yuv422_yuy2 support

[TomoyukiaW]

I'm using the YUYV format output camera and usb_cam package as a driver. usb_cam uses image_transport::CameraPublisher to publish the image. Published raw_image encoding was "encoding: yuv422_yuy2" and raw_image/compressed format was "format: yuv422_yuy2; jpeg compressed mono8".
raw_image could be shown in RViz2. But raw_image/compressed couldn't shown in RViz2, and we got some errors in some image conversion processes (e.g. compressing with ffmpeg). It seems that the compression process for yuv image has a problem.

Then, the following modification worked in my environment. How about this?
cv_bridge supports to convert yuv422_yuy2 and yuv422 format image to cv Mat. For compressed_subscriber.cpp, I couldn't find the suitable conversion pattern and I thought probably most application don't use yuv format in its processing.

--- a/compressed_image_transport/src/compressed_publisher.cpp
+++ b/compressed_image_transport/src/compressed_publisher.cpp
@@ -201,7 +201,7 @@ void CompressedPublisher::publish(
       {
         // Target image format
         std::string targetFormat;
-        if (enc::isColor(message.encoding))
+        if (enc::isColor(message.encoding) || message.encoding == enc::YUV422 || message.encoding == enc::YUV422_YUY2)
         {
           // convert color images to BGR8 format
           targetFormat = "bgr8";

--- a/compressed_image_transport/src/compressed_subscriber.cpp
+++ b/compressed_image_transport/src/compressed_subscriber.cpp
@@ -143,7 +143,7 @@ void CompressedSubscriber::internalCallback(const CompressedImage::ConstSharedPt
 
       cv_ptr->encoding = image_encoding;
 
-      if ( enc::isColor(image_encoding))
+      if ( enc::isColor(image_encoding) || image_encoding == enc::YUV422 || image_encoding == enc::YUV422_YUY2)
       {
         std::string compressed_encoding = message->format.substr(split_pos);
         bool compressed_bgr_image = (compressed_encoding.find("compressed bgr") != std::string::npos);
@@ -160,6 +160,9 @@ void CompressedSubscriber::internalCallback(const CompressedImage::ConstSharedPt
 
           if ((image_encoding == enc::BGRA8) || (image_encoding == enc::BGRA16))
             cv::cvtColor(cv_ptr->image, cv_ptr->image, CV_BGR2BGRA);
+
+          if ((image_encoding == enc::YUV422) || (image_encoding == enc::YUV422_YUY2))
+            cv_ptr->encoding = enc::BGR8;
         } else
         {
           // if necessary convert colors from rgb to bgr
@@ -171,6 +174,9 @@ void CompressedSubscriber::internalCallback(const CompressedImage::ConstSharedPt
 
           if ((image_encoding == enc::RGBA8) || (image_encoding == enc::RGBA16))
             cv::cvtColor(cv_ptr->image, cv_ptr->image, CV_RGB2RGBA);
+
+          if ((image_encoding == enc::YUV422) || (image_encoding == enc::YUV422_YUY2))
+            cv_ptr->encoding = enc::RGB8;
         }
       }
       if (message->format.find("jpeg") != std::string::npos && enc::bitDepth(image_encoding) == 16) {

[milidam]

The compression of yuv422_yuy2-encoded images seems to be OK; the problem, to me, seems to be on the decoder side, where format yuv422_yuy2; jpeg compressed... does not seem to be handled properly; see my analysis in ros-perception/image_common#293 (comment).