c++ - 在 OpenCV 中分析 sampleBuffer 时出现 EXC_Bad_Access

Question

我正在尝试使用 OpenCV 进行实时相机处理。在 AVCaptureVideoDataOutputSampleBufferDelegate 的 didOutputSampleBuffer 方法中，我从 sampleBuffer 创建了一个矩阵(可以正常工作)。但是在执行某些方法时，例如cv::GaussianBlur，应用程序会因为“exc_bad_access code=1, address = 0x10 ......”而崩溃。你知道为什么吗？

cv::Mat matrix(bufferHeight, bufferWidth, CV_8UC4, baseAddress);

cv::GaussianBlur(matrix, matrix, cvSize(5,5), 0); // Crahes here

__ 编辑:

基址计算如下(在将这些变量传递给 objective-c++ 之前，这是在 Swift 中的 didOutputSampleBuffer 方法中完成的)

    var pixelBuffer: CVImageBuffer = CMSampleBufferGetImageBuffer(sampleBuffer)!
    CVPixelBufferLockBaseAddress(pixelBuffer, CVPixelBufferLockFlags(rawValue: 0))

    var baseAddress = CVPixelBufferGetBaseAddress(pixelBuffer)

____ 编辑 2:

baseAddress 的值:0x0000000107790000

pixelBuffer 的值:

<CVPixelBuffer 0x17413aae0 width=1920 height=1080 pixelFormat=420v iosurface=0x1700039e0 planes=2>
<Plane 0 width=1920 height=1080 bytesPerRow=1920>
<Plane 1 width=960 height=540 bytesPerRow=1920>
<attributes=<CFBasicHash 0x17426e640 [0x1b8d18bb8]>{type = immutable dict, count = 5,
entries =>
    0 : <CFString 0x1b300de28 [0x1b8d18bb8]>{contents = "PixelFormatDescription"} = <CFBasicHash 0x17426a080 [0x1b8d18bb8]>{type = immutable dict, count = 10,
entries =>
    0 : <CFString 0x1b300e088 [0x1b8d18bb8]>{contents = "Planes"} = (
        {
        BitsPerBlock = 8;
        BlackBlock = <10>;
        FillExtendedPixelsCallback = <00000000 00000000 b840aa95 01000000 00000000 00000000>;
    },
        {
        BitsPerBlock = 16;
        BlackBlock = <8080>;
        FillExtendedPixelsCallback = <00000000 00000000 443faa95 01000000 00000000 00000000>;
        HorizontalSubsampling = 2;
        VerticalSubsampling = 2;
    }
)
    2 : <CFString 0x1b300dd68 [0x1b8d18bb8]>{contents = "IOSurfaceOpenGLESFBOCompatibility"} = <CFBoolean 0x1b8d19110 [0x1b8d18bb8]>{value = true}
    3 : <CFString 0x1b300e228 [0x1b8d18bb8]>{contents = "ContainsYCbCr"} = <CFBoolean 0x1b8d19110 [0x1b8d18bb8]>{value = true}
    4 : <CFString 0x1b300dd48 [0x1b8d18bb8]>{contents = "IOSurfaceOpenGLESTextureCompatibility"} = <CFBoolean 0x1b8d19110 [0x1b8d18bb8]>{value = true}
    5 : <CFString 0x1b300e288 [0x1b8d18bb8]>{contents = "ComponentRange"} = <CFString 0x1b300e2a8 [0x1b8d18bb8]>{contents = "VideoRange"}
    6 : <CFString 0x1b300e008 [0x1b8d18bb8]>{contents = "PixelFormat"} = <CFNumber 0xb000000343230762 [0x1b8d18bb8]>{value = +875704438, type = kCFNumberSInt32Type}
    7 : <CFString 0x1b300dd28 [0x1b8d18bb8]>{contents = "IOSurfaceCoreAnimationCompatibility"} = <CFBoolean 0x1b8d19110 [0x1b8d18bb8]>{value = true}
    9 : <CFString 0x1b300e068 [0x1b8d18bb8]>{contents = "ContainsAlpha"} = <CFBoolean 0x1b8d19120 [0x1b8d18bb8]>{value = false}
    10 : <CFString 0x1b300e248 [0x1b8d18bb8]>{contents = "ContainsRGB"} = <CFBoolean 0x1b8d19120 [0x1b8d18bb8]>{value = false}
    11 : <CFString 0x1b300dd88 [0x1b8d18bb8]>{contents = "OpenGLESCompatibility"} = <CFBoolean 0x1b8d19110 [0x1b8d18bb8]>{value = true}
}

    2 : <CFString 0x1b300dbe8 [0x1b8d18bb8]>{contents = "ExtendedPixelsRight"} = <CFNumber 0xb000000000000002 [0x1b8d18bb8]>{value = +0, type = kCFNumberSInt32Type}
    3 : <CFString 0x1b300dbc8 [0x1b8d18bb8]>{contents = "ExtendedPixelsTop"} = <CFNumber 0xb000000000000002 [0x1b8d18bb8]>{value = +0, type = kCFNumberSInt32Type}
    4 : <CFString 0x1b300dba8 [0x1b8d18bb8]>{contents = "ExtendedPixelsLeft"} = <CFNumber 0xb000000000000002 [0x1b8d18bb8]>{value = +0, type = kCFNumberSInt32Type}
    5 : <CFString 0x1b300dc08 [0x1b8d18bb8]>{contents = "ExtendedPixelsBottom"} = <CFNumber 0xb000000000000082 [0x1b8d18bb8]>{value = +8, type = kCFNumberSInt32Type}
}
 propagatedAttachments=<CFBasicHash 0x17426e900 [0x1b8d18bb8]>{type = mutable dict, count = 4,
entries =>
    0 : <CFString 0x1b300d7c8 [0x1b8d18bb8]>{contents = "CVImageBufferYCbCrMatrix"} = <CFString 0x1b300d7e8 [0x1b8d18bb8]>{contents = "ITU_R_709_2"}
    1 : <CFString 0x1b300d928 [0x1b8d18bb8]>{contents = "CVImageBufferTransferFunction"} = <CFString 0x1b300d7e8 [0x1b8d18bb8]>{contents = "ITU_R_709_2"}
    2 : <CFString 0x1b3044fa0 [0x1b8d18bb8]>{contents = "MetadataDictionary"} = <CFBasicHash 0x170077840 [0x1b8d18bb8]>{type = mutable dict, count = 3,
entries =>
    0 : <CFString 0x1b304d100 [0x1b8d18bb8]>{contents = "SNR"} = <CFNumber 0x170036300 [0x1b8d18bb8]>{value = +28.30700356903138370512, type = kCFNumberFloat64Type}
    1 : <CFString 0x1b304b2e0 [0x1b8d18bb8]>{contents = "ExposureTime"} = <CFNumber 0x170033d00 [0x1b8d18bb8]>{value = +0.01000000000000000021, type = kCFNumberFloat64Type}
    2 : <CFString 0x1b304d0e0 [0x1b8d18bb8]>{contents = "SensorID"} = <CFNumber 0xb000000000002472 [0x1b8d18bb8]>{value = +583, type = kCFNumberSInt32Type}
}

    5 : <CFString 0x1b300d8a8 [0x1b8d18bb8]>{contents = "CVImageBufferColorPrimaries"} = <CFString 0x1b300d7e8 [0x1b8d18bb8]>{contents = "ITU_R_709_2"}
}
 nonPropagatedAttachments=<CFBasicHash 0x17426e8c0 [0x1b8d18bb8]>{type = mutable dict, count = 0,
entries =>
}

Answer 1

啊 - 您的视频数据不是 4 分量 RGBA(或其他)，而是“1.5”分量 YUV。您应该在 YUV 中执行模糊处理，或者更容易地将捕获 session 切换为 RGBA。

YUV 是默认格式，其中有两个“平面”。

平面 0 是“Y”，一个 1920x1080 的 8 位位图，平面 1 是“UV”，一个 960x540 的 16 位位图(实际上是两个并排的 960x540 8 位位图，U 和 V，不知道为什么它们实际上并没有分成 3 个平面)。

在任何情况下，您的代码都需要一个 1920x1080 的 32 位位图，并且会耗尽 Y channel 内存的末端。

如果你想切换到 RGBA，请执行(我想 - 我永远不记得 iOS 使用哪种 4 分量格式):

output.videoSettings = [kCVPixelBufferPixelFormatTypeKey as AnyHashable: kCVPixelFormatType_32BGRA]

如果您喜欢冒险，可以对 yuv 数据进行模糊处理 - 它小 2.666666667 倍，您的代码可以快 2.666667 倍。