项目作者: ptyshevs

项目描述 :
Align Lidar Pointcloud on RGB camera 🥶
高级语言: Jupyter Notebook
项目地址: git://github.com/ptyshevs/lidar_align.git
创建时间: 2019-10-21T14:12:34Z
项目社区:https://github.com/ptyshevs/lidar_align

开源协议:

下载


Data

Data is collected by KITTI project.
You need to download 2011_09_26_drive_0002 and extract it in this directory. (synced+rectified data)
You will also need calibration parameters.

Paper: link

Experiment

The idea was simple - take one frame from KITTI data and introduce artificial bias in rotation and translation of velo2cam. Then use
the process decribed in a paper to find correction in R and T.
Simulated annealing approach is used (Random Search + Suboptimal solution acceptance with diminishing probability).

Results

I’ve failed to reproduce the results. It seems that my implementation of loss function is not correct. Euclidean distance between true rotation/translation and predicted ones increases as the number of iterations raises.