Plume Volume Estimation from Multi-View Imagery
As an undergraduate research assistant in the RIT Digital Imaging and Remote Sensing Lab (DIRS) Laboratory, I made significant contributions to a collaborative research project between DIRS and the Savannah River National Laboratory, led by PhD student Ryan Connal and PI Dr. Carl Salvaggio
​
This project was focused on volume estimation of condensed water vapor plumes from mechanical draft cooling towers.
SPIE is Paywalled
By Dr. Ryan Connal
The initial data collection for the project was performed using remotely operated ground camera stations placed around a test site with draft cooling towers. These stations were designed to capture synchronized imagery of condensed water vapor plumes under varying environmental conditions over the course of 17 months.
​
I soldered the boards and assembled the camera units for this part of the project. Each unit consisted of a Canon M200 mirrorless camera paired with a 10-18 mm wide-angle lens, controlled by a Raspberry Pi running gPhoto2 to manage firing and exposure settings. A Verizon 4G LTE modem provided remote connectivity for data transfer and system updates. All components were housed inside a weather-proof enclosure equipped with an internal heater and temperature monitoring to prevent fogging.
​
​
Image from Ryan Connal
The goal of the research was to develop an automated plume-segmentation method. As data came in, the images needed to be manually annotated to create reliable ground truth masks for training neural networks to separate the plumes from the rest of the scene.
​
​
Image from Ryan Connal
Image from Ryan Connal
I spent several hours a week annotating hundreds of plume images for training.
Ryan wrapped up his dissertation in May of 2024, with the project delivering a complete imaging and reconstruction pipeline that produced hundreds of 3D plume volumes from long-term multi-camera data collected from both the ground camera stations and from synchronized drone imagery, and demonstrated that the reconstructed plume volumes closely matched real-world power measurements at the cooling tower test site.
​
Contributing to this project as an undergraduate gave me a real sense of how all the pieces of research fit together. I saw how ideas turn into systems, how real-world data becomes models, and how every decision along the way matters.