This thesis investigates a workflow for vibration measurement and modal interpretation of structures using UAV-acquired videos and motion magnification. The proposed workflow combines four main stages: Video stabilization through a SURF--RANSAC approach with features selected only on the background; learning-based motion magnification in selected global frequency bands; and optical-flow for static reporting of the amplified motion, while accelerometer measurements are used in this thesis to support the selection of relevant frequency bands. In addition, displacement tracking and spectrogram analysis of selected points in the magnified videos were used as a verification step by comparing the reconstructed motion with the corresponding accelerometer measurements. The methodology was applied to two staircases at Politecnico di Milano. The results show that reliable visualization of structural motion is possible when the videos contain sufficient and well-distributed static background features, when the UAV recordings include sufficiently stable time windows, and when drone motion remains limited. Under these conditions, the magnified responses and static reports reveal consistent motion patterns across different acquisitions and viewpoints of the same structure, with lower bands involving broader structural participation and higher bands becoming more localized on slender members. In contrast, acquisitions with a limited number of background features due to close standoffs and datasets with intentional drone motion highlight the main limitations of the method. Overall, the thesis shows that the proposed framework can support vibration-based structural monitoring, while also identifying the experimental and processing conditions required to obtain reliable results.
Questa tesi analizza una procedura per la misura delle vibrazioni e l’interpretazione modale di strutture mediante video acquisiti da UAV e motion magnification. La procedura proposta combina quattro fasi principali: stabilizzazione video mediante un approccio SURF--RANSAC con feature selezionate esclusivamente sullo sfondo; motion magnification di tipo learning-based in bande di frequenza globali selezionate; e optical flow per la rappresentazione statica del moto amplificato, mentre le misure accelerometriche sono utilizzate in questa tesi per supportare la selezione delle bande di frequenza più rilevanti. Inoltre, il tracciamento dello spostamento e l’analisi tramite spettrogramma di punti selezionati nei video magnified sono stati utilizzati come fase di verifica, confrontando il moto ricostruito con le corrispondenti misure accelerometriche. La metodologia è stata applicata a due scale presso il Politecnico di Milano. I risultati mostrano che una visualizzazione affidabile del moto strutturale è possibile quando i video contengono un numero sufficiente di feature statiche di sfondo ben distribuite, quando le registrazioni UAV includono intervalli temporali sufficientemente stabili e quando il moto del drone rimane limitato. In queste condizioni, le risposte magnificate e le rappresentazioni statiche evidenziano pattern di moto coerenti tra acquisizioni e punti di vista differenti della stessa struttura, con le bande più basse associate a una partecipazione strutturale più estesa e le bande più alte a una risposta più localizzata su elementi snelli. Al contrario, acquisizioni con un numero limitato di feature di sfondo dovuto a ridotte distanze di ripresa e dataset con moto intenzionale del drone evidenziano i principali limiti del metodo. Nel complesso, la tesi mostra che la procedura proposta può supportare il monitoraggio strutturale basato sulle vibrazioni, individuando al tempo stesso le condizioni sperimentali e di elaborazione necessarie per ottenere risultati affidabili.
Vibration measurement and modal analysis of structures based on UAV camera and video motion magnification
RASI REZVANI, SOHEIL
2025/2026
Abstract
This thesis investigates a workflow for vibration measurement and modal interpretation of structures using UAV-acquired videos and motion magnification. The proposed workflow combines four main stages: Video stabilization through a SURF--RANSAC approach with features selected only on the background; learning-based motion magnification in selected global frequency bands; and optical-flow for static reporting of the amplified motion, while accelerometer measurements are used in this thesis to support the selection of relevant frequency bands. In addition, displacement tracking and spectrogram analysis of selected points in the magnified videos were used as a verification step by comparing the reconstructed motion with the corresponding accelerometer measurements. The methodology was applied to two staircases at Politecnico di Milano. The results show that reliable visualization of structural motion is possible when the videos contain sufficient and well-distributed static background features, when the UAV recordings include sufficiently stable time windows, and when drone motion remains limited. Under these conditions, the magnified responses and static reports reveal consistent motion patterns across different acquisitions and viewpoints of the same structure, with lower bands involving broader structural participation and higher bands becoming more localized on slender members. In contrast, acquisitions with a limited number of background features due to close standoffs and datasets with intentional drone motion highlight the main limitations of the method. Overall, the thesis shows that the proposed framework can support vibration-based structural monitoring, while also identifying the experimental and processing conditions required to obtain reliable results.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_02_Rasi_Rezvani_Thesis_01.pdf
accessibile in internet per tutti
Descrizione: Master's Thesis – Soheil Rasi Rezvani
Dimensione
47.78 MB
Formato
Adobe PDF
|
47.78 MB | Adobe PDF | Visualizza/Apri |
|
2026_02_Rasi_Rezvani_Executive Summary_02.pdf
accessibile in internet per tutti
Descrizione: Executive Summary – Soheil Rasi Rezvani
Dimensione
8.45 MB
Formato
Adobe PDF
|
8.45 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/252527