Algorithm optimization for real-time volumetric particle tracking velocimetry under non-uniform illumination for airflow studies

Zhao, Yu

Permalink

https://hdl.handle.net/2142/115570 Copy

Description

Title

Algorithm optimization for real-time volumetric particle tracking velocimetry under non-uniform illumination for airflow studies

Author(s)

Zhao, Yu

Issue Date

2022-04-22

Director of Research (if dissertation) or Advisor (if thesis)

Zhang, Yuanhui

Doctoral Committee Chair(s)

Zhang, Yuanhui

Committee Member(s)

• Grift, Tony
• Wang, Xinlei
• Sun, Yigang

Department of Study

Engineering Administration

Discipline

Agricultural & Biological Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Building Environment
• Particle Tracking
• Real-time Visualization
• Flow Pattern

Abstract

Volumetric Particle Tracking Velocimetry (VPTV) is a scalable real-time method for measuring 3D velocity fields. Based on the previous work in our group, grayscale cameras equipped with Field Programmable Gate Arrays (FPGA) were used to reduce the data flow rate and achieve real-time 3D measurement. However, the centroids of particle segments were used in FPGA cameras for particle detection, which could be hindered when particles were streak-shaped due to non-uniform illumination and long exposure time. To improve particle detection, a hybrid VPTV system was proposed containing both regular grayscale cameras and FPGA grayscale cameras. In this hybrid system, the exposure time was extended, and the “good features to track (GFTT)” approach was used to find the corners of particle streaks. The Graphics Processing Units (GPU) were used to process the grayscale images and perform the GFTT. This approach resulted in a 65% enhancement of particle detection ratio compared to the previous algorithm. With GPU acceleration, the processing time per frame was accelerated 15 times compared to the CPU version. In addition, a data-analysis sequence correspondence (DSC) algorithm was developed to improve the correspondence task in this hybrid system. The DSC constructed datasets to train a logistic regression classifier to identify the correct correspondence using only two cameras in actual measurements. The results showed that this approach could achieve an 82% correct prediction ratio and thus can improve the correspondence task in VPTV.

Graduation Semester

2022-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2022 Yu Zhao

Owning Collections