A few clarification here,
. The default OV5693 camera is bayer format sensor. You could use NV nvgstcapture-1.0 (under /usr/bin and source also available) to perform capture function. It will use Tegra ISP function as the capture is a raw image. This is described in the first section of SCF user mode data path of V4L2 Sensor Driver programming Guide.
Without going through ISP, it won’t be human consumable image. This will be the case for standard V4L2-based application as v4l2 capture won’t have Tegra ISP support and the image is raw data.
. To use standard v4l2 application, you could use YUV camera module by writing your own media controller sensor driver (such as some of the Omnivision YUV camera sensor). Again above default bayer camera won’t have this data pass. One thing to note is, with this YUV sensor driver, you will need to configure V4L2 media controller driver as VI non-bypass mode which I will cover at the very end of my comment about documentation.
. From r23.2 to r24.1, there is slight camera driver model changes - we revamp nvhost-vi (bypass) driver and v4l2 soc_camera (non-bypass) driver into one v4l2 media controller driver model. That’s why nvhost-vi driver is gone for r24.1. Tegra-camera driver is used for loadable v4l2 driver to support both L4T and Android. But using one media controller driver, there is no more need for it. So it’s gone for r24.1 too.
. if cheese use gstreamer framework, it should work for default camera but might not be accelerated by TX1 unless you use NV gstreamer plug-in components. Can refer to Multimedia User Guide for details. If cheese uses standard v4l2 IOCtl, you won’t be able to use it with default bayer camera.
We can explore the missing cluttervideosink error message if you could provide us more details.
. As for r24.1 documentation confusion, we will review, clarify and update both ‘Video for Linux User Guide’ and ‘V4L2 Sensor Driver Programming Guide’ ASAP. Thanks for your patience.