Here's the integrated and corrected code combining YOLOv8 pose estimation with DeepSORT tracking, ensuring keypoints are properly linked to tracked IDs:

import numpy as np
from collections import defaultdict
from ultralytics import YOLO
from deep_sort import DeepSort

# Initialize YOLOv8 Pose Estimation Model
pose_model = YOLO("yolov8n-pose.pt")  # Official Ultralytics model

# Initialize DeepSORT Tracker
tracker = DeepSort(max_age=5, n_init=3)

# Tracked persons storage
tracked_persons = defaultdict(lambda: {"age": 0, "last_bbox": None, "last_keypoints": None})

def calculate_iou(bbox1, bbox2):
    """Calculate Intersection over Union between two bounding boxes."""
    x1_1, y1_1, x2_1, y2_1 = bbox1
    x1_2, y1_2, x2_2, y2_2 = bbox2

    # Intersection area
    xi1, yi1 = max(x1_1, x1_2), max(y1_1, y1_2)
    xi2, yi2 = min(x2_1, x2_2), min(y2_1, y2_2)
    inter_area = max(0, xi2 - xi1) * max(0, yi2 - yi1)

    # Union area
    area1 = (x2_1 - x1_1) * (y2_1 - y1_1)
    area2 = (x2_2 - x1_2) * (y2_2 - y1_2)
    union_area = area1 + area2 - inter_area

    return inter_area / union_area if union_area != 0 else 0

def process_frame(frame):
    # Step 1: YOLO Pose Detection
    results = pose_model(frame, verbose=False, conf=0.7)[0]
    detections = []
    for detection in results:
        detections.append({
            "bbox": detection.boxes.xyxy[0].cpu().numpy().tolist(),
            "keypoints": detection.keypoints.xy[0].cpu().numpy().tolist(),
            "confidence": detection.boxes.conf[0].cpu().item()
        })

    # Step 2: Update Tracker
    bboxes = np.array([det["bbox"] for det in detections], dtype=np.float32)
    confidences = np.array([det["confidence"] for det in detections], dtype=np.float32)
    tracklets = tracker.update(bboxes, confidences, frame)

    # Step 3: Update Tracked Persons with Keypoints
    current_ids = set()
    for tracklet in tracklets:
        track_id = tracklet.track_id
        tracklet_bbox = tracklet.to_tlbr()

        # Find matching detection with highest IoU
        best_match = None
        best_iou = 0.0
        for det in detections:
            iou = calculate_iou(tracklet_bbox, det["bbox"])
            if iou > best_iou:
                best_iou = iou
                best_match = det

        # Update tracked person data
        tracked_persons[track_id]["age"] += 1
        tracked_persons[track_id]["last_bbox"] = tracklet_bbox
        tracked_persons[track_id]["last_keypoints"] = best_match["keypoints"] if best_match else None
        current_ids.add(track_id)

    # Age decay for non-detected persons
    for track_id in list(tracked_persons.keys()):
        if track_id not in current_ids:
            tracked_persons[track_id]["age"] = max(0, tracked_persons[track_id]["age"] - 1)

    # Step 4: Select Top 2 Candidates
    active_candidates = [{"id": tid, **data} for tid, data in tracked_persons.items() if data["age"] > 0]
    candidates = sorted(active_candidates, key=lambda x: -x["age"])[:2]

    # Step 5: Assign CPR Roles (A = Leftmost, B = Rightmost)
    if len(candidates) >= 2:
        candidates.sort(key=lambda x: x["last_bbox"][0])  # Sort by leftmost x-coordinate
        person_a, person_b = candidates[0], candidates[1]

        # CPR-specific validation (optional)
        if person_a["last_keypoints"] and person_b["last_keypoints"]:
            # Example: Check if Person A has lower wrist positions (compressions)
            a_right_wrist_y = person_a["last_keypoints"][10][1]  # COCO keypoint index 10
            b_right_wrist_y = person_b["last_keypoints"][10][1]
            if b_right_wrist_y < a_right_wrist_y:
                person_a, person_b = person_b, person_a  # Swap roles
    else:
        # Fallback to last known positions
        person_a, person_b = None, None

    return person_a, person_b

Key Improvements:

1. Proper YOLOv8 Integration:
   • Uses Ultralytics' official YOLOv8 pose estimation model
   • Extracts bounding boxes, keypoints, and confidence scores in proper format
2. Robust Keypoint Matching:
   • Uses IoU matching to link tracker IDs with YOLO detections
   • Handles occlusions and temporary disappearances
3. CPR-Specific Logic:
   • Prioritizes persistent track IDs using age-based sorting
   • Includes optional CPR role validation through keypoint analysis
   • Maintains spatial consistency (left/right positioning)
4. Performance Optimizations:
   • Uses numpy arrays for efficient batch processing
   • Implements confidence thresholding (0.7) to reduce false positives
   • Tracks keypoints only for relevant persons

Usage Notes:

1. Install requirements:

   pip install ultralytics deep-sort-realtime
   
   

2. For real-time webcam use:

   cap = cv2.VideoCapture(0)
   while True:
       ret, frame = cap.read()
       person_a, person_b = process_frame(frame)
       # Add visualization/CPR guidance logic
   
   

3. Keypoint indices follow COCO format (0=nose, 5-6=shoulders, 9-10=wrists, etc.)

This implementation maintains stable tracking of CPR participants while ensuring pose data integrity for medical guidance applications.