This thesis presents a comparison of four Optical Flow algorithms, Lucas-Kanade, Horn-Schunck, Farnebäck, and RAFT, for the kinematic analysis of steam flows through image processing. The study was conducted using two datasets acquired under different experimental conditions: the first, at 2 fps, involving a kitchen extractor hood integrated into a cooktop; the second, at 200 fps, showing a steam flow driven by a fan. Image processing was performed in MATLAB to estimate velocity fields and calculate flow-related quantities for comparison among the algorithms. The results highlighted the limitations associated with the low acquisition frequency of the first dataset and considerable differences between the estimates obtained from the second. In particular, RAFT provided flow estimates that were more consistent with an indicative estimate derived from the direct observation of two consecutive frames. The frame rate sensitivity analysis also showed a progressive stabilization of RAFT estimates at higher sampling frequencies, which was not observed for the classical algorithms. Overall, the study highlights the importance of acquisition frequency and algorithm selection in steam motion analysis, while emphasizing the need for independent experimental measurements to assess estimation accuracy.
La presente tesi propone un confronto tra quattro algoritmi di Optical Flow, Lucas-Kanade, Horn-Schunck, Farnebäck e RAFT, per l’analisi cinematica di flussi di vapore mediante elaborazione di immagini. Lo studio è stato condotto su due dataset acquisiti in differenti condizioni sperimentali: il primo, a 2 fps, relativo a una cappa aspirante integrata nel piano di cottura; il secondo, a 200 fps, relativo a un flusso di vapore movimentato mediante una ventola. Attraverso l’elaborazione delle immagini in MATLAB sono stati stimati i campi di velocità e ricavate grandezze di flusso utili al confronto tra gli algoritmi. I risultati hanno evidenziato le criticità associate alla ridotta frequenza di acquisizione del primo dataset e differenze significative tra le stime ottenute sul secondo. In particolare, RAFT ha fornito valori di flusso più coerenti con una stima indicativa ricavata dall’osservazione diretta di due fotogrammi consecutivi. L’analisi di sensibilità al frame rate ha inoltre mostrato una progressiva stabilizzazione delle stime di RAFT alle frequenze più elevate, non riscontrata negli algoritmi classici. Il lavoro evidenzia quindi l’importanza della frequenza di acquisizione e della scelta dell’algoritmo nell’analisi del moto del vapore, pur sottolineando la necessità di misure sperimentali indipendenti per valutarne l’accuratezza.
CONFRONTO TRA ALGORITMI DI OPTICAL FLOW PER L’ANALISI CINEMATICA DI FLUSSI DI VAPORE
DEL PAPA, MARCO
2025/2026
Abstract
This thesis presents a comparison of four Optical Flow algorithms, Lucas-Kanade, Horn-Schunck, Farnebäck, and RAFT, for the kinematic analysis of steam flows through image processing. The study was conducted using two datasets acquired under different experimental conditions: the first, at 2 fps, involving a kitchen extractor hood integrated into a cooktop; the second, at 200 fps, showing a steam flow driven by a fan. Image processing was performed in MATLAB to estimate velocity fields and calculate flow-related quantities for comparison among the algorithms. The results highlighted the limitations associated with the low acquisition frequency of the first dataset and considerable differences between the estimates obtained from the second. In particular, RAFT provided flow estimates that were more consistent with an indicative estimate derived from the direct observation of two consecutive frames. The frame rate sensitivity analysis also showed a progressive stabilization of RAFT estimates at higher sampling frequencies, which was not observed for the classical algorithms. Overall, the study highlights the importance of acquisition frequency and algorithm selection in steam motion analysis, while emphasizing the need for independent experimental measurements to assess estimation accuracy.| File | Dimensione | Formato | |
|---|---|---|---|
|
MarcoDelPapa_tesi_versionefinalissimissima.pdf
embargo fino al 22/10/2029
Dimensione
3.06 MB
Formato
Adobe PDF
|
3.06 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.12075/29708