ABSTRACT
Digital camera capacity and quality have advanced greatly in recent decades, providing the growing field of camera trap research with massive amounts of photographic data. The advances in camera technology have not been matched by advances in the use of modern methods of data management and analysis. The analysis and management of data is a bottleneck in the animal population monitoring process. In this chapter, we argue that computer vision technologies can be adapted for animal identification and used in combination with models for estimating animal density to automatically estimate animal density using camera trap technology. Computer vision technologies can be combined with camera traps for other purposes as well. We give an overview of the relevant computer vision technologies, focusing on convolutional neural networks. We discuss how they can be adapted to work with camera traps, and how they can cope with issues such as transient physical markings and the underrepresentation of endangered and elusive species. We include experiments with state-of-the-art convolutional networks that assess their effectiveness in scenarios similar to the camera trap scenario. We give an overview of several population density models that can be used in conjunction with computer vision technologies. Several considerations are discussed for the use of these techniques in ecological studies.
