Face Counting using Haarcascade and YOLOv3

Preface

This blog will guide you through creating a face counting system using Haarcascade and YOLOv3. The system reads a video file, detects faces using both methods, and counts the faces only when the bounding boxes of both algorithms intersect.

The repository for this tutorial:

https://github.com/fiqgant/face_counting_haarcascade_yolov3

Project Structure

haarcascade_frontalface_default.xml

yolov3.onnx

face_counting.py

video.mp4

How to Create the Face Counting System

First, ensure you have the necessary libraries installed:

Terminal

pip install numpy opencv-python opencv-python-headless onnxruntime

Next, we will create a Python script to implement the face counting system.

Importing Libraries and Loading Models

We begin by importing the required libraries and loading the Haarcascade and YOLOv3 models.

face_counting.py

import cv2
import numpy as np
import onnxruntime as ort

# Load the Haarcascade model
haarcascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')

# Load the YOLOv3 model
ort_session = ort.InferenceSession('yolov3.onnx')
input_name = ort_session.get_inputs()[0].name

Processing the Video

Read the video file and initialize variables for counting faces.

face_counting.py

# Read the video file
cap = cv2.VideoCapture('video.mp4')

# Initialize the face count
face_count = 0

Detecting Faces

Create functions to detect faces using Haarcascade and YOLOv3.

face_counting.py

def detect_faces_haarcascade(frame):
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    faces = haarcascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5)
    return faces

def detect_faces_yolo(frame):
    blob = cv2.dnn.blobFromImage(frame, 1/255.0, (416, 416), swapRB=True, crop=False)
    ort_inputs = {input_name: blob}
    detections = ort_session.run(None, ort_inputs)
    faces = []
    for detection in detections[0]:
        confidence = detection[4]
        if confidence > 0.5:
            x, y, w, h = (detection[0:4] * np.array([frame.shape[1], frame.shape[0], frame.shape[1], frame.shape[0]])).astype(int)
            faces.append((x, y, w, h))
    return faces

Counting Faces

Process each frame of the video and count the faces only if both Haarcascade and YOLOv3 detect and intersect the same face.

face_counting.py

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

    # Detect faces using both methods
    faces_haarcascade = detect_faces_haarcascade(frame)
    faces_yolo = detect_faces_yolo(frame)

    # Count faces based on intersection of both methods
    for (x1, y1, w1, h1) in faces_haarcascade:
        for (x2, y2, w2, h2) in faces_yolo:
            if (x1 < x2 + w2 and x1 + w1 > x2 and y1 < y2 + h2 and y1 + h1 > y2):
                face_count += 1
                cv2.rectangle(frame, (x1, y1), (x1 + w1, y1 + h1), (0, 255, 0), 2)
                break

    cv2.imshow('Face Counting', frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()
print(f'Total faces counted: {face_count}')

Conclusion

This tutorial demonstrated how to create a face counting system by combining Haarcascade and YOLOv3. By leveraging the strengths of both methods, we can achieve more accurate face detection and counting. Feel free to explore the GitHub repository for more details and additional functionalities.

Useful links

Feel free to reach out with any questions or comments about this tutorial!