Deep IoU Network for Dense Rebar Object Detection

Typically, dense rebar detection scenes comprise cross-sections of hundreds or even thousands of rebars. We demonstrate that most commonly used object detectors still have trouble detecting objects accurately in such settings. We present a novel deep-learning-based approach for tackling this problem...

Full description

Saved in:
Bibliographic Details
Published in2022 IEEE International Conference on e-Business Engineering (ICEBE) pp. 45 - 50
Main Authors Zhong, Xiaojing, Hu, Hao, Li, Li, Cen, Junhua, Wu, Qingyao
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.10.2022
Subjects
Corporate acquisitions
Dense Rebar Detection
Detectors
Image resolution
Merging Duplicate Detections
Object detection
Online AccessGet full text

Cover

More Information
Summary:Typically, dense rebar detection scenes comprise cross-sections of hundreds or even thousands of rebars. We demonstrate that most commonly used object detectors still have trouble detecting objects accurately in such settings. We present a novel deep-learning-based approach for tackling this problem, which is combined with a useful soft-Iou layer to predict the Iou of a detected bounding box and its ground truth and an efficient EM - Merger unit to resolve a single detection per object, enabling the accurate detection of the bounding box of tiny objects such as rebars. Experiments show that the proposed method can achieve excellent results in our collected rebar images while our network is trained on the RebarDSC dataset.
DOI:10.1109/ICEBE55470.2022.00018