ios - 将 CVImageBuffer 转换为 YUV420 对象

Question

我想保留来自摄像机的流式视频的 YUV420 格式以避免转换为灰度时的损失，但我也想保留颜色分量。最终目标是使用像 OpenCV 这样的计算机视觉库进行处理。虽然我最终可能会选择 BGRA，但我仍然希望能够使用 YUV 测试一个可行的解决方案。那么如何将像素格式为 kCVPixelFormatType_420YpCbCr8BiPlanarFullRange 的 CVImageBuffer 转换为单个内存块？

被拒绝的解决方案:

• CIImage 非常方便，但不允许渲染为 YUV 格式的位图。
• cv::Mat 用 C++ 污染你的 Obj-C 代码

Answer 1

AVCaptureSessionDelegate

这将根据指定的像素格式将数据填充到包含字节的 NSObject 中。我继续为 BGRA 或 YUV 像素格式提供检测和 malloc 内存的能力。所以这个解决方案非常适合测试这两者。

- (void)captureOutput:(AVCaptureOutput *)captureOutput didOutputSampleBuffer:(CMSampleBufferRef)sampleBuffer fromConnection:(AVCaptureConnection *)connection
{
    CVImageBufferRef videoImageBuffer = CMSampleBufferGetImageBuffer(sampleBuffer);

    CVPixelBufferLockBaseAddress(videoImageBuffer, 0);

    void *baseAddress = NULL;
    NSUInteger totalBytes = 0;
    size_t width = CVPixelBufferGetWidth(videoImageBuffer);
    size_t height = 0;
    size_t bytesPerRow = CVPixelBufferGetBytesPerRow(videoImageBuffer);
    OSType pixelFormat = CVPixelBufferGetPixelFormatType(videoImageBuffer);
    if (pixelFormat == kCVPixelFormatType_420YpCbCr8BiPlanarVideoRange ||
        pixelFormat == kCVPixelFormatType_420YpCbCr8BiPlanarFullRange) {
        size_t planeCount = CVPixelBufferGetPlaneCount(videoImageBuffer);
        baseAddress = CVPixelBufferGetBaseAddressOfPlane(videoImageBuffer, 0);

        for (int plane = 0; plane < planeCount; plane++) {
            size_t planeHeight = CVPixelBufferGetHeightOfPlane(videoImageBuffer, plane);
            size_t bytesPerRow = CVPixelBufferGetBytesPerRowOfPlane(videoImageBuffer, plane);
            height += planeHeight;
            totalBytes += (int)planeHeight * (int)bytesPerRow;
        }
    } else if (pixelFormat == kCVPixelFormatType_32BGRA) {
        baseAddress = CVPixelBufferGetBaseAddress(videoImageBuffer);
        height = CVPixelBufferGetHeight(videoImageBuffer);
        size_t bytesPerRow = CVPixelBufferGetBytesPerRow(videoImageBuffer);
        totalBytes += (int)height * (int)bytesPerRow;
    }

    // Doesn't have to be an NSData object
    NSData *rawPixelData = [NSData dataWithBytes:baseAddress length:totalBytes];

    // Just a plain-ol-NSObject with the following properties
    NTNUVideoFrame *videoFrame = [[NTNUVideoFrame alloc] init];
    videoFrame.width = width;
    videoFrame.height = height;
    videoFrame.bytesPerRow = bytesPerRow;
    videoFrame.pixelFormat = pixelFormat; 
    // Alternatively if you switch rawPixelData to void *
    // videoFrame.rawPixelData = baseAddress;
    videoFrame.rawPixelData = rawPixelData;
    [self.delegate didUpdateVideoFrame:videoFrame];

    CVPixelBufferUnlockBaseAddress(videoImageBuffer, 0);
}

你唯一需要记住的是，如果你计划切换线程，你将需要 malloc 和 memcpy 基地址，或者 dispatch_async 而你不这样做使用 NSData。解锁基地址后，像素数据将不再有效。

void *rawPixelData = malloc(totalBytes);
memcpy(rawPixelData, baseAddress, totalBytes);

此时您需要考虑在完成后对该内存块调用 free 的问题。