RDK X5 4-Channel 1080P Camera Recording Test

General
1

We ran a 4-channel 1080P camera recording test on RDK X5, including S230AI stereo + two USB cameras. The goal was to check long-run recording stability, H.264 saving, storage behavior, frame-rate behavior, and camera timing difference.

img_v3_02130_ed8efb8c-8d57-4509-8dc9-073a809f33eg
img_v3_02130_ed8efb8c-8d57-4509-8dc9-073a809f33eg1278×1256 248 KB

Test Setup

Board: RDK X5
Duration: 4 hours
Resolution: 1920x1080 for all 4 channels
Output format: H.264
Storage policy: rolling cleanup around 10 GB

Camera configuration:

Channel Camera Input Save / Encode Path
1 S230AI stereo left NV12 H.264 via hobot_codec
2 S230AI stereo right NV12 H.264 via hobot_codec
3 UGREEN USB Camera MJPEG hardware MJPEG decode + hardware H.264 encode via sp_codec
4 Huawei USB Camera H.264 H.264 stream copy

4-Hour Recording Result

The 4-hour run completed successfully.

2026-06-25 15:23:10 duration reached
2026-06-25 15:23:16 finished

Final storage status:

  • Retained video directory: about 11 GB
  • Deleted by rolling cleanup: 254 files, 96.68 GB
  • Root filesystem after test: 58 GB total, 32 GB used, 24 GB available

Frame / stream summary:

Channel Result
S230AI left avg 28.995 fps, 415,509 frames
S230AI right avg 28.929 fps, 414,156 frames
UGREEN USB 355,024 captured / decoded / encoded, avg 24.654 fps, 0 errors
Huawei USB ran about 4 hours, speed=1x, normal exit

UGREEN single-camera test:

duration=180s
captured=5261
decoded=5261
encoded=5261
avg_fps=29.224
dq_err=0
dec_err=0
enc_err=0

So the UGREEN camera can run close to 30 fps alone, but drops to about 24.65 fps under the full 4-channel workload. This looks more like multi-camera resource contention than a basic single-camera failure.

Storage / System Observations

iostat:

  • Average write throughput: 7.57 MB/s
  • Peak write throughput: 50.74 MB/s
  • Average write await: 654.78 ms
  • Peak write await: 5361.93 ms
  • Peak disk utilization: 101.64%

mpstat:

  • Average user CPU: 23.90%
  • Average system CPU: 5.82%
  • Average iowait: 5.776%
  • Peak iowait: 57.11%

CPU load became acceptable after replacing software x264 with hardware encode for UGREEN. The main concern is storage latency spikes.

Camera Timing / Web Mosaic Test

We also added a 4-camera web mosaic tool. It displays:

  • age=xxxms: freshness of latest frame
  • rx=HH:MM:SS.mmm: board-side receive time
  • board_time=...mmm: board time when mosaic is rendered

Observed examples:

  • Board-side receive-time difference across 4 cameras: about 27-31 ms
  • Web display freshness difference: about 148-206 ms
  • Visible stopwatch comparison showed UGREEN could differ from S230AI by around 100 ms in one sample

Important note: rx is board receive time, not true exposure timestamp. For real exposure-level sync, hardware timestamps or a dedicated visual timing target are needed.

Conclusion

The 4-channel recording pipeline completed the 4-hour test successfully. S230AI stereo and the Huawei native H.264 USB camera were stable. The UGREEN USB camera worked correctly through hardware MJPEG decode and H.264 encode with no decode/encode errors, but it did not sustain 30 fps in the 4-channel test.

For a stable production 4x1080P30 H.264 setup, using USB cameras with native H.264 output is recommended. Storage latency should also be monitored, since write-await spikes were observed during the long run.

Code Package

The test scripts and startup guide are attached:

  • rdk_4cam_storage_test.sh
  • h26x_segment_writer.py
  • ugreen_hw_transcode.cpp
  • four_cam_web_mosaic.py
  • README with build/start/monitor instructions
    RDK_X5_4Cam_Test_Package.zip (14.9 KB)