Why does the bayes_people_tracker use non-linear filters (EKF,UKF) for linear models

asked 2019-04-18 03:42:32 -0600

stuggibo gravatar image

updated 2019-04-18 03:49:58 -0600


I'm curious of why the bayes_people_tracker package of the strands project is using non-linear filters for estimating the motion of humans, as they have a linear motion and observation model.


In the official paper they also state:

"In all of the examined tracking systems, person detections arrive in their sensor-specific coordinate frame and are in- stantaneously transformed into a locally fixed frame (based upon robot odometry) that does not move with the robot. This ensures that the motion prediction of tracked persons is independent from the robot’s ego-motion. In the resulting set of measurements Z = {z1, ..., zn} ⊂ R2, we drop the z coordinate as we only track in 2D world coordinates."

Link to paper: https://ieeexplore.ieee.org/abstract/...

So why aren't they just using the "normal" Kalman filter, instead of the particle fitler, or extended kalman filters. Or am I missing something? ?

Any help would be greatly appreciated

edit retag flag offensive close merge delete