Augmenting Humans in the Loop: Towards an Augmented Reality Object Labeling Application for Crowdsourcing Communities

Convolutional neural networks (CNNs) offer great potential for business applications because they enable real-time object recognition. However, their training requires structured data. Crowdsourcing constitutes a popular approach to obtain large databases of manually-labeled images. Yet, the process of labeling objects is a time-consuming and cost-intensive task. In this context, augmented reality provides promising solutions by allowing an end-to-end process of capturing objects, directly labeling them and immediately embedding the data in training processes. Consequently, this paper deals with the development of an object labeling application for crowdsourcing communities following the design science research paradigm. Based on seven issues and twelve corresponding meta-requirements, we developed an AR-based prototype and evaluated it in two evaluation cycles. The evaluation results reveal that the prototype facilitates the process of object detection, labeling and training of CNNs even for inexperienced participants. Thus, our prototype can help crowdsourcing communities to render labeling tasks more efficient.