ABNORMAL BEHAVIOUR DETECTION

Abstract: Modern security teams in busy public spaces often struggle to keep up with ever-growing video surveillance demands, missing crucial incidents due to human fatigue and excessive false alarms. This project puts forward a smart, automated surveillance tool that uses only the movement patterns captured by cameras-never faces or identifying features-to spot genuinely suspicious activities like fights or panic situations in crowds. Built using open-source tools on regular computers, the system breaks up video footage, analyzes how people move, and recognizes the difference between everyday strolls and dangerous behavior, all while preserving privacy. Its machine learning algorithms learn from thousands of real-life examples, so it's accurate and alert only when it really matters, drawing attention to anomalies with clear boxes and labels on the screen. In essence, this research marks a step toward smarter, fairer public safety, making advanced surveillance accessible and ethical for crowded places everywhere.

Keywords: Abnormal Behaviour Detection, Computer Vision, Optical Flow Analysis, Machine Learning, Crowd Surveillance, SVM (Support Vector Machine), KNN (K-Nearest Neighbours), Logistic Regression.