Adaptive real-time clustering method for dynamic visual tracking of very flexible wings

Abstract

Advancements in intelligent aircraft controller design, paired with increasingly flexible and efficient aircraft concepts, create the need for the development of novel (smart) adaptive sensing suitable for aeroelastic state estimation. In contrast to rigid states, aeroelastic state estimation requires more measurement points (displacements and forces) across the span to capture the vibrational shapes of the wing undergoing excitations. A potentially universal and non-invasive approach is visual tracking. However, many tracking methods require manual selection of initial marker locations as the start of a tracking sequence. This study is part of a larger study in the field of smart sensing and aims to cover the gap by investigating a robust machine learning approach for unsupervised automatic labelling of visual markers. The method utilizes fast DBSCAN and adaptive image segmentation pipeline with HSV colour filter to extract and label the marker centres under the presence of marker failure. A comparison is made with Disjoint-set data structure for clustering of the data. The segmentation-clustering pipeline with DBSCAN shows the capability to act as a visual tracking method on its own, capable of running real-time at 250 fps on an image sequence of a single camera with a resolution of 1088×600 pixels. To increase the robustness against noise, a novel formulation of DBSCAN better suited against noise, the inverse DBSCAN (DBSCAN
−1), is proposed, allowing to cast the clustering problem into noise filtering problem with an additional MaxPts parameter. Furthermore, observations are made regarding the frequency content of the image pixel intensities across time, and how this can be utilized to estimate the natural frequency of the system and adjust the segmentation-clustering pipeline with a sliding DFT (Discrete Fourier Transform).