Zhang, Jing; Haas, Carl; Hanna, Sean; (2023) Comparative Study of Automatic Multi-class Object Detection Algorithms with Transfer Learning based on a Dataset from Construction Sites. In: EG-ICE 2023 Conference Papers. University College London (UCL): London, UK.

Preview

PDF EG-ICE_2023_paper_179.pdf - Published Version Download (909kB) | Preview

Abstract

Advanced technologies, such as Computer Vision, are helping to transform the traditional Architecture, Engineering, and Construction (AEC) industry. Although this cutting-edge artificial intelligence technology has begun to enter more construction sites, automated methods of data collection for construction site management have room for further development. Therefore, this paper compares two vision-based automatic detection algorithms for multiple categories of moving target objects and their improvement based on the construction site dataset (MOCS). The study methodology used includes two stages: (1) Basic Model: The first stage compares the detection performance of mainstream target detection algorithms (Faster R-CNN and YOLOv7) on the same database, and (2) Transfer Learning: The second phase added the pre-training network by transfer learning strategy, anticipating that it may improve detection performance in the initial algorithms. This digital route relies on an HD camera installed on UAVs to achieve monitor automation in the construction process to support supervision engineers.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item