Slow rendering when streaming a video using image object

General discussion
1

Hello,

I am new to the LVGL library and forum, so please bear with me in case of incomplete details in my question.

Basically, I am trying to stream a video on the display with a raspberry pi 4, but with the current approach/configuration I am able to render only ~5-6 FPS, which adds lagging as time increases.
Here is the setup and implementation:

  • I am using RPi 4, 4 GB variant and using Frame buffer for rending on the HDMI display.
  • I have created two threads, 1st for receiving the RTSP frame and pushing it into the queue.
  • The second thread has the below tasks in the main loop
    1. It polls the queue, copies the frame, and updates the image source
    2. calls lv_timer_handler
    3. sleep for 1 millisecond

Problem and Observations so far:

  • lv_timer_handler() takes around 200-400 msec random delay in each iteration after updating the image source. Due to this thread 2 is not able to read at the required FPS.
  • Polling, copying, and updating lv_img_set_src is taking around 4 msec (based on the lv_log time difference) in thread 2. After this event lv_timer_handler takes a lot of time to render the new image on the screen.
  • I have removed the polling call and tested thread two without updating any image, in that case, lv_timer_handler() is taking 20-25 msec.
  • Checked the flush_cb time as well, it is taking 3-4 msec

Firstly, is it an expected delay in lv_timer_handler?
Has anyone faced similar latency in rending, if not please suggest how to reduce the delay.
Has anyone already tried video streaming?

Thanks in advance,

Best regards,
Fakhkrudin

2

Hi,

This random delay sounds very strange and it’s certainly not the expected behavior.

Can you use ffmpeg? If so LVGL has a FFmpeg support — LVGL documentation. Could you it try out?

3

Hi @kisvegabor,

Thanks for the reply, ill check out the FFmpeg feature and see if that also helps.

I figured out the mistake I had made in my code by referring to this post. I created a new image object on each RTSP frame received instead of just calling the lv_obj_invalidate after updating the source. Now I can see better rendering with some delay around 500-600ms.

Do you suggest FFmpeg will give better results?

Thanks again,
Fakhruddin

4

It’s great!

FFmpeg already works and integrated, so there shouldn’t be issues similar to what you have just fixed.

1 Like