Integrated YOLO and CNN Algorithms for Evaluating Degree of Walkway Breakage

• Published in: KSCE journal of civil engineering Vol. 26; no. 8; pp. 3570 - 3577
• Main Authors: Choi, Min Je, Ku, Dong Gyun, Lee, Seung Jae
• Format: Journal Article
• Language: English
• Published: Seoul Korean Society of Civil Engineers 01.08.2022; Springer Nature B.V; 대한토목학회
• Subjects: Algorithms; Artificial neural networks; Breakage; Civil Engineering; Damage; Deep learning; Detection; Economic impact; Engineering; Geotechnical Engineering & Applied Earth Sciences; Industrial Pollution Prevention; Machine learning; Model accuracy; Neural networks; Object recognition; Restoration; Transportation Engineering; Walkways; 토목공학; Image deep learning; YOLO algorithm; Pedestrian safety; Evaluation indicators; Breakage detection; Walking environment assessment; Walkway
• ISSN: 1226-7988; 1976-3808
• DOI: 10.1007/s12205-022-1017-1

The focus of policymaking in Korea has changed from vehicle-centric road environments to people-centric environments. As the importance of walking has increased, the construction of pedestrian paths and interest in pedestrian environments have also increased. However, problem recognition and resolution require considerable time in the event of a problem in a pedestrian path. People with reduced mobility tend to resist changes in roads that they use. Thus, damaged pedestrian paths and obstacles pose a considerable risk and economic loss to transportation. In this study, we aimed to minimize the time and cost required for the evaluation of pedestrian paths by developing an automatic system for determining damage using integrated You Only Look Once (YOLO) and convolutional neural network (CNN) image deep learning algorithms. We constructed a model using image deep learning by dividing the steps into walkway breakage detection and score evaluation according to the degree of breakage. The accuracy of the model was determined to be 92%. In the future, the evaluation of pedestrian path damage is expected to be automated using images and videos, thereby reducing the time required for the detection and restoration of damage.