International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 10 Issue: 05 | May 2023
p-ISSN: 2395-0072
www.irjet.net
Real-time Anomaly Detection and Alert System for Video Surveillance Seemantula Nischal1, Bhunesh.k2, Ashwin Sathappan v3 1,2 B.E Student, Computer Science, RMD Engineering College, Chennai, India
3 B.Tech Student, Information Technology, RMD Engineering College, Chennai, India
-----------------------------------------------------------------------***--------------------------------------------------------------------Abstract - The objective of this project is to develop a real-time video surveillance system that can detect and classify various
types of anomalies such as theft, street crime, unauthorized access and burglary. The system will use a combination of deep learning models, such as CenterNet and Graph Convolutional Networks (GCN), to detect and classify the anomalies in real-time CCTV footage. Once an anomaly is detected, the system will trigger an alert to the nearest police station the system is incorporated with Twilio Video API to send alerts to officials in real-time and send a video message along with the location information of the anomaly. The system stores information about the anomaly in a database [MindsDB], including the type of anomaly, severity level, and longitude and latitude of the location. Overall, the real-time anomaly detection and alert system for video surveillance will provide an efficient way to detect and track criminal activities in real-time, enhancing public safety and security.
Key Words: Real-time anomaly Detection, UCF Dataset, MindsDB, Crime India Dataset, Twilo Video-based API 1. INTRODUCTION Video surveillance systems are critical to public safety and security. Real-time monitoring and analysis of video footage have become more important in detecting and responding to potential threats or unlawful behavior. Traditional surveillance systems rely heavily on manual monitoring, which is time-consuming and prone to human error. As a result, there is a growing demand for automated systems that can detect and categorize anomalies in real-time video streams. Detecting and categorizing a wide range of anomalies such as theft, street crime, illegal access, and burglary. By using breakthroughs in deep learning models, we want to increase video surveillance capabilities for efficient anomaly identification. To achieve this goal, we employ a range of deep learning models This research project's purpose is to develop a real-time video surveillance system capable of, including CenterNet and Graph Convolutional Networks (GCN). These models have proven to be extremely effective in object detection and identifying spatial relationships in complex circumstances. CenterNet's object recognition is precise and efficient, allowing us to discover anomalies inside video frames. GCN, on the other hand, captures contextual information and object relationships using the power of graph-based representations, allowing for robust anomaly categorization. When an anomaly is detected, our system uses the Twilio Video API to notify the nearest police station. Because the warnings include real-time video messaging and location information, authorities can respond and interfere immediately. Furthermore, precise information about each abnormality is saved in a database (MindsDB). such as its classification, severity degree, and exact geographical location. This allows for comprehensive post-analysis and a better knowledge of the patterns and trends related to criminal activities. Standard surveillance techniques are outperformed by the proposed real-time anomaly identification and alert system. It boosts the efficiency and effectiveness of video surveillance by automating the detection and classification of abnormalities. The ability to swiftly alert authorities to potential threats enables faster response times and proactive intervention, resulting in a safer and more secure environment. In the next parts, we will look at the technical aspects of our system, such as the deep learning models employed, the alert generating process, database storage, and so on. is category, severity degree, and specific geographical location. This allows for comprehensive post-analysis and a better knowledge of the patterns and trends related to criminal activities.
© 2023, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 1493