使用 AVMutableVideoComposition 导出时的视频方向问题

Question

这是我用来导出视频的功能：

- (void) videoOutput
{
//1 - Early exit if there's no video file selected
if (!self.videoAsset) {
    UIAlertView *alert = [[UIAlertView alloc] initWithTitle:@"Error" message:@"Please Load a Video Asset First"
                                                   delegate:nil cancelButtonTitle:@"OK" otherButtonTitles:nil];
    [alert show];
    return;
}

// 2 - Create AVMutableComposition object. This object will hold your AVMutableCompositionTrack instances.
AVMutableComposition *mixComposition = [[AVMutableComposition alloc] init];

// 3 - Video track
AVMutableCompositionTrack *videoTrack = [mixComposition addMutableTrackWithMediaType:AVMediaTypeVideo
                                                                    preferredTrackID:kCMPersistentTrackID_Invalid];
[videoTrack insertTimeRange:CMTimeRangeMake(kCMTimeZero, self.videoAsset.duration)
                    ofTrack:[[self.videoAsset tracksWithMediaType:AVMediaTypeVideo] objectAtIndex:0]
                     atTime:kCMTimeZero error:nil];

// 3.1 - Create AVMutableVideoCompositionInstruction
AVMutableVideoCompositionInstruction *mainInstruction = [AVMutableVideoCompositionInstruction videoCompositionInstruction];
mainInstruction.timeRange = CMTimeRangeMake(kCMTimeZero, self.videoAsset.duration);

// 3.2 - Create an AVMutableVideoCompositionLayerInstruction for the video track and fix the orientation.
AVMutableVideoCompositionLayerInstruction *videolayerInstruction = [AVMutableVideoCompositionLayerInstruction videoCompositionLayerInstructionWithAssetTrack:videoTrack];
AVAssetTrack *videoAssetTrack = [[self.videoAsset tracksWithMediaType:AVMediaTypeVideo] objectAtIndex:0];

BOOL isVideoAssetPortrait_  = NO;
CGAffineTransform videoTransform = videoAssetTrack.preferredTransform;
if (videoTransform.a == 0 && videoTransform.b == 1.0 && videoTransform.c == -1.0 && videoTransform.d == 0) {
    isVideoAssetPortrait_ = YES;
}
if (videoTransform.a == 0 && videoTransform.b == -1.0 && videoTransform.c == 1.0 && videoTransform.d == 0) {

    isVideoAssetPortrait_ = YES;
}
if (videoTransform.a == 1.0 && videoTransform.b == 0 && videoTransform.c == 0 && videoTransform.d == 1.0) {
    isVideoAssetPortrait_  = NO;
}
if (videoTransform.a == -1.0 && videoTransform.b == 0 && videoTransform.c == 0 && videoTransform.d == -1.0) {
    isVideoAssetPortrait_  = NO;
}
[videolayerInstruction setTransform:videoAssetTrack.preferredTransform atTime:kCMTimeZero];
[videolayerInstruction setOpacity:0.0 atTime:self.videoAsset.duration];

// 3.3 - Add instructions
mainInstruction.layerInstructions = [NSArray arrayWithObjects:videolayerInstruction,nil];

AVMutableVideoComposition *mainCompositionInst = [AVMutableVideoComposition videoComposition];

CGSize naturalSize;
if(isVideoAssetPortrait_){
    naturalSize = CGSizeMake(videoAssetTrack.naturalSize.height, videoAssetTrack.naturalSize.width);
} else {
    naturalSize = videoAssetTrack.naturalSize;
}

mainCompositionInst.renderSize = naturalSize;
mainCompositionInst.instructions = [NSArray arrayWithObject:mainInstruction];
mainCompositionInst.frameDuration = CMTimeMake(1, 30);

// 4 - Get path
NSArray *paths = NSSearchPathForDirectoriesInDomains(NSDocumentDirectory, NSUserDomainMask, YES);
NSString *documentsDirectory = [paths objectAtIndex:0];
NSString *myPathDocs =  [documentsDirectory stringByAppendingPathComponent:
                         [NSString stringWithFormat:@"FinalVideo-%d.mov",arc4random() % 1000]];
NSURL *url = [NSURL fileURLWithPath:myPathDocs];

// 5 - Create exporter
AVAssetExportSession *exporter = [[AVAssetExportSession alloc] initWithAsset:mixComposition
                                                                  presetName:AVAssetExportPresetHighestQuality];
exporter.outputURL=url;
exporter.outputFileType = AVFileTypeQuickTimeMovie;
exporter.shouldOptimizeForNetworkUse = YES;
exporter.videoComposition = mainCompositionInst;
[exporter exportAsynchronouslyWithCompletionHandler:^{
    dispatch_async(dispatch_get_main_queue(), ^{
        [self exportDidFinish:exporter];
    });
}];
}

问题是我第一次用这个函数导出人像视频时，变量videoTransform(videoAssetTrack.preferredTransform)是：

videoTransform.a == 0 && videoTransform.b == 1.0 && videoTransform.c == -1.0 && videoTransform.d == 0

并且变量 isVideoAssetPortrait_ 等于 YES。一切都是对的。但是，在导出完成并保存到相机胶卷后，我使用此功能重新加载结果视频。这一次，videoTransform 发生了变化：

videoTransform.a == 1.0 && videoTransform.b == 0 && videoTransform.c == 0 && videoTransform.d == 1.0

并且 isVideoAssetPortrait_ 等于 NO。这意味着在一次导出后，videoTransform 改变了它的值（方向从纵向 -> 横向）

我在谷歌上搜索了很多关于 AV Foundation 的视频方向的问题，但还没有找到解决方案。

感谢您阅读我的长篇解释。如果您有任何问题，请告诉我。

Answer 1

没有关于视频轨道的纵向/横向概念。它只有它的尺寸和应用的转换，以便正确呈现它。默认情况下，“人像”视频在由相机（比如说风景）产生时被编码，并且用于正确呈现它的 90 度旋转。

当您导出它时，它的方向没有改变，它只是重新编码物理旋转，所以不需要旋转来正确呈现它。 这就是为什么你第二次得到单位矩阵（这并不意味着它从纵向变为横向），但是这次它的自然大小也被交换了，根据你的代码一切都应该没问题。

请说明后一种情况的问题。