kitti_depthmap

Module for converting raw lidar in kitti to depthmaps in C++/Python.

Why I wrote this

Surprisingly, I wasn't able to find any decent/simple/readable code online to figure out how to generate registered depthmaps from the kitti pointcloud data. And those that did, gave garbage results (when actually visualized as a pointcloud), or were written with for loops in python which were way too slow for a data pipeline. This codebase handles parallax errors that come from a large transformation between a sparse 3d sensor and an rgb image, and also performs resampling in spherical coordinate space, leading to more accurate results. .

To Run

mkdir build;
cd build;
cmake ..;
make
python3 testing.py # Change the path in the file to point to the kitti data

Large Image

Accounting for parallax distortion etc.

Small Image

Works well for small images too

Sampling Modes (Lower)

Selectively simulate a higher or lower resolution lidar in spherical space

Upsampling (Higher)

Selectively simulate a higher or lower resolution lidar in spherical space