Date of Award
Spring 5-2011
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Computing
School
Computing Sciences and Computer Engineering
Committee Chair
Andrew Strelzoff
Committee Chair Department
Computing
Committee Member 2
Ray Seyfarth
Committee Member 2 Department
Computing
Committee Member 3
Todd Holland
Committee Member 3 Department
Marine Science
Committee Member 4
Beddhu Murali
Committee Member 4 Department
Computing
Committee Member 5
Joe Calantoni
Committee Member 5 Department
Marine Science
Abstract
A modular particle image velocimetry program was developed and optimized to read and process video of river surface flows from different sensor types. The program was designed for long-term deployment with the ability to sample data continuously in realtime and save the results in a compact format. The time needed to compute a velocity measurement from video input was reduced by using concurrent processing techniques, multi-threading, and a graphics hardware-based correlation algorithm. When used to process field data on a low power Intel Atom based computer the PIV system was capable of computing up to 64 velocity measurements at a rate of 7.5 frames per second. A more powerful computer equipped with a discrete GPU was capable of computing 4800 velocity measurements at a rate of 7.5 frames per second when using the same PIV data and settings. Processing speed of the GPU correlation module was analyzed using a number of different benchmarks. Results show that the GPU-based correlation algorithm has the potential to improve the PIV processing speed of high-end workstations by as much as 2x and low-end portable computers by 10-20x. Methods were also introduced to improve the quality of PIV measurements on river currents. Processing video of river currents with the standard particle image velocimetry technique produced a large number of inaccurate vectors. Most of these inaccurate vectors were correctly identified and removed by using different confidence scoring and filtering techniques. Results from three different experiments were compared to the velocity measurements of other devices to verify the accuracy of the program. These measurements agree to within 16% difference. These results show that accurate PIV measurements of river surface velocity may be computed in real time even on low end and portable computer hardware.
Copyright
2011, David William Dobson
Recommended Citation
Dobson, David William, "Real-Time Riverine Particle Image Velocimetry" (2011). Dissertations. 571.
https://aquila.usm.edu/dissertations/571