Abstract
With the advent of wearable devices in Computer Vision, a new first person perspective in video sequences is possible. By studying this innovative perspective, a new approach to human behaviour analysis and social interactions is made possible.
This thesis proposes to study, project and implement a computer vision framework capable of estimating and classifying human interactions in a video sequence recorded by wearable cameras.
The proposed method uses face detection and tracking tools in order to recognize and follow the presence of human beings in the scene, SVM classifiers based on HOG features and HMM stateful models in order to estimate head pose and SVR regressors in the distance estimation.
Such informations are then used in the estimation of social interactions between the subjects participating in the recorded event.
|