Paper Info
Title
Deep generative adversarial network to enhance image quality for fast object detection in construction sites
Abstract
ABSTRACTVisual recognition of the content and actions that take place in a construction site is important in many applications such as data-driven simulation, autonomous systems, and intelligent machinery. Construction project, however, are dynamic and complex, and often take place in harsh environments. This may hinder the ability to collect good quality, well-lit, and occlusion-free imagery, which in turn, can lower the performance of computer vision models for fast and reliable object detection. In this paper, we propose and validate a deep convolutional neural network (CNN)-based generative adversarial network (GAN) trained and tested on construction site photos from two in-house datasets to increase image resolution by generating missing pixel information. Results show that using GAN-enhanced images can improve the average precision of pre-trained models for detecting objects such as building, equipment, worker, hard hat, and safety vest by up to 32% while maintaining the overall processing time for real-time object detection.
Year
DOI
Venue
2020
10.5555/3466184.3466463
Winter Simulation Conference
DocType
Citations 
PageRank 
Conference
0
0.34
References 
Authors
0
2
Name
Order
Citations
PageRank
Nipun D. Nath100.68
Amir H. Behzadan212217.55