Improve depth sensing to match Intel L515 result

Hello,

In order to replace Intel L515, I test the ZED 2i.
Basically our scene is a green plant seen from top.

With L515 with the help of the distance it was easy to filter accurately the green plant because L515 give sharp measure of the distance.

With the test of the ZED 2i , I receive a depth image that is not very sharp compared to L515.

Here a comparison :

Does it exist better settings to improve depth image to reach the performance of L515.?

Hello,

What is the program you are running on the ZED ? You can tune the parameters to get a better depth : minimal distance, tolerance, etc. And of course, NEURAL depth mode is the way to go in your situation.

Antoine

Thanks Antoine for your answer,

I built my own program based on ZED SDK 3.7

I tested the following sense mode :

  • ultra ( the one you see on the picture)
  • quality (not successfull because it looked like the image was blurrred and I wanted more details)
  • neural with no result at all because i received only NAN the the resulting matrix. (I mentionned it in this post : ZED Depth Mode Neural with Jetson Nano)

I did not set minimum and maximum distance (I only did it after acquisition using numpy clip function)

Here the images with the different mode (also tested with the standard / fill flag):

Hello Antoine,

I fixed the issue with the Neural, I will repeat the same experiment to see how it improves the result.

Should I set the minimum and maximum distance to improve the results or does it only reduce the computation time ?

Hi @bpasserat
minimum and maximum values do not impact the quality of the depth map.
They are only thresholding filters to remove values outside the range of work of your algorithm.
For your tests, I suggest you also upgrade to ZED SDK v3.8 which add improvements to AI Depth estimation (NEURAL mode).