File size: 15,141 Bytes
98a79a7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
YOLOv8 Crowd Detection Module - Enhanced for Small Objects
Implements REQ-1, REQ-2, REQ-3: Person detection with bounding boxes and counting
"""

import cv2
import numpy as np
from ultralytics import YOLO
import time
from typing import List, Tuple, Dict
import logging

logger = logging.getLogger(__name__)


class CrowdDetector:
    """
    Main crowd detection class using YOLOv8 - Enhanced for small objects
    
    Satisfies SRS Requirements:
    - REQ-1: Detect individuals in video frames using pre-trained model
    - REQ-2: Display bounding boxes around detected individuals
    - REQ-3: Update count continuously as frames are processed
    """
    
    def __init__(self, config: Dict):
        """
        Initialize the YOLOv8 detector
        
        Args:
            config: Configuration dictionary from config.yaml
        """
        self.config = config
        self.model_name = config['model']['name']
        self.confidence_threshold = config['model']['confidence_threshold']
        self.iou_threshold = config['model']['iou_threshold']
        self.device = config['model']['device']
        self.class_filter = config['model']['class_filter']
        self.min_size = config['crowd']['min_detection_size']
        
        # Optimization parameters
        self.small_object_mode = config['model'].get('small_object_mode', True)
        self.imgsz = 416  # Lower resolution for faster TensorRT inference
        
        # Dynamic mode parameters (can be updated via API)
        self.max_det = 300  # Default max detections
        self.second_pass_conf = 0.05  # Default second pass confidence
        self.duplicate_threshold = 30  # Default duplicate detection threshold
        self.min_box_size = 5  # Default minimum box size
        
        # Performance tracking
        from collections import deque
        self.frame_times = deque(maxlen=30)  # Keep last 30 frame times
        self.detection_count = 0
        self.frame_count = 0  # Track frames for logging throttle
        
        logger.info(f"Initializing YOLOv8 Detector with model: {self.model_name}")
        logger.info(f"Device: {self.device}, Confidence: {self.confidence_threshold}")
        logger.info(f"Small object mode: {self.small_object_mode}")
        
        # Check for TensorRT optimized model first
        tensorrt_model = self.model_name.replace('.pt', '.engine')
        use_tensorrt = False
        
        try:
            import os
            if os.path.exists(tensorrt_model):
                logger.info(f"Loading TensorRT optimized model: {tensorrt_model}")
                self.model = YOLO(tensorrt_model)
                use_tensorrt = True
            else:
                logger.info(f"Loading PyTorch model: {self.model_name}")
                logger.info(f"TIP: Export to TensorRT for 2-3x speedup: yolo export model={self.model_name} format=engine half=True device=0")
                self.model = YOLO(self.model_name)
                self.model.to(self.device)
                use_tensorrt = False
            
            logger.info("YOLOv8 model loaded successfully")
            logger.info(f"*** USING {'TensorRT ENGINE' if use_tensorrt else 'PyTorch FP16'} for inference ***")
            
            # GPU Warmup - run dummy inference to compile CUDA kernels
            logger.info("Warming up GPU (this may take a few seconds)...")
            dummy_frame = np.zeros((416, 416, 3), dtype=np.uint8)
            for _ in range(5):  # Run 5 warmup passes for better optimization
                self.model(dummy_frame, conf=0.5, verbose=False, device=self.device, half=(self.device != "cpu"), imgsz=self.imgsz)
            logger.info("GPU warmup complete - ready for fast inference")
            
        except Exception as e:
            logger.error(f"Failed to load YOLOv8 model: {e}")
            raise

    def preprocess_frame(self, frame: np.ndarray) -> np.ndarray:
        """
        Light preprocessing - only applied if needed
        Returns frame as-is for speed (YOLO handles normalization)
        """
        return frame  # Skip preprocessing for speed

    def detect_additional_pass(self, frame: np.ndarray, existing_detections: List[Dict]) -> List[Dict]:
        """
        Additional detection pass with very low confidence for missed small objects
        """
        try:
            # Second pass with lower confidence threshold for small/distant objects
            results = self.model(
                frame,
                conf=self.second_pass_conf,  # Use dynamic second pass confidence
                iou=self.iou_threshold,
                classes=self.class_filter,
                verbose=False,
                imgsz=self.imgsz,
                device=self.device,
                half=(self.device != "cpu")
            )
            
            additional_detections = []
            existing_centers = [(d['center'][0], d['center'][1]) for d in existing_detections]
            
            for result in results:
                boxes = result.boxes
                
                for box in boxes:
                    x1, y1, x2, y2 = box.xyxy[0].cpu().numpy()
                    confidence = float(box.conf[0].cpu().numpy())
                    class_id = int(box.cls[0].cpu().numpy())
                    
                    width = x2 - x1
                    height = y2 - y1
                    center_x = int((x1 + x2) / 2)
                    center_y = int((y1 + y2) / 2)
                    
                    # Filter out very small noise using dynamic min_box_size
                    if width < self.min_box_size or height < self.min_box_size:
                        continue
                    
                    # Check if this is a duplicate (near existing detection)
                    is_duplicate = False
                    for ex, ey in existing_centers:
                        distance = ((center_x - ex)**2 + (center_y - ey)**2)**0.5
                        if distance < self.duplicate_threshold:  # Use dynamic threshold
                            is_duplicate = True
                            break
                    
                    if not is_duplicate:
                        detection = {
                            'bbox': [int(x1), int(y1), int(x2), int(y2)],
                            'confidence': confidence,
                            'class_id': class_id,
                            'class_name': 'person',
                            'center': [center_x, center_y],
                            'size': 'tiny' if (width < 10 or height < 10) else ('small' if (width < 50 or height < 50) else 'normal')
                        }
                        additional_detections.append(detection)
            
            return additional_detections
            
        except Exception as e:
            logger.error(f"Additional detection pass error: {e}")
            return []
    
    def detect(self, frame: np.ndarray, resize_factor: float = 1.0, 
               confidence_threshold: float = None) -> Tuple[List[Dict], int, float]:
        """
        Detect people in the frame using YOLOv8 with TensorRT
        
        Args:
            frame: Input frame (BGR format from OpenCV)
            resize_factor: Ignored - always uses full resolution for best accuracy
            confidence_threshold: Optional override for detection threshold
        
        Returns:
            detections: List of all detected people (primary + second pass)
            count: Total number of people detected
            processing_time: Time taken for detection
        """
        start_time = time.time()
        detections = []
        
        if confidence_threshold is None:
            confidence_threshold = self.confidence_threshold
        
        try:
            # Primary detection with CUDA using configured device
            results = self.model(
                frame,
                conf=confidence_threshold,
                iou=self.iou_threshold,
                classes=self.class_filter,
                verbose=False,
                imgsz=self.imgsz,
                device=self.device,
                half=(self.device != "cpu"),  # FP16 inference for 2x speedup on RTX 3050
                max_det=self.max_det,  # Use dynamic max detections
                agnostic_nms=False,
                retina_masks=False  # Disable for speed
            )
            
            # Extract primary detections
            for result in results:
                boxes = result.boxes
                
                for box in boxes:
                    x1, y1, x2, y2 = box.xyxy[0].cpu().numpy()
                    confidence = float(box.conf[0].cpu().numpy())
                    class_id = int(box.cls[0].cpu().numpy())
                    
                    width = x2 - x1
                    height = y2 - y1
                    
                    # Filter by minimum size to remove noise
                    if width >= self.min_size and height >= self.min_size:
                        detection = {
                            'bbox': [int(x1), int(y1), int(x2), int(y2)],
                            'confidence': confidence,
                            'class_id': class_id,
                            'class_name': 'person',
                            'center': [int((x1 + x2) / 2), int((y1 + y2) / 2)],
                            'size': 'tiny' if (width < 10 or height < 10) else ('small' if (width < 50 or height < 50) else 'normal')
                        }
                        detections.append(detection)
            
            # Second pass for better small object detection
            if self.small_object_mode:
                additional = self.detect_additional_pass(frame, detections)
                detections.extend(additional)
            
            processing_time = time.time() - start_time
            self.frame_times.append(processing_time)
            self.detection_count += len(detections)
            self.frame_count += 1
            
            count = len(detections)
            
            # Log detection count occasionally (every 30 frames) to avoid log spam
            if self.frame_count % 30 == 0 or count > 0:
                logger.debug(f"Detected {count} people in {processing_time:.3f}s (frame {self.frame_count})")
            
            return detections, count, processing_time
            
        except Exception as e:
            logger.error(f"Detection error: {e}")
            import traceback
            logger.error(traceback.format_exc())
            return [], 0, 0.0

    def _calculate_iou(self, box1: List[int], box2: List[int]) -> float:
        """
        Calculate Intersection over Union for two bounding boxes with edge case handling
        
        Args:
            box1: [x1, y1, x2, y2]
            box2: [x1, y1, x2, y2]
            
        Returns:
            IoU value between 0.0 and 1.0
        """
        # Calculate intersection area
        x1_inter = max(box1[0], box2[0])
        y1_inter = max(box1[1], box2[1])
        x2_inter = min(box1[2], box2[2])
        y2_inter = min(box1[3], box2[3])
        
        inter_area = max(0, x2_inter - x1_inter) * max(0, y2_inter - y1_inter)
        
        # Calculate union area
        box1_area = (box1[2] - box1[0]) * (box1[3] - box1[1])
        box2_area = (box2[2] - box2[0]) * (box2[3] - box2[1])
        
        # Handle edge cases
        if box1_area <= 0 or box2_area <= 0:
            return 0.0
        
        union_area = box1_area + box2_area - inter_area
        
        # Avoid division by zero
        if union_area <= 0:
            return 0.0
        
        return inter_area / union_area
    
    def draw_detections(self, frame: np.ndarray, detections: List[Dict], 
                       show_confidence: bool = True) -> np.ndarray:
        """
        Draw bounding boxes with high visibility for crowd detection
        Color-coded by confidence level
        
        Args:
            frame: Input frame
            detections: List of detections
            show_confidence: Whether to display confidence scores
        
        Returns:
            frame: Frame with drawn bounding boxes
        """
        frame_copy = frame.copy()
        
        for i, det in enumerate(detections):
            x1, y1, x2, y2 = det['bbox']
            confidence = det['confidence']
            is_small = det.get('size') == 'small'
            
            # Color by confidence: Green (high) -> Yellow (medium) -> Orange (low)
            if confidence >= 0.5:
                color = (0, 255, 0)  # Green - high confidence
            elif confidence >= 0.25:
                color = (0, 255, 255)  # Yellow - medium
            elif confidence >= 0.15:
                color = (0, 165, 255)  # Orange - lower
            else:
                color = (0, 128, 255)  # Light orange - very low
            
            thickness = 1 if is_small else 2
            
            # Draw bounding box
            cv2.rectangle(frame_copy, (x1, y1), (x2, y2), color, thickness)
            
            # Draw center dot for all detections
            center_x, center_y = det['center']
            cv2.circle(frame_copy, (center_x, center_y), 2, color, -1)
        
        # Draw prominent count display in top-left corner
        count = len(detections)
        count_text = f"PEOPLE: {count}"
        
        # Background box for count
        (text_w, text_h), _ = cv2.getTextSize(count_text, cv2.FONT_HERSHEY_SIMPLEX, 1.0, 2)
        cv2.rectangle(frame_copy, (5, 5), (text_w + 15, text_h + 15), (0, 0, 0), -1)
        cv2.rectangle(frame_copy, (5, 5), (text_w + 15, text_h + 15), (0, 255, 0), 2)
        
        # Count text
        cv2.putText(frame_copy, count_text, (10, text_h + 8), 
                   cv2.FONT_HERSHEY_SIMPLEX, 1.0, (0, 255, 0), 2)
        
        return frame_copy
    
    def get_statistics(self) -> Dict:
        """
        Get detection statistics
        
        Returns:
            stats: Dictionary with performance metrics
        """
        if not self.frame_times:
            return {
                'avg_processing_time': 0.0,
                'fps': 0.0,
                'total_detections': 0,
                'frames_processed': 0
            }
        
        avg_time = np.mean(self.frame_times[-100:])  # Last 100 frames
        fps = 1.0 / avg_time if avg_time > 0 else 0.0
        
        return {
            'avg_processing_time': avg_time,
            'fps': fps,
            'total_detections': self.detection_count,
            'frames_processed': len(self.frame_times),
            'max_processing_time': max(self.frame_times) if self.frame_times else 0.0,
            'min_processing_time': min(self.frame_times) if self.frame_times else 0.0
        }
    
    def reset_statistics(self):
        """Reset detection statistics"""
        from collections import deque
        self.frame_times = deque(maxlen=30)
        self.detection_count = 0
        self.frame_count = 0
        logger.info("Detection statistics reset")