Applying Deep Learning and Single Shot Detection in Construction Site Image Recognition

Abstract

A construction site features an open field and complexity and relies mainly on manual labor for construction progress, quality, and field management to facilitate job site coordination and productive results. It has a tremendous impact on the effectiveness and efficiency of job site supervision. However, most job site workers take photos of the construction activities. These photos serve as aids for project management, including construction history records, quality, and schedule management. It often takes a great deal of time to process the many photos taken. Most of the time, the image data are processed passively and used only for reference, which could be better. For this, a construction activity image recognition system is proposed by incorporating image recognition through deep learning, using the powerful image extraction ability of a convolution neural network (CNN) for automatic extraction of contours, edge lines, and local features via filters, and feeding feature data to the network for training in a fully connected way. The system is effective in image recognition, which is in favor of telling minute differences. The parameters and structure of the neural network are adjusted for using a CNN. Objects like construction workers, machines, and materials are selected for a case study. A CNN is used to extract individual features for training, which improves recognizability and helps project managers make decisions regarding construction safety, job site configuration, progress control, and quality management, thus improving the efficiency of construction management.

Keywords

• construction image
• artificial intelligence
• deep learning
• object detection
• single shot multibox detector (SSD)