Animal Wildlife Population Estimation Using Social Media Images Collections

We propose a method to estimate the number of animals of a given species using social media images collections. Traditional methods used to provide this estimate are expensive and time-consuming. Our solution aims at solving these issues by providing a species independent framework capable of dealing with social media biases. In fact, while biologists are usually in charge of the data collection process, when using social media as a source of data, we have to deal with biased datasets. A photographer may choose to share only its best pictures, hiding us information useful to estimate the number of animals. Previous works have shown that there is indeed a bias related to individual pictures. In our approach, we extend these researches by studying the bias at the level of images collections. We propose an approach based on two steps. The first step consists in the use of a regression model to estimate the number of animals photographed by a social media user, but not shared on the social media platform. The second step is the use of traditional wildlife estimator to predict the number of animals of a given species. This traditional model will be fed with data predicted by the regression model when applying it on several images collections retrieved from social media.