main.py

## ====================================================================
## ADVANCED HUB ATTENDANCE (Gender Voting + Blur Filter)
## ====================================================================
from ultralytics import YOLO
import cv2
import os
import csv
from datetime import datetime
from deepface import DeepFace
import numpy as np
import warnings
from collections import Counter

# Suppress TensorFlow noise
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
warnings.filterwarnings("ignore")

# --- Config ---
SNAPSHOTS_DIR = "people_snapshots"
LOG_FILE = "attendance_gender.csv"
VOTES_REQUIRED = 5    # We need 5 good samples before we "confirm" a person
os.makedirs(SNAPSHOTS_DIR, exist_ok=True)

# Load YOLO model (Small is better for movement)
model = YOLO('yolov8s.pt')

def find_camera():
    for index in range(10):
        cap = cv2.VideoCapture(index)
        if cap.isOpened():
            print(f"[#] Camera found at index {index}")
            return cap
        cap.release()
    return None

def is_blurry(image, threshold=50):
    """Returns True if the image is too blurry for accurate analysis."""
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    score = cv2.Laplacian(gray, cv2.CV_64F).var()
    return score < threshold

cap = find_camera()
if not cap:
    print("[!] Error: Could not access webcam")
    exit()

# State Management
gender_votes = {}     # track_id -> [list of gender guesses]
final_results = {}    # track_id -> confirmed gender
processed_ids = set() # track_id -> bool (logged to CSV)

# CSV Setup
with open(LOG_FILE, "a", newline="") as f:
    f.seek(0, 2)
    if f.tell() == 0:
        csv.writer(f).writerow(["timestamp", "person_id", "gender", "snapshot"])

print("-" * 50)
print(" GENDER ANALYSIS (Voting System enabled) ")
print(" Analysis will confirm after 5 clear frames ")
print(" Press 'q' to quit ")
print("-" * 50)

while True:
    ret, frame = cap.read()
    if not ret: break

    # Track bodies
    results = model.track(frame, persist=True, classes=[0], conf=0.3, tracker="bytetrack.yaml", verbose=False)

    for r in results:
        if r.boxes.id is None: continue
        
        boxes = r.boxes.xyxy.cpu().numpy().astype(int)
        track_ids = r.boxes.id.cpu().numpy().astype(int)

        for box, track_id in zip(boxes, track_ids):
            x1, y1, x2, y2 = box
            
            # 1. Skip if already finalized
            if track_id in final_results and track_id in processed_ids:
                color, label = (0, 255, 0), f"ID:{track_id} | {final_results[track_id]}"
            else:
                color, label = (0, 255, 255), f"Analyzing ID:{track_id}..."

                # 2. Try to analyze this person's gender
                person_crop = frame[max(0, y1):y2, max(0, x1):x2]
                
                # Check quality before analyzing
                if person_crop.size > 1000 and not is_blurry(person_crop, threshold=40):
                    try:
                        # Analyze Gender
                        # Use RetinaFace for detection inside DeepFace for better small-face results
                        res = DeepFace.analyze(person_crop, actions=['gender'], enforce_detection=True, detector_backend='opencv', silent=True)
                        gender = res[0]['dominant_gender']
                        
                        # Add to voting pool
                        if track_id not in gender_votes: gender_votes[track_id] = []
                        gender_votes[track_id].append(gender)
                        
                        # 3. If we have enough votes, finalize!
                        if len(gender_votes[track_id]) >= VOTES_REQUIRED:
                            # Take the most frequent result
                            vote_counts = Counter(gender_votes[track_id])
                            confirmed_gender = vote_counts.most_common(1)[0][0]
                            final_results[track_id] = confirmed_gender
                            
                            # Log and Snapshot
                            if track_id not in processed_ids:
                                ts = datetime.now().strftime("%H%M%S")
                                snap_name = f"person_{track_id}_{confirmed_gender}_{ts}.jpg"
                                cv2.imwrite(os.path.join(SNAPSHOTS_DIR, snap_name), person_crop)
                                with open(LOG_FILE, "a", newline="") as f:
                                    csv.writer(f).writerow([datetime.now().strftime("%Y-%m-%d %H:%M:%S"), track_id, confirmed_gender, snap_name])
                                
                                print(f" [✓] Finalized ID {track_id}: {confirmed_gender}")
                                processed_ids.add(track_id)
                    except:
                        pass # No face found or blurry - wait for a better frame

            # Drawing
            cv2.rectangle(frame, (x1, y1), (x2, y2), color, 2)
            cv2.putText(frame, label, (x1, y1-10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)

    cv2.imshow("Hub Attendance Analysis", frame)
    if cv2.waitKey(1) & 0xFF == ord("q"): break

cap.release()
cv2.destroyAllWindows()