Toward automated sidewalk pavement condition elevaluation utilizing bike mounted camera

Abstract

This study proposes a methodology to expedite sidewalk condition assessments and develops a numerical rating system for ranking concrete sidewalk conditions. The goal is to identify hazardous and distressed areas within sidewalk networks to better address maintenance needs and accessibility concerns. The methodology introduces an automated Concrete Sidewalk Condition Index (CSCI) based on observable surface distresses and detects other hazards based on compliance with standards. Surface distresses are captured by an off-the-shelf camera mounted on a bicycle to collect video data of the sidewalks of interest. Images extracted from these videos fine-tune a lightweight deep learning object detection model (YOLOv8m) to detect and categorize various concrete sidewalk pavement distresses (shattered slab, transverse cracks, scaling, heavy scaling, and corner breaks) solely from RGB images. Inertial sensors within the camera are used to detect vertical faults, quantify their severity, and flag tripping hazards. Additionally, cross slopes and running slopes can be captured using the inertial sensors within the camera. The sidewalk pavement conditions extracted from the low-cost camera are weighted and fused to develop a numerical rating for the CSCI. Sidewalk sections CSCI scores and other factors of compliance can then be used to produce visual aids of sidewalk conditions (such as heatmaps) and help make maintenance related decisions. The results demonstrate the effectiveness of the proposed methodology with the object detection model achieving a mean average precision at 50 percent intersection over union (mAP50) of 97.6 percent on the validation data, a detection accuracy of 87.7 percent for flagging tripping hazards, an accuracy of nearly 97 percent for detecting running slopes, and an accuracy of 97.5 percent for detecting cross slopes. A case study utilizing the methods is provided.