I'm working on the Intel's RealSense SDK. For signal processing purposes, I have to convert it into an OpenCV format.
我正在开发英特尔的RealSense SDK。为了信号处理的目的,我必须把它转换成OpenCV格式。
Do you have a function to do such a conversion?
你有一个函数来做这样的转换吗?
1 个解决方案
#1
2
Here is the conversion (some cases aren't managed but feel free to adapt it):
这里是转换(有些情况没有管理,但可以随意调整):
void ConvertPXCImageToOpenCVMat(PXCImage *inImg, Mat *outImg) {
int cvDataType;
int cvDataWidth;
PXCImage::ImageData data;
inImg->AcquireAccess(PXCImage::ACCESS_READ, &data);
PXCImage::ImageInfo imgInfo = inImg->QueryInfo();
switch (data.format) {
/* STREAM_TYPE_COLOR */
case PXCImage::PIXEL_FORMAT_YUY2: /* YUY2 image */
case PXCImage::PIXEL_FORMAT_NV12: /* NV12 image */
throw(0); // Not implemented
case PXCImage::PIXEL_FORMAT_RGB32: /* BGRA layout on a little-endian machine */
cvDataType = CV_8UC4;
cvDataWidth = 4;
break;
case PXCImage::PIXEL_FORMAT_RGB24: /* BGR layout on a little-endian machine */
cvDataType = CV_8UC3;
cvDataWidth = 3;
break;
case PXCImage::PIXEL_FORMAT_Y8: /* 8-Bit Gray Image, or IR 8-bit */
cvDataType = CV_8U;
cvDataWidth = 1;
break;
/* STREAM_TYPE_DEPTH */
case PXCImage::PIXEL_FORMAT_DEPTH: /* 16-bit unsigned integer with precision mm. */
case PXCImage::PIXEL_FORMAT_DEPTH_RAW: /* 16-bit unsigned integer with device specific precision (call device->QueryDepthUnit()) */
cvDataType = CV_16U;
cvDataWidth = 2;
break;
case PXCImage::PIXEL_FORMAT_DEPTH_F32: /* 32-bit float-point with precision mm. */
cvDataType = CV_32F;
cvDataWidth = 4;
break;
/* STREAM_TYPE_IR */
case PXCImage::PIXEL_FORMAT_Y16: /* 16-Bit Gray Image */
cvDataType = CV_16U;
cvDataWidth = 2;
break;
case PXCImage::PIXEL_FORMAT_Y8_IR_RELATIVE: /* Relative IR Image */
cvDataType = CV_8U;
cvDataWidth = 1;
break;
}
// suppose that no other planes
if (data.planes[1] != NULL) throw(0); // not implemented
// suppose that no sub pixel padding needed
if (data.pitches[0] % cvDataWidth!=0) throw(0); // not implemented
outImg->create(imgInfo.height, data.pitches[0] / cvDataWidth, cvDataType);
memcpy(outImg->data, data.planes[0], imgInfo.height*imgInfo.width*cvDataWidth*sizeof(pxcBYTE));
inImg->ReleaseAccess(&data);
}
#1
2
Here is the conversion (some cases aren't managed but feel free to adapt it):
这里是转换(有些情况没有管理,但可以随意调整):
void ConvertPXCImageToOpenCVMat(PXCImage *inImg, Mat *outImg) {
int cvDataType;
int cvDataWidth;
PXCImage::ImageData data;
inImg->AcquireAccess(PXCImage::ACCESS_READ, &data);
PXCImage::ImageInfo imgInfo = inImg->QueryInfo();
switch (data.format) {
/* STREAM_TYPE_COLOR */
case PXCImage::PIXEL_FORMAT_YUY2: /* YUY2 image */
case PXCImage::PIXEL_FORMAT_NV12: /* NV12 image */
throw(0); // Not implemented
case PXCImage::PIXEL_FORMAT_RGB32: /* BGRA layout on a little-endian machine */
cvDataType = CV_8UC4;
cvDataWidth = 4;
break;
case PXCImage::PIXEL_FORMAT_RGB24: /* BGR layout on a little-endian machine */
cvDataType = CV_8UC3;
cvDataWidth = 3;
break;
case PXCImage::PIXEL_FORMAT_Y8: /* 8-Bit Gray Image, or IR 8-bit */
cvDataType = CV_8U;
cvDataWidth = 1;
break;
/* STREAM_TYPE_DEPTH */
case PXCImage::PIXEL_FORMAT_DEPTH: /* 16-bit unsigned integer with precision mm. */
case PXCImage::PIXEL_FORMAT_DEPTH_RAW: /* 16-bit unsigned integer with device specific precision (call device->QueryDepthUnit()) */
cvDataType = CV_16U;
cvDataWidth = 2;
break;
case PXCImage::PIXEL_FORMAT_DEPTH_F32: /* 32-bit float-point with precision mm. */
cvDataType = CV_32F;
cvDataWidth = 4;
break;
/* STREAM_TYPE_IR */
case PXCImage::PIXEL_FORMAT_Y16: /* 16-Bit Gray Image */
cvDataType = CV_16U;
cvDataWidth = 2;
break;
case PXCImage::PIXEL_FORMAT_Y8_IR_RELATIVE: /* Relative IR Image */
cvDataType = CV_8U;
cvDataWidth = 1;
break;
}
// suppose that no other planes
if (data.planes[1] != NULL) throw(0); // not implemented
// suppose that no sub pixel padding needed
if (data.pitches[0] % cvDataWidth!=0) throw(0); // not implemented
outImg->create(imgInfo.height, data.pitches[0] / cvDataWidth, cvDataType);
memcpy(outImg->data, data.planes[0], imgInfo.height*imgInfo.width*cvDataWidth*sizeof(pxcBYTE));
inImg->ReleaseAccess(&data);
}