This is an interdisciplinary project in collaboration with School of Psychology.

Human babies/young children show fascinating performance when trying to learn from the real-world environment which can often be cluttered, using a combination of their personal behaviour (e.g., body movement, picking up/moving objects, etc.) and vision. This is of great interest both to psychologists to understand human behaviour development, and also computer scientists, as their capabilities to learn from cluttered environments can far exceed the current advanced computer vision/machine learning techniques.

To better understand how human children interact with the environment, first-person videos are captured. However, to understand what they see and interact, it is essential to identify objects in such videos accurately. Although manual labelling can provide accurate solutions, it is not feasible to scale up to the amount of videos available. The task of this project is to develop automatic computer vision techniques to extract useful information, including individual objects being seen and manipulated, along with other analysis, with an aim to lead to some interdisciplinary findings.