Understanding the state of the Art in Animal detection and classification using computer vision technologies

Abstract

This work presents the results of a survey through the analysis of studies published between January 2017 and May 2021, aiming to compose a broader view of the state of the art in the field of animal detection and classification using computer vision technologies in urban environments, and also the majors researches gaps available to address. We conducted an automatic search through two digital knowledge bases identifying 146 studies in the subject, among them 20 were selected for our analysis and data extraction. Further, the 20 studies were classified into 6 categories: (i) studies using SVM, (ii) studies using HOG, (iii) studies using SIFT, (iv) studies using PCA, (v) studies using CNN, and (vi) DFDL. As a result, it can be noted that the use of CNN is predominant concerning other approaches and that there are also combinations to improve the accuracy of classification models. In conclusion, it is possible to observe that the state-of-the-art approaches have been used in different situations, however, in the context of animal detection and classification in intelligent urban environments, there is still a lack of specific architectures to improve results.