High CPU Usage with 2 Femto Mega Cameras - GPU Acceleration Options?

[ozeryeden]

System Environment

• Hardware Info:
  • CPU - ARMv8 Processor rev 1 (v8l) × 8
  • Memory size - 64GB
  • GPU - NVIDIA Tegra Orin (nvgpu)/integrated
• Ubuntu Version: Ubuntu 20.04.6 LTS
• ROS Version: ROS Noetic
• Camera Model: 2 × Femto Mega i
• Firmware Version: 2.0.4

Description

I’m encountering high CPU usage when running two Femto Mega i cameras with the OrbbecSDK ROS1 wrapper and would like to explore options for GPU acceleration or other performance optimizations.

Depth-only streaming performs well at 30 Hz using roughly one CPU core per camera. However, enabling color or registered streams significantly increases CPU usage (over three cores per camera) and causes frame rate drops. My goal is to achieve stable registered (depth + color) output at around 10 Hz without excessive CPU load.

Does the Femto Mega support GPU-accelerated depth registration or colored point-cloud generation?

Can my Jetson’s GPU be utilized for registration/alignment tasks?

Does the camera offer any onboard processing capabilities to offload these computations?

Thank you for your help and any guidance you can provide.

[obshenlanzhou]

@ozeryeden I tried to reproduce your problem using the following configuration, but everything seems to be working fine on my end; the CPU usage of both camera nodes is about 9% after enabling the depth_registration parameter. Below is my configuration information.

Can you provide the commit ID? And what configurations did you make to your camera?