Digital Twins (DTs) in contemporary urban planning and management is urgent, as they provide virtual interaction and simulation that overshadows traditional reliance on on-site visits. This technological shift demands advanced services from the mere passive storage and visualization of conventional geoportals. OpenStreetMap (OSM) is one open global scale platform of geospatial data, storing a great number of civil infrastructures, particularly buildings and streets (Zhou et al., 2022). Recent studies show that OSM elements with highway tags are potentially used as digital twin street models (Duque et Brovelli, 2025; Scalas et al., 2022). However, they usually missed width and lane number attributes, only 1.14% element with “highway tags” has also included “width” tag, and 6.61% with “lanes” tag(Taginfo. ,n.d.).

OSM elements are usually mapped from satellite images, which has certain limitations for measuring detailed dimensions with low resolution. To overcome this problem, there are studies extracting street characters as materials and dimensions from crowdsourcing streetview imageries (SVI) using mostly image segmentation and photogrammetry (Kapp et al., 2025; ). A critical need is reliable remote physical measurement technique, working with even a small amount of streetview images where traditional photogrammetry usually fails. To bridge this data gap, this research proposes state-of-the-art Monocular Depth Estimation (MDE) technologies with careful evaluations.

A novel, scalable framework, called StreetMeasure, is developed to support 3D reconstruction of the street environment. The system is built around a crowdsourcing model, allowing users to contribute data through cloud based or locally-hosted SVI services (e.g., Mapillary, Panoramax). Upon requesting a reconstruction, the system processes the image via certain MDE models, aligning with existing photogrammetric sparse models or low-resolution objects detected from satellite images, providing visualization of the results and showing them on a web map, alongside comparative metrics on model inference time and precision. A crucial part of the framework is the integration of a validation loop, where users can perform and input on-site physical measurements to iteratively validate and refine the inferred depth data. These results could be reused to enhance future predictions. Furthermore, to guarantee operational flexibility and responsiveness across diverse field conditions, the framework is designed to operate efficiently on edge devices.

As part of a strategic transition towards open-source ecosystems, the Province of Piacenza (Italy) implemented a digital road cadaster using PostGIS, LizMap, QGIS and QField technologies. to optimize the management of provincial infrastructure (Gaspari et al., 2023). Following the release of Italian guidelines for extraordinary transport, there is an increasing need for innovative, low-cost approaches to updating relevant characteristics (such as width, surfaces, curves angles) of road graphs to develop better strategies in identifying optimal routes.
Taking the 16 km-long SP6 between the municipalities of Piacenza and Carpaneto Piacentino as the main case study, this project aims to integrate traditional mapping approaches by integrating open, crowdsourced data and AI, while measuring the width of each lane in different road cross-sections (see Figure 1). Although street-level imagery from Mapillary and 20-cm resolution orthophotos from Regione Emilia Romagna are openly available, the 360° sequences captured from car-mounted cameras at 13-m intervals lack the required density of mutual points for traditional photogrammetric reconstruction. To overcome these technical limitations, this research explores AI-driven alternatives to extract measurements from existing open data, enriching OSM data as the basis for the updated graph of the Province. All digital outputs will be validated through in-situ measurements to ensure regulatory compliance.
Initially, a set of 714 panoramic images have been downloaded and ordered sequentially (see Figure 2). Then, using Segment Anything Model 2 (SAM 2), we masked the “road” element in both streetview and satellite images. Then streetview images are passed through Depth Anything in 360°(DA360, cite) to reconstruct a 3D scaled point cloud model (see Figure 3). A number of transects are set along the street in both satellite and street view images in order to recover the scale. To evaluate the quality of this framework, the study compared the estimated width of the lane with ground truth measurement. Later, from the sequence of point clouds, it is possible to update width (for total width), lanes( for number of lanes) tags on OSM.

This framework establishes a highly scalable not only in mapping other areas, but also measuring different civil elements, e.g. bridges or tunnels. Moreover, the blended data of OSM and point clouds can serve as a powerful simulating environment capable of hosting virtual field trips and immersive planning exercises. Ultimately, StreetMeasure can support urban planning workshop tools, uniquely enabling comparative analysis and rapid assessment of project impact by simulating clear ‘before-and-after’ scenarios in a collaborative virtual space.

This study proposes a framework using up-to-date open source technology extracting information from SVI and SI in order to fulfill OSM data. It has been applied in a practical case of SP6, with actual problems of both quantity and quality of crowdsourcing data. Study found that in the field of DT, OSM can be placed in the centre of a framework. At one point OSM stores data from various public surveys, while from another point OSM becomes a primary dataset for DT applications.