def test_ObjectTracker_prune_too_invisible(object_now, tracker_now):
now = datetime.now()
object_now.side_effect = [
now, # create
now, # create
now, # create
now, # create
]
tracker_now.side_effect = [
# update 1
now, # _prune
# update 2
now + timedelta(seconds=10), # _prune
]
ot = ObjectTracker(5)
tos = [TrackedObject(np.array([0, 1, 2])), TrackedObject(np.array([0, 1, 2]))]
ot.update(copy.deepcopy(tos))
assert len(ot._objects) == 2
ot._objects[0].last_seen = now + timedelta(seconds=9)
ot._objects[1].last_seen = now + timedelta(seconds=3.9)
ot.update([])
assert len(ot._objects) == 1
def test_ObjectTracker_update_multiple():
ot = ObjectTracker(5)
tos = [TrackedObject(np.array([0, 1, 2])), TrackedObject(np.array([0, 1, 2]))]
ot.update(tos)
assert len(ot._objects) == 2
assert ot._objects[0].id == 0
assert ot._objects[1].id == 1
def test_simple_update_10_pct_delta(object_now, tracker_now):
now = datetime.datetime.now()
object_now.side_effect = [
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
]
tracker_now.side_effect = [
now + datetime.timedelta(seconds=1), # _track
now + datetime.timedelta(seconds=1), # _prune
now + datetime.timedelta(seconds=1), # _track
now + datetime.timedelta(seconds=1), # _prune
]
objects1 = [
TrackedObject(np.array([0, 0, 0])),
TrackedObject(np.array([1, 0, 0])),
TrackedObject(np.array([0, 1, 0])),
TrackedObject(np.array([1, 1, 0]))
]
objects2 = [
TrackedObject(np.array([0.1, 0, 0])),
TrackedObject(np.array([0.9, 0, 0])),
TrackedObject(np.array([0.1, 1, 0])),
TrackedObject(np.array([0.9, 1, 0]))
]
tracker = SimpleTracker(deletion_threshold=5)
tracker.update(copy.deepcopy(objects1))
tracked_objects1 = copy.deepcopy(tracker.get_objects())
tracker.update(copy.deepcopy(objects2))
tracked_objects2 = copy.deepcopy(tracker.get_objects())
assert len(tracked_objects1) == len(tracked_objects2)
obj_map = {obj.id: obj for obj in tracked_objects2}
for i in range(len(objects1)):
assert np.linalg.norm(tracked_objects1[i].pos -
obj_map[tracked_objects1[i].id].pos) < .2
def test_ObjectTracker_prune_too_old():
dt = datetime.now()
ot = ObjectTracker(5)
tos = [TrackedObject(np.array([0, 1, 2])), TrackedObject(np.array([0, 1, 2]))]
ot.update(tos)
assert len(ot._objects) == 2
ot._objects[1].created = dt - timedelta(seconds=100)
ot._objects[1].last_seen = dt - timedelta(seconds=6)
ot.update([])
assert len(ot._objects) == 1
def _update(update: Dict, clients: List[socket.socket]):
people = [
TrackedObject(np.asarray([person["x"], person["y"], person["z"]]))
for person in update
]
tracking_state.update(people)
tracked = tracking_state.get_objects()
update_dict = {
"objects": [{
"x": person.pos[0],
"y": person.pos[1],
"z": person.pos[2],
"id": person.id
} for person in tracked]
}
bson_data = bson.dumps(update_dict)
length = len(bson_data)
size_data = struct.pack(size_fmt, length)
data = size_data + bson_data
excepted = []
for client in clients:
try:
client.sendall(data)
except Exception:
excepted.append(client)
for client in excepted:
clients.remove(client)
def test_kalman_update_stable_association(object_now, tracker_now):
now = datetime.datetime.now()
object_now.side_effect = [
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
]
tracker_now.side_effect = [
now + datetime.timedelta(seconds=1), # _predict
now + datetime.timedelta(seconds=1), # _track
now + datetime.timedelta(seconds=1), # _prune
now + datetime.timedelta(seconds=2), # _predict
now + datetime.timedelta(seconds=2), # _track
now + datetime.timedelta(seconds=2), # _prune
]
objects = [
TrackedObject(np.array([0, 0, 0])),
TrackedObject(np.array([1, 0, 0])),
TrackedObject(np.array([0, 1, 0])),
TrackedObject(np.array([1, 1, 0]))
]
tracker = KalmanTracker(deletion_threshold=5)
tracker.update(copy.deepcopy(objects))
tracked_objects1 = copy.deepcopy(tracker.get_objects())
tracker.update(copy.deepcopy(objects))
tracked_objects2 = copy.deepcopy(tracker.get_objects())
assert len(tracked_objects1) == len(tracked_objects2)
obj_map = {obj.id: obj for obj in tracked_objects1}
for i in range(len(objects)):
assert np.array_equal(tracked_objects2[i].pos,
obj_map[tracked_objects2[i].id].pos)
def test_TrackedObject_creation():
time_thresh = timedelta(milliseconds=100)
to = TrackedObject(np.array([0, 1, 2]))
dt = datetime.now()
assert np.array_equal(to.pos, np.array([0, 1, 2]))
assert dt - to.last_seen < time_thresh
assert to.id is None
assert to.tracking_ctx is None
assert dt - to.created < time_thresh
assert to.history == []
def test_TrackedObject_notequal_different_ids():
to1 = TrackedObject(np.array([0, 1, 2]))
to2 = TrackedObject(np.array([0, 1, 2]))
to1.id = 1
to2.id = 2
assert to1 != to2
def test_TrackedObject_equal():
to1 = TrackedObject(np.array([0, 1, 2]))
to2 = TrackedObject(np.array([0, 1, 2]))
to1.id = 1
to2.id = 1
assert to1 == to2
def test_kalman_update(object_now, tracker_now):
now = datetime.datetime.now()
object_now.side_effect = [
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
]
tracker_now.side_effect = [
now + datetime.timedelta(seconds=1), # _predict
now + datetime.timedelta(seconds=1), # _track
now + datetime.timedelta(seconds=1), # _prune
]
objects = [
TrackedObject(np.array([0, 0, 0])),
TrackedObject(np.array([1, 0, 0])),
TrackedObject(np.array([0, 1, 0])),
TrackedObject(np.array([1, 1, 0]))
]
tracker = KalmanTracker(deletion_threshold=5)
tracker.update(copy.deepcopy(objects))
tracked_objects = tracker.get_objects()
assert len(tracked_objects) == len(objects)
for i in range(len(objects)):
assert any(
np.array_equal(obj.pos, objects[i].pos) for obj in tracked_objects)
def test_TrackedObject_notequal_different_types():
to1 = TrackedObject(np.array([0, 1, 2]))
to2 = object()
assert to1 != to2
def test_kalman_update_predictive_tracking(object_now, tracker_now):
now = datetime.datetime.now()
object_now.side_effect = [
now, # create 1
now, # create 1
now, # create 1
now, # create 1
now, # create 1
now, # create 1
now, # create 1
now, # create 1
now + datetime.timedelta(seconds=1), # create 2
now + datetime.timedelta(seconds=1), # create 2
now + datetime.timedelta(seconds=1), # create 2
now + datetime.timedelta(seconds=1), # create 2
now + datetime.timedelta(seconds=1), # create 2
now + datetime.timedelta(seconds=1), # create 2
now + datetime.timedelta(seconds=1), # create 2
now + datetime.timedelta(seconds=1), # create 2
now + datetime.timedelta(seconds=2), # create 3
now + datetime.timedelta(seconds=2), # create 3
now + datetime.timedelta(seconds=2), # create 3
now + datetime.timedelta(seconds=2), # create 3
now + datetime.timedelta(seconds=2), # create 3
now + datetime.timedelta(seconds=2), # create 3
now + datetime.timedelta(seconds=2), # create 3
now + datetime.timedelta(seconds=2), # create 3
]
tracker_now.side_effect = [
now + datetime.timedelta(seconds=1), # _predict
now + datetime.timedelta(seconds=1), # _track
now + datetime.timedelta(seconds=1), # _prune
now + datetime.timedelta(seconds=2), # _predict
now + datetime.timedelta(seconds=2), # _track
now + datetime.timedelta(seconds=2), # _prune
now + datetime.timedelta(seconds=3), # _predict
now + datetime.timedelta(seconds=3), # _track
now + datetime.timedelta(seconds=3), # _prune
]
objects1 = [
TrackedObject(np.array([0, 0, 0])),
TrackedObject(np.array([1, 0, 0])),
TrackedObject(np.array([0, 1, 0])),
TrackedObject(np.array([1, 1, 0]))
]
objects2 = [
TrackedObject(np.array([0, .3, 0])),
TrackedObject(np.array([1, .3, 0])),
TrackedObject(np.array([0, .7, 0])),
TrackedObject(np.array([1, .7, 0]))
]
objects3 = [
TrackedObject(np.array([0, .6, 0])),
TrackedObject(np.array([1, .6, 0])),
TrackedObject(np.array([0, .4, 0])),
TrackedObject(np.array([1, .4, 0]))
]
tracker = KalmanTracker(deletion_threshold=5)
tracker.update(copy.deepcopy(objects1))
tracked_objects_start = copy.deepcopy(tracker.get_objects())
tracker.update(copy.deepcopy(objects2))
tracker.update([])
tracked_objects_final = copy.deepcopy(tracker.get_objects())
assert len(tracked_objects_start) == len(tracked_objects_final)
index_map = {}
for obj in tracked_objects_start:
index = [
i for i, elem in enumerate(objects1)
if np.array_equal(elem.pos, obj.pos)
][0]
index_map[obj.id] = index
for obj in tracked_objects_final:
index = index_map[obj.id]
assert np.all(abs(objects3[index].pos - obj.pos) < .001)
def test_kalman_update_10_pct_delta(object_now, tracker_now):
'''
This tests the case where an update is closer to the a different object in the previous test_kalman_update
'''
now = datetime.datetime.now()
object_now.side_effect = [
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
now, # create
]
tracker_now.side_effect = [
now + datetime.timedelta(seconds=1), # _predict
now + datetime.timedelta(seconds=1), # _track
now + datetime.timedelta(seconds=1), # _prune
now + datetime.timedelta(seconds=2), # _predict
now + datetime.timedelta(seconds=2), # _track
now + datetime.timedelta(seconds=2), # _prune
]
objects1 = [
TrackedObject(np.array([0, 0, 0])),
TrackedObject(np.array([1, 0, 0])),
TrackedObject(np.array([0, 1, 0])),
TrackedObject(np.array([1, 1, 0]))
]
objects2 = [
TrackedObject(np.array([0.1, 0, 0])),
TrackedObject(np.array([0.9, 0, 0])),
TrackedObject(np.array([0.1, 1, 0])),
TrackedObject(np.array([0.9, 1, 0]))
]
tracker = KalmanTracker(deletion_threshold=5)
tracker.update(copy.deepcopy(objects1))
tracked_objects1 = copy.deepcopy(tracker.get_objects())
tracker.update(copy.deepcopy(objects2))
tracked_objects2 = copy.deepcopy(tracker.get_objects())
assert len(tracked_objects1) == len(tracked_objects2)
obj_map = {obj.id: obj for obj in tracked_objects2}
for i in range(len(objects1)):
assert np.linalg.norm(tracked_objects1[i].pos -
obj_map[tracked_objects1[i].id].pos) < .2
