Road quality recognition system in mining quarry based on ZED 2i + jetson nano + ZED SDK + YOLO + SLAM + IMU

Hello, I’m working on a project to detect road obstacles and surface defects in a mining quarry. I chose to use a combination of ZED 2i and Jetson Nano, but I’m having trouble configuring and running the necessary services (YOLO + SLAM + IMU) so that they work together and send the output to a cloud server.

I’m looking for help or consultation both in choosing the right libraries and in configuring the ZED 2i + Jetson Nano setup to ensure that the following processing pipeline works with sufficient performance:

  1. The camera captures video and streams it to the Jetson Nano → SDK
  2. The SDK integrates YOLO + SLAM + IMU to detect road defects in the background
  3. Road section parameters (coordinates, timestamp, vehicle speed) + an image of the detected defect are transmitted to a cloud server

Hi @aslevandovsky
Welcome to the Stereolabs community.

What performance should be expected from this application?
The Jetson Nano appears insufficiently powerful to support this demanding configuration.

Hi,
It is necessary for the hardware to process data from the camera + IMU in the background, in parallel with the SLAM process. Other users of the system say that for such performance the camera and Jetson Nano need to be switched to a special operating mode, but I do not understand how exactly to do this physically.

Hi @aslevandovsky
the Jetson Nano is an old entry-level model with low compute power capability.
You can maximize the performance by setting the power mode to 0:
sudo nvpmodel -m 0
and running the jetsonclocks script:
sudo jetson_clocks.sh

Thank you very much!

Is it possible to contact you to ask more questions on this topic? I understand that the level of knowledge of my team is currently quite low and the questions may be too general and we need a person who will point the questions in the right direction.

You can contact us at support@stereolabs.com for direct support.