# Segments metric

The idea of this metric is to divide the estimated trajectory into segments. Each segment is ended when the error between the estimation and the groundtruth surpasses some error tolerance, here are two examples with tolerances 5cm and 1cm. The idea is to study these segments: how many are there in a dataset? what's their length? are there any particular "type of segment ending" that is recurrent and would need fixing in the algorithm? etc I think this metric will be quite better than the ATE/RTE ones, although I still need to finish the details

Hopefully, this can give us a better idea of tracking accuracy than usual ATE/RTE metrics.