October 2019
tl;dr: Summary of 3DOD methods based on monocular images, lidars and sensor fusion methods of the two.
The review is updated as of 2018. However there have been a lot of progress of mono 3DOD in 2019. I shall write a review of mono 3DOD soon.
- The main drawback of monocular methods is the lack of depth cues, which limits detection and localization accuracy, especially for far and occluded objects.
- Most mono 3DOD methods have shifted towards a learned paradigm for RPN and second stage of of 3D model matching and reprojection to obtain 3D Box.