Hi Chris,
I've tweaked the capture command to not skip any ( I had originally tried not skipping ) but unfortunately it is still garbled:
insmod /mnt/apps/r8a66597-hcd.ko
cd /sys/module/uvcvideo/parameters
echo 1 > nodrop
echo 15000 > timeout
echo 640 > quirks
echo 0xffff > trace
fsck.fat -a /dev/sda
mkdir -p /tmp/usb
mount /dev/sda /tmp/usb
nice -n -19 chrt -f 99 fswebcam -v --device /dev/video0 -D 1 --input 0 --resolution 320x240 --no-banner -p YUYV --save /tmp/vl.jpeg
cp /tmp/vl.jpeg /tmp/usb/
umount /tmp/usb
The image does look just slightly better ( at least half of it is not green ) - but it is still not coherent:
I downloaded and looked at the example code from the App note you posted, it is very similar to my custom C++ code I had written to capture from the V4L2 device. In my app I capture 1 frame by queuing and dequeuing 1 V4L2 MMAP buffer.
I was using fswebcam as a sanity check on that capture code because it was a widely used application.