A Big Data-Driven Unsupervised Indoor and Outdoor Detection Approach Using Deep Contrastive Learning

This With the rapid development of mobile internet, people's lives have become more convenient, such as taking taxis with phones and so on. In the process of people enjoying convenience, user needs have evolved from basic positioning ability to perception ability. IOD (Indoor and outdoor detection), as an important perception ability, has been studied by many scholars in academia and industry. However, there are still many problems such as difficulty in obtaining supervision information and poor model robustness. In this work, we propose an unsupervised label generation approach that obtains IOD labels by mining massive localization logs, effectively solving the problem of expensive manual labelling for IOD labels. In addition, we propose a deep contrastive learning based indoor and outdoor detection network, called CLIODNet, which fully enhances the robustness and correctness of model detection. Finally, we conduct extensive experiments in four real environments. The experimental results show that our model has greatly improved compared to the baseline.