Smartphone region-wise image indoor localization using deep learning for indoor tourist attraction

Area of Science:

• Computer Science
• Artificial Intelligence
• Computer Vision

Background:

• Smart indoor tourist attractions require precise indoor localization.
• Global Positioning Systems (GPS) are ineffective indoors due to signal obstruction.
• Existing indoor localization solutions often involve significant infrastructure investment.

Purpose of the Study:

• To propose a deep learning-based method for indoor localization using smartphone images.
• To reduce the cost and time for developing smart indoor tourist attractions.
• To offer an infrastructure-free localization solution for museums and aquariums.

Main Methods:

• Utilized deep learning algorithms for image classification to determine location.
• Collected a dataset of 3654 images from ten smartphones at the Pantanal Biopark.
• Tested seven state-of-the-art neural networks, including transformer-based models.

Main Results:

• Achieved an average precision of approximately 90%.
• Obtained an average recall and F-score of approximately 89%.
• Demonstrated the suitability of the proposed method in a real-world scenario.

Conclusions:

• Deep learning with smartphone images is a viable and cost-effective solution for indoor localization.
• The proposed method eliminates the need for specialized indoor localization hardware.
• This approach can significantly accelerate the transformation of tourist attractions into smart venues.