| In this thesis, a new vision-based hybrid particle tracking velocimetry (VB-PTV) technique is described and methods of processing randomly scattered velocity data investigated. The VB-PTV technique uses a feature matching method from computer vision theory which relies on the principles of proximity, similarity, and exclusion, meaning that it seeks to match one feature to one feature in subsequent images, and it favors matches which are close to one another and "look" similar. By constructing a matrix which takes these principles into account and performing singular value decomposition, a straightforward method of matching is developed which can give accurate matching results in a wide variety of flows. PIV velocity information is used to provide guidance to the matching algorithm. In addition, matches are made iteratively and validated by an outlier detection scheme. When this method is tested on synthetic images it results in matches which are typically reliable more than 98% of the time. A simple modification to the principle of proximity is introduced which reduces the PTV method's errors in highly shearing flow, as well as improving performance in general for various flow types.;Finally, a natural neighbor-based interpolation technique is investigated for use in estimating flow derivatives using scattered velocity data. This interpolation method is compared with other existing techniques in terms of accuracy, sensitivity to noise, computational efficiency, and spatial resolution. It is found that the natural neighbor interpolation is less accurate than RBF and kriging interpolation methods, and more sensitive to noise, despite the use of a denoising technique. |
