Python Perception Types

# simplified
import horus
from horus import Image, Detection, DetectionList, BoundingBox2D, Topic

model = load_yolo("model.pt")
sub_image = Topic(Image, "camera.rgb")
pub_detections = Topic(DetectionList, "detections")

def detector_tick(node):
    img = sub_image.recv()
    if img is not None:
        results = model.predict(img.to_numpy())

        detections = DetectionList()
        for r in results:
            det = Detection(
                label=r.class_name,
                confidence=r.score,
                bbox=BoundingBox2D(
                    x=r.x, y=r.y,
                    width=r.w, height=r.h,
                ),
            )
            detections.add(det)

        pub_detections.send(detections)

detector = horus.Node(name="detector", tick=detector_tick, rate=30,
                      subs=["camera.rgb"], pubs=["detections"])

# simplified
from horus import (
    BoundingBox2D, Detection, DetectionList,
    PointXYZ, PointXYZRGB, PointCloudBuffer,
    Landmark, Landmark3D, LandmarkArray,
    TrackedObject, COCOPose,
)

# simplified
bbox = BoundingBox2D(x=10.0, y=20.0, width=100.0, height=200.0)
bbox = BoundingBox2D.from_center(cx=60.0, cy=120.0, width=100.0, height=200.0)

# simplified
det = Detection(class_name="person", confidence=0.95,
                x=10.0, y=20.0, width=100.0, height=200.0)

det = Detection.from_bbox(bbox, class_name="car", confidence=0.87)

# simplified
detections = DetectionList()
detections.append(Detection("person", 0.95, 10, 20, 100, 200))
detections.append(Detection("car", 0.72, 300, 150, 80, 60))
detections.append(Detection("person", 0.45, 500, 100, 90, 180))

# Filter by confidence
confident = detections.filter_confidence(0.7)  # 2 detections

# Filter by class
people = detections.filter_class("person")  # 2 detections

# Iterate
for det in detections:
    print(f"{det.class_name}: {det.confidence:.2f}")

# Index access
first = detections[0]
count = len(detections)

# Convert to dicts (for JSON/logging)
dicts = detections.to_dicts()

# Serialization
data = detections.to_bytes()
restored = DetectionList.from_bytes(data)

# simplified
point = PointXYZ(x=1.0, y=2.0, z=3.0)
print(point.distance())                  # Distance from origin
print(point.distance_to(other_point))    # Distance between points
np_arr = point.to_numpy()                # [1.0, 2.0, 3.0]

colored = PointXYZRGB(x=1.0, y=2.0, z=3.0, r=255, g=0, b=0)
print(colored.rgb())    # (255, 0, 0)
print(colored.xyz())    # PointXYZ(1.0, 2.0, 3.0)

# simplified
buffer = PointCloudBuffer(capacity=10000, frame_id="lidar_front")

# Add points one at a time
buffer.add_point(1.0, 2.0, 3.0)
buffer.add_point(4.0, 5.0, 6.0)

# From NumPy — shape (N, 3)
buffer = PointCloudBuffer.from_numpy(np_points, frame_id="lidar")

# Access
point = buffer[0]           # PointXYZ
count = len(buffer)
np_arr = buffer.to_numpy()  # Shape (N, 3)
data = buffer.to_bytes()    # Serialization

# simplified
tracked = TrackedObject(
    track_id=42,
    bbox=BoundingBox2D(10, 20, 100, 200),
    class_name="person",
    confidence=0.95,
)

# simplified
from horus import DetectionList, TrackedObject, Topic

tracks: dict[int, TrackedObject] = {}

def tracker_tick(node):
    detections = det_topic.recv()
    if not detections:
        return

    # Simple nearest-neighbor matching
    matched = match_detections(tracks, detections)

    for track_id, det in matched.items():
        tracks[track_id].update(det.bbox, det.confidence)

    for track_id in unmatched_tracks:
        tracks[track_id].mark_missed()
        if tracks[track_id].age > 30:
            tracks[track_id].delete()

# simplified
from horus import COCOPose

# Keypoint indices
COCOPose.NOSE           # 0
COCOPose.LEFT_EYE       # 1
COCOPose.RIGHT_EYE      # 2
COCOPose.LEFT_EAR       # 3
COCOPose.RIGHT_EAR      # 4
COCOPose.LEFT_SHOULDER  # 5
COCOPose.RIGHT_SHOULDER # 6
COCOPose.LEFT_ELBOW     # 7
COCOPose.RIGHT_ELBOW    # 8
COCOPose.LEFT_WRIST     # 9
COCOPose.RIGHT_WRIST    # 10
COCOPose.LEFT_HIP       # 11
COCOPose.RIGHT_HIP      # 12
COCOPose.LEFT_KNEE      # 13
COCOPose.RIGHT_KNEE     # 14
COCOPose.LEFT_ANKLE     # 15
COCOPose.RIGHT_ANKLE    # 16
COCOPose.NUM_KEYPOINTS  # 17

# simplified
from horus import Landmark, LandmarkArray, COCOPose

# Create landmarks from pose model output
landmarks = LandmarkArray(num_landmarks=17, dimension=2)
landmarks.confidence = 0.92

# Access specific keypoints
nose = Landmark(x=320.0, y=200.0, visibility=0.99, index=COCOPose.NOSE)
left_wrist = Landmark(x=450.0, y=380.0, visibility=0.85, index=COCOPose.LEFT_WRIST)

if nose.is_visible(0.5) and left_wrist.is_visible(0.5):
    dist = nose.distance_to(left_wrist)
    print(f"Nose to wrist: {dist:.1f}px")

# simplified
# Safe detection pipeline
try:
    detections = DetectionList.from_bytes(raw_data)
    confident = detections.filter_confidence(0.7)
except ValueError as e:
    print(f"Bad detection data: {e}")
    confident = DetectionList()  # Empty fallback