Hello all,
I’m pretty new to the topic of computer vision but am diving in by trying to create my own VIO algorithm based on the Fast-MSCKF implementation published here ([2210.08117] An Improved Multi-State Constraint Kalman Filter for Visual-Inertial Odometry). I’m trying to use the triangulate points function from the sfm module to estimate the 3d positions of features that I’m tracking in the scene. the documentation on the modules is a bit sparse so I’m confused on what the input to the function should be. namely should the vector of points be undistorted, normalized points, or raw pixel coordinates.
I’ve tried both options and the results do not look good.
Looking for anyone out there that has used this module before!