Efficient and Sustainable Video Surveillance Using CNN-LSTM Model for Suspicious Activity Detection

Abstract

This study presents a novel approach for enhancing the automation and effectiveness of real-time threat detection in video surveillance systems. Traditional surveillance methods require continuous human monitoring, are resource-intensive, and often fail to consistently identify suspicious activities with precision. Addressing these challenges, we propose the Mono-Scale CNN-LSTM Fusion Network, an advanced deep-learning model designed for automated, sustainable, and high-accuracy CCTV systems. The model utilizes Convolutional Neural Networks (CNN) in combination with Long Short-Term Memory (LSTM) networks to improve recognition capabilities by capturing tempo-ral and spatial features. For feature extraction, the Oriented FAST and Rotated BRIEF (ORB) techniques are employed to enhance detection efficiency. The model was tested using the UCF crime image dataset and achieved an accuracy rate of approximately 99%, surpassing traditional models like CNN, VGG-16, VGG-19, ResNet-50, and DenseNet. This study highlights the contributions of our approach, which offers a significant reduction in the need for human oversight and sets new standards in the field of automatic threat detection. Furthermore, it emphasizes the model’s capability to support contemporary security systems with high precision, reliability, and scalability, making it a valuable tool for the next generation of intelligent surveillance systems.