Recently, automated observation systems for animals using artificial intelligence have been proposed. In the wild, animals are difficult to detect and track automatically because of lamination and occlusions. Our study proposes a new approach to automatically detect and track wild Japanese macaques (Macaca fuscata) using deep learning and a particle filter algorithm. Macaque likelihood is derived through deep learning and used as an observation model in a particle filter to predict the macaques’ position and size in an image. By using deep learning as an observation model, it is possible to simplify the observation model and improve the accuracy of the classifier. We investigated whether the algorithm could find body regions of macaques in video recordings of free‐ranging groups at Katsuyama, Japan to evaluate our model. Experimental results showed that our method with deep learning as an observation model had higher tracking accuracy than a method that uses a support vector machine. More generally, our study will help researchers to develop automatic observation systems for animals in the wild.

Automatic individual recognition techniques can support data collection in the field of ethology. Recent studies have contributed to development of automatic individual recognition techniques using machine learning and deep learning. However, varied conditions in the wild, such as the presence of occlusions and head rotations of individuals, can lower the accuracy of automatic recognition techniques. Thus, there is requirement for improvement in the accuracy and robustness of these techniques. In this study, we have used previously observed information updated with given current observation by Bayesian inference to improve the automatic individual recognition of free‐ranging Japanese macaques (Macaca fuscata) at Katsuyama, Japan. We collected static images and video footage of 51 adult individuals. Using the static images, we created eight individual recognition systems (classifiers), using GoogLeNet and ResNet‐18 as convolutional neural network models. Additionally, sequential data of the faces of the macaques were obtained from 86 video recordings of the 51 individuals to evaluate the classifiers. We were able to successfully recognize 90% or more individuals with each classifier through the combination of the sequential Bayesian filter and the classifier. Eighty‐five percent or more of the individuals had posterior probabilities of 90% or above when conducting recognition tests using the sequential Bayesian filter with 10 images. The best classifier recognized 98% of individuals using 10 images and all individuals using 50 images. Recognition was also successful when the sequential Bayesian filter was applied to cases in which the recognition rate was <50% with test data when the filter was not applied. Based on the above results, we posit that the accuracy of individual recognition systems can be improved using a sequential Bayesian filter that considers past information for individuals.