BUILDING DEFECT RECOGNITION FOR DEEP LEARNING MOBILE VISUAL INSPECTION

 BUILDING DEFECT RECOGNITION FOR DEEP LEARNING MOBILE VISUAL INSPECTION APPLICATION

 ZI HAO LIM1, CHU KIONG LOO2, HASNIYATI HAMZAH3

1,2 Faculty of Computer Science and Information Technology

Department Of Artificial Intelligence

3 Faculty of Built Environment

Department of Real Estate

UNIVERSITI MALAYA (UM)

ABSTRACT

A frequent building inspection and monitoring for building defects are critical to ensuring the structural health and dependability of the structures. Human visual examinations have traditionally been used to detect building defects. However, there are some drawbacks to this strategy which are time-consuming, high labour cost and may yield inconsistent results because the inspector lacks experience and the risk of inspection in limited spots. Therefore, an automated building defects inspection system is necessary to ensure the safety of people. This research aims to design an automated building defects recognition system based on the implementation of YOLOv5 models, this technique for identifying defects in building parts as interior in images or videos. The design of mobile application utilized the best performing inferencing result for building defects in ceiling, door, floor, wall, and windows with values of 0.72, 0.891, 0.873, 0.853, 0.727 in mAP with limited datasets in real-time for mobile system.

Keyword: Building defects, deep learning, convolutional neural network, computer vision, mobile application