Zed2i odometry on an empty parking


I’m using Zed2i camera outdoor with the zed_ros_wrapper (ros melodic). When I measure the estimated distance traveled by the camera in an empty parking, the estimated distance is half the real distance.
When I do the same test on a parking with cars, the estimated distance is correct.

I guess the estimated odometry is not correct because there is no reference point in the empty parking while cars are good reference points for the computation when the parking is not empty. Is it correct ?
What is the best position for the camera to compute this visual odometry ?


Van Malleghem Antoine

Hi @avanmalleghem
Welcome to the Stereolabs community.

What you say is correct, when you perform a visual odometry task, the more visual features are available the most the final result is good.

You can try to improve the performance in the empty park by increasing the maximum depth value:

Thanks for your answer !
I tried but it still gives the same issue.
What I don’t understand is that the odometry covariance stays quite low (x10^-6 or 7) even when it drifts a lot. Is it normal ? What do you suggest ?

@avanmalleghem can you provide two frames, one of the empty parking and the other of the full parking, so I can compare them and understand how to improve the results in the first condition?

Here is the empty parking situation :

If needed, I can provide a full rosbag of each situation. In this case, provide an email address to me.

Thanks in advance for your answer.

Here is the full parking situation (can’t provide 2 embedded images in the same comment)

Hi @avanmalleghem
you can send comparative SVOs to support@stereolabs.com.
Please link this thread in the text of the email.

Done @Myzhar
Thanks for your time :slight_smile:

1 Like