Automatic concrete sidewalk deficiency detection and mapping with deep learning

Vertical displacement is a common concrete slab sidewalk deficiency, which may cause trip hazards and reduce wheelchair accessibility. This paper presents an automatic approach for trip hazard detection and mapping based on deep learning. A low-cost mobile LiDAR scanner was used to obtain full-width as-is conditions of sidewalks, after which a method was developed to convert the scanned 3D point clouds into 2D RGB orthoimages and elevation images. Then, a deep learning-based model was developed for pixelwise segmentation of concrete slab joints. Algorithms were developed to extract different types of joints of straight and curved sidewalks from the segmented images. Vertical displacement was evaluated by measuring elevation differences of adjacent concrete slab edges parallel to the boundaries of joints, based on which potential trip hazards were identified. In the end, the detected trip hazards and normal sidewalk joints were geo-visualized with specific information on Web GIS. Experiments demonstrated the proposed approach performed well for segmenting joints from images, with a highest segmentation IoU (Intersection over Union) of 0.88, and achieved similar results compared with manual assessment for detecting and mapping trip hazards but with a higher efficiency. The developed approach is cost- and time-effective, which is expected to enhance sidewalk assessment and improve sidewalk safety for the general public.