Résumé:
While Pavement Management Systems are commonly used for effective roadway inspection and maintenance, similarly efficient solutions for sidewalk assessment remain underdeveloped. This study introduces a semi-automated method for sidewalk evaluation based on a mobile device equipped with a 2D LiDAR scanner, and image processing techniques. The proposed approach includes the design of a cost-effective sensor-equipped vehicle, data acquisition and processing workflows, elevation barrier detection methods, and a comprehensive validation framework. The system integrates data from a 2D LiDAR scanner, wheel encoders, and a localization unit to generate a local digital elevation model (DEM) via constrained spline interpolation. Barrier detection is performed using both a focal range thresholding method and a machine learning-based segmentation approach. Validation is conducted using high spatial resolution reference DEMs (1 mm and 2 cm) derived from Terrestrial Laser Scanning across two pilot sites. Detection quality is evaluated using three key metrics: Coefficient of Area Correspondence, Identification Accuracy, and Point Accuracy. The focal range method achieved overall accuracies of 91% and 92% at the two test sites, while the machine learning approach reached only 36% and 38%. Spatial coincidence indicators revealed reduced agreement, particularly due to shape distortions and segmentation inconsistencies; however, Point Accuracy proved to be the most robust validation metric. The results confirm that the proposed 2D LiDAR-based system and focal range method are effective in identifying significant elevation barriers on sidewalks. The compact design, low operational cost, and ease of use make the solution suitable for widespread deployment in municipal infrastructure assessment. © 2013 IEEE.
