Autonomous driving represents one of the most promising technological advancements in modern mobility. The development of reliable self-driving systems is driving the traditional transportation to a renewed concept of mobility. The objective of this thesis was to design and validate a novel method for the identification, fitting, and tracking of road lines based on a geometric reference extracted from a constructed graph. The method integrates both perception data, obtained from exteroceptive sensors, and topological in formation, derived from OpenStreetMap (OSM), into a unified framework that enables a precise and consistent representation of the road structure surrounding the vehicle. The proposed methodology provides an efficient, accurate, and reproducible means of detecting and tracking lane lines using publicly available data and moderate computational resources. The novelty of the proposed approach lies in the introduction of a reference driven fitting and tracking framework, that leverages a dynamically constructed graph from OSM data. This graph represented the base to a new detection an tracking algorithm, based on the parallelism between road lines and road centerline, extracted from the graph. The other novelty lies in a close loop communication between the fitting and tracking modules of the algorithm, that enables temporal consistency and robustness against transient errors improving the continuity of the detected lines. Furthermore, the algorithm shown positive results also in the identification and tracking of road boundaries. This result, combined with the line detection, enables an accurate and comprehensive reconstruction of the road structure. In summary, the proposed method bridges the gap between geometric and data-driven techniques, offering several conceptual and practical advantages compared to other state of the art methods. The obtained results demonstrate that the combination of a digital road graph and probabilistic fitting can yield accurate, interpretable, and computationally efficient results suitable for real-time autonomous driving applications.
La guida autonoma rappresenta uno degli sviluppi tecnologici più promettenti della mobilità moderna, rinnovandone il concetto stesso. Questo lavoro vede la progettazione e validazione di un nuovo metodo per l’identificazione, il fitting e il tracciamento delle linee stradali; basati sull’utilizzo di un riferimento geometrico estratto da un grafo costruito a partire da dati open source. Il metodo integra in un unico framework sia i dati di percezione, ottenuti da sensori esteroceptivi, sia le informazioni topologiche, derivate da OSM, consentendo una rappresentazione completa della struttura stradale. La metodologia proposta fornisce un approccio accurato e riproducibile per il rilevamento e il tracciamento delle linee stradali, utilizzando dati pubblicamente disponibili e risorse computazionali moderate. L’elemento innovativo del metodo risiede nell’introduzione di un framework di fitting e tracking guidato da un riferimento, ottenuto da un grafo costruito dinamicamente a partire da dati ottenuti tramite OSM. Questo grafo costituisce la base per un nuovo algoritmo di rilevamento e tracciamento, fondato sul parallelismo tra le linee stradali e la linea centrale della carreggiata. Un’ulteriore innovazione consiste nella comunicazione bidirezionale tra i moduli di fitting e di tracking, che garantisce coerenza temporale e robustezza rispetto a errori transitori, migliorando la continuità delle linee rilevate. Inoltre, l’algoritmo ha mostrato ottimi risultati anche nell’identificazione e nel tracciamento dei margini stradali. Questo risultato, combinato con il rilevamento delle linee di corsia, consente una ricostruzione accurata e completa della struttura stradale. In sintesi, il metodo proposto colma il divario tra le tecniche di rilevamento delle corsie basate su modelli geometrici e data-driven, offrendo numerosi vantaggi rispetto allo stato dell’arte. I risultati ottenuti dimostrano che la combinazione di un grafo stradale digitale e di un fitting probabilistico può fornire risultati accurati, interpretabili ed efficienti dal punto di vista computazionale, adatti ad applicazioni di guida autonoma in tempo reale.
Development of a road structure reconstruction and line detection algorithm, based on Open Street Map data and onboard measurements
DELFRATE, LUCA
2024/2025
Abstract
Autonomous driving represents one of the most promising technological advancements in modern mobility. The development of reliable self-driving systems is driving the traditional transportation to a renewed concept of mobility. The objective of this thesis was to design and validate a novel method for the identification, fitting, and tracking of road lines based on a geometric reference extracted from a constructed graph. The method integrates both perception data, obtained from exteroceptive sensors, and topological in formation, derived from OpenStreetMap (OSM), into a unified framework that enables a precise and consistent representation of the road structure surrounding the vehicle. The proposed methodology provides an efficient, accurate, and reproducible means of detecting and tracking lane lines using publicly available data and moderate computational resources. The novelty of the proposed approach lies in the introduction of a reference driven fitting and tracking framework, that leverages a dynamically constructed graph from OSM data. This graph represented the base to a new detection an tracking algorithm, based on the parallelism between road lines and road centerline, extracted from the graph. The other novelty lies in a close loop communication between the fitting and tracking modules of the algorithm, that enables temporal consistency and robustness against transient errors improving the continuity of the detected lines. Furthermore, the algorithm shown positive results also in the identification and tracking of road boundaries. This result, combined with the line detection, enables an accurate and comprehensive reconstruction of the road structure. In summary, the proposed method bridges the gap between geometric and data-driven techniques, offering several conceptual and practical advantages compared to other state of the art methods. The obtained results demonstrate that the combination of a digital road graph and probabilistic fitting can yield accurate, interpretable, and computationally efficient results suitable for real-time autonomous driving applications.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_Delfrate_Tesi_01.pdf
non accessibile
Descrizione: tesi
Dimensione
13.12 MB
Formato
Adobe PDF
|
13.12 MB | Adobe PDF | Visualizza/Apri |
|
2025_12_Delfrate_Executive Summary_02.pdf
non accessibile
Descrizione: executive summary
Dimensione
2.53 MB
Formato
Adobe PDF
|
2.53 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/10589/246631