Animals in the Wild: Using Crowdsourcing to Enhance the Labelling of Camera Trap Images

Artificial intelligence and its subarea of computer vision became one of the top research areas in the past decade. Gathering information from all kinds of images is beneficial for various applications. In most of those applications, the images are either collected in a controlled (i.e. with known light conditions) environment or the object of interest is the dominant part of the image. Detecting small and only partly visible objects in the background within varying lighting and environmental conditions requires a well-labelled training and validation dataset. This work combines automatic labelling with a web app for label confirmation. It offloads the tiring manual labelling and checking process using a simple webpage and crowdsourcing to validate the image labels. Depending on the domain, even an inexperienced user can quickly check several hundreds of images within a couple of minutes. In this work, we present and evaluate the concept and our toolchain for this assisted labelling approach.