Updated tracking to support multi-lemon tracking

- Improved tracking logic to handle multiple lemons simultaneously
- Added necessary YAML configuration for BoT-SORT

Higher Sensitivity to Minor Defects:
- Weighted “DefectiveLemon” more heavily
- Extended HISTORY_LENGTH for improved tracking stability

Test_Track_updated.py

@@ -0,0 +1,125 @@
+import cv2
+from ultralytics import YOLO
+from collections import deque
+import os
+
+# Allow duplicate loading of OpenMP runtime
+os.environ["KMP_DUPLICATE_LIB_OK"] = "True"
+
+# Define the official YAML configuration file path (adjust as needed)
+yaml_path = "botsort.yaml"
+
+# Camera index (default camera index, 1 indicates an external camera)
+camera_index = 1
+
+# Load the YOLO model
+model = YOLO(r"D:\AIM\lemon\runs\detect\train4\weights\best.pt")  # Load custom model
+
+# Define class labels
+class_labels = {
+    0: "Bruised",
+    1: "DefectiveLemon",
+    2: "GoodLemon",
+    3: "NotRipeLemon",
+    4: "Rotten"
+}
+
+# Apply smoothing to "DefectiveLemon", "GoodLemon", and "NotRipeLemon"
+smoothing_labels = ["DefectiveLemon", "GoodLemon", "NotRipeLemon"]
+
+# Smoothing parameters for sliding window
+HISTORY_LENGTH = 20  # Number of recent frames
+DEFECT_THRESHOLD = 0.3  # Threshold for "DefectiveLemon" proportion
+GOOD_THRESHOLD = 0.7    # Threshold for "GoodLemon" and "NotRipeLemon" proportion
+
+# State history for each target (used for smoothing), format: {ID: deque([...], maxlen=HISTORY_LENGTH)}
+lemon_history = {}
+
+# Set the display window to be resizable
+cv2.namedWindow("Live Detection", cv2.WINDOW_NORMAL)
+
+# Smoothing function:
+# If the current detected label is not in smoothing_labels, clear the target's history and return the current label;
+# Otherwise, add the current label to the history and return a smoothed label based on the proportion.
+def get_smoothed_label(obj_id, current_label):
+    if current_label not in smoothing_labels:
+        if obj_id in lemon_history:
+            lemon_history[obj_id].clear()
+        return current_label
+
+    if obj_id not in lemon_history:
+        lemon_history[obj_id] = deque(maxlen=HISTORY_LENGTH)
+    lemon_history[obj_id].append(current_label)
+
+    history = lemon_history[obj_id]
+    defect_count = history.count("DefectiveLemon")
+    good_count = history.count("GoodLemon")
+    notripe_count = history.count("NotRipeLemon")
+    total = len(history)
+
+    if total == 0:
+        return current_label
+    if defect_count / total >= DEFECT_THRESHOLD:
+        return "DefectiveLemon"
+    elif good_count / total >= GOOD_THRESHOLD:
+        return "GoodLemon"
+    elif notripe_count / total >= GOOD_THRESHOLD:
+        return "NotRipeLemon"
+    else:
+        return history[-1]
+
+# Use streaming tracking mode to maintain tracker state
+results = model.track(
+    source=camera_index,    # Get video stream directly from the camera
+    conf=0.5,
+    tracker=yaml_path,      # Use the YAML configuration file
+    persist=True,           # Persist tracking (do not reset)
+    stream=True,            # Stream processing, not frame-by-frame calling
+    show=False
+)
+
+# Iterate over streaming tracking results
+for result in results:
+    frame = result.orig_img  # Current frame
+    detections = result.boxes  # Detection box information
+
+    for box in detections:
+        x1, y1, x2, y2 = map(int, box.xyxy[0])  # Detection box coordinates
+        obj_id = int(box.id) if box.id is not None else -1  # Tracking ID
+        class_id = int(box.cls)  # Class ID
+        score = box.conf  # Confidence
+        label = class_labels.get(class_id, "Unknown")  # Get class name
+
+        # If target ID is valid
+        if obj_id != -1:
+            # If the detected label requires smoothing, use the smoothing function
+            if label in smoothing_labels:
+                final_label = get_smoothed_label(obj_id, label)
+                display_text = f"ID {obj_id} | {final_label}"
+                # Only print for targets with smoothed labels (only care about these three classes)
+                if final_label in smoothing_labels:
+                    position = f"({x1}, {y1}, {x2}, {y2})"
+                    print(f"ID: {obj_id}, Position: {position}, Label: {display_text}")
+            else:
+                # For other classes, display the current detection result directly and clear history (if exists)
+                if obj_id in lemon_history:
+                    lemon_history[obj_id].clear()
+                display_text = label
+        else:
+            display_text = label
+
+        # Draw detection box and label
+        cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
+        cv2.putText(frame, display_text, (x1, y1 - 10),
+                    cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0), 2)
+
+    # Display the processed frame
+    cv2.imshow("Live Detection", frame)
+
+    # Exit program when ESC key is pressed
+    if cv2.waitKey(1) & 0xFF == 27:
+        print("ESC key detected. Exiting the program.")
+        break
+
+cv2.destroyAllWindows()
+print("Camera video processing complete. Program terminated.")

botsort.yaml

@@ -0,0 +1,21 @@
+# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
+
+# Default Ultralytics settings for BoT-SORT tracker when using mode="track"
+# For documentation and examples see https://docs.ultralytics.com/modes/track/
+# For BoT-SORT source code see https://github.com/NirAharon/BoT-SORT
+
+tracker_type: botsort # tracker type, ['botsort', 'bytetrack']
+track_high_thresh: 0.25 # threshold for the first association
+track_low_thresh: 0.1 # threshold for the second association
+new_track_thresh: 0.4 # threshold for init new track if the detection does not match any tracks
+track_buffer: 30 # buffer to calculate the time when to remove tracks
+match_thresh: 0.7 # threshold for matching tracks
+fuse_score: True # Whether to fuse confidence scores with the iou distances before matching
+# min_box_area: 10  # threshold for min box areas(for tracker evaluation, not used for now)
+
+# BoT-SORT settings
+gmc_method: sparseOptFlow # method of global motion compensation
+# ReID model related thresh (not supported yet)
+proximity_thresh: 0.5
+appearance_thresh: 0.25
+with_reid: False