Date of Award
Open Access Dissertation
Doctor of Philosophy in Electrical Engineering (PhD)
Administrative Home Department
Department of Electrical and Computer Engineering
Dr. Michael C. Roggemann
Committee Member 1
Dr. Laura Brown
Committee Member 2
Dr. Timothy Havens
Committee Member 3
Dr. Christopher Middlebrook
It is critical for defense and security applications to have a high probability of detection and low false alarm rate while operating over a wide variety of conditions. Sensor fusion, which is the the process of combining data from two or more sensors, has been utilized to improve the performance of a system by exploiting the strengths of each sensor. This dissertation presents algorithms to fuse multi-sensor data that improves system performance by increasing detection rates, lowering false alarms, and improving track performance. Furthermore, this dissertation presents a framework for comparing algorithm error for image registration which is a critical pre-processing step for multi-spectral image fusion.
First, I present an algorithm to improve detection and tracking performance for moving targets in a cluttered urban environment by fusing foreground maps from multi-spectral imagery. Most research in image fusion consider visible and long-wave infrared bands; I examine these bands along with near infrared and mid-wave infrared. To localize and track a particular target of interest, I present an algorithm to fuse output from the multi-spectral image tracker with a constellation of RF sensors measuring a specific cellular emanation. The fusion algorithm matches the Doppler differential from the RF sensors with the theoretical Doppler Differential of the video tracker output by selecting the sensor pair that minimizes the absolute difference or root-mean-square difference. Finally, a framework to quantify shift-estimation error for both area- and feature-based algorithms is presented. By exploiting synthetically generated visible and long-wave infrared imagery, error metrics are computed and compared for a number of area- and feature-based shift estimation algorithms.
A number of key results are presented in this dissertation. The multi-spectral image tracker improves the location accuracy of the algorithm while improving the detection rate and lowering false alarms for most spectral bands. All 12 moving targets were tracked through the video sequence with only one lost track that was later recovered. Targets from the multi-spectral tracking algorithm were correctly associated with their corresponding cellular emanation for all targets at lower measurement uncertainty using the root-mean-square difference while also having a high confidence ratio for selecting the true target from background targets. For the area-based algorithms and the synthetic air-field image pair, the DFT and ECC algorithms produces sub-pixel shift-estimation error in regions such as shadows and high contrast painted line regions. The edge orientation feature descriptors increase the number of sub-field estimates while improving the shift-estimation error compared to the Lowe descriptor.
Demars, Casey D., "Target detection, tracking, and localization using multi-spectral image fusion and RF Doppler differentials", Open Access Dissertation, Michigan Technological University, 2018.