5.2 Lessons from methods comparison
The comparison led to several conclusions:
1. Smartphone survey method achieves large proportion of data collection needs, but remains time-consuming. Using our current data collection instrument with all the questions included, up to 12 minutes is required to capture measurements for a single intersection. Field testing led us to conclude that the time requirements remain onerous where volunteer-led data collection is concerned. A priority in refining the tool should be to reduce the time requirement per intersection.
2. LiDAR scanner is of highest, but excessive, quality; while data quality from the low-cost Structure Sensor is sufficient for our needs. The device records points density at one million points per second. During post-processing, the superior quality of the point could for our target curb cuts was clearly apparent, when compared with Structure Sensor. However, the point density is far above requirements. Moreover, the Structure Sensor provided a sufficiently detailed point cloud to extract our required measurements at the necessary data quality. As such, Structure Sensor is our preferred 3D image acquisition method. Given its cost benefit and presumed similarity to the devices that may ship in 2019 iPhones, the technology is considered highly suitable for crowd-sourced street imagery in future.
3. The smartphone survey is not currently suitable for crowd-sourcing purposes. While conducting the smartphone-based survey, team members noted a self-conscious feeling that accompanied taking physical measurements. The process required crouching or bending down to foot level, extending a tape measure, and recording the dimensions noted. In a busy streetscape, there was a risk of tripping up pedestrians with the tape measure. Moreover, the measurements were time-consuming: for an intersection with 8 curb cuts it required at least 12 minutes. By contrast, taking smartphone photos produced little feeling of self-consciousness.
4. Three-dimensional sensors may help overcome behavioral constraints among crowd-sourcing volunteers. It was noted that 3D sensing technologies in smartphones may help volunteers overcome the behavioral constraint induced by feeling awkward when using a tape measure. Moreover, rapid acquisition of 3D images may help reduce the crucial time-per-observation criterion. Since physical measurements can be extracted during post-processing, a survey instrument incorporating 3D imaging could reduce the number of data points required - potentially limiting these to location fields, the 3D scan, and the minimum necessary qualitative fields where subjective judgment is required (such as overall curb condition, presence of detectable warning, and danger factor of curb cut location). Moreover, replacing hand measurements with 3D sensing could make it possible for wheelchair users to participate in data acquisition.
5. Good weather is required for 3D sensing. Field testing revealed that rainy days invalidate 3D scans of curbs. Puddles are recorded as physical surface, rendering the slope angle and lip / bump measurement inaccurate.
6. Query functionality realises significant value from sensor-acquired data. The ability to query the survey results through ArcGIS and a web map interface realized significant value. In our work, end users may call up photos of specific curbs based on a geographic region of interest, or any query term combining the attributes under consideration (examples include showing all curbs with crumbling concrete value "severe", all curbs with condition "very poor", or all curbs on a specific stretch of Broadway with no detectable warnings for the visually impaired). It was noted that 3D scans could be attached to the same Geodatabase structure. Through use of standardized 3D viewing platforms such as SketchFab, DOT or advocacy group users could view 3D representations of the curbs under question and conduct further exploration of their physical form such as extracting new measurements or verifying measurements from a printed report prior to sending field staff to make repairs.
7. 3D imagery may improve data credibility. After team members surveyed 40+ curb cuts using tape measures and photos, repetitiveness of the task led to a desire to speed things up by using rules of thumb such as estimating ramp width based on visual recollection for specific types of ramp slab. It was noted that quality of measurements made by hand may suffer as the number of curbs surveyed rises. Final data quality may suffer from credibility issues if volunteers are known to stray from the instructions. While 3D imaging brings its own challenges - such as partial scans in cases where a pedestrian or street furniture obstructed the view - we noted advantages. Extracting measurements from 3D scans circumvents the tendency of volunteers to cut corners, and may increase the trust that decision-makers place in the resulting measurements. Besides providing great vision aid for inspections, 3D models can be easily converted to Computer Aided Drawings (CAD) which is popular used in engineering world.
8. GPS inaccuracy is a key operational constraint. Coordinates generated from smartphone GPS are not usable for placing curb cut records on a map. To conduct future curb cut surveys in an organizational context, volunteers could be equipped with accurate GPS devices. For crowd-sourced efforts, manual entry of cross-street and curb orientation is a workable solution, albeit adding to post-processing time.
6. Recommendation and preliminary design for 3D-enabled smartphone app
Our study leads to the following recommendation: develop a smartphone app for 3D-enabled collection of streetscape accessibility data, combining survey methodology with 3D scan capability.
Preliminary design for the app is shown in Figure 13. The design responds to the following design criteria: (i) benefit from anticipated deployment of Structure Sensor-style technology in smartphones from 2019; (ii) minimize time requirement per curb cut; (iii) incorporate the required measurement data and qualitative data; (iv) facilitate the merging of information from multiple users into a queryable Geodatabase; and (v) overcome GPS location constraints through snapping of curbs to their position on each intersection.
The smartphone app interface would resemble the Survey123 instrument developed for this project. Users would have an additional option to capture 3D imagery at each curb cut that they inspect. The 3D imagery component would allow the number of data points entered manually to be reduced, eliminating use of tape measures on street corners.
In light of GPS location problems with smartphone data and to minimize post-processing time, the app should automatically snap curb records to a representation of their location. A proposed method is to use the Google Maps API to locate the intersection dimensions, and place curb cuts according to their cross street and orientation.
A visualization interface should be developed allowing simple queries of the resulting Geodatabase. Alternatively, the data could be exported as a Shapefile for use in ArcGIS or QGIS alongside other datasets. We note that the method could equally be extended to other urban accessibility problems for which 3D imagery adds value. Wheelchair users are unlikely to use tape measures to confirm lack of turning space in movie theater corridors; and the slope measurement problems noted above mean they are unlikely to record objective evidence of excessive ramp slope at public libraries, hospitals, schools, airports or other categories of building. A crowd-sourcing effort using the architecture described here may add significant value to data generation, helping draw attention to these barriers and pressure organizations into making the necessary investments to become ADA compliant.