Wheely Easy: Creating a Wheelability Network for Bradford

In partnership with Healthy Urban Places, Population Health Improvement UK and Visualising Heritage, Department of Archaeology & Forensic Sciences, Institute of Digital & Sustainable Futures, University of Bradford.

Cities may appear easy to navigate, but for wheelchair users, everyday journeys are often shaped by barriers that most people never notice. 

Project overview 

Accessible description: Figure 1 illustration compares pavement use by an able-bodied pedestrian and a wheelchair user. Features such as narrow pavements, steep slopes and high curbs are shown as manageable for pedestrians but challenging for wheelchair users. The figure highlights the importance of dropped curbbs, wider pavements and gentler slopes for accessible mobility. 

Urban accessibility research has long focused on walkability, assessing how easily people can move through cities on foot. However, this perspective often assumes an able-bodied pedestrian and overlooks the specific challenges faced by wheelchair users. As a result, streets that are considered accessible in conventional assessments may still present significant barriers for those relying on wheeled mobility. 

Recent advances in data collection, particularly mobile LiDAR scanning, have made it possible to capture cities in far greater detail than before. This has enabled researchers to begin measuring key aspects of accessibility such as pavement width and slope. While these developments mark important progress, they do not fully account for one of the most critical elements of wheelchair movement: the ability to transition between the pavement and the road. This transition depends on the presence of dropped curbs, which are often missing, inconsistently designed, or poorly documented in existing datasets.  

Accessible description: Figure 2 illustration shows how sidewalk width, sidewalk slope and dropped curbs affect whether a wheelchair user can safely cross a street. The lower part of the figure presents the proposed “wheelability network”, where accessible sidewalks and crossing points are connected into a navigable street network for wheelchair users. 

This project addresses this gap by developing a new approach to modelling urban accessibility from a wheelchair user’s perspective. Using high-resolution street data from Bradford, we construct a wheelability network that reflects how accessible the street environment is in practice. By combining information on pavement conditions with the location of dropped curbs, the project moves beyond traditional pedestrian-focused measures and provides a more realistic representation of how cities can be navigated using a wheelchair. 

Data and methods  

Understanding wheelchair accessibility requires more detail than traditional urban datasets can provide. Common sources such as street networks or street-view imagery offer largely two-dimensional representations, which miss the fine-grained structure of pavements, curbs, and crossings that determine real accessibility. 

Accessible description: Figure 3 illustration shows the process of collecting street data using a vehicle-mounted LiDAR system. A scanning vehicle captures laser measurements of the surrounding street environment, producing dense point cloud data. These point clouds are then processed to create a 3D street model used for accessibility analysis. 

To address this, we use high-resolution mobile LiDAR data collected across Bradford. Captured by a vehicle-mounted system at up to 4 million point per second, the data form a dense three-dimensional point cloud. This level of detail enables key features such as curb height, pavement slope, and width to be measured directly, while also supporting analysis across entire street networks rather than isolated locations. As a result, accessibility can be assessed not just in terms of infrastructure presence, but in how routes connect in practice, allowing barriers and accessible pathways to be identified systematically. 

Method

To understand wheelchair accessibility in a meaningful way, we began by reframing the problem from the perspective of real-world mobility constraints. For wheelchair users, accessibility is not simply determined by the presence of pavements, but by whether routes are physically navigable. This depends on two interrelated conditions: whether a pavement is usable (in terms of slope and width), and whether it is connected through safe crossing points. In practice, these connections are enabled by dropped curbs. This insight shaped the design of our analytical workflow. 

Accessible description: Figure 4 diagram presents the analytical workflow used to build the wheelability network. Mobile LiDAR point cloud data are first processed using a pre-trained road classification model to extract road surface data. The workflow then branches into two analyses: dropped curb detection for identifying accessible crossings, and sidewalk extraction for assessing sidewalk width and slope. These outputs are combined to create a connected wheelability network representing wheelchair accessibility across the street system. 

Building on the high-resolution point cloud data described above, we developed a structured workflow that translates these constraints into measurable features of the street environment. The workflow is centred on two complementary components: the detection of dropped curbs, which determine network connectivity, and the assessment of sidewalks, which determine route quality. 

Detecting dropped curbs: identifying where crossings are possible 

The detection of dropped curbs forms the core of the analysis, as these features determine whether different parts of the street network are connected for wheelchair users. Rather than relying on pre-classified curb labels, which are often unreliable in complex urban environments, we adopt a geometry-based approach grounded in the structure of the road surface. 

The process begins by constructing a stable representation of the road from the point cloud data. An initial classification is used to identify road-related surfaces, but instead of directly trusting these outputs, we consolidate them into a unified road layer and focus on geometric properties. This allows the method to remain robust to inconsistencies in the underlying classification model. 

Accessible description: Figure 5 image shows a 3D surface model reconstructed from mobile LiDAR data. The street layout and surrounding surface geometry are represented as a continuous textured mesh. Variations in elevation and surface form are visible across the model, demonstrating the level of spatial detail captured during the reconstruction process. 

Within this representation, potential curb structures are identified by analysing local variations in surface height. Grid cells with a local height range exceeding 5 cm changes are flagged as curb candidates, as these discontinuities correspond to the physical edges between the road and pavement. 

Accessible description: Figure 6 shows a binary spatial map highlighting areas where local surface height variation exceeds 5 centimetres. White pixels represent locations with larger height changes, while black areas indicate relatively flat surfaces. Linear patterns can be observed along the street structure, revealing continuous surface boundaries and elevation transitions across the scene. 

To make these features easier to interpret and analyse, the detected curb regions are then reduced to continuous linear structures using a skeletonisation process. This step simplifies complex surface geometry into a network of curb lines. 

Accessible description: Figure 7 shows continuous linear segments extracted from the binary height variation map. Different colours represent separate connected line segments identified during the extraction process. The extracted lines follow the spatial structure of the street environment and highlight continuous surface boundaries across the scene. 

Once curb segments are established, the method focuses on identifying where these segments are interrupted in ways that indicate a possible crossing point. For each curb segment, we analyse its endpoints and compute spatial relationships to nearby segments. Dropped curbs are characterised by pairs of curb segments that terminate in close proximity, forming a gap that allows a wheelchair to pass between pavement and road. 

Accessible description: Figure 8 shows detected dropped curb locations across the street network. Black markers represent the endpoints of extracted curb line segments, while red points indicate the estimated centre locations of detected dropped curbs. The detected locations are distributed along both sides of the street and at crossing areas where gaps between line segments fall within the defined distance threshold. 

To distinguish meaningful crossings from noise, we apply distance-based criteria to these gaps. Gaps smaller than 1.5m are treated as artefacts of the data, while gaps larger than 5.0m are associated with junctions or unrelated street features. Intermediate distances, consistent with accessibility design standards, are classified as dropped curbs. Using this approach, we are able to systematically detect dropped curb locations from the underlying geometry. 

Extracting and assessing sidewalks: measuring route quality 

While dropped curbs determine whether routes are connected, sidewalks determine how usable those routes are. Extracting sidewalks from point cloud data, however, presents several challenges in real-world environments. 

A common approach is to separate sidewalks based on elevation differences from the road surface. However, this is unreliable in many streets where the road surface is cambered, with the centre of the road higher than the edges. In such cases, elevation-based methods can incorrectly classify parts of the road as sidewalks, leading to systematic errors. 

To address this, we developed a curb-driven approach to sidewalk extraction that builds directly on the detected curb structures. Instead of identifying sidewalks globally, we reconstruct them locally from their boundaries. 

Accessible description: Figure 9 illustrates the process used to identify sidewalk regions and measure local accessibility conditions. The left panel shows how probe points are used to distinguish the lower road surface from the higher sidewalk side along a curb segment. The right panel shows sampled sidewalk boundary points and measurement lines used to estimate local sidewalk width and slope. Different coloured markers and lines represent sampled points, sidewalk boundaries and local measurement locations across the street segment. 

For each curb segment, we sample points on both sides of the boundary to determine which side corresponds to the sidewalk surface. This probing step allows us to distinguish between road and pavement even when elevation differences are ambiguous. Once the sidewalk side is identified, we generate perpendicular sampling lines extending outward from the curb. These lines act as local scans, capturing the extent and geometry of the sidewalk in a structured way. 

This boundary-driven reconstruction produces a more stable representation of sidewalks across varying street conditions. To further improve robustness, we apply location-based filtering to remove spurious geometries, including central islands and fragmented segments that do not correspond to true curb-defined sidewalks. This filtering step reduces noise and ensures that only meaningful sidewalk segments are retained.  

Accessible description: Figure 10 shows the filtering results applied to extracted curb line segments across the street network. Green lines represent retained segments used for further analysis, red lines indicate segments removed near the boundary of the dataset, and blue regions represent detected central islands. Grey points in the background show the underlying road surface data used during the filtering process. 

 

Accessible description: Figure 11 shows filtered sidewalk measurement strips distributed along the street network. Red lines represent extracted curb segments, while green perpendicular strips indicate local sidewalk width measurements extending from the curb toward the sidewalk boundary. The lower panel overlays these measurements on the underlying road surface data, and orange markers indicate locations where valid sidewalk boundaries could not be identified. 

With sidewalks reconstructed, we compute two geometric measures for each segment: slope and width. Slope is calculated along the centreline based on local elevation changes, while width is measured as the shortest distance from the centreline to the sidewalk boundary. Each measure is then mapped to a continuous score between 0 and 1 using threshold ranges derived from the Department for Transport’s Inclusive Mobility: A Guide to Best Practice on Access to Pedestrian and Transport Infrastructure. These mappings translate guideline-based criteria into quantitative scores that can be applied consistently across the network. The final wheelability index is computed by combining the two scores multiplicatively, producing a composite measure of accessibility for each sidewalk segment. 

Accessible description: Figure 12 contains two line graphs showing how the wheelability index changes with sidewalk width and slope. The left graph shows the wheelability index increasing as sidewalk width becomes larger, reaching the highest values at wider widths. The right graph shows the wheelability index decreasing as sidewalk slope becomes steeper, approaching zero at high slope values. Together, the graphs illustrate how wider and flatter sidewalks correspond to higher accessibility scores. 

From local analysis to network: constructing the wheelability framework 

The final stage of the workflow integrates dropped curb detections and sidewalk assessments into a unified network representation. In this framework, sidewalks are represented as links, while dropped curbs define connection points between segments. 

Accessible description: Figure 13 shows the calculated wheelability index along sidewalk segments within the street network. Colours represent accessibility scores ranging from low to high values, as indicated by the colour bar on the right side of the figure. Higher scores are concentrated along some continuous sidewalk sections, while lower scores appear in narrower or steeper areas. The map demonstrates how accessibility conditions vary spatially across the network. 

This produces a connected network in which each segment is associated with an accessibility score, combining both connectivity and sidewalk conditions. 

Key findings  

Applying the workflow to a 200-metre street segment shows that wheelchair accessibility can be identified at a fine spatial scale, including both sidewalk conditions and crossing points. The dropped curb detection component has also been extended to a 3 km section of Bradford’s street network, with results showing strong agreement with manual inspection. 

Accessible description: Figure 14 shows a larger test area of approximately 3 kilometres used to evaluate the methodology. In the main map, grey lines show the extracted street structure and red markers show detected dropped curbs. A blue rectangle highlights one selected section of the larger sample. The inset image on the right shows this highlighted section in more detail, with the wheelability index mapped along the sidewalk segments. 

A key finding is that accessibility depends less on the presence of pavements and more on how they connect. Missing or misaligned dropped curbs can prevent movement between sidewalks, breaking the network even where both sides of a street are usable. As a result, routes that appear continuous on a map may not be navigable in reality. 

Value of the research  

This research provides a practical way to translate detailed street-level data into actionable insights for accessibility. By constructing a wheelability network that combines sidewalk conditions with crossing points, the approach enables accessibility to be understood as a connected system rather than isolated features. This creates immediate opportunities for real-world applications, including accessible route planning for wheelchair users, the development of pathfinding tools, and the integration of accessibility metrics into digital mapping services through APIs. 

Accessible description: Figure 15 illustrates potential applications of the wheelability network. The left panel shows a simplified accessible street network with connected sidewalk segments and crossing points. The centre panel presents a conceptual pathfinding API designed to support wheelchair users in route planning. The right panel highlights the use of accessibility information in inclusive transport policy and urban planning decisions. 

Beyond navigation, the framework also supports evidence-based decision-making for urban planning and transport policy. By identifying where connectivity breaks down, such as missing or poorly aligned dropped curbs, the analysis highlights priority locations for intervention. This allows local authorities and partner organisations to move from general accessibility guidelines to targeted, data-driven improvements, contributing to more inclusive street design and more equitable urban mobility. 

The project has also demonstrated impact within the geospatial research community. The methodology was presented as an oral presentation at GISRUK 2026, where it was recognised with the GISRUK and OSGeo:UK GoFundGeo award. This award supports early-career researchers developing tools or methods with strong potential for adoption within the open-source geospatial community, providing funding to help translate research into reusable code and practical tools. The recognition reflects the potential of this work to be developed into widely accessible, open-source solutions for analysing and improving wheelchair accessibility at scale. 

Quote from project partner

Insights 

• Accessibility is a network problem, not just a property of individual streets.  
• Dropped curbs are critical connection points that determine whether routes are usable in practice. 
• Streets that appear continuous in standard maps can function as disconnected networks for wheelchair users.  
• High-resolution point cloud data enables accessibility to be measured directly, rather than inferred.  
• A curb-driven approach provides a more robust way to identify sidewalks in complex street environments.  
• The methodology has strong potential to be developed into open-source tools for wider use in planning and navigation. 

Team 

• Chenrui Xiao, Data Scientist, Leeds Institute for Data Analytics, University of Leeds 
• Victoria Houlden, Associate Professor, School of Geography, University of Leeds 
• Jennie Gray, Research Fellow, School of Geography, University of Leeds 
• Markus Billeter, Lecturer, School of Computer Science, University of Leeds 
• Tom Sparrow, Senior Scientist, Institute of Digital & Sustainable Futures, University of Bradford 
• Andrew S. Wilson, Professor, Institute of Digital & Sustainable Futures at the University of Bradford 

Funder 

This research is funded by the Healthy Urban Places (HUP) consortium. The study uses high-resolution mobile mapping data provided by the Visualising Heritage team at the University of Bradford, using infrastructure funded by AHRC CapCo AH/V01255X/1. 

This work has been facilitated by the Leeds Institute for Data Analytics (LIDA) Data Scientist Development Programme, which employs early-career data scientists to deliver real-world data-driven impact in the interests of the public good. 

References 

• Department for Transport. (2021). Inclusive Mobility: A Guide to Best Practice on Access to Pedestrian and Transport Infrastructure. London: Department for Transport. 
• Deturbide, K., & Terashima, M. (2025). Enhancing walkability/rollability audit tools to address qualitative measures for accessibility. Journal of Transport & Health, 36, 101682. 
•  Hou, Q., & Ai, C. (2020). A network-level sidewalk inventory method using mobile LiDAR and deep learning. Transportation Research Part C: Emerging Technologies, 117, 102255. 
•  Meng, S., Su, X., Sun, G., Li, M., & Xue, F. (2025). From 3D pedestrian networks to wheelable networks: An automatic wheelability assessment method for high-density urban areas using contrastive deep learning of smartphone point clouds. Computers, Environment and Urban Systems, 117, 102255.  
•  Ning, H. (2021). Neighborhood mobility auditing and assessment for wheelchair users based on street view imagery and artificial intelligence (Doctoral dissertation). The Pennsylvania State University. 
•  Shields, R., Gomes da Silva, E. J., Lima e Lima, T., & Osorio, N. (2023). Walkability: A review of trends. Journal of Urbanism: International Research on Placemaking and Urban Sustainability, 16(1), 19–41.  
•  Xiao, C., Houlden, V., Gray, J., Billeter, M., Sparrow, T., & Wilson, A. S. (2026). Wheely Easy: Creating a Wheelability Network for Bradford. The 34th Annual GIS Research UK (GISRUK) Conference (GISRUK), The University of Birmingham, United Kingdom.
•  Zhang, T. Y., & Suen, C. Y. (1984). A fast parallel algorithm for thinning digital patterns. Communications of the ACM, 27(3), 236–239.