Spaces:
Sleeping
Sleeping
File size: 9,847 Bytes
9f43980 c64a277 9f43980 c64a277 9f43980 c64a277 5bba11c c64a277 9f43980 94741c6 9f43980 94741c6 9f43980 94741c6 9f43980 94741c6 9f43980 94741c6 9f43980 94741c6 9f43980 94741c6 9f43980 94741c6 9f43980 | 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 | import os
from pathlib import Path
import cv2
import pickle
import numpy as np
import logging
from ultralytics import YOLO
from deepface import DeepFace
from pathlib import Path
logger = logging.getLogger(__name__)
class SingleTracker:
def __init__(self):
logger.info("Initializing Single Tracker (Face Priority)")
# Configuration matches face_model.py
# self.base_dir = "/Users/adisankarlalan/Documents/GitHub/afs-fl/Model"
base_dir = Path(__file__).parent
self.base_dir = base_dir.parent / "Model"
print(self.base_dir,"base")
self.reference_video_path = os.path.join(self.base_dir, 'my_scan.mp4')
self.model_name = "ArcFace"
self.detector_model_path = os.path.join(self.base_dir, "yolov8n-face.pt")
self.cache_file = os.path.join(self.base_dir, "embeddings_cache.pkl")
# State
self.priority_track_id = None
self.known_tracks = {} # {track_id: is_main_user}
self.track_retries = {} # {track_id: retry_count}
self.max_retries = 20
self.similarity_threshold = 0.70
self.main_user_embeddings = []
self._load_embeddings()
try:
self.model = YOLO(self.detector_model_path)
logger.info("Loaded YOLO model")
except Exception as e:
logger.error(f"Failed to load YOLO model: {e}")
self.model = None
def _is_cache_valid(self, cache_data):
if not cache_data:
return False
if cache_data.get('video_path') != 'my_scan.mp4' and cache_data.get('video_path') != self.reference_video_path:
return False
if cache_data.get('model_name') != self.model_name:
return False
if cache_data.get('version', 1) < 2:
return False
return True
def _load_embeddings(self):
logger.info("Loading main user embeddings...")
cache_loaded = False
if os.path.exists(self.cache_file):
try:
with open(self.cache_file, 'rb') as f:
cache_data = pickle.load(f)
if self._is_cache_valid(cache_data):
self.main_user_embeddings = cache_data['embeddings']
logger.info("Loaded master signature from cache")
cache_loaded = True
except Exception as e:
logger.error(f"Could not load cache: {e}")
if not cache_loaded:
logger.warning(f"Cache invalid or not found at {self.cache_file}. Returning empty embeddings. Please run Model/face_model.py to generate cache.")
def process_frame(self, frame, custom_embeddings=None):
"""
Process a single BGR image frame for single face tracking.
Returns a dictionary with tracking results.
"""
results_data = {
"boxes": [],
"priority_id": self.priority_track_id,
"error": None,
"frame_width": int(frame.shape[1]),
"frame_height": int(frame.shape[0])
}
if self.model is None:
results_data["error"] = "Model not initialized"
return results_data
try:
# RUN BYTETRACK
results = self.model.track(frame, persist=True, tracker="bytetrack.yaml", verbose=False)
if results and len(results) > 0 and results[0].boxes.id is not None:
boxes = results[0].boxes.xyxy.cpu().numpy().astype(int)
track_ids = results[0].boxes.id.cpu().numpy().astype(int)
keypoints = None
if hasattr(results[0], 'keypoints') and results[0].keypoints is not None:
keypoints = results[0].keypoints.xy.cpu().numpy()
for idx, (box, track_id) in enumerate(zip(boxes, track_ids)):
x1, y1, x2, y2 = box.tolist()
track_id = int(track_id)
max_similarity = 0.0
# Compute Head Pose
yaw = 0.0
pitch = 0.0
if keypoints is not None and len(keypoints) > idx:
kpts = keypoints[idx]
if len(kpts) >= 5:
lex, ley = kpts[0]
rex, rey = kpts[1]
nx, ny = kpts[2]
lmx, lmy = kpts[3]
rmx, rmy = kpts[4]
# Yaw: (-) turned left, (+) turned right
l_nose = abs(nx - lex)
r_nose = abs(nx - rex)
yaw = (l_nose - r_nose) / (l_nose + r_nose + 1e-6)
# Pitch: (-) looking up, (+) looking down
eye_cy = (ley + rey) / 2
mouth_cy = (lmy + rmy) / 2
n_eye = ny - eye_cy
n_mouth = mouth_cy - ny
pitch = (n_eye - n_mouth) / (n_eye + n_mouth + 1e-6)
# Lock resolution logic
embeddings_to_check = custom_embeddings if custom_embeddings is not None and len(custom_embeddings) > 0 else self.main_user_embeddings
if track_id not in self.known_tracks and len(embeddings_to_check) > 0:
if track_id not in self.track_retries:
self.track_retries[track_id] = 0
# Crop face
face_crop = frame[y1:y2, x1:x2]
try:
# Strict check
current_face = DeepFace.represent(face_crop, model_name=self.model_name, enforce_detection=False)[0]["embedding"]
embeddings_to_check = custom_embeddings if custom_embeddings is not None and len(custom_embeddings) > 0 else self.main_user_embeddings
for user_embedding in embeddings_to_check:
sim = np.dot(user_embedding, current_face) / (np.linalg.norm(user_embedding) * np.linalg.norm(current_face))
if sim > max_similarity:
max_similarity = sim
max_similarity = float(max_similarity)
if max_similarity > self.similarity_threshold:
self.known_tracks[track_id] = True
self.priority_track_id = track_id
results_data["priority_id"] = track_id
if track_id in self.track_retries:
del self.track_retries[track_id]
else:
self.track_retries[track_id] += 1
if self.track_retries[track_id] > self.max_retries:
self.known_tracks[track_id] = False
del self.track_retries[track_id]
except Exception as e:
# Exception means no face/blur => skip for this frame but count retry
logger.error(f"DeepFace failed on track_id {track_id}: {e}")
self.track_retries[track_id] += 1
if self.track_retries[track_id] > self.max_retries:
self.known_tracks[track_id] = False
del self.track_retries[track_id]
else:
# Ensures unknown tracks still get registered for scanning display
if track_id not in self.known_tracks and track_id not in self.track_retries:
self.track_retries[track_id] = 0
# Determine label and color representation
is_target = self.known_tracks.get(track_id, False)
if is_target:
label = f"TARGET LOCKED"
results_data["boxes"].append({
"id": track_id,
"x1": x1, "y1": y1,
"x2": x2, "y2": y2,
"is_target": True,
"label": label,
"similarity": max_similarity if 'max_similarity' in locals() else -1.0,
"yaw": float(yaw),
"pitch": float(pitch)
})
elif track_id in self.track_retries:
# Draw scanning box
label = f"SCANNING"
results_data["boxes"].append({
"id": track_id,
"x1": x1, "y1": y1,
"x2": x2, "y2": y2,
"is_target": False,
"label": label,
"similarity": max_similarity if 'max_similarity' in locals() else -1.0,
"yaw": float(yaw),
"pitch": float(pitch)
})
except Exception as e:
logger.error(f"Error during SingleTrack: {e}")
results_data["error"] = str(e)
return results_data
|