DEVELOPMENT AND VALIDATION OF A SYSTEM FOR OSCILLATION MONITORING USING SINGLE NON-METRIC CAMERA

Abstract

This research aims the development of in-plane oscillation – deformation monitoring technique using available on shelf cameras and taking advantage of the high optical zoom offered by bridge type cameras and DSLR lenses. Consequently, a photogrammetric model is developed to accommodate the lens distortions effect, which requires forth degree polynomial function to be used for image projection. The proposed technique relies on a predefined gridded target pattern that is used as control points to rectify the initial frame, resulting in projected ortho-photo and eliminating both radial and tangential lens distortions effect. The target is detected by Harris corner detector (C.Harris, 1998), and target points detection precision is achieved in sub-pixel accuracy using neighbor gradient technique. The limitation of testing in controlled lighting conditions is handled by AI techniques that are adopted by implementing Faster RCNN network for detecting a tracking pattern located at the targets’ center. The deep network is trained to detect a track point at various lighting conditions. The resulting monitoring using AI is compared to the previously proposed technique, and the impact of using AI on measurement precision is evaluated. The approach is validated by a set of tests comparing the monitored oscillation in the time domain with different measuring techniques; starting by the usage of predefined moving patterns, then using an oscillating dynamic actuator, and shaking table, the resulting monitoring data is compared to measured data, then a field test is performed to measure the oscillations of an industrial chimney. The proposed monitoring technique proved to be precise and robust in different environmental conditions.