Person Tracker

This project uses multiple independent object
tracking algorithms as inputs to a single Kalman filter. A
function for estimating each algorithm's error from related
features is trained using linear regression. This error is used
as the algorithm's measurement variance. With a dynamic
measurement error covariance computed from these estimates,
we attempt to produce an overall object tracking filter that
combines each algorithm's best-case behavior while diminishing
worst-case behavior. This filter is intended to be robust without
being programmed with any environment-specific rules.

Category:

People & Blogs