Learning-Deep-Learning

DisNet: A novel method for distance estimation from monocular camera

October 2019

tl;dr: Summary of the main idea.

Overall impression

Use bbox H, W, D (diagnal), average size of object h, w, b (breadth, along depth dimension) is good enough to regress the distance, with relative error ~10%, up to 300 meters.

This method seems much more promising than the one presented in object distance estimation.

This idea is quite similar to the more elaborate ICCV 2019 paper monoloco.

Key ideas

Technical details

Notes