Visual Attention for Robotic Manipulations
DescriptionThis project aims to develop a robust 3D object tracking method in gaze tracking by combining the merits of visual attention mechanism and deep feature learning. In our proposed tracking framework, visual attention controls where to look and select candidate regions which are more likely to contain the target object and deep feature learning provides a representation of what is seen. Specifically, visual attention mechanism outputs a small set of potentially most relevant candidate regions. These candidate regions are evaluated by image features learned by deep learning. Hence, a robust HVS-like 3D object tracking can be achieved. The study will have significant impact on the theoretical studies in robust 3D tracking and the practical applications in gaze tracking.
|Effective start/end date||1/09/18 → …|