Download PDFOpen PDF in browser

Reliable Night Bear and Boar Detection Based on Training with Pseudo Infrared Images

10 pagesPublished: September 20, 2022

Abstract

In recent years, accidents and damages caused by wild animals have been serious prob- lems. It has become important to detect wild animals accurately at an early stage. A sufficient number of training infrared images is required to detect wild animals taking various postures at night time using deep learning techniques. In this study, we propose a method to increase appropriate training samples for night wild animal detection using annotated daytime images. We employ a model based on Cycle Generative Adversarial Network (CycleGAN) to be able to generate pseudo infrared images from daytime images. In our experiments, we apply the proposed method to bear and boar detection. The exper- imental results show that the proposed method achieves significant improvements in bear detection accuracy taking various postures.

Keyphrases: generative adversarial network, infrared image, object detection, wild animal

In: Tokuro Matsuo (editor). Proceedings of 11th International Congress on Advanced Applied Informatics, vol 81, pages 101-110.

BibTeX entry
@inproceedings{IIAIAAI2021-Winter:Reliable_Night_Bear_Boar,
  author    = {Keigo Fusaka and Yoichi Tomioka and Hiroshi Saito and Yukihide Kohira},
  title     = {Reliable Night Bear and Boar Detection Based on Training with Pseudo Infrared Images},
  booktitle = {Proceedings of 11th International Congress on Advanced Applied Informatics},
  editor    = {Tokuro Matsuo},
  series    = {EPiC Series in Computing},
  volume    = {81},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2398-7340},
  url       = {/publications/paper/Spht},
  doi       = {10.29007/xghf},
  pages     = {101-110},
  year      = {2022}}
Download PDFOpen PDF in browser