Towards Automatic Detection of Animals in Camera-Trap Images

In recent years the world's biodiversity is declining on an unprecedented scale. Many species are endangered and remaining populations need to be protected. To overcome this agitating issue, biologist started to use remote camera devices for wildlife monitoring and estimation of remaining population sizes. Unfortunately, the huge amount of data makes the necessary manual analysis extremely tedious and highly cost intensive. In this paper we re-train and apply two state-of-the-art deep-learning based object detectors to localize and classify Serengeti animals in camera-trap images. Furthermore, we thoroughly evaluate both algorithms on a self-established dataset and show that the combination of the results of both detectors can enhance overall mean average precision. In contrast to previous work our approach is not only capable of classifying the main species in images but can also detect them and therefore count the number of individuals which is in fact an important information for biologists, ecologists, and wildlife epidemiologists.