Autotracking tweaks (#8400)

* optimize motion and velocity estimation

* change recommended fps and fix config validate

* remove unneeded var

* process at most 3 objects per second

* fix test
This commit is contained in:
Josh Hawkins 2023-11-01 06:12:43 -05:00 committed by GitHub
parent d1620b4e39
commit af24eb7dbf
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 39 additions and 40 deletions

View File

@ -113,7 +113,7 @@ If you initially calibrate with zooming disabled and then enable zooming at a la
Every PTZ camera is different, so autotracking may not perform ideally in every situation. This experimental feature was initially developed using an EmpireTech/Dahua SD1A404XB-GNR.
The object tracker in Frigate estimates the motion of the PTZ so that tracked objects are preserved when the camera moves. In most cases (especially for faster moving objects), the default 5 fps is insufficient for the motion estimator to perform accurately. 10 fps is the current recommendation. Higher frame rates will likely not be more performant and will only slow down Frigate and the motion estimator. Adjust your camera to output at least 10 frames per second and change the `fps` parameter in the [detect configuration](index.md) of your configuration file.
The object tracker in Frigate estimates the motion of the PTZ so that tracked objects are preserved when the camera moves. In most cases 5 fps is sufficient, but if you plan to track faster moving objects, you may want to increase this slightly. Higher frame rates (> 10fps) will only slow down Frigate and the motion estimator and may lead to dropped frames, especially if you are using experimental zooming.
A fast [detector](object_detectors.md) is recommended. CPU detectors will not perform well or won't work at all. You can watch Frigate's debug viewer for your camera to see a thicker colored box around the object currently being autotracked.

View File

@ -171,7 +171,7 @@ class PtzAutotrackConfig(FrigateBaseModel):
timeout: int = Field(
default=10, title="Seconds to delay before returning to preset."
)
movement_weights: Optional[Union[float, List[float]]] = Field(
movement_weights: Optional[Union[str, List[str]]] = Field(
default=[],
title="Internal value used for PTZ movements based on the speed of your camera's motor.",
)

View File

@ -60,7 +60,7 @@ REQUEST_REGION_GRID = "request_region_grid"
# Autotracking
AUTOTRACKING_MAX_AREA_RATIO = 0.5
AUTOTRACKING_MAX_AREA_RATIO = 0.6
AUTOTRACKING_MOTION_MIN_DISTANCE = 20
AUTOTRACKING_MOTION_MAX_POINTS = 500
AUTOTRACKING_MAX_MOVE_METRICS = 500

View File

@ -248,10 +248,8 @@ class TrackedObject:
if self.obj_data["frame_time"] - self.previous["frame_time"] > 60:
significant_change = True
# update autotrack at half fps
if self.obj_data["frame_time"] - self.previous["frame_time"] > (
1 / (self.camera_config.detect.fps / 2)
):
# update autotrack at most 3 objects per second
if self.obj_data["frame_time"] - self.previous["frame_time"] >= (1 / 3):
autotracker_update = True
self.obj_data.update(obj_data)

View File

@ -215,8 +215,8 @@ class PtzAutoTracker:
maxlen=round(camera_config.detect.fps * 1.5)
)
self.tracked_object_metrics[camera] = {
"max_target_box": 1
- (AUTOTRACKING_MAX_AREA_RATIO ** self.zoom_factor[camera])
"max_target_box": AUTOTRACKING_MAX_AREA_RATIO
** (1 / self.zoom_factor[camera])
}
self.calibrating[camera] = False
@ -268,6 +268,10 @@ class PtzAutoTracker:
if camera_config.onvif.autotracking.movement_weights:
if len(camera_config.onvif.autotracking.movement_weights) == 5:
camera_config.onvif.autotracking.movement_weights = [
float(val)
for val in camera_config.onvif.autotracking.movement_weights
]
self.ptz_metrics[camera][
"ptz_min_zoom"
].value = camera_config.onvif.autotracking.movement_weights[0]
@ -521,7 +525,11 @@ class PtzAutoTracker:
camera_height = camera_config.frame_shape[0]
# Extract areas and calculate weighted average
areas = [obj["area"] for obj in self.tracked_object_history[camera]]
# grab the largest dimension of the bounding box and create a square from that
areas = [
max(obj["box"][2] - obj["box"][0], obj["box"][3] - obj["box"][1]) ** 2
for obj in self.tracked_object_history[camera]
]
filtered_areas = (
remove_outliers(areas)
@ -686,19 +694,20 @@ class PtzAutoTracker:
camera_height = camera_config.frame_shape[0]
camera_fps = camera_config.detect.fps
# estimate_velocity is a numpy array of bbox top,left and bottom,right velocities
velocities = obj.obj_data["estimate_velocity"]
logger.debug(f"{camera}: Velocities from norfair: {velocities}")
# if we are close enough to zero, return right away
if np.all(np.round(velocities) == 0):
return True, np.zeros((2, 2))
return True, np.zeros((4,))
# Thresholds
x_mags_thresh = camera_width / camera_fps / 2
y_mags_thresh = camera_height / camera_fps / 2
dir_thresh = 0.93
delta_thresh = 12
var_thresh = 5
delta_thresh = 20
var_thresh = 10
# Check magnitude
x_mags = np.abs(velocities[:, 0])
@ -722,7 +731,6 @@ class PtzAutoTracker:
np.linalg.norm(velocities[0]) * np.linalg.norm(velocities[1])
)
dir_thresh = 0.6 if np.all(delta < delta_thresh / 2) else dir_thresh
print(f"cosine sim: {cosine_sim}")
invalid_dirs = cosine_sim < dir_thresh
# Combine
@ -752,10 +760,10 @@ class PtzAutoTracker:
)
)
# invalid velocity
return False, np.zeros((2, 2))
return False, np.zeros((4,))
else:
logger.debug(f"{camera}: Valid velocity ")
return True, np.mean(velocities, axis=0)
return True, velocities.flatten()
def _get_distance_threshold(self, camera, obj):
# Returns true if Euclidean distance from object to center of frame is
@ -836,7 +844,7 @@ class PtzAutoTracker:
# ensure object is not moving quickly
below_velocity_threshold = np.all(
np.abs(average_velocity)
< np.array([velocity_threshold_x, velocity_threshold_y])
< np.tile([velocity_threshold_x, velocity_threshold_y], 2)
) or np.all(average_velocity == 0)
below_area_threshold = (
@ -938,7 +946,7 @@ class PtzAutoTracker:
camera_height = camera_config.frame_shape[0]
camera_fps = camera_config.detect.fps
average_velocity = np.zeros((2, 2))
average_velocity = np.zeros((4,))
predicted_box = obj.obj_data["box"]
centroid_x = obj.obj_data["centroid"][0]
@ -966,7 +974,6 @@ class PtzAutoTracker:
# this box could exceed the frame boundaries if velocity is high
# but we'll handle that in _enqueue_move() as two separate moves
current_box = np.array(obj.obj_data["box"])
average_velocity = np.tile(average_velocity, 2)
predicted_box = (
current_box
+ camera_fps * predicted_movement_time * average_velocity
@ -1010,7 +1017,10 @@ class PtzAutoTracker:
zoom = 0
result = None
current_zoom_level = self.ptz_metrics[camera]["ptz_zoom_level"].value
target_box = obj.obj_data["area"] / (camera_width * camera_height)
target_box = max(
obj.obj_data["box"][2] - obj.obj_data["box"][0],
obj.obj_data["box"][3] - obj.obj_data["box"][1],
) ** 2 / (camera_width * camera_height)
# absolute zooming separately from pan/tilt
if camera_config.onvif.autotracking.zooming == ZoomingModeEnum.absolute:
@ -1061,24 +1071,15 @@ class PtzAutoTracker:
< self.tracked_object_metrics[camera]["max_target_box"]
else self.tracked_object_metrics[camera]["max_target_box"]
)
zoom = (
2
* (
limit
/ (
self.tracked_object_metrics[camera]["target_box"]
+ limit
)
)
- 1
)
ratio = limit / self.tracked_object_metrics[camera]["target_box"]
zoom = (ratio - 1) / (ratio + 1)
logger.debug(f"{camera}: Zoom calculation: {zoom}")
if not result:
# zoom out with special condition if zooming out because of velocity, edges, etc.
zoom = -(1 - zoom) if zoom > 0 else -(zoom + 1)
zoom = -(1 - zoom) if zoom > 0 else -(zoom * 2 + 1)
if result:
# zoom in
zoom = 1 - zoom if zoom > 0 else (zoom + 1)
zoom = 1 - zoom if zoom > 0 else (zoom * 2 + 1)
logger.debug(f"{camera}: Zooming: {result} Zoom amount: {zoom}")
@ -1199,8 +1200,8 @@ class PtzAutoTracker:
)
self.tracked_object[camera] = None
self.tracked_object_metrics[camera] = {
"max_target_box": 1
- (AUTOTRACKING_MAX_AREA_RATIO ** self.zoom_factor[camera])
"max_target_box": AUTOTRACKING_MAX_AREA_RATIO
** (1 / self.zoom_factor[camera])
}
def camera_maintenance(self, camera):

View File

@ -1651,11 +1651,11 @@ class TestConfig(unittest.TestCase):
runtime_config = frigate_config.runtime_config()
assert runtime_config.cameras["back"].onvif.autotracking.movement_weights == [
0,
1,
1.23,
2.34,
0.50,
"0.0",
"1.0",
"1.23",
"2.34",
"0.5",
]
def test_fails_invalid_movement_weights(self):