Study of footpaths in Singapore using image recognition model

Abstract

With the rise of alternative transportation means such as cycling and PMDs, pedestrian footpaths face a rapidly changing demographic as more users switch to these vehicles for their first and last mile solutions. Comfortability of footpaths for users, thus, becomes an increasingly important factor for users who travel every day. While conventional indexes exist for common footpaths, Singapore’s footpaths are out of the norm due to the existence of drainage covers. As such, drainage covers may pose as an additional factor for pavement comfortability, especially for cyclists whose journeys are the most affected by evenness of the pavement. This project is the initial exploration of programming an automatic system for measuring comfortability indexes or level of service of footpaths for cyclists. Image recognition models are used to identify the existence of drainage covers along footpaths as drainage covers are factors contributing to the comfortability index of footpaths. The model is trained on a series of data images collected around Singapore and adjusted to determine the most optimal outcome. It was found that a larger dataset, coupled with a smaller learning rate and training batch size, significantly improved the classification accuracy of the model while saving computational time. The model was able to efficiently achieve an accuracy of 80% with an average analysis time of 5 min.