TDA4VH-Q1: uyvy cam to U16

Kim Semin

Prodigy 150 points

Part Number: TDA4VH-Q1

Tool/software:

Hello, I am using sdk 9.2 version.My camera format is a uyvy format

But I want to access each pixel through vxMapImagePatch and process the image I want

before I enter the display node after capturing in single camera application

obj->capt_yuv_image = (vx_image)vxGetObjectArrayItem(obj->cap_frames[0], 0);

...


    for(i=0; i<frm_loop_cnt; i++)
     {
         vx_image test_image;
         appPerfPointBegin(&obj->total_perf);
         graph_parameter_num = 0;
         if(status == VX_SUCCESS)
         {
             status = vxGraphParameterDequeueDoneRef(obj->graph, graph_parameter_num, (vx_reference*)&out_capture_frames, 1, &num_refs_capture);
         }
         graph_parameter_num++;
         if((status == VX_SUCCESS) && (obj->test_mode == 1))
         {
             status = vxGraphParameterDequeueDoneRef(obj->graph, 1, (vx_reference*)&test_image, 1, &num_refs_capture);
         }
 
         if((status == VX_SUCCESS) && (obj->mosaic_enable == 1) && obj->mosaic_enqueue == 1)
         {
             status = vxGraphParameterDequeueDoneRef(obj->graph, 1, (vx_reference*)&obj->imgMosaicObj.output_image, 1, &num_refs_capture);
             graph_parameter_num++;
         }
 

     vx_image image = (vx_image)vxGetObjectArrayItem(out_capture_frames, 0);
     
     converToU16(image);
     
     
...



obj->display_image = obj->capt_yuv_image;

However, as a result of the code modification,

the display screen remains unchanged, and the afterimage remains as much as 4 frames, which is the number of buffers

If i capture using the 's' key,

the first image applies well and the second image comes out as the uyvy image and repeats with this rule

I want to modify the result of the capture node and put it in the display node and spray it on the display. Which part is wrong?

thank you

3 months ago

0 Gang Hua 3 months ago

TI__Guru 59740 points

Hi Kim,

What do you mean by "uyvy cam to U16"?

Kim Semin said:
But I want to access each pixel through vxMapImagePatch and process the image I want

That is possible and typically done by creating a new openvx node for your desired processing and add the processing node between capture and display nodes.

Kim Semin said:
the display screen remains unchanged, and the afterimage remains as much as 4 frames, which is the number of buffers

vx_image image = (vx_image)vxGetObjectArrayItem(out_capture_frames, 0);
 
 converToU16(image);
...

obj->display_image = obj->capt_yuv_image;

I suppose you would have to "get", "process", and then "release" those images so that the entire pipeline can keep running.
In your code above, I am not sure if you have done that properly.
If you create a new opnevx node for your own processing, the buffer management can be handled by the openvx framework automatically.

Our S/W expert Gokul S is copied for his comments and any further questions you may have.

0 Kim Semin 3 months ago in reply to Gang Hua

Prodigy 150 points

Hello
As you suggested, I created a custom node between the capture node and the display node and applied the algorithm I wanted.

As a result, I confirmed that it comes out on the display as I want it to.

However, there is a small problem, but there is an issue where a slight horizontal line only appears and disappears in the area where there is movement.

Custom nodes are in the form of vx_image, both input and output.

capture node -> obj->cap_frames[0] -> vx_image -> custom node -> vx_image -> display node

What would be a problem

Thank you

0 Gang Hua 3 months ago in reply to Kim Semin

TI__Guru 59740 points

Hi Kim,

Kim Semin said:
As you suggested, I created a custom node between the capture node and the display node and applied the algorithm I wanted.

Thanks for the update and that is good news!

Kim Semin said:
but there is an issue where a slight horizontal line only appears and disappears in the area where there is movement.

That sounds like a synchronization or performance overload issue to me.
Theoretically, the vx_image should be completely updated before it is sent to display to avoid these horizontal lines.

Hi Gokul S,
Could you please advise?

0 Kim Semin 3 months ago in reply to Gang Hua

Prodigy 150 points

Hello
The camera frame is normally 30fps and there seems to be no delay on the display.

What I'm curious about is that if I convert capture output vx_object_array to vx_image,

put it as node input, receive the output as vx_image,

and hand it over to the display, will the contents of the four capture buffers be delivered to the display?

0 Gang Hua 3 months ago in reply to Kim Semin

TI__Guru 59740 points

Hi Kim,

If you looked the sample code below, you can see that we take the input image of a node from the capture array and create an image for the node output.
That should work well for your node as well.

https://git.ti.com/cgit/processor-sdk/vision_apps/tree/apps/utilities/app_heterogeneous/app_heterogeneous.c?h=main#n539

            raw_viss[ci] = (tivx_raw_image) vxGetObjectArrayItem(capture_frames[ci][0], 0);
            yuv_viss[ci] = vxCreateImage(context, width[ci], height[ci], VX_DF_IMAGE_NV12);

Your video problem is typically a s/w issue in your node implementation, e.g., cache coherence.

Processors

Processors forum

TDA4VH-Q1: uyvy cam to U16