def match(self, new_human_detections, new_pose_estimations,
new_face_detections, new_facial_landmarks, new_facial_emotions):
face_indices = self.match_faces_with_body(new_human_detections,
new_face_detections)
distances = np.array([[Util.bbox_distance_iou(new_human_detections[i].bounding_box, self.humans[j].human_bounding_box_path[-1].bounding_box) \
for j in range(len(self.humans))]
for i in range(len(new_human_detections))])
if distances.shape[0] == 0:
row_ind, col_ind = [], []
else:
row_ind, col_ind = scipy.optimize.linear_sum_assignment(-distances)
unmatched_tracking_indices = []
for j in range(len(self.humans)):
# unmatched tracking
if j not in col_ind:
unmatched_tracking_indices.append(j)
matched_detection_indices, unmatched_detection_indices = [], []
for i in range(len(new_human_detections)):
# matched detection
if i in row_ind:
j = col_ind[np.where(row_ind == i)[0][0]]
matched_detection_indices.append([i, j])
# unmatched detection
else:
unmatched_detection_indices.append(i)
humans = []
for inds in matched_detection_indices:
i, j = inds
human = self.humans[j]
human.human_bounding_box_path.append(new_human_detections[i])
human.human_pose_path.append(new_pose_estimations[i])
human.face_bounding_box_path.append(
new_face_detections[face_indices[i]])
human.facial_landmarks_path.append(
new_facial_landmarks[face_indices[i]])
human.facial_emotion_path.append(
new_facial_emotions[face_indices[i]])
humans.append(human)
for ind in unmatched_detection_indices:
bounding_box = new_human_detections[ind]
pose = new_pose_estimations[ind]
face_bounding_box = new_face_detections[face_indices[ind]]
facial_landmarks = new_facial_landmarks[face_indices[ind]]
facial_emotion = new_facial_emotions[face_indices[ind]]
humans.append(
Person(bounding_box, pose, face_bounding_box, facial_landmarks,
facial_emotion))
self.humans = humans
def match(self, new_face_bboxes, new_facial_landmarks):
distances = np.array([[Util.bbox_distance_iou(new_face_bboxes[i].bounding_box, self.humans[j].face_bounding_box_path[-1].bounding_box) \
for j in range(len(self.humans))]
for i in range(len(new_face_bboxes))])
if distances.shape[0] == 0:
row_ind, col_ind = [], []
else:
row_ind, col_ind = linear_sum_assignment(- distances)
unmatched_tracking_indices = []
for j in range(len(self.humans)):
# unmatched tracking
if j not in col_ind:
unmatched_tracking_indices.append(j)
matched_detection_indices, unmatched_detection_indices = [], []
for i in range(len(new_face_bboxes)):
# matched detection
if i in row_ind:
j = col_ind[np.where(row_ind == i)[0][0]]
matched_detection_indices.append([i, j])
# unmatched detection
else:
unmatched_detection_indices.append(i)
humans = []
for inds in matched_detection_indices:
i, j = inds
human = self.humans[j]
human.face_bounding_box_path.append(new_face_bboxes[i])
human.facial_landmarks_path.append(new_facial_landmarks[i])
humans.append(human)
self.humans = humans
