| 13 | |
| 14 | |
| 15 | class STrack(BaseTrack): |
| 16 | shared_kalman = KalmanFilter() |
| 17 | def __init__(self, tlwh, score, temp_feat, buffer_size=30): |
| 18 | |
| 19 | # wait activate |
| 20 | self._tlwh = np.asarray(tlwh, dtype=np.float) |
| 21 | self.kalman_filter = None |
| 22 | self.mean, self.covariance = None, None |
| 23 | self.is_activated = False |
| 24 | |
| 25 | self.score = score |
| 26 | self.tracklet_len = 0 |
| 27 | |
| 28 | self.smooth_feat = None |
| 29 | self.update_features(temp_feat) |
| 30 | self.features = deque([], maxlen=buffer_size) |
| 31 | self.alpha = 0.9 |
| 32 | |
| 33 | def update_features(self, feat): |
| 34 | feat /= np.linalg.norm(feat) |
| 35 | self.curr_feat = feat |
| 36 | if self.smooth_feat is None: |
| 37 | self.smooth_feat = feat |
| 38 | else: |
| 39 | self.smooth_feat = self.alpha * self.smooth_feat + (1 - self.alpha) * feat |
| 40 | self.features.append(feat) |
| 41 | self.smooth_feat /= np.linalg.norm(self.smooth_feat) |
| 42 | |
| 43 | def predict(self): |
| 44 | mean_state = self.mean.copy() |
| 45 | if self.state != TrackState.Tracked: |
| 46 | mean_state[7] = 0 |
| 47 | self.mean, self.covariance = self.kalman_filter.predict(mean_state, self.covariance) |
| 48 | |
| 49 | @staticmethod |
| 50 | def multi_predict(stracks): |
| 51 | if len(stracks) > 0: |
| 52 | multi_mean = np.asarray([st.mean.copy() for st in stracks]) |
| 53 | multi_covariance = np.asarray([st.covariance for st in stracks]) |
| 54 | for i, st in enumerate(stracks): |
| 55 | if st.state != TrackState.Tracked: |
| 56 | multi_mean[i][7] = 0 |
| 57 | multi_mean, multi_covariance = STrack.shared_kalman.multi_predict(multi_mean, multi_covariance) |
| 58 | for i, (mean, cov) in enumerate(zip(multi_mean, multi_covariance)): |
| 59 | stracks[i].mean = mean |
| 60 | stracks[i].covariance = cov |
| 61 | |
| 62 | def activate(self, kalman_filter, frame_id): |
| 63 | """Start a new tracklet""" |
| 64 | self.kalman_filter = kalman_filter |
| 65 | self.track_id = self.next_id() |
| 66 | self.mean, self.covariance = self.kalman_filter.initiate(self.tlwh_to_xyah(self._tlwh)) |
| 67 | |
| 68 | self.tracklet_len = 0 |
| 69 | self.state = TrackState.Tracked |
| 70 | if frame_id == 1: |
| 71 | self.is_activated = True |
| 72 | # self.is_activated = True |