Make a face tracking camera on Raspberry Pi !

This article is recommended for such people!

Someone who wants to do something with Raspberry Pi
Someone who wants to do something with Python or machine learning
Someone who wants to make a face tracking camera

I tried to build a face tracking system based on Raspberry Pi and Python, machine learning technology.
I’m going to share to you how to make the face tracking camera today.

If you want to see more detail, please see the YouTube video for about 20 seconds.
You can see that the camera attached to the Raspberry Pi is chasing Mozart’s photo.
The camera has a servomotor which moves horizontally and vertically.
It is directed towards the face of Mozart.

I will write out the framework of this system, including Python Code, used materials, and so on…

Python Code

This system works with Python code.
I mainly referred to this web page.
(Reference：Raspberry Pi3 とカメラモジュールで顔追跡カメラを作る (Making face tracking camera with Raspberry Pi and camera module))

I further improved the code.

# -*- coding: UTF-8 -*-
import cv2
from picamera import PiCamera 
import io 
import numpy as np 
import os
import time
import pigpio
 
 
pi = pigpio.pi()
 
def move(x_move, y_move):
    pi.set_servo_pulsewidth(4, x_move)
    pi.set_servo_pulsewidth(17, y_move)
 
 
#Servomotor movable range of X, Y
X_MAX = 2500
X_MIN = 500
X_HOME = 1500 #Horizontal 
 
Y_MAX = 2500
Y_MIN = 1500
Y_HOME =2100 #Vertical
 
 
#Initialization of camera position
move(X_HOME, Y_HOME)
 
#Path of the library for face detection
cascade_path = "/home/pi/opencv-3.1.0/data/haarcascades/haarcascade_frontalface_alt.xml"
 
cascade = cv2.CascadeClassifier(cascade_path)
 
stream = io.BytesIO() #
 
camera=PiCamera()
 
W = 500
H = 500
 
camera.resolution = (W, H)
color = (255, 255, 255) #white
 
 
#Assign the initial position to the current position
now_degree_x, now_degree_y, move_degree_x, move_degree_y = X_HOME, Y_HOME, 0, 0
 
while(True):
    camera.capture(stream, format='jpeg') #
    data = np.fromstring(stream.getvalue(), dtype=np.uint8) #
    image = cv2.imdecode(data, 1) #
    
    frame = image
    facerect = cascade.detectMultiScale(frame, scaleFactor=1.2, minNeighbors=2, minSize=(10, 10))
    if len(facerect) > 0:
        for rect in facerect:
            img_x = rect[0]+rect[2]/2
            img_y = rect[1]+rect[3]/2
            print(img_x, img_y)
            #Calculation the between the face and the camera position.(250 at (W,H)/2)
            move_degree_x = now_degree_x + (-img_x+250)*0.8
            move_degree_y = now_degree_y + (img_y-250)*0.5
 
            #Move the camera only within the movable range of the motor
            #if ((move_degree_x >= 500 and move_degree_x <= 2500) and (move_degree_y >= 500 and move_degree_y <= 2500)):
            if move_degree_x >= X_MIN:
                if move_degree_x <= X_MAX:
                    if move_degree_y >= Y_MIN:
                        if move_degree_y <= Y_MAX:
                            #Move the camera towards the face
                            move(move_degree_x, move_degree_y)
                            now_degree_x = move_degree_x
                            now_degree_y = move_degree_y
                        else:
                            print("Can't see the bottom")
                    else:
                        print("Can't see the top")
                else:
                    print("Can't see the right")
            else:
                print("Can't see the left")
            #A circle describes the face recognition center point
            cv2.circle(frame, (int(img_x), int(img_y)), 10, color, -1)
            #The face recognition part is enclosed in a rectangle
            cv2.rectangle(frame, tuple(rect[0:2]),tuple(rect[0:2] + rect[2:4]), color, thickness=3)
    else:
        print("Where is your face?")
 
    cv2.imshow("Show FLAME Image", frame)
    k = cv2.waitKey(1)
    if k == ord('q'):
        break
    stream.seek(0)
 
cv2.destroyAllWindows()

# -*- coding: UTF-8 -*-

import cv2

from picamera import PiCamera

import io

import numpy as np

import os

import time

import pigpio

pi = pigpio.pi()

def move(x_move, y_move):

pi.set_servo_pulsewidth(4, x_move)

pi.set_servo_pulsewidth(17, y_move)

#Servomotor movable range of X, Y

X_MAX = 2500

X_MIN = 500

X_HOME = 1500 #Horizontal

Y_MAX = 2500

Y_MIN = 1500

Y_HOME =2100 #Vertical

#Initialization of camera position

move(X_HOME, Y_HOME)

#Path of the library for face detection

cascade_path = "/home/pi/opencv-3.1.0/data/haarcascades/haarcascade_frontalface_alt.xml"

cascade = cv2.CascadeClassifier(cascade_path)

stream = io.BytesIO() #

camera=PiCamera()

W = 500

H = 500

camera.resolution = (W, H)

color = (255, 255, 255) #white

#Assign the initial position to the current position

now_degree_x, now_degree_y, move_degree_x, move_degree_y = X_HOME, Y_HOME, 0, 0

while(True):

camera.capture(stream, format='jpeg') #

data = np.fromstring(stream.getvalue(), dtype=np.uint8) #

image = cv2.imdecode(data, 1) #

frame = image

facerect = cascade.detectMultiScale(frame, scaleFactor=1.2, minNeighbors=2, minSize=(10, 10))

if len(facerect) > 0:

for rect in facerect:

img_x = rect[0]+rect[2]/2

img_y = rect[1]+rect[3]/2

print(img_x, img_y)

#Calculation the between the face and the camera position.(250 at (W,H)/2)

move_degree_x = now_degree_x + (-img_x+250)*0.8

move_degree_y = now_degree_y + (img_y-250)*0.5

#Move the camera only within the movable range of the motor

#if ((move_degree_x >= 500 and move_degree_x <= 2500) and (move_degree_y >= 500 and move_degree_y <= 2500)):

if move_degree_x >= X_MIN:

if move_degree_x <= X_MAX:

if move_degree_y >= Y_MIN:

if move_degree_y <= Y_MAX:

#Move the camera towards the face

move(move_degree_x, move_degree_y)

now_degree_x = move_degree_x

now_degree_y = move_degree_y

else:

print("Can't see the bottom")

else:

print("Can't see the top")

else:

print("Can't see the right")

else:

print("Can't see the left")

#A circle describes the face recognition center point

cv2.circle(frame, (int(img_x), int(img_y)), 10, color, -1)

#The face recognition part is enclosed in a rectangle

cv2.rectangle(frame, tuple(rect[0:2]),tuple(rect[0:2] + rect[2:4]), color, thickness=3)

else:

print("Where is your face?")

cv2.imshow("Show FLAME Image", frame)

k = cv2.waitKey(1)

if k == ord('q'):

break

stream.seek(0)

cv2.destroyAllWindows()

I thought that some people might be interested in the code and write it first.
I’m going to introduce “The framework of the system” with Python Code.

System structure

The program is divided into three parts.
as illustration above,

Library import
Definition of parameters and various characters, initialization, etc.
Processing part (from “while(True)”)

I will explain important things in the system.

Library import

Definition of parameters and various characters, initialization, etc.

Installing of 2-axis servo motors!

In this system, two servomotors (2 SG90s) are used.

There are three lines per servomotors.
Each line is below.

Yellow : GPIO number 4 (X) and 17 (Y)
Red : To 5V power supply
Brown : To GND

Power and ground can be anywhere.
It is ok if you connect as shown below.

Determine the movable range of X-axis and Y-axis on the program

Set this movable range empirically while actually moving the servomotor !
(It’s safer!!)

So, please change X_MAX and Y_MAX directly at the final adjustment stage.

After that, it works by posting the program.

Although the product link is attached, the camera kit here is a little substandard.

In addition, there is no connector part linking the base stage and the servomotor! ! Lol

I was patient because the price was cheap.
I made it by processing the connecting part.

I also attach a camera module link, but if you have a webcam, that will do.

Important point

Please keep in mind that the connection confirmation of the camera and the GPIO pin can be used.

On each terminal,

vcgencmd get_camera
sudo pigpiod

Just press enter!

There’s Endless possibility with Raspberry Pi!!!!

You can do anything with Raspberry Pi!
The face tracking camera was also fairly easy to make.
Why not to make some systems using Raspberry Pi for everyone.