This research focuses on developing an algorithm set to track the vine tendril motion of a passiflora incarnate, commonly referred to as the passion fruit plant, to facilitate research into if there is a correlation between plant motion and plant health. An evaluation was done of clustering based color segmentation with a focus on K-means, feature / texture segmenta- tion utilizing Scale Invariant Feature Transforms (SIFT), and temporal based segmentation using Gaussian Mixture Model Background Subtraction to segment out the tendril in each video frame. Morphological image processing methods, such as dilation and connected com- ponent analysis, were used to clean up the segmentation results to give an estimate of the vine tendril’s location at each frame. Kalman filtering was then used to track the tendril’s location through the different frames dealing with large jumps in tendril location, cases where the tendril remained stationary between frames, and cases where there was error in the segmentation process. The resulting algorithm set was successful at tracking the tendril during times when the tendril had large jumps in position and it almost always succeeded in keeping track of the tendril during errors in the segmentation due to lack of tendril motion. The few cases that were not successful were evaluated and suggestions were made to resolve these issues in future data collection. / Master of Science / This research focused on developing an algorithm sequence that could find the tendril of a passiflora incarnate, commonly referred to as the passion fruit plant, in a single frame of a video and then track that tendril through the different frames in the video. Having the ability to track a plant tendril through a video allows biologists to research if there is a link between the amount a plant moves and the plant’s health. The algorithms evaluated for finding the plant in the image used color, features and motion to try and distinguish the tendril from the rest of the image. After the tendril was found, a tracking algorithm that combined a prediction from a model for the tendril’s location with the measured location was used to deal with noise and errors in the measurement. It was found that using the motion based algorithm worked the best to find the tendril (with the addition of some image processing to remove noise). This combined with the tracking algorithm allowed for the tendril to be successfully tracked through the different frames with one exception. Future work and recommendations were made to deal with this exception.
| Identifer | oai:union.ndltd.org:VTETD/oai:vtechworks.lib.vt.edu:10919/83446 |
| Date | January 2018 |
| Creators | Moser, Joshua N. |
| Contributors | Mechanical Engineering, Wicks, Alfred L., Abbott, A. Lynn, Asbeck, Alan T. |
| Publisher | Virginia Tech |
| Source Sets | Virginia Tech Theses and Dissertation |
| Language | en_US |
| Detected Language | English |
| Type | Thesis |
| Format | ETD, application/pdf |
| Rights | Creative Commons Attribution-ShareAlike 3.0 United States, http://creativecommons.org/licenses/by-sa/3.0/us/ |
Page generated in 0.0021 seconds