Invariant Features-Based Fuzzy Inference System for Animal Detection and Recognition Using Thermal Images
- PDF / 3,291,275 Bytes
- 12 Pages / 595.276 x 790.866 pts Page_size
- 97 Downloads / 191 Views
Invariant Features-Based Fuzzy Inference System for Animal Detection and Recognition Using Thermal Images S. Divya Meena1
•
L. Agilandeeswari1
Received: 1 August 2019 / Revised: 13 May 2020 / Accepted: 15 June 2020 Taiwan Fuzzy Systems Association 2020
Abstract Human–Animal Conflict (HAC) is one of the primary threats to the continued survival of animal species and it has also impacted the lives of humans drastically. In this paper, we propose an efficient animal detection and recognition system with invariant features and fuzzy logic using thermal images. The proposed system exploits various features like Zernike, shape, texture and skeleton path. Cumulatively, these features are invariant to rotation, scaling, translation, illumination, and partly posture. The proposed model is robust to several challenging image conditions like low contrast/illumination, haze/blur, occlusion, camouflage, background clutter, and poses variation. The model is tested on our thermal animal dataset that has 1862 images and 12 different animal species. Experimental results validate the significance of thermal images for animal-based applications. Besides, the proposed fuzzy system has achieved an average accuracy of 97% which is equivalent to the accuracy produced by domain experts in identifying the animals from our thermal dataset. Keywords Fuzzy logic Thermal images Skeleton and shape features Zernike moments
& L. Agilandeeswari [email protected] 1
School of Information Technology and Engineering, Vellore Institute of Technology, Vellore, India
1 Introduction Wild animal monitoring systems are gaining importance due to the number of Human–Animal Conflict (HAC) that has been occurring over the decades. With visible images, it is tedious to detect animals during night time due to the engulfed darkness. Besides, animals could self-mask with their flexible structure and their cluttered background adds to the complexity. However, thermal imaging cameras are one of the perfect tools for night vision applications. They work on the principle of heat energy and so they can detect hot-blooded organisms like human, animals very easily, besides ignoring camouflage. Although camera traps are one of the best tools for capturing the animals, they do not always capture perfect images and they do pose several challenges like, noise, low contrast/illumination, haze/blur, occlusion, camouflage, background clutter, and pose variations leading to poor interpretation. As a counter measure, we extract the invariant features which are not deterred by these challenges. Invariant features are a special type of features which can identify the animals precisely even when they appear slightly different from their original form, thus making them robust to any challenging image conditions. The number of animal detection models with thermal images is relatively low and the primary reason is the lack of publicly available thermal animal dataset. Yet, thermal images have been used in few animal studies like studying the population of Brazilian free-t
Data Loading...
