Best brutal force SLAM configuration about pyslam HOT 1 CLOSED

Hi,
Thanks for your feedback. As explained in the main README, the script main_vo.py is a first start to understanding the basics of inter-frame feature tracking and camera pose estimation (there is no triangulation and mapping there). You cannot use it without a ground truth. Not sure about the results you are getting.

Pure rotations do break monocular VO/SLAM (such as pyslam). Indeed, pure rotations prevent triangulation, so no new map points if you look at unexplored/unmapped areas. That means, no map no tracking.

Low-light condition is another great challenge for indirect (feature-based) tracking. Many SLAM systems have a difficult time to track in such conditions. In some cases, direct (based on photometric error) tracking (or a mixed approach) can give you a more robust performance.

Unfortunately, I didn't yet have time to identify the BEST brutal force SLAM configuration. It's a work in progress. Feature selection is just a part of the full system. There are many other parameters to tune. As you may know, SLAM systems are complex.
Please, consider that pyslam is a playground to test local features within a VO/SLAM system. You can have fun yourselves by testing, tweaking, and improving. Let me know.

Cheers,
Luigi

from pyslam.