fix-bug/frontal-detector

.gitignore

@@ -15,3 +15,5 @@ feeder/
 .venv/
 .vscode/
 dist/
+websocket_comm.log
+temp_debug/

siwatsystem/pympta.py

@@ -61,12 +61,27 @@ def crop_region_by_class(frame, regions_dict, class_name):
     
     bbox = regions_dict[class_name]['bbox']
     x1, y1, x2, y2 = bbox
+    
+    # TEMP DEBUG: Diagnostic logging for crop issues
+    frame_h, frame_w = frame.shape[:2]
+    logger.info(f"🔍 CROP DEBUG: Frame dimensions: {frame_w}x{frame_h}")
+    logger.info(f"🔍 CROP DEBUG: Original bbox: {bbox}")
+    logger.info(f"🔍 CROP DEBUG: Bbox dimensions: {x2-x1}x{y2-y1}")
+    
+    # Check if bbox is within frame bounds
+    if x1 < 0 or y1 < 0 or x2 > frame_w or y2 > frame_h:
+        logger.warning(f"🔍 CROP DEBUG: Bbox extends beyond frame! Clipping...")
+        x1, y1 = max(0, x1), max(0, y1)
+        x2, y2 = min(frame_w, x2), min(frame_h, y2)
+        logger.info(f"🔍 CROP DEBUG: Clipped bbox: ({x1}, {y1}, {x2}, {y2})")
+    
     cropped = frame[y1:y2, x1:x2]
     
     if cropped.size == 0:
         logger.warning(f"Empty crop for class '{class_name}' with bbox {bbox}")
         return None
     
+    logger.info(f"🔍 CROP DEBUG: Successful crop shape: {cropped.shape}")
     return cropped
 
 def format_action_context(base_context, additional_context=None):
@@ -113,6 +128,7 @@ def load_pipeline_node(node_config: dict, mpta_dir: str, redis_client, db_manage
         "crop": node_config.get("crop", False),
         "cropClass": node_config.get("cropClass"),
         "minConfidence": node_config.get("minConfidence", None),
+        "frontalMinConfidence": node_config.get("frontalMinConfidence", None),
         "minBboxAreaRatio": node_config.get("minBboxAreaRatio", 0.0),
         "multiClass": node_config.get("multiClass", False),
         "expectedClasses": node_config.get("expectedClasses", []),
@@ -634,8 +650,7 @@ def run_detection_with_tracking(frame, node, context=None):
                 logger.info(f"Camera {camera_id}: 🔄 Reset YOLO tracker - new cars will get fresh track IDs")
             session_state["reset_tracker_on_resume"] = False  # Clear the flag
         
-        # Get tracking zone from runtime context (camera-specific)
+        # Tracking zones removed - process all detections
-        tracking_zone = context.get("trackingZone", []) if context else []
         
         # Prepare class filtering
         trigger_class_indices = node.get("triggerClassIndices")
@@ -643,19 +658,13 @@ def run_detection_with_tracking(frame, node, context=None):
         
         logger.debug(f"Running detection for {node['modelId']} - tracking: {tracking_enabled}, stability_threshold: {stability_threshold}, classes: {node.get('triggerClasses', 'all')}")
         
-        if tracking_enabled and tracking_zone:
+        # Use predict for detection-only models (frontal detection), track for main detection models
-            # Use tracking with zone validation
+        model_id = node.get("modelId", "")
-            logger.debug(f"Using tracking with ReID config: {reid_config_path}")
+        use_tracking = tracking_enabled and not ("frontal" in model_id.lower() or "detection" in model_id.lower())
-            res = node["model"].track(
+        
-                frame,
+        if use_tracking:
-                stream=False,
+            # Use tracking for main detection models (yolo11m, etc.)
-                persist=True,
+            logger.debug(f"Using tracking for {model_id}")
-                tracker=reid_config_path,
-                **class_filter
-            )[0]
-        elif tracking_enabled:
-            # Use tracking without zone restriction
-            logger.debug("Using tracking without zone restriction")
             res = node["model"].track(
                 frame,
                 stream=False,
@@ -663,8 +672,8 @@ def run_detection_with_tracking(frame, node, context=None):
                 **class_filter
             )[0]
         else:
-            # Use detection only (no tracking)
+            # Use detection only for frontal detection and other detection-only models
-            logger.debug("Using detection only (tracking disabled)")
+            logger.debug(f"Using prediction only for {model_id}")
             res = node["model"].predict(
                 frame,
                 stream=False,
@@ -673,6 +682,12 @@ def run_detection_with_tracking(frame, node, context=None):
         
         # Process detection results
         candidate_detections = []
+        # Use frontalMinConfidence for frontal detection models, otherwise use minConfidence
+        model_id = node.get("modelId", "")
+        if "frontal" in model_id.lower() and "frontalMinConfidence" in node:
+            min_confidence = node.get("frontalMinConfidence", 0.0)
+            logger.debug(f"Using frontalMinConfidence={min_confidence} for {model_id}")
+        else:
             min_confidence = node.get("minConfidence", 0.0)
         
         if res.boxes is None or len(res.boxes) == 0:
@@ -716,15 +731,7 @@ def run_detection_with_tracking(frame, node, context=None):
                 logger.debug(f"❌ Camera {camera_id}: Detection {i+1} REJECTED - confidence {conf:.3f} < {min_confidence}")
                 continue
             
-            # Apply tracking zone validation if enabled
+            # Tracking zone validation removed - process all detections
-            if tracking_enabled and tracking_zone:
-                bbox_center_x = (x1 + x2) // 2
-                bbox_center_y = (y1 + y2) // 2
-                
-                # Check if detection center is within tracking zone
-                if not _point_in_polygon((bbox_center_x, bbox_center_y), tracking_zone):
-                    logger.debug(f"❌ Camera {camera_id}: Detection {i+1} REJECTED - outside tracking zone")
-                    continue
             
             # Create detection object
             detection = {
@@ -757,6 +764,11 @@ def run_detection_with_tracking(frame, node, context=None):
         for i, detection in enumerate(candidate_detections):
             logger.debug(f"🏆 Camera {camera_id}: Candidate {i+1}: {detection['class']} conf={detection['confidence']:.3f} track_id={detection['id']}")
         
+        # TEMP DEBUG: Show all candidate detections before selection
+        logger.info(f"🔍 TEMP DEBUG: Found {len(candidate_detections)} candidate detections:")
+        for i, det in enumerate(candidate_detections):
+            logger.info(f"🔍 TEMP DEBUG: Candidate {i+1}: {det['class']} conf={det['confidence']:.3f} bbox={det['bbox']}")
+        
         # Find the single highest confidence detection across all detected classes
         best_detection = max(candidate_detections, key=lambda x: x["confidence"])
         original_class = best_detection["class"]
@@ -808,6 +820,25 @@ def run_detection_with_tracking(frame, node, context=None):
         logger.info(f"✅ Camera {camera_id}: DETECTION COMPLETE - tracking single car: track_id={track_id}, conf={best_detection['confidence']:.3f}")
         logger.debug(f"📊 Camera {camera_id}: Detection summary: {len(res.boxes)} raw → {len(candidate_detections)} candidates → 1 selected")
         
+        # TEMP DEBUG: Save vehicle crop immediately after yolo detection
+        if node.get("modelId") in ["yolo11n", "yolo11m"] and regions_dict:
+            try:
+                import datetime
+                os.makedirs("temp_debug", exist_ok=True)
+                timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S_%f")[:-3]
+                
+                for class_name, region_data in regions_dict.items():
+                    bbox = region_data['bbox']
+                    x1, y1, x2, y2 = bbox
+                    cropped = frame[y1:y2, x1:x2]
+                    if cropped.size > 0:
+                        model_name = node.get("modelId", "yolo")
+                        debug_path = f"temp_debug/{model_name}_{class_name}_crop_{timestamp}.jpg"
+                        cv2.imwrite(debug_path, cropped)
+                        logger.info(f"🚗 TEMP DEBUG: Saved {model_name} {class_name} crop to {debug_path}")
+            except Exception as e:
+                logger.error(f"🚗 TEMP DEBUG: Failed to save {node.get('modelId', 'yolo')} crop: {e}")
+        
         # Update track-based stability tracking for the single selected car
         camera_id = context.get("camera_id", "unknown") if context else "unknown"
         model_id = node.get("modelId", "unknown")
@@ -826,28 +857,6 @@ def run_detection_with_tracking(frame, node, context=None):
         logger.debug(f"Detection error traceback: {traceback.format_exc()}")
         return [], {}, {"validation_complete": False, "stable_tracks": [], "current_tracks": []}
 
-def _point_in_polygon(point, polygon):
-    """Check if a point is inside a polygon using ray casting algorithm."""
-    if not polygon or len(polygon) < 3:
-        return True  # No zone restriction if invalid polygon
-    
-    x, y = point
-    n = len(polygon)
-    inside = False
-    
-    p1x, p1y = polygon[0]
-    for i in range(1, n + 1):
-        p2x, p2y = polygon[i % n]
-        if y > min(p1y, p2y):
-            if y <= max(p1y, p2y):
-                if x <= max(p1x, p2x):
-                    if p1y != p2y:
-                        xinters = (y - p1y) * (p2x - p1x) / (p2y - p1y) + p1x
-                    if p1x == p2x or x <= xinters:
-                        inside = not inside
-        p1x, p1y = p2x, p2y
-    
-    return inside
 
 def get_camera_stability_data(camera_id, model_id):
     """Get or create stability tracking data for a specific camera and model."""
@@ -1424,6 +1433,49 @@ def run_pipeline(frame, node: dict, return_bbox: bool=False, context=None, valid
             # Normal detection stage - Using structured detection function
             all_detections, regions_dict, track_validation_result = run_detection_with_tracking(frame, node, context)
         
+        # TEMP DEBUG: Save only specific crops
+        if regions_dict:
+            try:
+                import datetime
+                os.makedirs("temp_debug", exist_ok=True)
+                timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
+                model_id = node.get("modelId", "unknown")
+                
+                # Save vehicle crop from yolo model (any vehicle: car, truck, bus, motorcycle)
+                if model_id in ["yolo11n", "yolo11m"]:
+                    # Look for any vehicle class in regions_dict
+                    vehicle_classes = ["car", "truck", "bus", "motorcycle"]
+                    found_vehicle = None
+                    for vehicle_class in vehicle_classes:
+                        if vehicle_class in regions_dict:
+                            found_vehicle = vehicle_class
+                            break
+                    
+                        bbox = regions_dict[found_vehicle]['bbox']
+                        x1, y1, x2, y2 = bbox
+                        cropped = frame[y1:y2, x1:x2]
+                        if cropped.size > 0:
+                            debug_path = f"temp_debug/{found_vehicle}_crop_{timestamp}.jpg"
+                            cv2.imwrite(debug_path, cropped)
+                            logger.info(f"🚗 TEMP DEBUG: Saved {found_vehicle} crop to {debug_path}")
+                        else:
+                            logger.warning(f"🚗 TEMP DEBUG: Empty {found_vehicle} crop with bbox {bbox}")
+                    else:
+                        logger.warning(f"🚗 TEMP DEBUG: {model_id} detected but no vehicle classes found. Available: {list(regions_dict.keys())}")
+                
+                # Save frontal crop from frontal_detection_v1
+                elif model_id == "frontal_detection_v1" and "frontal" in regions_dict:
+                    bbox = regions_dict["frontal"]['bbox']
+                    x1, y1, x2, y2 = bbox
+                    cropped = frame[y1:y2, x1:x2]
+                    if cropped.size > 0:
+                        debug_path = f"temp_debug/frontal_crop_{timestamp}.jpg"
+                        cv2.imwrite(debug_path, cropped)
+                        logger.info(f"🔍 TEMP DEBUG: Saved frontal crop to {debug_path}")
+                        
+            except Exception as e:
+                logger.error(f"🔍 TEMP DEBUG: Failed to save crops: {e}")
+        
         if not all_detections:
             logger.debug("No detections from structured detection function - sending 'none' detection")
             none_detection = {