International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 04 | Apr 2025
p-ISSN: 2395-0072
www.irjet.net
HUMAN VIOLENT ACTIVITY RECOGNIZATION Prof. S. V. Patil1, Shubham Lanke2, Shubham Mukhekar3, Amol Marathe4, Manav Chauhan5 *1Assistant Professor, Department Of Computer Engineering, Sinhgad College Of Engineering, Pune, India. *2,3,4,5Department Of Computer Engineering, Sinhgad College Of Engineering, Pune, India.
DOI : https://www.doi.org/10.56726/IRJMETS48058 ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The conventional approaches to violence
unsupervised learning techniques are incorporated to handle violent behavior detection without the need for extensive labeled datasets. The system is designed as a cross-platform application, compatible with both mobile and desktop devices, allowing administrators to receive alerts and manage incidents remotely. Its adaptability also allows it to run on low-cost hardware, including embedded systems and mobile devices, making it user-friendly and accessible. In terms of motivation, violence detection has been a challenging area in computer vision, with applications in surveillance, behavior analysis, and human-computer interaction. The project aims to improve real-time surveillance by overcoming limitations related to low-cost sensors through the use of deep learning and CNNs, thereby enhancing public safety in high-risk environments. The main objectives of the project are to detect violent activities in real-time and provide an easy-to-use interface for administrators to receive timely alerts.
detection primarily focus on predicting body part or joint locations from images or videos. However, these methods face limitations, such as difficulty estimating human poses due to low-resolution and noisy data from depth sensors, especially in real-time settings like CCTV footage. Human violence detection has been a key problem in computer vision for over last many years due to its numerous applications, such as video surveillance, human-computer interaction, and behavior analysis. This project aims to address these challenges by using neural networks to detect violent human activity from realtime surveillance footage. The primary features of this system include: 1. Real-time Violence Detection 2. Violence Activity Monitoring 3. Automated Alerts 4. Categorization of Violence and Non-Violence Activities Our intelligent video surveillance system can be applied in public areas like airports, malls, schools, and parking lots to monitor human activities and prevent incidents such as theft, accidents, and vandalism. The system provides real-time monitoring and can automatically generate alerts when unusual or violent activities are detected, helping enhance public safety.
2. LITERATURE REVIEW In the paper Real-Time Violence Detection and Localization in Crowded Scenes by Mohammad Sabokrou and Mahmood Fathy, a method is proposed for real-time detection and localization of violent activities in crowded areas. The approach segments each video into non-overlapping cubic patches, using two types of descriptors—local and global—to capture various video properties. These descriptors analyze structural similarities between adjacent patches to differentiate normal activities from anomalies. By using costeffective Gaussian classifiers and learning HUMAN VIOLENT ACTIVITY RECOGNIZATION Shubham Lanke, Shubham Mukhekar, Amol Marathe, Manav Chauhan Pranali D. Dahiwal B.E. Computer Engineering, Sinhgad College of Engineering, Pune, India features in an unsupervised manner with a sparse autoencoder, the system can efficiently identify violent actions. Experimental results demonstrate the algorithm’s effectiveness and comparable accuracy to state-of-the-art methods, with improved time efficiency, especially on the UCSD ped2 and UMN benchmark datasets. However, a notable limitation is that scalability to larger real-world environments may affect performance [1]. In the paper Learning Temporal Regularity in Video Sequences by Mahmudul Hasan and Jonghyun Choi, the authors explore methods to identify meaningful activities in long, cluttered video sequences. This challenge is heightened by the subjective nature of what constitutes ‘meaningful’ activity. To address this, they propose a generative model that learns regular motion patterns, or ‘regularity,’ using multiple data
Key Words: Violence Detection, Real-time Monitoring, Violence Activity Recognition, Neural Networks, Intelligent Surveillance.
1.INTRODUCTION The project focuses on developing a real-time violence detection system for public places such as malls, airports, and railway stations, utilizing deep learning, neural networks, and computer vision. To understand the entire project, familiarity with deep learning, neural networks, computer vision, video analysis, and desktop application development is essential. Deep learning is used to enable the model to learn from extensive datasets and recognize patterns in video frames, with neural networks helping to identify human body parts and detect violent movements. The project leverages these techniques to analyze real-time CCTV footage and trigger alerts upon detecting violent activities. The application has potential uses beyond security, such as monitoring patient movements in healthcare facilities to prevent injuries. Computer vision plays a vital role in detecting and classifying human poses within video data, introducing temporal complexity that requires advanced techniques to track movements over time. By analyzing video frames, the system can efficiently detect violent actions in real time, essential for its deployment in diverse public spaces. Furthermore,
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 258