학술논문
Small Object Detection on Road by Embedding Focal-Area Loss
Document Type
Chapter
Author
Goos, Gerhard, Founding Editor; Hartmanis, Juris, Founding Editor; Bertino, Elisa, Editorial Board Member; Gao, Wen, Editorial Board Member; Steffen, Bernhard, Editorial Board Member; Woeginger, Gerhard, Editorial Board Member; Yung, Moti, Editorial Board Member; Zhao, Yao, Editor; Barnes, Nick, Editor; Chen, Baoquan, Editor; Westermann, Rüdiger, Editor; Kong, Xiangwei, Editor; Lin, Chunyu, Editor; Wang, Zijie; Fang, Jianwu; Dou, Jian; Xue, Jianru
Source
Image and Graphics : 10th International Conference, ICIG 2019, Beijing, China, August 23–25, 2019, Proceedings, Part I. 11/28/2019. 11901:657-665
Subject
Language
English
ISSN
0302-9743
1611-3349
1611-3349
Abstract
In recent years, with the continuous popularity of deep learning, the research on artificial intelligence has boosted the progress of many new applications, such as the autonomous driving. At present, the detection methods of vehicles, pedestrians and other objects in the self-driving technology have been investigated numerously, but there is no good solution for the detection of small objects such as stones on road. However, small targets on road seriously affects the stability of automated vehicle system. Therefore, it is important to carry out the detection of small targets on road.
This paper designs a focal-area loss function which is learned by focusing the area change of small targets. The contribution of small object is weighted more in learning. We embed this focal-area loss into a newly proposed Scale Normalization for Image Pyramids (SNIP). Exhaustive experiments on Lost And Found (LAF) dataset show that our method can significantly boost the performance of state-of-the-art.
This paper designs a focal-area loss function which is learned by focusing the area change of small targets. The contribution of small object is weighted more in learning. We embed this focal-area loss into a newly proposed Scale Normalization for Image Pyramids (SNIP). Exhaustive experiments on Lost And Found (LAF) dataset show that our method can significantly boost the performance of state-of-the-art.