File size: 13,683 Bytes
8a74c03
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python3
"""
Comprehensive test of all videos in test folder to create best person-on-track implementation
"""
import sys
import os
from io import BytesIO
import glob

# Add current directory to path
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))

def test_all_videos_person_on_track():
    """Test all videos in test folder for person-on-track scenarios"""
    print("COMPREHENSIVE PERSON-ON-TRACK DETECTION ANALYSIS")
    print("=" * 70)
    print("Testing all videos in test folder to find best implementation")
    print()
    
    try:
        from local_models import get_local_model_manager
        from app import extract_frames_from_video, process_image_locally
        print("+ Components loaded successfully")
    except ImportError as e:
        print(f"- Import error: {e}")
        return
    
    # Find all videos in test folder
    test_videos = glob.glob("test\\*.mp4")
    if not test_videos:
        print("- No MP4 files found in test folder")
        return
    
    print(f"+ Found {len(test_videos)} test videos: {[os.path.basename(v) for v in test_videos]}")
    
    # Initialize models
    try:
        local_manager = get_local_model_manager()
        print("+ All models ready for testing")
    except Exception as e:
        print(f"- Model initialization error: {e}")
        return
    
    # Test different approaches
    approaches = {
        "Approach 1 - People Counter": {
            "model": "People Counter",
            "prompt": "Track Safety Analysis"
        },
        "Approach 2 - Direct CNN": {
            "model": "CNN (BLIP)", 
            "prompt": "Is there a person standing on train tracks? Answer yes or no."
        },
        "Approach 3 - Detailed Transformer": {
            "model": "Transformer (ViT-GPT2)",
            "prompt": "Describe people and train tracks in this image"
        },
        "Approach 4 - Safety Focus": {
            "model": "CNN (BLIP)",
            "prompt": "Describe any safety concerns with people near train tracks"
        }
    }
    
    all_results = {}
    
    # Test each video with each approach
    for video_idx, video_path in enumerate(test_videos):
        video_name = os.path.basename(video_path)
        print(f"\n" + "=" * 70)
        print(f"TESTING VIDEO {video_idx + 1}: {video_name}")
        print("=" * 70)
        
        try:
            # Extract frames
            with open(video_path, 'rb') as f:
                video_data = f.read()
            
            video_file = BytesIO(video_data)
            frames = extract_frames_from_video(video_file, fps=0.5)  # Every 2 seconds
            
            if not frames:
                print(f"- No frames extracted from {video_name}")
                continue
            
            print(f"+ Extracted {len(frames)} frames from {video_name}")
            
            # Test 2-3 frames per video to get representative sample
            test_frames = frames[:min(3, len(frames))]
            video_results = {}
            
            # Test each approach on this video
            for approach_name, config in approaches.items():
                print(f"\n  Testing {approach_name}:")
                print(f"  {'-' * 40}")
                
                approach_results = []
                
                for frame_idx, frame_data in enumerate(test_frames):
                    frame_num = frame_idx + 1
                    timestamp = frame_data['timestamp']
                    
                    try:
                        result = process_image_locally(
                            frame_data['frame'],
                            config["prompt"],
                            config["model"],
                            local_manager
                        )
                        
                        # Analyze result for person-on-track
                        person_on_track_analysis = analyze_for_person_on_track(result, config["model"])
                        
                        approach_results.append({
                            'frame': frame_num,
                            'timestamp': timestamp,
                            'raw_result': result,
                            'person_on_track': person_on_track_analysis['on_track'],
                            'confidence': person_on_track_analysis['confidence'],
                            'reasoning': person_on_track_analysis['reasoning']
                        })
                        
                        status = "ON TRACK" if person_on_track_analysis['on_track'] else "SAFE"
                        print(f"    Frame {frame_num} ({timestamp:.1f}s): {status} - {person_on_track_analysis['confidence']:.0%} confidence")
                        print(f"      Reasoning: {person_on_track_analysis['reasoning'][:80]}...")
                        
                    except Exception as e:
                        approach_results.append({
                            'frame': frame_num,
                            'timestamp': timestamp,
                            'raw_result': {'error': str(e)},
                            'person_on_track': False,
                            'confidence': 0,
                            'reasoning': f"Error: {str(e)}"
                        })
                        print(f"    Frame {frame_num} ({timestamp:.1f}s): ERROR - {str(e)}")
                
                video_results[approach_name] = approach_results
            
            all_results[video_name] = video_results
            
        except Exception as e:
            print(f"- Failed to process {video_name}: {e}")
            continue
    
    # Comprehensive analysis and recommendation
    analyze_all_approaches(all_results, approaches)
    
    return all_results

def analyze_for_person_on_track(result, model_type):
    """Analyze model result to determine if person is on train tracks"""
    
    if 'error' in result:
        return {
            'on_track': False,
            'confidence': 0,
            'reasoning': f"Error in processing: {result['error']}"
        }
    
    # Handle different result types
    if 'people_analysis' in result:
        # People Counter result
        analysis = result['people_analysis']
        on_track = analysis.get('on_tracks', False) or analysis.get('safety_risk', False)
        confidence = analysis.get('confidence', 0)
        reasoning = analysis.get('analysis_summary', 'People Counter analysis')
        
        return {
            'on_track': on_track,
            'confidence': confidence,
            'reasoning': reasoning
        }
    
    elif 'yes_no_detection' in result:
        # Yes/No detector result
        detection = result['yes_no_detection']
        # For track detection, we need more than just person presence
        return {
            'on_track': False,  # Yes/No detector doesn't check tracks specifically
            'confidence': 0.3,
            'reasoning': "Yes/No detector not suitable for track-specific detection"
        }
    
    elif 'generated_text' in result:
        # Text analysis result
        text = result['generated_text'].lower()
        
        # Keywords for person on tracks
        person_keywords = ['person', 'people', 'man', 'woman', 'human', 'individual']
        track_keywords = ['track', 'tracks', 'rail', 'rails', 'railway']
        position_keywords = ['on', 'standing', 'walking', 'sitting', 'crossing']
        danger_keywords = ['danger', 'unsafe', 'risk', 'hazard', 'warning']
        
        # Strong indicators
        strong_patterns = [
            'person on track', 'man on track', 'woman on track',
            'standing on track', 'walking on track', 'person crossing',
            'on the tracks', 'on train tracks', 'on railway'
        ]
        
        # Count indicators
        person_mentions = sum(1 for kw in person_keywords if kw in text)
        track_mentions = sum(1 for kw in track_keywords if kw in text)
        position_mentions = sum(1 for kw in position_keywords if kw in text)
        danger_mentions = sum(1 for kw in danger_keywords if kw in text)
        strong_indicators = sum(1 for pattern in strong_patterns if pattern in text)
        
        # Decision logic
        if strong_indicators > 0:
            on_track = True
            confidence = min(0.8 + strong_indicators * 0.1, 1.0)
            reasoning = f"Strong indicators: {strong_indicators} pattern matches"
            
        elif person_mentions > 0 and track_mentions > 0 and position_mentions > 0:
            on_track = True
            confidence = 0.6 + min(person_mentions + track_mentions + position_mentions, 3) * 0.1
            reasoning = f"Person + track + position keywords: {person_mentions}+{track_mentions}+{position_mentions}"
            
        elif danger_mentions > 0 and (person_mentions > 0 or track_mentions > 0):
            on_track = True
            confidence = 0.5 + danger_mentions * 0.1
            reasoning = f"Safety concern mentioned with people/tracks: {danger_mentions} danger keywords"
            
        else:
            on_track = False
            confidence = 0.7 if person_mentions == 0 else 0.4
            reasoning = f"No clear person-on-track indicators. Person:{person_mentions}, Track:{track_mentions}"
        
        return {
            'on_track': on_track,
            'confidence': confidence,
            'reasoning': reasoning
        }
    
    else:
        return {
            'on_track': False,
            'confidence': 0,
            'reasoning': "Unknown result format"
        }

def analyze_all_approaches(all_results, approaches):
    """Analyze all approaches and provide recommendations"""
    
    print(f"\n" + "=" * 80)
    print("COMPREHENSIVE ANALYSIS OF ALL APPROACHES")
    print("=" * 80)
    
    # Calculate performance metrics for each approach
    approach_metrics = {}
    
    for approach_name in approaches.keys():
        total_frames = 0
        on_track_detections = 0
        avg_confidence = 0
        error_count = 0
        
        for video_name, video_results in all_results.items():
            if approach_name in video_results:
                for frame_result in video_results[approach_name]:
                    total_frames += 1
                    if frame_result['person_on_track']:
                        on_track_detections += 1
                    avg_confidence += frame_result['confidence']
                    if 'error' in frame_result.get('raw_result', {}):
                        error_count += 1
        
        if total_frames > 0:
            avg_confidence = avg_confidence / total_frames
            detection_rate = on_track_detections / total_frames * 100
            error_rate = error_count / total_frames * 100
        else:
            avg_confidence = 0
            detection_rate = 0
            error_rate = 100
        
        approach_metrics[approach_name] = {
            'total_frames': total_frames,
            'on_track_detections': on_track_detections,
            'detection_rate': detection_rate,
            'avg_confidence': avg_confidence,
            'error_rate': error_rate
        }
    
    # Display results table
    print(f"\nAPPROACH PERFORMANCE COMPARISON:")
    print("-" * 80)
    print(f"{'Approach':<25} {'Frames':<8} {'On-Track':<10} {'Rate':<8} {'Confidence':<12} {'Errors':<8}")
    print("-" * 80)
    
    for approach, metrics in approach_metrics.items():
        print(f"{approach:<25} {metrics['total_frames']:<8} {metrics['on_track_detections']:<10} "
              f"{metrics['detection_rate']:<8.1f}% {metrics['avg_confidence']:<12.0%} {metrics['error_rate']:<8.1f}%")
    
    # Find best approach
    best_approach = max(approach_metrics.items(), 
                       key=lambda x: x[1]['avg_confidence'] * (100 - x[1]['error_rate']) / 100)
    
    print(f"\n" + "=" * 80)
    print("RECOMMENDATIONS")
    print("=" * 80)
    
    print(f"BEST APPROACH: {best_approach[0]}")
    print(f"  - Average Confidence: {best_approach[1]['avg_confidence']:.0%}")
    print(f"  - Detection Rate: {best_approach[1]['detection_rate']:.1f}%")
    print(f"  - Error Rate: {best_approach[1]['error_rate']:.1f}%")
    print(f"  - Total Frames Tested: {best_approach[1]['total_frames']}")
    
    # Detailed recommendations
    print(f"\nDETAILED ANALYSIS:")
    
    if best_approach[0] == "Approach 1 - People Counter":
        print("+ People Counter is most effective for track safety")
        print("+ Uses specialized multi-prompt analysis")
        print("+ Provides detailed safety risk assessment")
        
    elif "CNN" in best_approach[0]:
        print("+ CNN model provides good balance of speed and accuracy")
        print("+ Direct prompting works well for specific scenarios")
        print("+ Consider using for real-time applications")
        
    elif "Transformer" in best_approach[0]:
        print("+ Transformer model provides detailed scene understanding")
        print("+ Better for complex scene analysis")
        print("+ Higher computational cost but more accurate descriptions")
        
    # Video-by-video breakdown
    print(f"\nPER-VIDEO ANALYSIS:")
    print("-" * 50)
    
    for video_name, video_results in all_results.items():
        print(f"\n{video_name}:")
        for approach_name, results in video_results.items():
            on_track_frames = sum(1 for r in results if r['person_on_track'])
            total_frames = len(results)
            print(f"  {approach_name}: {on_track_frames}/{total_frames} frames with person on track")

if __name__ == "__main__":
    test_all_videos_person_on_track()