Spaces:
Running
Running
File size: 9,473 Bytes
88f1cd0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 |
import numpy as np
import time
from typing import Dict, List, Optional
import threading
import torch
from sklearn.cluster import DBSCAN
# -- Local Imports --
from config import APP_CONFIG, TUNING, INACTIVITY_CFG, DENSITY_DBSCAN_CFG
from .frame_reader import FrameReader
from .tracker import MOTTracker
class StreamProcessor:
def __init__(self,
video_url: str,
model,
device: str = "cpu",
half: bool = False,
infer_lock=None):
# -- Inisialisasi --
self.video_url = video_url
self.model = model
self.device = device
self.detection_interval = max(1, int(TUNING["DETECTION_INTERVAL_FRAMES"]))
self.imgsz = int(TUNING["YOLO_IMG_SIZE"])
self.half = bool(half and device == "cuda")
self._infer_lock = infer_lock
self.frame_reader = FrameReader(video_url)
self.tracker = MOTTracker(tracker_type="bytetrack", device=device)
self._running = False
self._thread = threading.Thread(target=self._run, daemon=True)
self._lock = threading.Lock()
self._latest_payload = None
self._frame_idx = 0
self._last_dets = None
self.id_state: Dict[int, Dict] = {}
def start(self):
if self._running:
return
self._running = True
self.frame_reader.start()
self._thread.start()
def stop(self):
self._running = False
self.frame_reader.stop()
self._thread.join()
def get_latest(self) -> Optional[Dict]:
with self._lock:
if self._latest_payload is None:
return None
return {
"frame" : self._latest_payload["frame"].copy(),
"tracks" : list(self._latest_payload["tracks"]),
"timestamp" : self._latest_payload["timestamp"],
"frame_idx" : self._latest_payload["frame_idx"],
"stats" : dict(self._latest_payload.get("stats", {}))
}
def _size_norm(self, box: List[float]) -> float:
x1, y1, x2, y2 = box
w = max(1.0, x2- x1)
h = max(1.0, y2 - y1)
return (w*w + h*h) ** 0.5
# -- Inactivity Logic --
def _update_inactivity(self, tracks: List[Dict], now: float):
current_ids = set(t["id"] for t in tracks)
for t in tracks:
tid = t["id"]
cx, cy = self._center(t["box"])
diag = self._size_norm(t["box"])
st = self.id_state.get(tid)
if st is None:
# please refer to "config.py" for each definition
st = {"pos": (cx, cy), "t": now, "ema_v": 0.0, "inactive": False, "since": None, "last_seen": now}
self.id_state[tid] = st
t["inactive"] = False
continue
dt = max(1e-3, now - st["t"])
dx = cx - st["pos"][0]
dy = cy - st["pos"][1]
v_norm = ((dx*dx + dy * dy)**0.5 / dt) / max(1.0, diag)
alpha = INACTIVITY_CFG["EMA_ALPHA"]
ema_v = alpha * v_norm + (1.0 - alpha) * st["ema_v"]
entry = INACTIVITY_CFG["ENTER_THRESH_NORM_SPEED"]
exit_ = INACTIVITY_CFG["EXIT_THRESH_NORM_SPEED"]
dwell = INACTIVITY_CFG["MIN_DURATION_S"]
if st["inactive"]:
if ema_v > exit_:
st["since"] = st.get("since") or now
if (now - st["since"]) >= dwell:
st["inactive"] = False
st['since'] = None
else:
st["since"] = None
else:
if ema_v < entry:
st["since"] = st.get("since") or now
if (now - st["since"]) >= dwell:
st["inactive"] = True
st["since"] = None
else:
st["since"] = None
st.update(pos = (cx, cy), t = now, ema_v = ema_v, last_seen = now)
t["inactive"] = st["inactive"]
# ensure old unseen ID removed
stale = [
tid for tid, st in list(self.id_state.items())
if tid not in current_ids and (now - st.get("last_seen", now)) > INACTIVITY_CFG["MAX_UNSEEN_GAP_S"]
]
for tid in stale:
self.id_state.pop(tid, None)
def _center(self, box: List[float]) -> tuple[float, float]:
x1, y1, x2, y2 = box
return ((x1 + x2) * 0.5, (y1 + y2) * 0.5)
# -- Density Logic --
def _compute_density_dbscan(self, tracks: List[Dict]) -> set:
if not tracks:
return set(), 0
centers = np.array([self._center(t["box"]) for t in tracks], dtype=np.float32)
# please refer to "config.py" for each definition
min_samples = max(1, DENSITY_DBSCAN_CFG["MIN_NEIGHBORS"] + 1)
labels = DBSCAN(eps=DENSITY_DBSCAN_CFG["EPS_PX"], min_samples=min_samples).fit_predict(centers)
# below to count cluster that happend ( -1 is noise )
n_clusters = int(len(set(lbl for lbl in labels if lbl != -1)))
dense_ids = {t["id"] for t, lbl in zip(tracks, labels) if lbl != -1}
return dense_ids, n_clusters
# return {t["id"] for t, lbl in zip(tracks, labels) if lbl != -1}
def _run(self):
""" Detect Object from The Frame and add its metadata (Inactivity / Density) """
while self._running:
frame = self.frame_reader.read()
if frame is None:
time.sleep(0.01)
continue
self._frame_idx += 1
# ensure frame to detect only on the interval
if self._frame_idx % self.detection_interval == 1:
try:
with torch.no_grad():
res = self.model.predict(
frame,
imgsz = self.imgsz,
device = self.device,
half = self.half,
verbose = False
)[0]
boxes = res.boxes
if boxes is not None and len(boxes) > 0:
dets = np.concatenate([
boxes.xyxy.cpu().numpy(),
boxes.conf.cpu().numpy()[:, None],
boxes.cls.cpu().numpy()[:, None]
], axis = 1).astype("float32")
else:
dets = np.empty((0, 6), dtype="float32")
self._last_dets = dets
except Exception as e:
print(f"[StreamProcessor] detection error: {e}")
self._last_dets = None
dets = self._last_dets # use last know tracking to ensure trackig is keept
# Update tracker
try:
tracks = self.tracker.update(dets, frame)
except Exception as e:
print(f"[StreamProcessor] tracking error: {e}")
tracks = []
# Update metadata
now = time.time()
self._update_inactivity(tracks, now)
dense_ids, n_clusters = self._compute_density_dbscan(tracks)
for t in tracks:
t["dense"] = (t["id"] in dense_ids)
stats = {
"detected": len(tracks),
"inactive": sum(1 for t in tracks if t.get("inactive")),
"dense_clusters": n_clusters
}
# Store latest result
with self._lock:
self._latest_payload = {
"frame" : frame,
"tracks" : tracks,
"timestamp" : now,
"frame_idx" : self._frame_idx,
"stats" : stats
}
class StreamRegistry:
def __init__(self):
self._by_url: Dict[str, StreamProcessor] = {}
self._ref_count: Dict[str, int] = {}
self._lock = threading.Lock()
def get(self, url: str, model, device="cpu", half=False) -> StreamProcessor:
""" Gather and made new StreamProcessor from each Video URL"""
with self._lock:
sp = self._by_url.get(url)
if sp is None:
print(f"[Registry] Creating new stream processor for: {url}")
sp = StreamProcessor(url, model=model, device=device, half=half)
sp.start()
self._by_url[url] = sp
self._ref_count[url] = 0
self._ref_count[url] += 1
print(f"[Registry] URL {url} ref count is now {self._ref_count[url]}")
return sp
def release(self, url: str):
""" Ensure when stream video stop, threading is stop too """
with self._lock:
if url in self._by_url:
self._ref_count[url] -= 1
print(f"[Registry] URL {url} ref count is now {self._ref_count[url]}")
if self._ref_count[url] <= 0:
print(f"[Registry] Stopping and removing processor for {url}")
try:
self._by_url[url].stop()
finally:
self._by_url.pop(url, None)
self._ref_count.pop(url, None) |