A predictive model for recognizing human behaviour based on trajectory representation

The automatic understanding of the behaviour conducted by humans in scenarios using images as input of the system is a very important and challenging problem involving different areas of computational intelligence. In this paper human activity recognition is studied from a prediction point of view. We propose a model that, in addition to the capabilities of it to predict behaviour from new inputs, it is able to detect behaviour using a portion of the input. Specifically, we propose a prediction activity method based on the Activity Description Vector (ADV) to early detect the behaviour performed by a person in a scene. ADV is used to extract features that are normalized to be the cue of behaviour classifiers. We use complete sequences for training and partial sequences to evaluate the prediction capabilities having a specific observation time of the scene. CAVIAR dataset and different classic classifiers have been used for experimentation in order to evaluate the proposal obtaining great accuracy on the early recognition.