L'électron rare
0497a8951a
New standalone visualizer: OSC listener, pose bridge, Apple Vision / CoreML / YOLO pose backends, Euro filter, fine analysis, mesh topology, holistic renderer. Metal shader (scene.metal) for GPU-accelerated drawing.
141 lines
5.0 KiB
Python
141 lines
5.0 KiB
Python
"""Webcam + YOLOv8-pose integre au visualizer Metal.
|
|
|
|
Pourquoi ici plutot que dans data_feeds/feeds/pose.py :
|
|
- Un seul process = une seule webcam (macOS interdit la double ouverture)
|
|
- GPU partage : inference MPS et rendu Metal sur le meme device
|
|
- TCC : si data_only_viz est lance par le launcher bundle, le subprocess
|
|
Python herite (au moins une fois) du contexte camera autorise.
|
|
|
|
Met a jour directement state.pose_kp[17] sous lock. Le shader Metal lit
|
|
ces valeurs a chaque frame via renderer._update_skeleton.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import logging
|
|
import threading
|
|
import time
|
|
|
|
from .state import PoseKp, State
|
|
|
|
LOG = logging.getLogger("pose")
|
|
|
|
|
|
class PoseWorker:
|
|
"""Thread daemon de capture + inference."""
|
|
|
|
def __init__(
|
|
self,
|
|
state: State,
|
|
model_name: str = "yolov8n-pose.pt",
|
|
device: str = "mps",
|
|
camera_index: int = 0,
|
|
target_fps: float = 20.0,
|
|
conf_thresh: float = 0.35,
|
|
max_persons: int = 4,
|
|
) -> None:
|
|
self.state = state
|
|
self.model_name = model_name
|
|
self.device = device
|
|
self.camera_index = camera_index
|
|
self.period = 1.0 / max(1.0, target_fps)
|
|
self.conf_thresh = conf_thresh
|
|
self.max_persons = max_persons
|
|
self._thread: threading.Thread | None = None
|
|
self._stop = threading.Event()
|
|
|
|
def start(self) -> None:
|
|
self._thread = threading.Thread(
|
|
target=self._run, name="pose", daemon=True)
|
|
self._thread.start()
|
|
|
|
def stop(self) -> None:
|
|
self._stop.set()
|
|
|
|
def _run(self) -> None:
|
|
try:
|
|
import cv2
|
|
import numpy as np
|
|
from ultralytics import YOLO
|
|
except ModuleNotFoundError as e:
|
|
LOG.error("dependances manquantes : %s — uv sync --extra pose", e)
|
|
return
|
|
|
|
LOG.info("loading %s on %s", self.model_name, self.device)
|
|
try:
|
|
model = YOLO(self.model_name)
|
|
except Exception as e: # noqa: BLE001
|
|
LOG.error("YOLO load failed: %s", e)
|
|
return
|
|
|
|
cap = cv2.VideoCapture(self.camera_index)
|
|
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
|
|
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 480)
|
|
if not cap.isOpened():
|
|
LOG.error("camera index %d indisponible (TCC ?)", self.camera_index)
|
|
return
|
|
LOG.info("camera ouverte (index %d)", self.camera_index)
|
|
|
|
while not self._stop.is_set():
|
|
t0 = time.monotonic()
|
|
ok, frame = cap.read()
|
|
if not ok or frame is None:
|
|
time.sleep(self.period)
|
|
continue
|
|
h, w = frame.shape[:2]
|
|
try:
|
|
results = model.predict(
|
|
frame, device=self.device, conf=self.conf_thresh,
|
|
verbose=False, max_det=self.max_persons,
|
|
)
|
|
except Exception as e: # noqa: BLE001
|
|
LOG.warning("inference failed: %s", e)
|
|
time.sleep(self.period)
|
|
continue
|
|
if not results:
|
|
with self.state.lock():
|
|
self.state.pose_count = 0
|
|
self.state.pose_last_t = time.monotonic()
|
|
time.sleep(self.period)
|
|
continue
|
|
|
|
res = results[0]
|
|
kp_xy = getattr(res.keypoints, "xy", None)
|
|
kp_conf = getattr(res.keypoints, "conf", None)
|
|
n = 0 if kp_xy is None else int(len(kp_xy))
|
|
if n == 0:
|
|
with self.state.lock():
|
|
self.state.pose_count = 0
|
|
self.state.pose_last_t = time.monotonic()
|
|
time.sleep(self.period)
|
|
continue
|
|
|
|
# On suit le sujet 0 pour le squelette overlay (cf renderer).
|
|
pts = kp_xy[0].cpu().numpy()
|
|
cfs = (kp_conf[0].cpu().numpy() if kp_conf is not None
|
|
else np.ones(len(pts), dtype=float))
|
|
# Encode la frame en JPEG (qualite 70, ~30 ko) pour l'affichage
|
|
# NSImageView. Cheaper que d'envoyer du raw a Metal et marche
|
|
# peu importe le viz mode actif.
|
|
ok2, jpg = cv2.imencode(".jpg", frame,
|
|
[int(cv2.IMWRITE_JPEG_QUALITY), 70])
|
|
jpg_bytes = bytes(jpg) if ok2 else None
|
|
with self.state.lock():
|
|
self.state.pose_count = n
|
|
for k in range(min(17, len(pts))):
|
|
x, y = pts[k]
|
|
self.state.pose_kp[k] = PoseKp(
|
|
x=float(x) / max(1.0, w),
|
|
y=float(y) / max(1.0, h),
|
|
c=float(cfs[k]),
|
|
)
|
|
self.state.pose_last_t = time.monotonic()
|
|
if jpg_bytes:
|
|
self.state.last_webcam_jpeg = jpg_bytes
|
|
|
|
# cadence cible
|
|
dt = time.monotonic() - t0
|
|
if dt < self.period:
|
|
time.sleep(self.period - dt)
|
|
cap.release()
|
|
LOG.info("pose worker stopped")
|