Aplicación de técnicas de aprendizaje de máquina para la detección de aglomeraciones de personas y comportamientos anómalos mediante segmentación semántica y redes convolucionales : modelo de redes convolucionales para la detección de comportamientos anómalos en videos de video vigilancia aplicando aprendizaje de instancia múltiple para el entrenamiento.

Abstract

The detection of anomalous behavior in humans refers to the process of discriminating events that deviate from normality. The main objective of this component is to develop a model for detecting anomalous behavior in surveillance camera videos. A pre-trained model is used for feature extraction from the videos. The extracted features serve as input data for a multiple instance learning based model that will oversee anomaly detection, specifically assault and robbery, and thus perform video classification. This model is intended to facilitate the monitoring of the images generated by the video surveillance systems and to serve as a guide for the implementation of a detection software incorporated in the cameras' visualization systems.