Real-time scalable object detection

Real-time, scalable, multi-view object detection is an active area of research, particularly in robot vision community. An efficient template-based object detection algorithm has recently been proposed [1] that utilizes both color and depth information, and works on texture-less objects. However, the approach scales linearly with the number of objects and views. This project explores a variation of the algorithm that uses a fixed dictionary of templates instead of linearly large, sparse templates, which will allow us to use efficient approximate nearest neighbour methods for efficient matching instead of an exhaustive linear search.