All ETDs from UAB

Advisory Committee Chair

Arie Nakhmani

Advisory Committee Members

Mohammad Haider

Ismail Mohamed

Dalton Nelson

Nikolay M Sirakov

Document Type


Date of Award


Degree Name by School

Doctor of Philosophy (PhD) School of Engineering


A target tracking is a well-known computer vision problem due to its wide range of applications in industrial and biomedical domains, such as traffic planning and management, surveillance, robotics, economics, human-computer interaction prediction, and electrode tracking for surgical planning. A significant amount of research has been done to tackle the problem. However, it is still challenging to design a robust tracking system that can provide solutions for many real-world tracking applications and prediction. Most of the proposed tracking solutions are complex due to many model parameters. The overarching goal of this dissertation is to develop predictors and trackers that would be useful in many application and would include minimum of parameters that need to be manually tuned. The first part, of this thesis, presents a two-stage framework and its implementation for tracking multiple target trajectories in video sequences based on mathematical modeling and classification of curves. In the first stage, bidirectional Long Short-Term Memory Recurrent Neural Network is used to classify trajectories generated by the targets. In the second stage, the polynomial curve of 2nd degree is being fit to the location data of the previous six video frames, and the polynomial curve of 3rd degree is being fit to the velocity curve computed from the six past frames. This information helps to extrapolate the future velocity of the target and its future location. We demonstrate the tracking abilities of our new approach and estimate the error of prediction on simulated trajectory data and on real-world videos. Predicting various targets' motion using a generalized training data set with deep ii learning is a quite challenging problem and not much explored. In second part of this dissertation, to solve such a problem, we proposed a deep learning based trajectory prediction approach for predicting location of various kinds of targets. A deep neural network is trained on a large generalized training data set of linear and nonlinear trajectories. In other words, to design our data-driven prediction approach, we developed a freely available database of up to second-order algebraic curves uniformly distributed in a given domain. This database could be used for training and testing point-target tracking algorithms. Simulated noisy test sets of trajectories were produced using Gaussian noise for analyzing the forecasting performance and noise sensitivity of our model. Further, the newly designed LSTM-based network that uses polar coordinates for its training is capable of predicting the target's future locations on real world smooth trajectories. We compared the proposed predictor network with classical and state-of-the-art predictors based on average absolute and relative errors. The experimental results demonstrated that our novel predictor achieved significant improvement on test data sets with and without noise. Biomedical signals are used in clinical practice to determine and monitor the medical condition of patients. Continuous manual monitoring is laborious and causing high costs to healthcare for accurate diagnosis, early detection, or prediction of biological events. Heart rate and oxygen saturation monitoring have become a ubiquitous part of neonatal assessment. Apneic events in neonates correspond to decreased heart rate and blood oxygen levels below normal for the age. These events are called bradycardia and desaturation, respectively. The early prediction of these physiological events is essential for avoiding apnea that could cause death, neuropsychiatric disorders, and impaired cognitive functions in the long term. The problem of bradycardia and desaturation prediction could be thought as prediction and tracking of these signals in time. In the third part of our dissertation, we focus on proposing initial stage work of a deep learning based cost-effective buzzer system for the early prediction of bradycardia events in preterm infants. We study and compare various deep learning and statistical based predictive models to determine the most efficient one for predicting the bradycardia events. The main contribution in this study is the development of mathematical formalism that describes the clinical process of bradycardia events. To measure the effectiveness of event prediction, we formulated the notions true/false positive and true/false negative. The new theoretical concepts and contributions were validated with nine predictors, including deep neural networks on about 500 records of 24 hours records of signals.

Included in

Engineering Commons