Abstract The rapid evolution of Simultaneous Localization and Mapping (SLAM) algorithms integrated into handheld Mobile Mapping Systems (MMS) has significantly accelerated 3D spatial data acquisition, challenging the traditional role of static Terrestrial Laser Scanning (TLS) in architectural and structural documentation. However, assessing the absolute metric reliability of kinematic point clouds remains a critical concern, especially when surveys are executed within multi-scale indoor-outdoor transitions or geometrically challenging underground environments. This thesis establishes a comprehensive, multi-tier metrological validation framework to decouple localized absolute accuracy, continuous volumetric deformations, and the intrinsic noise floor of the handheld FJD Trion P1 SLAM platform against a high-precision TLS geodetic Ground Truth. The experimental pipeline is structured around three complementary validation methods. Method I evaluates localized positional accuracy by isolating homologous checkerboard targets. To bypass the projection and interpolation errors inherent to color-mapping algorithms, a radiometric extraction protocol was developed leveraging exclusively the sensor's backscattered LiDAR Intensity scalar field. Under short mapping windows, this discrete approach yielded absolute 3D global RMSE values of 0.041 ± 0.005 m for an open-loop path (Trajectory 1) and 0.033 ± 0.005 m for a closed-loop configuration (Trajectory 2), validating the metric integrity of the scan-matching backend and quantifying the sub-centimeter optimization benefit provided by automated place recognition. To evaluate macroscopic structural distortions across extended paths, Method II implements a continuous, surface-to-surface comparison utilizing the Multiscale Model to Model Cloud Comparison (M3C2) algorithm. The analysis of an extended outdoor-indoor loop (Trajectory 3) unmasked a critical statistical paradox: while the raw dataset exhibited a deceptive global mean distance near-zero (+0.0004 m ), the error envelope expanded to a standard deviation (σ) of 0.0781 m (7.81 cm ), masking systemic internal tracking drift that reached real local displacements of up to 15-20 cm. Directional residual analysis isolated this degradation along the vertical axis (Z), driven by the operator's gait oscillations and amplified by the low, uniform ceiling geometry of the underground parking lot, which deprived the SLAM algorithm of distinct vertical geometric features. To resolve this tracking failure, a manual post-processing optimization strategy was introduced. By partitioning the continuous kinematic stream into chronological rigid macro-blocks and executing a reversed sequential ICP alignment under strict structural tolerances, the systematic rotational drift was counteracted without resorting to artificial software scaling (s=1). This geodetically sound workflow achieved the total structural suppression of the re-entry misalignments, compressing the global M3C2 standard deviation by roughly 16% down to 0.0658 m (6.58 cm ). Finally, Method III isolated the sensor's intrinsic high-frequency stochastic measurement noise via local planarity tests, yielding an RMSE of just 0.0046 m (4.6 mm) on a sampled structural wall of the third trajectory. This final step highlights a fundamental Precision-Accuracy Paradox, proving that a mobile mapping cloud can maintain exceptional local structural sharpness while suffering from macroscopic global warping. In conclusion, while handheld SLAM systems provide unprecedented operational throughput, establishing an independent, high-order static control network remains an indispensable prerequisite to guarantee absolute metric accuracy in engineering applications.
Abstract La rapida evoluzione degli algoritmi di Localizzazione e Mappatura Simultanea (SLAM) integrati nei sistemi di mappatura mobile portatili (MMS) ha accelerato notevolmente l'acquisizione di dati spaziali 3D, mettendo in discussione il ruolo tradizionale del laser scanning terrestre statico (TLS) nella documentazione architettonica e strutturale. Tuttavia, valutare l'affidabilità metrica assoluta delle nuvole di punti cinematiche rimane una questione cruciale, specialmente quando i rilievi vengono eseguiti all'interno di transizioni interno-esterno a diverse scale o in ambienti sotterranei geometricamente complessi. Questa tesi definisce un framework di validazione metrologica completo e multi-livello per separare l'accuratezza assoluta localizzata, le deformazioni volumetriche continue e il rumore di fondo intrinseco della piattaforma SLAM portatile FJD Trion P1 rispetto a un Ground Truth geodetico TLS ad alta precisione. La pipeline sperimentale è strutturata attorno a tre metodi di validazione complementari. Il Metodo I valuta l'accuratezza posizionale localizzata isolando i target a scacchiera omologhi. Per aggirare gli errori di proiezione e interpolazione intrinseci agli algoritmi di mappatura del colore, è stato sviluppato un protocollo di estrazione radiometrica che sfrutta esclusivamente il campo scalare dell'intensità di retrodiffusione del LiDAR del sensore. All'interno di brevi finestre di mappatura, questo approccio discreto ha prodotto valori assoluti di RMSE globale 3D pari a 0.041 ± 0.005 m per un percorso ad anello aperto (Traiettoria 1) e 0.033 ± 0.005 m per una configurazione ad anello chiuso (Traiettoria 2), convalidando l'integrità metrica del backend di scan-matching e quantificando il beneficio dell'ottimizzazione sub-centimetrica fornito dal riconoscimento automatico dei luoghi. Per valutare le distorsioni strutturali macroscopiche lungo percorsi estesi, il Metodo II implementa un confronto continuo superficie-superficie utilizzando l'algoritmo Multiscale Model to Model Cloud Comparison (M3C2). L'analisi di un loop esteso esterno-interno (Traiettoria 3) ha svelato un paradosso statistico critico: sebbene il dataset grezzo mostrasse un'ingannevole distanza media globale vicina allo zero (+0.0004 m ), l'inviluppo dell'errore si espandeva fino a una deviazione standard (σ) of 0.0781 m (7.81 cm ), mascherando una deriva sistematica del tracking interno che raggiungeva spostamenti locali reali fino a 15-20 cm. L'analisi dei residui direzionali ha isolato questo degrado lungo l'asse verticale (Z), guidato dalle oscillazioni del passo dell'operatore e amplificato dalla geometria bassa e uniforme del soffitto del parcheggio sotterraneo, che ha privato l'algoritmo SLAM di caratteristiche geometriche verticali distinte. Per risolvere questo fallimento del tracking, è stata introdotta una strategia manuale di ottimizzazione in post-elaborazione. Suddividendo il flusso cinematico continuo in macro-blocchi rigidi cronologici ed eseguendo un allineamento ICP sequenziale invertito sotto rigide tolleranze strutturali, la deriva rotazionale sistematica è stata contrastata senza ricorrere a riscalamenti artificiali del software (s=1). Questo flusso di lavoro geodeticamente fondato ha ottenuto la totale soppressione strutturale dei disallineamenti di rientro, comprimendo la deviazione standard globale M3C2 di circa il 16% fino a 0.0658 m (6.58 cm ). Infine, il Metodo III ha isolato il rumore di misura stocastico ad alta frequenza intrinseco del sensore tramite test di planarità locale, producendo un RMSE di appena 0.0046 m (4.6 mm) nella traiettoria III su una colonna strutturale campione. Quest'ultimo passaggio evidenzia un fondamentale Paradosso Precisione-Accuratezza, dimostrando che una nuvola di mappatura mobile può mantenere un'eccezionale nitidezza strutturale locale pur soffrendo di un imbarcamento macroscopico globale. In conclusione, sebbene i sistemi SLAM portatili offrano una produttività operativa senza precedenti, l'istituzione di una rete di controllo statica indipendente d'ordine superiore rimane un prerequisito indispensabile per garantire l'accuratezza metrica assoluta nelle applicazioni ingegneristiche.
Comparative analysis of metric accuracy between Terrestrial Laser Scanner (TLS) systems and Handheld SLAM technology: the study case of the FJD Trion system in multivariate contexts
GOLETTI, VANESSA
2025/2026
Abstract
Abstract The rapid evolution of Simultaneous Localization and Mapping (SLAM) algorithms integrated into handheld Mobile Mapping Systems (MMS) has significantly accelerated 3D spatial data acquisition, challenging the traditional role of static Terrestrial Laser Scanning (TLS) in architectural and structural documentation. However, assessing the absolute metric reliability of kinematic point clouds remains a critical concern, especially when surveys are executed within multi-scale indoor-outdoor transitions or geometrically challenging underground environments. This thesis establishes a comprehensive, multi-tier metrological validation framework to decouple localized absolute accuracy, continuous volumetric deformations, and the intrinsic noise floor of the handheld FJD Trion P1 SLAM platform against a high-precision TLS geodetic Ground Truth. The experimental pipeline is structured around three complementary validation methods. Method I evaluates localized positional accuracy by isolating homologous checkerboard targets. To bypass the projection and interpolation errors inherent to color-mapping algorithms, a radiometric extraction protocol was developed leveraging exclusively the sensor's backscattered LiDAR Intensity scalar field. Under short mapping windows, this discrete approach yielded absolute 3D global RMSE values of 0.041 ± 0.005 m for an open-loop path (Trajectory 1) and 0.033 ± 0.005 m for a closed-loop configuration (Trajectory 2), validating the metric integrity of the scan-matching backend and quantifying the sub-centimeter optimization benefit provided by automated place recognition. To evaluate macroscopic structural distortions across extended paths, Method II implements a continuous, surface-to-surface comparison utilizing the Multiscale Model to Model Cloud Comparison (M3C2) algorithm. The analysis of an extended outdoor-indoor loop (Trajectory 3) unmasked a critical statistical paradox: while the raw dataset exhibited a deceptive global mean distance near-zero (+0.0004 m ), the error envelope expanded to a standard deviation (σ) of 0.0781 m (7.81 cm ), masking systemic internal tracking drift that reached real local displacements of up to 15-20 cm. Directional residual analysis isolated this degradation along the vertical axis (Z), driven by the operator's gait oscillations and amplified by the low, uniform ceiling geometry of the underground parking lot, which deprived the SLAM algorithm of distinct vertical geometric features. To resolve this tracking failure, a manual post-processing optimization strategy was introduced. By partitioning the continuous kinematic stream into chronological rigid macro-blocks and executing a reversed sequential ICP alignment under strict structural tolerances, the systematic rotational drift was counteracted without resorting to artificial software scaling (s=1). This geodetically sound workflow achieved the total structural suppression of the re-entry misalignments, compressing the global M3C2 standard deviation by roughly 16% down to 0.0658 m (6.58 cm ). Finally, Method III isolated the sensor's intrinsic high-frequency stochastic measurement noise via local planarity tests, yielding an RMSE of just 0.0046 m (4.6 mm) on a sampled structural wall of the third trajectory. This final step highlights a fundamental Precision-Accuracy Paradox, proving that a mobile mapping cloud can maintain exceptional local structural sharpness while suffering from macroscopic global warping. In conclusion, while handheld SLAM systems provide unprecedented operational throughput, establishing an independent, high-order static control network remains an indispensable prerequisite to guarantee absolute metric accuracy in engineering applications.| File | Dimensione | Formato | |
|---|---|---|---|
|
Comparative_Analysis_of_Metric_Accuracy_between_TLS_SLAM.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
9.05 MB
Formato
Adobe PDF
|
9.05 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/10589/261424