Smart Urban Mapping via Mobile LiDAR and AI for Rapid GIS Updates

Thesis Abstract

Urban environments are in a state of rapid change, demanding timely, accurate geospatial information to support planning, infrastructure management, and disaster resilience. Traditional GIS workflows struggle to keep pace with dynamic urban morphologies, resulting in lagged updates, high labor costs, and data silos. This study investigates a technology-driven solution that integrates mobile LiDAR data acquisition with artificial intelligence to enable rapid, scalable GIS updates for smart city applications. The aim is to develop and validate an end-to-end workflow that automates data collection, feature extraction, and GIS updating processes to produce high-fidelity urban models and semantic classifications in near real-time. Specific objectives are (i) to design a mobile LiDAR data capture protocol optimized for urban corridors with varying traffic and pedestrian activity; (ii) to develop AI-based algorithms for automated object detection and semantic segmentation of urban features (buildings, roads, street furniture, vegetation, and utilities) from multi-sensor streams; (iii) to integrate LiDAR-derived geometries with existing GIS datasets to generate consistent, up-to-date urban 3D models; (iv) to evaluate the accuracy, reliability, and update latency of the workflow across diverse city districts; and (v) to assess the scalability, cost-efficiency, and policy implications for municipal GIS operations. The methodological approach adopts a convergent mixed-methods design combining quantitative accuracy assessment with qualitative process evaluation. The population comprises urban areas within a mid-sized metropolitan region characterized by heterogeneous land uses. A stratified random sampling scheme selects 40 road segments and 15 intersections representing central business districts, residential neighborhoods, and peri-urban zones. Mobile LiDAR data are collected using a contemporary 360-degree LiDAR-enabled vehicle equipped with synchronized high-resolution cameras and GNSS/IMU sensors, yielding approximately 1.2 terabytes of point clouds per survey day. Data collection instruments include calibrated LiDAR scanners, calibration targets, high-definition imagery, and standardized metadata forms. For AI-driven analysis, deep learning models with architectures such as PointNet++, DeepLabv3+ (for image-augmented segmentation), and graph-based post-processing are trained on a labeled corpus of 2,500 manually annotated urban features. Validation employs field surveys, terrestrial laser scans, and cross-referencing with authoritative cadastral and utility datasets to compute metrics such as point-density accuracy, object-level IoU, completeness, correctness, and positional RMSE. Analytical methods encompass both geospatial analytics and machine learning evaluation. The accuracy assessment applies regression analyses to quantify deviations between LiDAR-derived surfaces and ground truth, along with ANOVA to test performance differences across feature classes and urban zones. AI models are evaluated using precision, recall, F1-score, and IoU, with k-fold cross-validation and ablation studies to determine the contribution of each sensor modality. The integration framework uses principled data fusion techniques, including co-registration error modeling and topology-aware mesh generation, to produce consistent 3D city models that seamlessly update existing GIS layers. The study additionally investigates update latency under different data throughput scenarios and conducts a cost-benefit analysis for municipal adoption. Ethical considerations address data privacy in public-rights-of-way environments and adherence to data governance standards. Expected findings indicate that the integrated workflow achieves a mean positional RMSE of less than 0.15 meters for major architectural facades and 0.25 meters for street-level features, with IoU scores exceeding 0.78 for building/road classes and 0.65 for street furniture and vegetation. Time-to-update from data collection to GIS publication is anticipated to reduce by at least 60% relative to conventional processes, while total cost per update cycle decreases by 25–40% when scaled across districts. The research contributes to knowledge by (i) providing a validated, scalable blueprint for mobile LiDAR-enabled rapid GIS updates; (ii) advancing AI-based semantic mapping in complex urban environments; and (iii) offering an integrative framework for harmonizing new sensory data with legacy GIS infrastructures, including standards-based data fusion and quality assurance protocols. The study concludes that mobile LiDAR coupled with AI can transform urban geospatial workflows, enabling agencies to maintain current, richly annotated 3D city models. Recommendations include the adoption of standardized data schemas, investment in continuous model retraining pipelines, governance mechanisms for privacy, and pilot deployment guidelines to scale the approach across multiple municipalities.