A Deep Learning based Scene Recognition Algorithm for Indoor Localization

In this paper, we make use of deep convolutional neural networks to fine tune ImageNet, as an object detection dataset to train a scene dataset that can recognize indoor environments within universities. To utilize the application of scene recognition in indoor environments, a high accuracy is needed, and the proposed scene recognition algorithm is tested with different models trained in Places365 to compare what works best for a new dataset specialized in indoor space. The proposed algorithm resulted in 96.43% accuracy in recognizing different indoor scenes, and it was able to achieve an average error distance of 1.64 meters in indoor localization.