def track(img):
"""
Tracks people without detection
Args:
img (np.array): current frame
start_time (float): time when the current frame was captured
Returns:
np: current frame with bounding boxes
"""
print('[TRACKING]')
global persons_list
# Predict state for each tracker without measurement
# TODO what to do if the box disappeared
for person in persons_list:
person.update_state(img)
x = int((person.box[0] + person.box[2]) / 2)
y = int((person.box[1] + person.box[3]) / 2)
center = (y, x)
img = cv2.circle(img, center, 7, (0, 255, 0), -1)
img = tools.draw_box(person.id, img, person.box)
return img
def predict(img, start_time):
"""
Tracks people without detection
Args:
img (np.array): current frame
start_time (float): time when the current frame was captured
Returns:
np: current frame with bounding boxes
"""
global persons_list
# Predict state for each tracker without measurement
for person in persons_list:
person.predict_state(time.time() - start_time)
x = person.x.T[0].tolist()
x = [
x[0] - int(person.height / 2), x[1] - int(person.width / 2),
x[0] + int(person.height / 2), x[1] + int(person.width / 2)
]
person.box = x
x = person.x.T[0].tolist()[0]
y = person.x.T[0].tolist()[1]
img = cv2.circle(img, (y, x), 4, (0, 255, 0), -1)
img = tools.draw_box(person.id, img, person.box)
return img
def detect(img):
"""
Runs person detection on img
Args:
img (np.array): input image
Returns:
np.array: image with bounding boxes
"""
detections = det.get_detections(img)
for detection in detections:
img = tools.draw_box(0, img, detection)
return img
def predict(img, start_time):
"""
Tracks people without detection
Args:
img (np.array): current frame
start_time (float): time when the current frame was captured
Returns:
np: current frame with bounding boxes
"""
# Predict state for each tracker without measurement
for person in persons_list:
person.predict_state(time.time() - start_time)
x = person.x
x = x.T[0].tolist()
x = [x[0], x[2], x[4], x[6]]
person.box = x
img = tools.draw_box(person.id, img, person.box)
return img
def track(img, start_time):
"""
Tracks people with detection
Args:
img (np.array): current frame
start_time (float): time when the current frame was captured
Returns:
np: current frame with bounding boxes
"""
global persons_list
global next_id
detections = det.get_detections(img)
persons_boxes = []
if len(persons_list) > 0:
for person in persons_list:
persons_boxes.append(person.box)
matched, unmatched_dets, unmatched_trks \
= assign_detections_to_trackers(persons_boxes, detections, iou_thrd=IOU_THRESH)
# Update trackers with measurements
if matched.size > 0:
for person_idx, det_idx in matched:
# Extract measurement
z = detections[det_idx]
z = np.expand_dims(z, axis=0).T
# Extract person tracker
person = persons_list[person_idx]
person.process_measurement(z, time.time() - start_time)
x = person.x.T[0].tolist()
x = [x[0], x[2], x[4], x[6]]
persons_boxes[person_idx] = x
person.box = x
person.hits += 1
person.misses = 0
# Update trackers without measurements
if len(unmatched_trks) > 0:
for person_idx in unmatched_trks:
person = persons_list[person_idx]
person.misses += 1
person.hits = 0
person.predict_state(time.time() - start_time)
x = person.x
x = x.T[0].tolist()
x = [x[0], x[2], x[4], x[6]]
person.box = x
persons_boxes[person_idx] = x
# Create tracker for new detections
if len(unmatched_dets) > 0:
for idx in unmatched_dets:
z = detections[idx]
z = np.expand_dims(z, axis=0).T
person = tracker.Tracker()
x = np.array([[z[0], 0, z[1], 0, z[2], 0, z[3], 0]]).T
person.x = x
# tmp_trk.predict_state(time.time() - start_time)
x = person.x
x = x.T[0].tolist()
x = [x[0], x[2], x[4], x[6]]
person.box = x
# Assign an ID for the tracker
next_id += 1
person.id = next_id
persons_list.append(person)
persons_boxes.append(x)
# Draw good boxes
for person in persons_list:
if ((person.hits >= MIN_HITS) and (person.misses <= MAX_MISSES)):
img = tools.draw_box(person.id, img, person.box)
# Remove lost trackers
persons_list = [i for i in persons_list if i.misses <= MAX_MISSES]
return img
def detect(img, start_time, det):
"""
Tracks people with detection
Args:
img (np.array): current frame
start_time (float): time when the current frame was captured
Returns:
np: current frame with bounding boxes
"""
print('[DETECTION]')
global persons_list
global next_id
# Get detections
detections = det.get_detections(img)
# Create list of existing persons' boxes
persons_boxes = []
if len(persons_list) > 0:
for person in persons_list:
persons_boxes.append(person.box)
# Solve data association problem
matched, unmatched_dets, unmatched_trks \
= assign_detections_to_trackers(persons_boxes, detections, dst_thrd=DISTANCE_THRESH)
# Update trackers with measurements
if matched.size > 0:
for person_idx, det_idx in matched:
# Extract measurement
z = detections[det_idx]
center = [
z[0] + int((z[2] - z[0]) / 2), z[1] + int((z[3] - z[1]) / 2)
]
img = cv2.circle(img, (center[1], center[0]), 6, (0, 0, 255), -1)
person.hits += 1
person.misses = 0
person.box = tuple(z) # NEW
# Update trackers without measurements
# TODO check removed unmatched trackers
for person_idx in unmatched_trks:
person = persons_list[person_idx]
person.misses += 1
person.hits = 0
# Create tracker for new detections
for idx in unmatched_dets:
z = detections[idx]
person = trc.Tracker(tuple(z), img)
#person.box = z
# Assign an ID for the tracker
next_id += 1
person.id = next_id
persons_list.append(person)
persons_boxes.append(z)
# Draw good boxes
for person in persons_list:
if ((person.hits >= MIN_HITS) and (person.misses <= MAX_MISSES)):
x = int((person.box[1] + person.box[3]) / 2)
y = int((person.box[0] + person.box[2]) / 2)
img = cv2.circle(img, (y, x), 4, (0, 255, 0), -1)
img = tools.draw_box(person.id, img, person.box)
# Remove lost trackers
persons_list = [i for i in persons_list if i.misses <= MAX_MISSES]
return img
def track(img, start_time):
"""
Tracks people with detection
Args:
img (np.array): current frame
start_time (float): time when the current frame was captured
Returns:
np: current frame with bounding boxes
"""
global persons_list
global next_id
# Get detections
detections = det.get_detections(img)
# Create list of existing persons' boxes
persons_boxes = []
if len(persons_list) > 0:
for person in persons_list:
persons_boxes.append(person.box)
# Solve data association problem
matched, unmatched_dets, unmatched_trks \
= assign_detections_to_trackers(persons_boxes, detections, dst_thrd=DISTANCE_THRESH)
# Update trackers with measurements
if matched.size > 0:
for person_idx, det_idx in matched:
# Extract measurement
z = detections[det_idx]
center = [
z[0] + int((z[2] - z[0]) / 2), z[1] + int((z[3] - z[1]) / 2)
]
img = cv2.circle(img, (center[1], center[0]), 6, (0, 0, 255), -1)
center = np.expand_dims(center, axis=0).T
# Extract person tracker
person = persons_list[person_idx]
person.hits += 1
person.misses = 0
person.process_measurement(center, time.time() - start_time)
person.width = z[3] - z[1]
person.height = z[2] - z[0]
x = person.x.T[0].tolist()
x = [
x[0] - int(person.height / 2), x[1] - int(person.width / 2),
x[0] + int(person.height / 2), x[1] + int(person.width / 2)
]
persons_boxes[person_idx] = x
person.box = x
# Update trackers without measurements
if len(unmatched_trks) > 0:
for person_idx in unmatched_trks:
person = persons_list[person_idx]
person.misses += 1
person.hits = 0
person.predict_state(time.time() - start_time)
x = person.x.T[0].tolist()
x = [
x[0] - int(person.height / 2), x[1] - int(person.width / 2),
x[0] + int(person.height / 2), x[1] + int(person.width / 2)
]
persons_boxes[person_idx] = x
person.box = x
# Create tracker for new detections
if len(unmatched_dets) > 0:
for idx in unmatched_dets:
z = detections[idx]
center = [
z[0] + int((z[2] - z[0]) / 2), z[1] + int((z[3] - z[1]) / 2)
]
center = np.expand_dims(center, axis=0).T
person = tracker_center.Tracker()
person.x = np.expand_dims([center[0], center[1], 0, 0], axis=0).T
person.width = z[3] - z[1]
person.height = z[2] - z[0]
x = person.x.T[0].tolist()
x = [
x[0] - int(person.height / 2), x[1] - int(person.width / 2),
x[0] + int(person.height / 2), x[1] + int(person.width / 2)
]
person.box = x
# Assign an ID for the tracker
next_id += 1
person.id = next_id
persons_list.append(person)
persons_boxes.append(x)
# Draw good boxes
for person in persons_list:
if ((person.hits >= MIN_HITS) and (person.misses <= MAX_MISSES)):
x = person.x.T[0].tolist()[0]
y = person.x.T[0].tolist()[1]
img = cv2.circle(img, (y, x), 4, (0, 255, 0), -1)
img = tools.draw_box(person.id, img, person.box)
# Remove lost trackers
persons_list = [i for i in persons_list if i.misses <= MAX_MISSES]
return img