Interpretation of deepsort source code (VII)

deepsort Project address

---

Catalog
deepsort Principle and analysis summary

---

tracker.py

# vim: expandtab:ts=4:sw=4
from __future__ import absolute_import
import numpy as np
from . import kalman_filter
from . import linear_assignment
from . import iou_matching
from .track import Track


class Tracker:
    """
    This is the multi-target tracker.

    Parameters
    ----------
    metric : nn_matching.NearestNeighborDistanceMetric
        A distance metric for measurement-to-track association.
    max_age : int
        Maximum number of missed misses before a track is deleted.
    n_init : int
        Number of consecutive detections before the track is confirmed. The
        track state is set to `Deleted` if a miss occurs within the first
        `n_init` frames.

    Attributes
    ----------
    metric : nn_matching.NearestNeighborDistanceMetric
        The distance metric used for measurement to track association.
    max_age : int
        Maximum number of missed misses before a track is deleted.
    n_init : int
        Number of frames that a track remains in initialization phase.
    kf : kalman_filter.KalmanFilter
        A Kalman filter to filter target trajectories in image space.
    tracks : List[Track]
        The list of active tracks at the current time step.

    """

    def __init__(self, metric, max_iou_distance=0.7, max_age=30, n_init=3):
        # metric Is a class , Used to calculate the distance ( Cosine distance or Mahalanobis distance )
        self.metric = metric
        # iou Maximum value when matching 
        self.max_iou_distance = max_iou_distance
        #  Unmatched track Maximum mismatch times , That is, directly specify cascading matching cascade_depth Parameters 
        self.max_age = max_age
        #  from unconfirmed track To confirmed track The minimum number of matches 
        self.n_init = n_init

        #  Kalman filter 
        self.kf = kalman_filter.KalmanFilter()
        #  Save a series of tracks 
        self.tracks = []
        #  Next assigned track id
        self._next_id = 1

    def predict(self):
        """Propagate track state distributions one time step forward.

        This function should be called once every time step, before `update`.
        """
        #  First frame ,tracks It's empty , Do not execute forecast 
        for track in self.tracks:
            track.predict(self.kf)

    def update(self, detections):
        """Perform measurement update and track management.

        Parameters
        ----------
        detections : List[deep_sort.detection.Detection]
            A list of detections at the current time step.

        """
        # Run matching cascade.
        matches, unmatched_tracks, unmatched_detections = \
            self._match(detections)

        # Update track set.
        #  For the result of matching , to update track
        for track_idx, detection_idx in matches:
            self.tracks[track_idx].update(
                self.kf, detections[detection_idx])
        #  For unmatched track, call mark_missed marked 
        for track_idx in unmatched_tracks:
            self.tracks[track_idx].mark_missed()
        #  For unmatched detection, Set up new track And initialize , Put in track list 
        for detection_idx in unmatched_detections:
            self._initiate_track(detections[detection_idx])
        #  Traverse to get the latest tracks list , What is kept is confirmed And unconfirmed State of tracks
        #  contain detection New track
        self.tracks = [t for t in self.tracks if not t.is_deleted()]

        # Update distance metric.
        #  Get all confirmed State of track id
        active_targets = [t.track_id for t in self.tracks if t.is_confirmed()]
        features, targets = [], []
        for track in self.tracks:
            if not track.is_confirmed():
                continue
            features += track.features
            targets += [track.track_id for _ in track.features]
            track.features = []
        self.metric.partial_fit(
            np.asarray(features), np.asarray(targets), active_targets)

    #  Firstly, the Mahalanobis distance based on appearance information is calculated tracks and detections The cost matrix of ,
    #  Then cascade matching and IOU matching , Finally, all matching pairs of the current frame are obtained 、 Unmatched tracks And unmatched detections
    def _match(self, detections):

        # track For all track
        # dets by detections
        # track_indices by 
        def gated_metric(tracks, dets, track_indices, detection_indices):
            #  Characteristics of all tested objects 
            features = np.array([dets[i].feature for i in detection_indices])
            #  Tracking object id
            targets = np.array([tracks[i].track_id for i in track_indices])
            #  What is returned here is the cost matrix calculated using cosine distance 
            cost_matrix = self.metric.distance(features, targets)
            # gate_cost_matrix, Calculate the Mahalanobis distance 
            cost_matrix = linear_assignment.gate_cost_matrix(
                self.kf, cost_matrix, tracks, dets, track_indices,
                detection_indices)
            #  Returns the cost matrix processed by cosine distance and Mahalanobis distance 
            return cost_matrix

        # Split track set into confirmed and unconfirmed tracks.
        confirmed_tracks = [
            i for i, t in enumerate(self.tracks) if t.is_confirmed()]
        unconfirmed_tracks = [
            i for i, t in enumerate(self.tracks) if not t.is_confirmed()]

        # Associate confirmed tracks using appearance features.
        #  Make a cascade match （ Special , about track It's empty time , Do not perform cascading matching ）
        matches_a, unmatched_tracks_a, unmatched_detections = \
            linear_assignment.matching_cascade(
                gated_metric, self.metric.matching_threshold, self.max_age,
                self.tracks, detections, confirmed_tracks)

        #  Conduct iou Matching calculation 
        # Associate remaining tracks together with unconfirmed tracks using IOU.
        #  Only calculate time_since_update be equal to 1 Of track（ Special , Don't do it for the first frame iou Handle ）
        iou_track_candidates = unconfirmed_tracks + [
            k for k in unmatched_tracks_a if
            self.tracks[k].time_since_update == 1]
        #  Screen out non-conforming products iou Matching track
        unmatched_tracks_a = [
            k for k in unmatched_tracks_a if
            self.tracks[k].time_since_update != 1]
        #  Conduct iou matching 
        matches_b, unmatched_tracks_b, unmatched_detections = \
            linear_assignment.min_cost_matching(
                iou_matching.iou_cost, self.max_iou_distance, self.tracks,
                detections, iou_track_candidates, unmatched_detections)

        matches = matches_a + matches_b
        unmatched_tracks = list(set(unmatched_tracks_a + unmatched_tracks_b))
        return matches, unmatched_tracks, unmatched_detections

    #  Initialize the first frame 
    def _initiate_track(self, detection):
        #  By detection target detection Coordinates of （(center x, center y, aspect ratio, height) structure 
        #  Mean vector （1×8 Vector , Initialize to [center_x,center_y,w/h,h,0,0,0,0] And covariance matrix (8*8 matrix , A diagonal matrix constructed according to the height of the target ）
        mean, covariance = self.kf.initiate(detection.to_xyah())
        # Track() Itself is a class , Here, each new target is characterized by a class , Put in tracks list 
        self.tracks.append(Track(
            mean, covariance, self._next_id, self.n_init, self.max_age,
            detection.feature))
        self._next_id += 1