DETECTING HEAD ORIENTATION

INTRODUCTION

The goal of this project is to train a neural network to classify the orientation of a centered image of a head as either left, right, up, or down. Head pose gives information about what someone is paying attention to, and as such is important for social interaction and for mediating learning. Estimating head pose is an active area of research in computer vision. In child development, sensitivity to head and eye position matures through a reasonably clear series of milestones. The earliest of these occurs within about six months, when an infant can decide whether someone is turned to gaze to their left or to their right, but cannot accurately determine their locus of attention. In this project, attempt to replicate this level of competence with a simple neural network that can distinguish between faces turned left, right, up and down 

First, a classifier is built which makes the same decision as the desired network, but does so using motion information – it needs to see the face turning from the frontal position to a new pose in order to classify that pose. This classifier is relatively straightforward to construct.

The motion-based classifier is used to collect and label examples for training the neural network. When the neural network has been trained, it has the same functionality as the original classifier, except it does not rely on motion information. It doesn’t have to see how the head moved to a given pose in order to classify that pose.

JOINT ATTENTION

It is important for joint visual attention – “looking where someone else is looking”. This is key to mediating social interaction and learning, currently topics of strong interest in humanoid robotics. Mechanisms of joint attention include:  

. Mutual gaze – maintaining eye contact.

. Gaze following – locating the target of another’s gaze.

. Imperative pointing – grasping towards an object that is out of reach. May be interpreted            by parent of a child as a request for that object, and soon comes to have that meaning for the        child too.

. Declarative pointing – drawing someone’s attention to an object by pointing; not necessarily         a request for the object.

. The first stage is sensitivity to field (left or right). At 6 months, children show sensitivity to           whether the care-giver is looking to their left or right.

. Then there is an “ecological stage” where the first object along the gaze direction as              projected onto the child’s point of view is chosen.

. This is followed by a “geometric stage” where full three-dimensional gaze information is     used, but the child will not turn to search behind it even when that is indicated by the geometry.

. Finally there is a “representational stage” where the child will for the first time turn to look behind her if necessary.