الفهرس 
Title : Fast CNN-Based Object Tracking with Online Training
SummaryOnly 14 pages are availabe for public view
 |  | from | 110 |  |  |
| | | from | 110 | | |
Abstract
In this thesis, we present the first survey in literature to review the hardware implementations of object trackers over the last two decades. We believe our survey would fill the gap and complete the picture with the previous surveys of how to design an efficient tracker. We tackle the speed limitation of CNN-based object trackers from the algorithm side and the implementation side. On the algorithm side, we adapt the CNN not only as a two-label classifier, object and background labeling, but also as a five-position classifier for the object position inside the candidate patch. Our tracker achieves competitive performance results with 8x speed improvements compared to the equivalent tracker. On the hardware implementation side, we performed design-space exploration of the different computation stages of the proposed tracker. Then, we design a fixed-point based hardware accelerator for the fully connected stages of the CNN network with online training capability