Feasibility of 3D size estimation of costum objects with a ZED camera


Currently I’m looking into the feasibility of 3D size estimation (Length, Width, and Height) of costum objects. The situation in which these objects occur is controlled (assembly line in a factory), and the objects are generally physical boxes with rectangular shapes.

According to (Using the Object Detection API with a Custom Detector | Stereolabs) “From there, the ZED can output the 2D bounding boxes of the objects and accurately compute the 3D bounding boxes with the help of the depth map.” 3D size estimation seems possible with any 2D costum object detector with a ZED camera. However, It is of great importance that the length, width, and height are within a margin of error of 1 millimeter.

Is it possible to achieve this with a costum 2D object detection model in combination with ZED (and the python ZED API)?
Moreover, is it also possible to get the dimensions from the inferenced 3D bounding box via the API? (

Kind Regards,


Hi @DatSplit
what you explain can be done with customized software that filters the 3D data.
The 1mm margin of error is truly stringent, maybe you can obtain such precision by fusion the information from multiple sensors.

1 Like

Thanks for your concise and clear reply!

1 Like