Problem

Public spaces can be prone to violence and other dangerous behaviours, making it difficult for security personnel to monitor and respond in real time. Traditional surveillance methods are often not enough to detect anomalies before they escalate into something more serious.

Our team was tasked with creating a solution that could accurately predict physical violence in real-time using AI and video analytics.

Solution

We developed an AI model using pedestrian detection, tracking, pose estimation, and neural networks to predict physical violence in public spaces. Our model achieved over 98% accuracy and could be extended to classify other types of behaviour as well.

We employed OpenCV's functions for Gaussian Blur, Sobel Operators, and Morphological Transformations to aid in object recognition and de-raining. We also utilized contours via OpenCV to further improve object recognition.

Anomaly Detection in Public Spaces