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

• Distance of 1/W (or 1/D, 1/H) are all approximately linear with distance.

Technical details

• 2000 bbox are used. Distance GT measured with laser scanner.

Notes

• We can add the backplane width for better estimation of depth.
• The method to extract GT information from point cloud may be noisy. But how to quantify and avoid this?