Deep Learning Approach for detecting Suspicious Activity in Surveillance Videos

Abstract

With the increasing need for public safety and security, the timely detection of suspicious activities in public spaces has become a critical task. This paper presents a novel approach for real-time suspicious activity detection from CCTV footage using neural networks. The proposed system leverages human pose estimation to get valuable aspect from video streams, which are then fed into a carefully designed neural network model for activity classification.

The methodology involves data collection, pre-processing, and feature extraction, followed by the construction of a neural network architecture optimized for activity recognition. The model is built using through a comprehensive dataset of labeled videos containing various normal and suspicious activities. Evaluation metrics are employed to evaluate the performance of system and compare it with existing methods.

Through extensive experimentation and analysis of results, our approach demonstrates promising accuracy in detecting suspicious activities, outperforming traditional methods. The system's architecture allows for real-time implementation, enabling swift responses to potential security threats.

The outcomes of this research give valuable perception into the effective utilization of neural networks for real-time suspicious activity detection, contributing to enhanced public safety measures. The proposed solution can find applications in surveillance systems, law enforcement, and public security domains, fostering a safer and more secure environment for everyone.