Development of Sidewalk Condition Assessment

With limited budget and resources, sidewalks are often overlooked. Damaged or unmaintained sidewalks may compromise a pedestrian’s safety and the accessibility of using public spaces. The importance of a well-maintained sidewalk network can affect many areas of the community, such as safety, the economy, socialization, and the environment. By addressing these issues, a municipality can avoid injury liabilities and encourage healthy lifestyles and social interactions by providing a sidewalk network that is accessible and free of tripping hazards and uneven surfaces.

Two of the most important aspects of geomatics engineering are fusion of sensor data and software analysis in the spatial domain. In this project, we took both of these areas and created two different solutions. The first solution is a 3D laser scanning assessment which utilizes the land surveying side of geomatics. The second solution uses machine learning to create a convolutional neural network which makes use of the programming and software development side of geomatics.

The most common method for assessing sidewalk maintenance conditions is a visual inspection. Although efficient in smaller municipalities, a visual inspection can be harder to implement in larger cities. Experience and a practical understanding of how sidewalk conditions can change over time are necessary for a reliable assessment. By testing multiple collection methods, our solution will provide detailed information about sidewalk condition that has a qualitative assessment for prioritizing sidewalk repair and a visual presentation of our results collected in the field. The development of our design should be feasible, efficient, mobile, and automated to take the guesswork out of condition assessment.

Each solution was compared to a baseline, which was our visual inspection. The team focused on the condition criteria from the baseline, which accounted for four defects; Deflection, Cracking, Spalling, and Settlement. These conditions were quantified from the results of each method and compared.

Each development method of data collection, CNN and 3D scanning, had pros and cons. The 3D scanning and modeling provided high resolution, measurable, and visual results, but was time consuming and would be better implemented with a dynamic setup with GPS, instead of multiple static setups. The Faro Scene software used to process the 3D data was very user friendly with automated registration and it was easy to combine the multiple scan frames. Although, there were data gaps directly underneath the instrument. The CNN method still required more development as it proved to be very complicated to create the machine learning algorithms. Our team was able to collect data for the recognition of cracks in the app quickly and the overall cost of the CNN implementation compared to the laser scanning was significantly less. Overall, each method could be used as an alternative to a visual inspection depending on the clients requirements.