requested time for efficiency.
   Pass kCMTimeZero for both toleranceBefore and toleranceAfter to request frame-accurate image generation; this may incur additional decoding delay.
   Default is kCMTimePositiveInfinity. */
@property (nonatomic) CMTime requestedTimeToleranceBefore NS_AVAILABLE(10_7, 5_0);
@property (nonatomic) CMTime requestedTimeToleranceAfter NS_AVAILABLE(10_7, 5_0);

在为两个属性设置kCMTimeZero之前，我为您提供了与您所经历的不同请求时间相同的图像。只需尝试以下代码即可。

self.imageGenerator = [AVAssetImageGenerator assetImageGeneratorWithAsset:myAsset];
self.imageGenerator.requestedTimeToleranceBefore = kCMTimeZero;
self.imageGenerator.requestedTimeToleranceAfter = kCMTimeZero;

Answer 2

我使用稍微不同的方式来计算CMTime请求，它似乎有效。这是代码（假设iOS）：

-(void)extractImagesFromMovie {

// set the asset
    NSString* path = [[NSBundle mainBundle] pathForResource:@"myMovie" ofType:@"MOV"];
    NSURL* movURL = [NSURL fileURLWithPath:path];

NSMutableDictionary* myDict = [NSMutableDictionary dictionaryWithObjectsAndKeys:[NSNumber numberWithBool:YES] , 
                                 AVURLAssetPreferPreciseDurationAndTimingKey , 
                                [NSNumber numberWithInt:0],
                                AVURLAssetReferenceRestrictionsKey, nil];

AVURLAsset* movie = [[AVURLAsset alloc] initWithURL:movURL options:myDict];


// set the generator
AVAssetImageGenerator* generator = [[AVAssetImageGenerator assetImageGeneratorWithAsset:movie] retain];
generator.requestedTimeToleranceBefore = kCMTimeZero;
generator.requestedTimeToleranceAfter = kCMTimeZero;

// look for the video track
AVAssetTrack* videoTrack;
bool foundTrack = NO;

for (AVAssetTrack* track in movie.tracks) {

    if ([track.mediaType isEqualToString:@"vide"]) {
        if (foundTrack) {NSLog (@"Error - - - more than one video tracks"); return(-1);}
        else {
            videoTrack = track;
            foundTrack = YES;
        }
    }
}
if (foundTrack == NO) {NSLog (@"Error - - No Video Tracks at all"); return(-1);}

// set the number of frames in the movie
int frameRate = videoTrack.nominalFrameRate;
float value = movie.duration.value;
float timeScale = movie.duration.timescale;
float totalSeconds = value / timeScale;
int totalFrames = totalSeconds * frameRate;

NSLog (@"total frames %d", totalFrames);

int timeValuePerFrame = movie.duration.timescale / frameRate;

NSMutableArray* allFrames = [[NSMutableArray new] retain];

// get each frame
for (int k=0; k< totalFrames; k++) {

    int timeValue = timeValuePerFrame * k;
    CMTime frameTime;
    frameTime.value = timeValue;
    frameTime.timescale = movie.duration.timescale;
    frameTime.flags = movie.duration.flags;
    frameTime.epoch = movie.duration.epoch;

    CMTime gotTime;

    CGImageRef myRef = [generator copyCGImageAtTime:frameTime actualTime:&gotTime error:nil];
    [allFrames addObject:[UIImage imageWithCGImage:myRef]];

    if (gotTime.value != frameTime.value) NSLog (@"requested %lld got %lld for k %d", frameTime.value, gotTime.value, k)

}

NSLog (@"got %d images in the array", [allFrames count]);
// do something with images here...
}

Answer 3

我遇到了同样的问题，但更明显的是，当两帧之间的间隔小于1.0秒时，重复发生了，我意识到这取决于我用来生成CMTime值的时间刻度。 / p>

在

CMTime requestTime = CMTimeMakeWithSeconds(imageTime, 1);

在

CMTime requestTime = CMTimeMakeWithSeconds(imageTime, playerItem.asset.duration.timescale);

......和Boom，不再重复：）

所以也许你可以尝试使用你的代码来增加时间尺度，加倍：

NSValue * time = [NSValue valueWithCMTime:CMTimeMakeWithSeconds(i*frameDuration, composition.frameDuration.timescale*2)]; // *2 at the end

以下是我的代码：

    playerItem = [AVPlayerItem playerItemWithURL:item.movieUrl];
    imageGenerator = [[AVAssetImageGenerator alloc] initWithAsset:playerItem.asset];
    imageGenerator.requestedTimeToleranceAfter = kCMTimeZero;
    imageGenerator.requestedTimeToleranceBefore = kCMTimeZero;

    float duration = item.duration;
    float interval = item.interval;

    NSLog(@"\nItem info:\n%f \n%f", duration,interval);

    NSString *srcPath = nil;
    NSString *zipPath = nil;

    srcPath = [item.path stringByAppendingPathComponent:@"info.json"];
    zipPath = [NSString stringWithFormat:@"/%@/info.json",galleryID];

    [zip addFileToZip:srcPath newname:zipPath level:0];

    NSTimeInterval frameNum = item.duration / item.interval;
    for (int i=0; i<=frameNum; i++)
    {
        NSArray* cachePathArray = NSSearchPathForDirectoriesInDomains(NSCachesDirectory, NSUserDomainMask, YES);
        NSString* cachePath = [cachePathArray lastObject];

        srcPath = [cachePath stringByAppendingPathComponent:@"export-tmp.jpg"];
        zipPath = [NSString stringWithFormat:@"/%@/%d.jpg",galleryID,i];

        float imageTime = ( i * interval );

        NSError *error = nil;
        CMTime requestTime = CMTimeMakeWithSeconds(imageTime, playerItem.asset.duration.timescale);
        CMTime actualTime;

        CGImageRef imageRef = [imageGenerator copyCGImageAtTime:requestTime actualTime:&actualTime error:&error];

        if (error == nil) {
            float req = ((float)requestTime.value/requestTime.timescale);
            float real = ((float)actualTime.value/actualTime.timescale);
            float diff = fabsf(req-real);

            NSLog(@"copyCGImageAtTime: %.2f, %.2f, %f",req,real,diff);
        }
        else
        {
            NSLog(@"copyCGImageAtTime: error: %@",error.localizedDescription);
        }



        // consider using CGImageDestination -> http://stackoverflow.com/questions/1320988/saving-cgimageref-to-a-png-file
        UIImage *img = [UIImage imageWithCGImage:imageRef];
        CGImageRelease(imageRef);  // CGImageRef won't be released by ARC



        [UIImageJPEGRepresentation(img, 100) writeToFile:srcPath atomically:YES];

        if (srcPath != nil && zipPath!= nil)
        {
            [zip addFileToZip:srcPath newname:zipPath level:0]; // 0 = no compression. everything is a jpg image
            unlink([srcPath UTF8String]);
        }

AVAssetImageGenerator有时返回来自2个连续帧的相同图像

3 个答案: