Activity detection using strings of symbols and Hidden Markov Models

Abstract

This paper presents a comprehensive approach to trajectory generation in computer vision, leveraging the representation of motion as compact chains. The proposed methodology incorporates a process to preserve motion information over time and identify areas of high probability of movement. Through morphological segmentation, simplified trajectories are constructed for a more compact representation. Hidden Markov Models (HMMs) are employed for activity modeling and recognition, benefiting from the reduction of visible states due to the compact representation of motion. This work demonstrates how this approach allows for precise and efficient detection of normal and abnormal activities, providing a robust framework for computer vision systems.