Towards Automatic Wild Animal Detection in Low Quality Camera-Trap Images Using Two-Channeled Perceiving Residual Pyramid Networks

Chunbiao Zhu, Thomas H. Li, Ge Li; The IEEE International Conference on Computer Vision (ICCV), 2017, pp. 2860-2864

Abstract


Monitoring animals in the wild without disturbing them is possible using camera trapping framework, which is a technique to study wildlife using automatically triggered cameras and produces great volumes of data. However, camera trapping collects images often result in low image quality and includes a lot of false positives (images without animals), which must be detection before the post-processing step. This paper presents a two-channeled perceiving residual pyramid networks(TPRPN) for camera-trap images objection. Our TPRPN model attends to generating high-resolution and high-quality results. In order to provide enough local information, we extract depth cue from the original images and use two-channeled perceiving model as input to training our networks. Finally, the proposed three-layer residual blocks learn to merge all the information and generate full size detection results. Besides, we construct a new high-quality dataset with the help of Wildlife Thailand's Community and eMammal Organization. Experimental results on our dataset demonstrate that our method is superior to the existing object detection methods.

Related Material


[pdf]
[bibtex]
@InProceedings{Zhu_2017_ICCV,
author = {Zhu, Chunbiao and Li, Thomas H. and Li, Ge},
title = {Towards Automatic Wild Animal Detection in Low Quality Camera-Trap Images Using Two-Channeled Perceiving Residual Pyramid Networks},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {Oct},
year = {2017}
}