Quick intro

The kinematic-tracker is available on PyPI.org

Having the tracker installed, we can use the tracker class NdKkfTracker. The class NdKkfTracker allows to define a state $\mathbf{x}$ composed of a set of N-dimensional point-motion models. For example, to track bounding boxes in 2D using constant-velocity for their centers and constant-position for their sizes we could play with the following script

from kinematic_tracker import NdKkfTracker
import numpy as np

tracker = NdKkfTracker([2, 1], [2, 2])
tracker.set_measurement_cov( 4 * np.eye(4))
tracker.advance(12345, np.array([[1.0, 2.0, 0.1, 0.2]]))
print(tracker.tracks_c[0])

In this script we defined the tracker to have states $\mathbf{x}$ consisting of two 2D parts: positions and sizes. The layout of the state $\mathbf{x}$ will be $\mathbf{x} = (p_x, v_x, p_y, v_y, s_x, s_y)$.

The measurement vector $\mathbf{z}$ excludes any derivatives by default $\mathbf{z} = (p_x, p_y, s_x, s_y)$.

We used a diagonal measurement noise covariance $\mathrm{R}=4.0 \mathrm{I}$.

At first step, we provided one measurement vector at time stamp 12345 nanoseconds.

As a result we should be able to see current list of targets which includes just one track

[TrackNdKkf(x = [1.  0.  2.  0.  0.1 0.2])]

The tracks are sorted by (object) class index. By default there is just one class, so the list of targets should be accessed via tracker.tracks_c[0].

Internally, we use OpenCV Kalman filters. We can access other variables of the target such as its prior and posterior error covariances

...
print(tracker.tracks_c[0][0].kkf.kalman_filter.errorCovPre)
print(tracker.tracks_c[0][0].kkf.kalman_filter.errorCovPost)