Hello Stereolabs team,
I am working with the ZED Depth SDK and would like to better understand how distance is calculated for detected objects, especially large objects such as buses.
When a bus is detected and a bounding box is generated around it, which point or region is used for the reported distance? Is it based on the center of the bounding box, the closest visible point, the object’s 3D centroid, or another method?
My concern is that if the distance is calculated from the bounding box center, then for a bus viewed at an angle, the reported distance may correspond to the side or middle of the bus rather than the closest part to the cameras.
Is it possible to choose which part of the object is used for distance measurement? For example, can I measure the distance specifically to the front of the bus, or to the closest valid depth point inside the ROI?
I also have a question regarding a custom stereo rig with a large baseline, for example 2 meters, using two ZED X One cameras. If the bus is very close to the cameras, the left camera may see more of the left side/front of the bus, while the right camera may see more of the right side/front of the bus. In this case, how does the Depth SDK handle the distance measurement? Could this cause the measured depth to correspond to different visible parts of the bus in each camera?
I reviewed the minimum distance table in this article:
If my setup respects the recommended minimum distance for the selected baseline, should I expect reliable depth measurements on large nearby objects such as buses?
Thank you.