Visual Odometry, Dataset

347 views Asked by At

I am working with VO (Visual Odometry) I don't understand many things, for example, is a dataset always needed, I want to use VO but I don't want to use a Kitti Dataset, I want to use the algorithm implemented in my drone, and my drone will be flying in my neighborhood (that's why I don't want to use Kitti Dataset), in case a dataset is always needed, how to do it, how to get the poses?

1

There are 1 answers

0
J. Smith On

The purpose of the KITTI dataset is two-fold. First, it's a standardized set of images and LIDAR data that researchers use in order to compare the relative performance of different algorithms. Second -- and most importantly for your case -- it's also a source of ground truth to debug or analyze your algorithm.

So, if you want to use visual odometry in your drone: pick a VO algorithm that will work on your drone hardware. Get it working on your desktop computer, using KITTI data to debug. That is: make sure your VO algorithm reports the same position as the KITTI ground truth for the sequence you are using. Note that most VO algorithms require stereo cameras, and many also use the IMU in order to generate better results.

This is a big project; don't expect quick results in a day or even a week. Work carefully, document your process, and be prepared to fail over and over again until it works.