Date of Award
Master of Science (MS)
Electrical and Computer Engineering
In this thesis, a novel method for tracker fusion is proposed and evaluated for vision-based tracking. This work combines three distinct popular techniques into a recursive Bayesian estimation algorithm. First, semi supervised learning approaches are used to partition data and to train a deep neural network that is capable of capturing normal visual tracking operation and is able to detect anomalous data. We compare various methods by examining their respective receiver operating conditions (ROC) curves, which represent the trade off between specificity and sensitivity for various detection threshold levels. Next, we incorporate the trained neural networks into an existing data fusion algorithm to replace its observation weighing mechanism, which is based on the Mahalanobis distance. We evaluate different semi-supervised learning architectures to determine which is the best for our problem. We evaluated the proposed algorithm on the OTB-50 benchmark dataset and compared its performance to the performance of the constituent trackers as well as with previous fusion. Future work involving this proposed method is to be incorporated into an autonomous following unmanned aerial vehicle (UAV).