SUSPICIOUS HUMAN ACTIVITY RECOGNITION FROM SURVEILLANCE VIDEOS USING DEEP LEARNING

COMPUTER SCIENCE AND ENGINEERING

Suspicious Human activity recognition (SHAR) is crucial for improving surveillance and security systems by recognizing and reducing possible hazards in different situations. Despite the abundance of research on the subject of SHAR, current methods frequently need to be revised with restricted levels of precision and efficiency. We aim to address the problem of inaccurate and inefficient activity recognition in surveillance systems through rigorous data collection, preparation, and model training. By leveraging Convolutional Neural Networks (CNNs) and deep learning architectures, including our time-distributed CNN and Conv3D models, it achieved improved accuracy rates of 90.14% and 88.23%. The exponential growth of CCTV installations in public and private spaces has revolutionized surveillance practices but also introduced challenges in monitoring and analysis. Traditionally, security personnel manually review surveillance footage, which is labor-intensive, time-consuming, and prone to errors caused by fatigue and oversight. To address these limitations, this project proposes an automated system powered by Machine Learning (ML) and Deep Learning (DL) technologies to detect suspicious activities in real-time, significantly enhancing efficiency and accuracy. The system utilizes Convolutional Neural Networks (CNNs), a proven tool in computer vision, to analyze frames extracted from video feeds. These frames are preprocessed for noise reduction and quality optimization before feeding into the CNN model. The trained model identifies unusual patterns based on spatial and temporal dynamics, classifying activities as either “Normal” or “Suspicious.” Suspicious activities such as theft, aggression, or unauthorized movements trigger instant alerts, enabling security personnel to respond swiftly. The system also securely stores processed data for later analysis, supporting forensic investigations and improving threat prevention mechanisms. Designed for scalability, the system can operate across diverse environments, including airports, malls, corporate offices, and residential areas. By automating labor-intensive tasks, it reduces human workload while improving detection reliability and response times. The integration of multimodal analysis, combining video and audio input, further refines its accuracy. Predictive modeling techniques offer proactive threat identification, and edge computing ensures decentralized, real-time data processing directly at surveillance sites. With its advanced capabilities and adaptive design, this system positions itself as a vital tool in modern security frameworks, offering improved monitoring, reduced operational costs, and enhanced safety across large-scale installations.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.