The rapid development of autonomous vehicle (AV) technology has made precise sensor integration essential for safe and efficient navigation. Sensor fusion, based on accurate extrinsic calibration, allows AVs to interpret complex environments and make critical real time decisions. However, in the automotive industry, calibration processes remain costly and complex, requiring controlled environments, specific targets, and skilled personnel. To address these challenges, this thesis proposes an automated approach to extrinsic sensor calibration that maintains high accuracy while significantly reducing the need for manual intervention and expensive setups. Specifically, the approach leverages metaheuristic algorithms—Simulated Annealing (SA), Genetic Algorithms (GA), and Particle Swarm Optimization (PSO)—to optimize the calibration process. By independently optimizing rotational and translational parameters, this method effectively avoids cross-compensation errors and simplifies calibration through the use of Euler angles. Additionally, a custom-built cost function, supported by Bayesian optimization, refines the search space, accelerating convergence and improving precision. Meanwhile, automation is further enhanced through computer vision techniques, which detect geometric features in camera and LiDAR frames using only a checkerboard refer ence and a Huber estimator to manage noise and outliers. Extensive experiments confirm the method’s feasibility and effectiveness, enabling reliable calibration with minimal poses and supporting multi-sensor configurations. Overall, this approach proves effective and is well-suited for practical applications, minimizing human intervention.
Il rapido sviluppo della tecnologia dei veicoli autonomi (AV) ha reso indispensabile una precisa integrazione dei sensori per garantire una navigazione sicura ed efficiente. La fu sione dei dati sensoriali, basata su una calibrazione accurata tra i sensori, consente agli AV di interpretare ambienti complessi e prendere decisioni critiche in tempo reale. Tuttavia, nell’industria automobilistica, i processi di calibrazione rimangono costosi e complessi, richiedendo ambienti controllati, target specifici e personale qualificato. Per rispondere a queste esigenze, questa tesi propone un approccio automatizzato alla calibrazione es terna dei sensori che mantiene elevata precisione, riducendo al contempo la necessità di intervento manuale e di costose configurazioni. In particolare, l’approccio sfrutta algoritmi metaeuristici—Simulated Annealing (SA), Genetic Algorithms (GA) e Particle Swarm Optimization (PSO)—per ottimizzare il pro cesso di calibrazione. Ottimizzando separatamente i parametri rotazionali e traslazionali, l’approccio evita errori di compensazione e semplifica la calibrazione tramite l’uso di angoli di Eulero. Inoltre, una funzione di costo progettata ad hoc, supportata dall’ottimizzazione bayesiana, restringe efficacemente lo spazio di ricerca, accelerando la convergenza e miglio rando la precisione. Parallelamente, l’automazione è potenziata attraverso tecniche di visione artificiale, che rilevano caratteristiche geometriche nei frame della telecamera e del LiDAR, utilizzando solo un riferimento a scacchiera e uno stimatore di Huber per gestire rumore e outlier. Esperimenti approfonditi dimostrano la validità e l’efficacia del metodo, consentendo una calibrazione affidabile con poche pose e supportando configurazioni multi-sensore. Com plessivamente, questo approccio si dimostra quindi efficace e si presta ad applicazioni pratiche, riducendo al minimo l’intervento umano.
Automatic multi-sensor calibration for autonomous vehicles: a rapid approach to LiDAR and camera data fusion
D'Amato, Francesca
2023/2024
Abstract
The rapid development of autonomous vehicle (AV) technology has made precise sensor integration essential for safe and efficient navigation. Sensor fusion, based on accurate extrinsic calibration, allows AVs to interpret complex environments and make critical real time decisions. However, in the automotive industry, calibration processes remain costly and complex, requiring controlled environments, specific targets, and skilled personnel. To address these challenges, this thesis proposes an automated approach to extrinsic sensor calibration that maintains high accuracy while significantly reducing the need for manual intervention and expensive setups. Specifically, the approach leverages metaheuristic algorithms—Simulated Annealing (SA), Genetic Algorithms (GA), and Particle Swarm Optimization (PSO)—to optimize the calibration process. By independently optimizing rotational and translational parameters, this method effectively avoids cross-compensation errors and simplifies calibration through the use of Euler angles. Additionally, a custom-built cost function, supported by Bayesian optimization, refines the search space, accelerating convergence and improving precision. Meanwhile, automation is further enhanced through computer vision techniques, which detect geometric features in camera and LiDAR frames using only a checkerboard refer ence and a Huber estimator to manage noise and outliers. Extensive experiments confirm the method’s feasibility and effectiveness, enabling reliable calibration with minimal poses and supporting multi-sensor configurations. Overall, this approach proves effective and is well-suited for practical applications, minimizing human intervention.File | Dimensione | Formato | |
---|---|---|---|
2024_12_DAmato_Executive Summary.pdf
non accessibile
Descrizione: Executive summary
Dimensione
4.78 MB
Formato
Adobe PDF
|
4.78 MB | Adobe PDF | Visualizza/Apri |
2024_12_DAmato_Tesi.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Thesis
Dimensione
27.49 MB
Formato
Adobe PDF
|
27.49 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/10589/230866