File size: 32,985 Bytes
fd50325 482aec2 fd50325 482aec2 fd50325 482aec2 908356e 482aec2 908356e 482aec2 908356e 482aec2 908356e 482aec2 908356e 482aec2 fd50325 482aec2 fd50325 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 | """
Behavior Analysis Integrator for DetectifAI
This module integrates behavior analysis (action recognition) into the video processing pipeline.
It processes video segments/keyframes to detect suspicious behaviors like fighting, accidents, and climbing.
Similar to ObjectDetectionIntegrator, it creates behavior-based events and identifies suspicious frames
for facial recognition processing.
"""
import os
import cv2
import time
import logging
import json
from typing import List, Dict, Any, Tuple, Optional
from dataclasses import dataclass, asdict
import numpy as np
import math
# Import behavior analysis module
from behavior_analysis.action_recognition import (
load_model, preprocess_clip, interpret_prediction,
MODEL_PATHS, RESNET_MODELS, YOLO_MODELS, ActionPrediction
)
logger = logging.getLogger(__name__)
@dataclass
class BehaviorDetectionResult:
"""Result of behavior detection on a frame or segment"""
frame_path: str
timestamp: float
frame_index: int
behavior_detected: str # "fighting", "accident", "climbing", or "no_action"
confidence: float
model_used: str
processing_time: float
@dataclass
class BehaviorEvent:
"""Behavior-based event created from detections"""
event_id: str
behavior_type: str
start_timestamp: float
end_timestamp: float
confidence: float
frame_indices: List[int]
keyframes: List[str]
model_used: str
importance_score: float
class BehaviorAnalysisIntegrator:
"""Integration layer between behavior analysis and video processing pipeline"""
def __init__(self, config):
self.config = config
self.enabled = getattr(config, 'enable_behavior_analysis', False)
logger.info(f"🔍 Initializing BehaviorAnalysisIntegrator - enabled: {self.enabled}")
# Initialize models if enabled
self.models = {}
self.device = None
if self.enabled:
try:
import torch
self.device = torch.device("cuda" if (torch.cuda.is_available() and getattr(config, 'use_gpu_acceleration', True)) else "cpu")
# Load all available models
logger.info(f"🔧 Attempting to load models from: {MODEL_PATHS}")
for model_name, model_path in MODEL_PATHS.items():
logger.info(f"📁 Checking model {model_name} at: {model_path}")
if os.path.exists(model_path):
try:
logger.info(f"⏳ Loading {model_name}...")
self.models[model_name] = load_model(model_path, self.device)
logger.info(f"✅ Loaded behavior analysis model: {model_name}")
except Exception as e:
logger.error(f"❌ Failed to load {model_name}: {e}")
else:
logger.error(f"❌ Model file not found: {model_path}")
if not self.models:
logger.warning("⚠️ No behavior analysis models loaded, disabling behavior analysis")
self.enabled = False
else:
logger.info(f"✅ Behavior analysis initialized with {len(self.models)} models")
except ImportError:
logger.warning("⚠️ PyTorch not available, disabling behavior analysis")
self.enabled = False
else:
logger.info("Behavior analysis disabled in config")
# General YOLO detector for Option B person/car validation
self._general_detector = None
if self.enabled:
try:
from ultralytics import YOLO as _YOLO
self._general_detector = _YOLO("yolov8n.pt")
logger.info("✅ General YOLO (yolov8n) loaded for fight/accident validation")
except Exception as e:
logger.warning(f"⚠️ General YOLO not available — fight/accident validation disabled: {e}")
def detect_behavior_in_frame(self, frame_path: str, timestamp: float, frame_index: int = 0) -> List[BehaviorDetectionResult]:
"""
Detect behaviors in a single frame
Args:
frame_path: Path to frame image
timestamp: Timestamp in seconds
frame_index: Frame index number
Returns:
List of BehaviorDetectionResult objects (one per model)
"""
if not self.enabled or not self.models:
return []
if not os.path.exists(frame_path):
logger.warning(f"Frame not found: {frame_path}")
return []
results = []
frame = cv2.imread(frame_path)
if frame is None:
logger.warning(f"Failed to read frame: {frame_path}")
return []
for model_name, model in self.models.items():
try:
start_time = time.time()
# YOLO models (wallclimb)
if model_name in YOLO_MODELS:
output = model.predict(frame, verbose=False)
# Use default per-action thresholds from ACTION_CONFIDENCE_THRESHOLDS
label, conf = interpret_prediction(model, output, model_name)
logger.info(f"🔍 YOLO model {model_name} prediction: {label} (confidence: {conf:.3f})")
if label != "no_action":
result = BehaviorDetectionResult(
frame_path=frame_path,
timestamp=timestamp,
frame_index=frame_index,
behavior_detected=label,
confidence=conf,
model_used=model_name,
processing_time=time.time() - start_time
)
results.append(result)
# 3D-ResNet models need clips of 16 frames
# For single frame detection, we'll need to handle this differently
# For now, skip 3D-ResNet models for single frame detection
# They should be used with video segments instead
except Exception as e:
logger.error(f"Error detecting behavior with {model_name}: {e}")
continue
return results
def detect_behavior_in_segment(self, video_path: str, start_time: float, end_time: float,
frame_indices: List[int] = None) -> List[BehaviorDetectionResult]:
"""
Detect behaviors in a video segment (for 3D-ResNet models that need temporal context)
Args:
video_path: Path to video file
start_time: Start timestamp in seconds
end_time: End timestamp in seconds
frame_indices: Optional list of frame indices to process
Returns:
List of BehaviorDetectionResult objects
"""
if not self.enabled or not self.models:
return []
if not os.path.exists(video_path):
logger.warning(f"Video not found: {video_path}")
return []
results = []
cap = cv2.VideoCapture(video_path)
if not cap.isOpened():
logger.error(f"Could not open video: {video_path}")
return []
fps = cap.get(cv2.CAP_PROP_FPS) or 25
start_frame = int(start_time * fps)
end_frame = int(end_time * fps)
# Read frames for the segment
frame_buffer = []
cap.set(cv2.CAP_PROP_POS_FRAMES, start_frame)
for idx in range(start_frame, min(end_frame, int(cap.get(cv2.CAP_PROP_FRAME_COUNT)))):
ret, frame = cap.read()
if not ret:
break
frame_buffer.append(frame)
cap.release()
# Calculate mid frame index
mid_frame_idx = (start_frame + end_frame) // 2 if end_frame > start_frame else start_frame
return self._process_frame_buffer(frame_buffer, start_time, end_time, mid_frame_idx, video_path)
def detect_behavior_in_segment_from_buffer(self, frame_buffer: List[np.ndarray],
start_time: float, end_time: float,
frame_indices: List[int] = None) -> List[BehaviorDetectionResult]:
"""
Detect behaviors in a frame buffer (for live streams)
Args:
frame_buffer: List of frames (numpy arrays)
start_time: Start timestamp in seconds
end_time: End timestamp in seconds
frame_indices: Optional list of frame indices
Returns:
List of BehaviorDetectionResult objects
"""
if not self.enabled or not self.models:
return []
if len(frame_buffer) < 16:
logger.debug(f"Frame buffer too short ({len(frame_buffer)} frames), skipping 3D-ResNet models")
return []
# Use last 16 frames from buffer
frames_to_process = frame_buffer[-16:] if len(frame_buffer) >= 16 else frame_buffer
mid_frame_idx = len(frame_buffer) // 2 if frame_indices is None else (frame_indices[len(frame_indices) // 2] if frame_indices else len(frame_buffer) // 2)
return self._process_frame_buffer(frames_to_process, start_time, end_time, mid_frame_idx, "live_stream")
def _process_frame_buffer(self, frame_buffer: List[np.ndarray], start_time: float,
end_time: float, frame_index: int, video_path: str = "live_stream") -> List[BehaviorDetectionResult]:
"""
Process frame buffer with behavior analysis models
Args:
frame_buffer: List of frames (numpy arrays)
start_time: Start timestamp
end_time: End timestamp
frame_index: Frame index for result
video_path: Path to video file or "live_stream" for live streams
Returns:
List of BehaviorDetectionResult objects
"""
if len(frame_buffer) < 16:
return []
results = []
# Process with 3D-ResNet models (need 16-frame clips)
for model_name, model in self.models.items():
if model_name not in RESNET_MODELS:
continue
try:
start_time_proc = time.time()
# Process last 16 frames from buffer
clip = preprocess_clip(frame_buffer[-16:], self.device)
import torch
model.eval()
with torch.no_grad():
output = model(clip)
# Use default per-action thresholds from ACTION_CONFIDENCE_THRESHOLDS
label, conf = interpret_prediction(model, output, model_name)
logger.info(f"🔍 Model {model_name} prediction: {label} (confidence: {conf:.3f})")
if label != "no_action":
# --- Option B: YOLO validation for fight / accident ---
mid_frame = frame_buffer[len(frame_buffer) // 2]
label_lower = label.lower()
if label_lower in ("fighting", "fight"):
valid, ctx = self._validate_persons(mid_frame, min_count=2)
if not valid:
logger.info(f"🚫 Fight suppressed by Option B: <2 persons close together ({label})")
continue
logger.info(f"✅ Fight validated by Option B: persons close together")
elif label_lower in ("accident", "road_accident"):
valid, ctx = self._validate_vehicles(mid_frame, min_count=2)
if not valid:
logger.info(f"🚫 Accident suppressed by Option B: <2 vehicles close together ({label})")
continue
logger.info(f"✅ Accident validated by Option B: vehicles close together")
# Use middle timestamp of the segment
mid_timestamp = (start_time + end_time) / 2
result = BehaviorDetectionResult(
frame_path="live_stream", # Live stream identifier
timestamp=mid_timestamp,
frame_index=frame_index,
behavior_detected=label,
confidence=conf,
model_used=model_name,
processing_time=time.time() - start_time_proc
)
results.append(result)
except Exception as e:
logger.error(f"Error detecting behavior with {model_name} in segment: {e}")
continue
return results
def detect_behavior_in_keyframes(self, keyframes: List, video_path: str = None) -> List[BehaviorDetectionResult]:
"""
Detect behaviors in keyframes
Args:
keyframes: List of KeyframeResult objects
video_path: Optional path to video file (needed for 3D-ResNet models)
Returns:
List of BehaviorDetectionResult objects
"""
if not self.enabled:
logger.info("🚫 Behavior analysis disabled, skipping")
return []
logger.info(f"🎬 Starting behavior detection on {len(keyframes)} keyframes")
logger.info(f"📹 Video path provided: {video_path}")
logger.info(f"🤖 Available models: {list(self.models.keys())}")
logger.info(f"🔍 Running behavior analysis on {len(keyframes)} keyframes...")
all_results = []
# Process YOLO models (single frame) - wallclimb
yolo_models_available = [m for m in self.models.keys() if m in YOLO_MODELS]
logger.info(f"🎯 Processing YOLO models (single frame): {yolo_models_available}")
for i, keyframe in enumerate(keyframes):
# Extract frame path and timestamp
frame_path = None
timestamp = 0.0
frame_index = i
if hasattr(keyframe, 'frame_data'):
frame_path = keyframe.frame_data.frame_path if hasattr(keyframe.frame_data, 'frame_path') else None
timestamp = keyframe.frame_data.timestamp if hasattr(keyframe.frame_data, 'timestamp') else 0.0
elif hasattr(keyframe, 'frame_path'):
frame_path = keyframe.frame_path
timestamp = getattr(keyframe, 'timestamp', 0.0)
if frame_path and os.path.exists(frame_path):
# Detect with YOLO models (single frame) - wallclimb
frame_results = self.detect_behavior_in_frame(frame_path, timestamp, frame_index)
all_results.extend(frame_results)
# Process 3D-ResNet models (need 16-frame clips) - fighting, road_accident
if video_path and os.path.exists(video_path) and RESNET_MODELS:
resnet_models_available = [m for m in self.models.keys() if m in RESNET_MODELS]
logger.info(f"🎬 Processing 3D-ResNet models using video segments...")
logger.info(f"📊 Available ResNet models: {resnet_models_available}")
logger.info(f"📊 Total ResNet models to process: {len(resnet_models_available)}")
# Group keyframes into temporal segments for 3D-ResNet processing
# Process segments of ~1 second (16 frames at ~30fps) around each keyframe
segment_window = 1.0 # 1 second window
processed_segments = set() # Track processed segments to avoid duplicates
for keyframe in keyframes:
timestamp = 0.0
if hasattr(keyframe, 'frame_data'):
timestamp = keyframe.frame_data.timestamp if hasattr(keyframe.frame_data, 'timestamp') else 0.0
elif hasattr(keyframe, 'timestamp'):
timestamp = getattr(keyframe, 'timestamp', 0.0)
if timestamp > 0:
# Create segment around this keyframe
start_time = max(0, timestamp - segment_window / 2)
end_time = timestamp + segment_window / 2
# Round to avoid processing same segment multiple times
segment_key = (int(start_time * 10), int(end_time * 10))
if segment_key not in processed_segments:
processed_segments.add(segment_key)
try:
logger.info(f"🎥 Processing video segment: {start_time:.1f}s - {end_time:.1f}s")
# Process segment with 3D-ResNet models
segment_results = self.detect_behavior_in_segment(
video_path=video_path,
start_time=start_time,
end_time=end_time,
frame_indices=None
)
logger.info(f"📈 Segment results: {len(segment_results)} detections")
for result in segment_results:
logger.info(f"🔍 Detected: {result.behavior_detected} (conf: {result.confidence:.3f})")
all_results.extend(segment_results)
except Exception as e:
logger.error(f"❌ Error processing segment {start_time:.1f}s-{end_time:.1f}s: {e}")
continue
logger.info(f"✅ Behavior analysis complete: {len(all_results)} behaviors detected")
return all_results
def create_behavior_events(self, detection_results: List[BehaviorDetectionResult],
temporal_window: float = 5.0) -> List[BehaviorEvent]:
"""
Create behavior-based events from detection results
Args:
detection_results: List of BehaviorDetectionResult objects
temporal_window: Time window in seconds for grouping detections
Returns:
List of BehaviorEvent objects
"""
if not detection_results:
return []
# Group detections by behavior type and temporal proximity
events = []
sorted_results = sorted(detection_results, key=lambda x: x.timestamp)
current_event = None
event_id_counter = 0
for result in sorted_results:
if result.behavior_detected == "no_action":
continue
if current_event is None:
# Start new event
event_id_counter += 1
current_event = {
'event_id': f"behavior_{result.behavior_detected}_{event_id_counter}",
'behavior_type': result.behavior_detected,
'start_timestamp': result.timestamp,
'end_timestamp': result.timestamp,
'confidences': [result.confidence],
'frame_indices': [result.frame_index],
'keyframes': [result.frame_path],
'model_used': result.model_used
}
elif (result.behavior_detected == current_event['behavior_type'] and
result.timestamp - current_event['end_timestamp'] <= temporal_window):
# Extend current event
current_event['end_timestamp'] = result.timestamp
current_event['confidences'].append(result.confidence)
current_event['frame_indices'].append(result.frame_index)
current_event['keyframes'].append(result.frame_path)
else:
# Finalize current event and start new one
avg_confidence = sum(current_event['confidences']) / len(current_event['confidences'])
importance = avg_confidence * (current_event['end_timestamp'] - current_event['start_timestamp'] + 1)
behavior_event = BehaviorEvent(
event_id=current_event['event_id'],
behavior_type=current_event['behavior_type'],
start_timestamp=current_event['start_timestamp'],
end_timestamp=current_event['end_timestamp'],
confidence=avg_confidence,
frame_indices=current_event['frame_indices'],
keyframes=current_event['keyframes'],
model_used=current_event['model_used'],
importance_score=importance
)
events.append(behavior_event)
# Start new event
event_id_counter += 1
current_event = {
'event_id': f"behavior_{result.behavior_detected}_{event_id_counter}",
'behavior_type': result.behavior_detected,
'start_timestamp': result.timestamp,
'end_timestamp': result.timestamp,
'confidences': [result.confidence],
'frame_indices': [result.frame_index],
'keyframes': [result.frame_path],
'model_used': result.model_used
}
# Finalize last event
if current_event:
avg_confidence = sum(current_event['confidences']) / len(current_event['confidences'])
importance = avg_confidence * (current_event['end_timestamp'] - current_event['start_timestamp'] + 1)
behavior_event = BehaviorEvent(
event_id=current_event['event_id'],
behavior_type=current_event['behavior_type'],
start_timestamp=current_event['start_timestamp'],
end_timestamp=current_event['end_timestamp'],
confidence=avg_confidence,
frame_indices=current_event['frame_indices'],
keyframes=current_event['keyframes'],
model_used=current_event['model_used'],
importance_score=importance
)
events.append(behavior_event)
logger.info(f"✅ Created {len(events)} behavior-based events")
return events
def process_keyframes_with_behavior_analysis(self, keyframes: List, video_path: str = None) -> Tuple[List[BehaviorDetectionResult], List[BehaviorEvent]]:
"""
Process keyframes with behavior analysis and create behavior-based events
Args:
keyframes: List of KeyframeResult objects
video_path: Optional path to video file (needed for 3D-ResNet models)
Returns:
Tuple of (detection_results, behavior_events)
"""
if not self.enabled:
logger.info("🚫 Behavior analysis disabled, skipping...")
return [], []
logger.info("🚀 ===== STARTING BEHAVIOR ANALYSIS INTEGRATION =====")
logger.info(f"📊 Input: {len(keyframes)} keyframes, video_path: {video_path}")
logger.info(f"🤖 Loaded models: {list(self.models.keys())}")
logger.info(f"⚙️ Confidence thresholds: fighting={getattr(self.config, 'fighting_detection_confidence', 0.5)}, accident={getattr(self.config, 'accident_detection_confidence', 0.6)}, climbing={getattr(self.config, 'climbing_detection_confidence', 0.7)}")
logger.info("🔍 Starting behavior analysis integration")
# Run behavior detection on keyframes (with video_path for 3D-ResNet models)
detection_results = self.detect_behavior_in_keyframes(keyframes, video_path=video_path)
# Create behavior-based events
temporal_window = getattr(self.config, 'behavior_event_temporal_window', 5.0)
logger.info(f"📅 Creating behavior events with temporal window: {temporal_window}s")
logger.info(f"📊 Total detections to process: {len(detection_results)}")
positive_detections = [r for r in detection_results if r.behavior_detected != "no_action"]
logger.info(f"✅ Positive detections: {len(positive_detections)}")
for detection in positive_detections:
logger.info(f" 🎯 {detection.behavior_detected} at {detection.timestamp:.1f}s (conf: {detection.confidence:.3f})")
behavior_events = self.create_behavior_events(detection_results, temporal_window)
# Store detection metadata
if hasattr(self.config, 'output_base_dir') and detection_results:
detection_metadata = {
'total_keyframes': len(keyframes),
'frames_with_behaviors': len([r for r in detection_results if r.behavior_detected != "no_action"]),
'behaviors_detected': {
'fighting': len([r for r in detection_results if r.behavior_detected == "fighting"]),
'accident': len([r for r in detection_results if r.behavior_detected == "accident"]),
'climbing': len([r for r in detection_results if r.behavior_detected == "climbing"])
},
'total_events': len(behavior_events),
'detection_summary': [asdict(r) for r in detection_results[:10]] # First 10 for summary
}
metadata_path = os.path.join(self.config.output_base_dir, 'behavior_analysis_metadata.json')
os.makedirs(os.path.dirname(metadata_path), exist_ok=True)
with open(metadata_path, 'w') as f:
json.dump(detection_metadata, f, indent=2, default=str)
logger.info(f"📊 Behavior analysis metadata saved: {metadata_path}")
logger.info("🏁 ===== BEHAVIOR ANALYSIS INTEGRATION COMPLETE =====")
logger.info(f"📈 Summary:")
logger.info(f" 📊 Total detections: {len(detection_results)}")
logger.info(f" ✅ Positive detections: {len([r for r in detection_results if r.behavior_detected != 'no_action'])}")
logger.info(f" 📅 Events created: {len(behavior_events)}")
for event in behavior_events:
logger.info(f" 🎬 Event: {event.behavior_type} ({event.start_timestamp:.1f}s-{event.end_timestamp:.1f}s, conf: {event.confidence:.3f})")
return detection_results, behavior_events
def get_suspicious_frames(self, detection_results: List[BehaviorDetectionResult]) -> List[BehaviorDetectionResult]:
"""
Get frames with suspicious behaviors (for facial recognition processing)
Args:
detection_results: List of BehaviorDetectionResult objects
Returns:
List of suspicious BehaviorDetectionResult objects
"""
suspicious = [r for r in detection_results if r.behavior_detected != "no_action"]
logger.info(f"🔍 Identified {len(suspicious)} suspicious frames from behavior analysis")
return suspicious
# ------------------------------------------------------------------ #
# Option B helpers — YOLO person / vehicle validation
# ------------------------------------------------------------------ #
def _validate_persons(self, frame: np.ndarray, min_count: int = 2) -> Tuple[bool, list]:
"""Return (True, detections) if ≥ min_count persons are close together."""
if self._general_detector is None:
return True, [] # no validator → allow
try:
preds = self._general_detector(frame, conf=0.35, verbose=False)
persons = []
for r in preds:
for box in r.boxes:
if int(box.cls) == 0: # COCO person
x1, y1, x2, y2 = box.xyxy[0].tolist()
persons.append({"class": "person", "confidence": float(box.conf),
"bbox": [x1, y1, x2, y2]})
if len(persons) < min_count:
logger.info(f"🚫 Person check: only {len(persons)} person(s) — need {min_count}+")
return False, persons
for i in range(len(persons)):
for j in range(i + 1, len(persons)):
if self._boxes_are_close(persons[i]["bbox"], persons[j]["bbox"], frame.shape):
logger.info(f"✅ Person check passed: {len(persons)} persons, pair ({i},{j}) close")
return True, persons
logger.info(f"🚫 Person check: {len(persons)} persons but none close together")
return False, persons
except Exception as e:
logger.warning(f"⚠️ Person validation error: {e}")
return True, []
def _validate_vehicles(self, frame: np.ndarray, min_count: int = 2) -> Tuple[bool, list]:
"""Return (True, detections) if ≥ min_count vehicles are close together."""
if self._general_detector is None:
return True, []
try:
preds = self._general_detector(frame, conf=0.35, verbose=False)
vehicle_classes = {2: "car", 3: "motorcycle", 5: "bus", 7: "truck"}
vehicles = []
for r in preds:
for box in r.boxes:
cid = int(box.cls)
if cid in vehicle_classes:
x1, y1, x2, y2 = box.xyxy[0].tolist()
vehicles.append({"class": vehicle_classes[cid],
"confidence": float(box.conf),
"bbox": [x1, y1, x2, y2]})
if len(vehicles) < min_count:
logger.info(f"🚫 Vehicle check: only {len(vehicles)} vehicle(s) — need {min_count}+")
return False, vehicles
for i in range(len(vehicles)):
for j in range(i + 1, len(vehicles)):
if self._boxes_are_close(vehicles[i]["bbox"], vehicles[j]["bbox"], frame.shape):
logger.info(f"✅ Vehicle check passed: {len(vehicles)} vehicles, pair ({i},{j}) close")
return True, vehicles
logger.info(f"🚫 Vehicle check: {len(vehicles)} vehicles but none close together")
return False, vehicles
except Exception as e:
logger.warning(f"⚠️ Vehicle validation error: {e}")
return True, []
@staticmethod
def _boxes_are_close(bbox1, bbox2, frame_shape, distance_ratio: float = 0.15) -> bool:
"""True if two boxes overlap or are within *distance_ratio* of frame diagonal."""
x1a, y1a, x2a, y2a = bbox1
x1b, y1b, x2b, y2b = bbox2
# overlap?
if min(x2a, x2b) > max(x1a, x1b) and min(y2a, y2b) > max(y1a, y1b):
return True
# edge-to-edge distance
dx = max(0, max(x1a, x1b) - min(x2a, x2b))
dy = max(0, max(y1a, y1b) - min(y2a, y2b))
dist = math.hypot(dx, dy)
diag = math.hypot(frame_shape[0], frame_shape[1])
return dist < diag * distance_ratio
def get_behavior_analysis_summary(self) -> Dict[str, Any]:
"""Get summary statistics of behavior analysis"""
return {
'enabled': self.enabled,
'models_loaded': list(self.models.keys()) if self.models else [],
'device': str(self.device) if self.device else None
}
|