question about Edge AI demo on TDA4

xu ji

Other Parts Discussed in Thread: SK-TDA4VM

I see the Edge AI demo introduce it support V4L2 capture driver,but I search from e2e case that said TDA4 Linux SDK not support V4L2 camera driver,camera driver in RTOS SDK.

How the Edge AI demo support V4L2?

If not support V4L2,how does the gstreamer get the camera source data?

Thanks

over 4 years ago

0 Takuma Fujiwara over 4 years ago

TI__Guru 61343 points

Hello,

The Edge AI demo supports V4L2, because in the Edge AI SDK we have moved the capture and display drivers to run on the A72 cores that run Linux.

As you read, until recently we only provided the camera drivers through the RTOS SDK to run on the R5 cores, and we did not provide V4L2 support through the Linux SDK. However, starting from 8.1 version of Linux SDK and Edge AI SDK version 8.0, we provide V4L2 through Linux.

For reference, here is documentation on V4L2 in Linux SDK 8.1: https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-jacinto7/08_01_00_07/exports/docs/linux/Foundational_Components_Multimedia_D5520_VXE384.html?highlight=v4l2

Regards,

Takuma

0 xu ji over 4 years ago in reply to Takuma Fujiwara

Intellectual 335 points

from the doc I see the V4L2 support encoder and decoder,how about camera and display,does they also driver in RTOS?

0 Takuma Fujiwara over 4 years ago in reply to xu ji

TI__Guru 61343 points

Hello,

Apologies, I had some misunderstanding.

Just for Edge AI SDK specifically, we enable camera and display from A72 using Linux which allows camera based V4L2.

On the other hand, for Linux SDK and RTOS SDK, the location of the camera and display drivers are still on the RTOS side, and only the V4L2 codec has moved from RTOS to Linux.

Regards,

Takuma

0 xu ji over 4 years ago in reply to Takuma Fujiwara

Intellectual 335 points

why only Edge AI SDK enable camera and display with V4L2 ? is it any technical limited on Linux side?only support one camera on Linux side can not support deserializer that multi cameras device?

0 Takuma Fujiwara over 4 years ago in reply to xu ji

TI__Guru 61343 points

Hello,

Edge AI SDK enables camera and display with V4L2, because Edge AI SDK and the SK-TDA4VM board that it was designed for is marketed more towards industrial customers that want to abstract the RTOS layer.

For automotive customers we have seen many use QNX+RTOS for production. And since QNX does not have V4L2 available, we have the display and capture drivers available through the RTOS SDK that are used by both Linux and QNX, instead of having display and capture drivers specifically for Linux and specifically for QNX.

So I am not sure if there is any technical limitation on the Linux side, and I will have to talk to a few more people to gather the whole picture.

What is the use case of your product (for example: automotive infotainment system, ADAS, industrial application, robotics)? And are there any requirements for display and capture like multi camera?

Regards,

Takuma

0 xu ji over 4 years ago in reply to Takuma Fujiwara

Intellectual 335 points

hello，

Thanks for your detail reply,our product is vehicle control system and develop NXP soc with Linux before,now change to TDA4

our application and framework about camera and display develop are based on Linux,so if we want to merged to TDA4 that used RTOS,it is difficult

is there any plan to support deserializer with multi cameras and serializer with multi display driver in TDA4 Linux release?

0 Takuma Fujiwara over 4 years ago in reply to xu ji

TI__Guru 61343 points

Hello,

What I have gathered so far from the team is this:

For Edge AI SDK, there is a dtbo file applied during u-boot that makes Linux not use the firmware on R5 for display and capture.
For Edge AI SDK, there is also a build option that enables display and capture in Linux, BUILD_EDGEAI=yes, and this build option is documented here: http://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-sk-tda4vm/08_00_01_10/exports/docs/development_flow.html#building-the-sdk-from-source

I'm still in the process of verifying whether there are technical limitations for multi cam on Linux... What I know for sure is we support multi camera on RTOS SDK and have examples for it.

To gather more specific information:

Are you trying to run an Edge AI SDK demo-like application on the SK-TDA4VM, TDA4VM-EVM, or a custom board?
Are you trying to run just Linux+RTOS SDK, or the Edge AI SDK?

Regards,

Takuma

0 xu ji over 4 years ago in reply to Takuma Fujiwara

Intellectual 335 points

Hello

I want to use Edge AI SDK demo run on our custom board that test get frame from camera then use gstreamer push streaming to ethernet

but our board just support deserializer with camera, so the Edge AI demo can not run

both Linux+RTOS SDK and Edge AI SDK I want to run on our board for diff scene needed

0 Takuma Fujiwara over 4 years ago in reply to xu ji

TI__Guru 61343 points

Hello,

Multi camera with V4L2 is technically possible, but software-wise the drivers for FPD Link serializer and deserializer are not present in the SDK yet.

We recommend waiting until PSDK 8.4 version release in which we plan to have this feature integrated.
As an alternative, we can share the patches that will allow multi cam through V4L2. We do not recommend this option because the patches are still a work in progress, and we will not be able to provide good support on it yet. However, if you cannot wait then I will notify the developer to share the patches with you.

Regards,

Takuma