Deep Neural Network Based Smart Intrusion Detection and Alerting System

With the increasing prevalence of surveillance systems, the need for an automated system that detects the presence of intruders from the surveillance camera and send an alert message to the user immediately has become very crucial. The research paper aims to provide a deep learning (DL) model-based solution to accurately detect the presence of human figures in the frame. The approach includes using YOLOv8 (You Only Look Once – Version 8) object detection model to detect human figures, paired with the implementation of the Non-Maximum Suppression technique to remove duplicate bounding boxes around humans. Additionally, Gaussian Noise Reduction technique has been used to remove Gaussian blur from the image for obtaining a clearer image of human figure present in the frame. The research paper demonstrates a procedure to quickly send alert message to the user when a human figure is detected in the camera using the Telegram application. The paper also demonstrates the implementation of frame skipping method based on periodic intervals for optimizing the computational efficiency. In addition to this, the paper also demonstrates a comparative study between the frequently used human detection algorithms namely YOLOv8, YOLOv4, SSD and HOG and provides statistical proof on how YOLOv8 outperforms the other standard models.