Overview / Usage

Using digital image processing moving objects are identified which are extracted and then fed to CNN to classify if motion is caused by humans or not.

Methodology / Approach

First the security feed is given as input where background subtraction algorithm is used to identify the change in pixel values. This results in white pixels, which are diluted to cover small area. Then based on some assigned threshold value covering minimum area that particular area is extracted from the image and stored. These extracted images which are continuously generated are continually fed into the neural network and then classified.

The model has 4 layer neural network with the hidden layers having activation functions "ReLU" and output layer "Sigmoid". The images are first converted into 64x64 single channel and fed into the network for training. As mentioned above the output is displayed on the security feed whether the motion is caused by human or not.

Technologies Used

For smooth working use 144p-240p video footage.

Tensorflow, Keras, and OpenCV is used.

Python packages used:

from imutils.video import VideoStream

import cv2

import tensorflow as tf

import argparse

import datetime

import imutils

import time

import cv2

import numpy as np

import matplotlib.pyplot as plt

import os

from tqdm import tqdm

import tensorflow as tf

from tensorflow.keras.datasets import cifar10

from tensorflow.keras.preprocessing.image import ImageDataGenerator

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten

from tensorflow.keras.layers import Conv2D, MaxPooling2D

import pickle

Repository

https://github.com/santos97/CNN-Keras-TF-Surveillance-System