Merge pull request #1488 from mikel-brostrom/fix-kf-name

xys now xysr kf

boxmot/motion/kalman_filters/xyah_kf.py

@@ -13,18 +13,24 @@ def __init__(self):
         super().__init__(ndim=4)
 
     def _get_initial_covariance_std(self, measurement: np.ndarray) -> np.ndarray:
+        # initial uncertainty in the aspect ratio is very low,
+        # suggesting that it is not expected to vary significantly.
         return [
-            2 * self._std_weight_position * measurement[3],
-            2 * self._std_weight_position * measurement[3],
-            1e-2,
-            2 * self._std_weight_position * measurement[3],
-            10 * self._std_weight_velocity * measurement[3],
-            10 * self._std_weight_velocity * measurement[3],
-            1e-5,
-            10 * self._std_weight_velocity * measurement[3]
+            2 * self._std_weight_position * measurement[3],     # x
+            2 * self._std_weight_position * measurement[3],     # y
+            1e-2,                                               # a (aspect ratio)
+            2 * self._std_weight_position * measurement[3],     # H
+            10 * self._std_weight_velocity * measurement[3],    # vx
+            10 * self._std_weight_velocity * measurement[3],    # vy
+            1e-5,                                               # va (aspect ration vel)
+            10 * self._std_weight_velocity * measurement[3]     # vh
         ]
 
     def _get_process_noise_std(self, mean: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
+        # very small process noise standard deviation assigned to the
+        # aspect ratio state and its velocity. Suggests
+        # that the aspect ratio is expected to remain relatively constant
+        # over time.
         std_pos = [
             self._std_weight_position * mean[3],
             self._std_weight_position * mean[3],
@@ -40,6 +46,9 @@ def _get_process_noise_std(self, mean: np.ndarray) -> Tuple[np.ndarray, np.ndarr
         return std_pos, std_vel
 
     def _get_measurement_noise_std(self, mean: np.ndarray, confidence: float) -> np.ndarray:
+        # small measurement noise standard deviation for
+        # aspect ratio state, indicating low expected measurement noise in
+        # the aspect ratio.
         std_noise = [
             self._std_weight_position * mean[3],
             self._std_weight_position * mean[3],

boxmot/motion/kalman_filters/xys_kf.py → boxmot/motion/kalman_filters/xysr_kf.py

@@ -1,6 +1,4 @@
-# -*- coding: utf-8 -*-
-# pylint: disable=invalid-name, too-many-arguments, too-many-branches,
-# pylint: disable=too-many-locals, too-many-instance-attributes, too-many-lines
+
 
 """
 This module implements the linear Kalman filter in both an object
@@ -106,9 +104,9 @@
 from filterpy.common import pretty_str, reshape_z
 from collections import deque
 
-
-# The aspect ratio r is assumed to be constant.
-class KalmanFilterXYS(object):
+# x_t is a seven-tuple, i.e. x_t =[u, v, s, r, u,˙ v,˙ s˙]^⊤,
+# r (aspect ratio) is assumed to be constant.
+class KalmanFilterXYSR(object):
     """ Implements a Kalman filter. You are responsible for setting the
     various state variables to reasonable values; the defaults  will
     not give you a functional filter.

boxmot/trackers/deepocsort/deep_ocsort.py

@@ -5,7 +5,7 @@
 
 from boxmot.appearance.reid_auto_backend import ReidAutoBackend
 from boxmot.motion.cmc import get_cmc_method
-from boxmot.motion.kalman_filters.xys_kf import KalmanFilterXYS
+from boxmot.motion.kalman_filters.xysr_kf import KalmanFilterXYSR
 from boxmot.utils.association import associate, linear_assignment
 from boxmot.utils.iou import get_asso_func
 from boxmot.trackers.basetracker import BaseTracker
@@ -107,7 +107,7 @@ def __init__(self, det, delta_t=3, emb=None, alpha=0, new_kf=False, max_obs=50):
         self.det_ind = det[6]
 
         if new_kf:
-            self.kf = KalmanFilterXYS(dim_x=8, dim_z=4, max_obs=max_obs)
+            self.kf = KalmanFilterXYSR(dim_x=8, dim_z=4, max_obs=max_obs)
             self.kf.F = np.array(
                 [
                     # x y w h x' y' w' h'

boxmot/trackers/hybridsort/hybridsort.py

@@ -131,8 +131,8 @@ def __init__(
         """
         # define constant velocity model
         # if not orig and not args.kalman_GPR:
-        from boxmot.motion.kalman_filters.xys_kf import KalmanFilterXYS
-        self.kf = KalmanFilterXYS(dim_x=9, dim_z=5, max_obs=max_obs)
+        from boxmot.motion.kalman_filters.xysr_kf import KalmanFilterXYSR
+        self.kf = KalmanFilterXYSR(dim_x=9, dim_z=5, max_obs=max_obs)
 
         # u, v, s, c, r, ~u, ~v, ~s, ~c
         self.kf.F = np.array([[1, 0, 0, 0, 0, 1, 0, 0, 0],

boxmot/trackers/ocsort/ocsort.py

@@ -7,7 +7,7 @@
 from collections import deque
 
 
-from boxmot.motion.kalman_filters.xys_kf import KalmanFilterXYS
+from boxmot.motion.kalman_filters.xysr_kf import KalmanFilterXYSR
 from boxmot.utils.association import associate, linear_assignment
 from boxmot.utils.iou import get_asso_func
 from boxmot.utils.iou import run_asso_func
@@ -80,7 +80,7 @@ def __init__(self, bbox, cls, det_ind, delta_t=3, max_obs=50):
         """
         # define constant velocity model
         self.det_ind = det_ind
-        self.kf = KalmanFilterXYS(dim_x=7, dim_z=4, max_obs=max_obs)
+        self.kf = KalmanFilterXYSR(dim_x=7, dim_z=4, max_obs=max_obs)
         self.kf.F = np.array(
             [
                 [1, 0, 0, 0, 1, 0, 0],