Scholarlink

In the construction industry, floor plans are widely used for sharing a building layout among various stakeholders such as clients, designers, builders and engineers. Automatically floor plan techniques can improve work accuracy and efficiency. For automatic analysis of a floorplan image, detection of various objects is an important step. However, few techniques have been proposed in the literature for automatic detection of objects in complex floor plan documents. A convolutional neural network (CNN) based architecture, referred to as the ObjNet, is proposed in this paper for detecting different objects/symbols, such as fridges, stoves and bathtub, in a complex architectural floor plan. The proposed ObjNet is based on YOLOv8, and has three main stages: Backbone, Neck and Detection. In the proposed architecture, a convolutional attention mechanism is used to improve feature learning in the Neck module. Experimental results obtained using a complex architectural floor plan dataset show that the ObjNet can provide a superior performance compared to the existing techniques with a mean average precision (mAP) of 88.9%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Object Detection on Complex Architectural Floor Plans with Efficient Attention Mechanisms

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zhongguo Xu,
  • Naresh Jha,
  • Syed Mehadi,
  • Santi P. Maity,
  • Mrinal Mandal

摘要

In the construction industry, floor plans are widely used for sharing a building layout among various stakeholders such as clients, designers, builders and engineers. Automatically floor plan techniques can improve work accuracy and efficiency. For automatic analysis of a floorplan image, detection of various objects is an important step. However, few techniques have been proposed in the literature for automatic detection of objects in complex floor plan documents. A convolutional neural network (CNN) based architecture, referred to as the ObjNet, is proposed in this paper for detecting different objects/symbols, such as fridges, stoves and bathtub, in a complex architectural floor plan. The proposed ObjNet is based on YOLOv8, and has three main stages: Backbone, Neck and Detection. In the proposed architecture, a convolutional attention mechanism is used to improve feature learning in the Neck module. Experimental results obtained using a complex architectural floor plan dataset show that the ObjNet can provide a superior performance compared to the existing techniques with a mean average precision (mAP) of 88.9%.