Crowd-sourced AI based Indoor Localization using Support Vector Regression and Pedestrian Dead Reckoning

Abstract

Aims and Background: Artificial intelligence (AI) is expanding in the market daily to assist humans in a variety of ways. However, as these models are expensive, there is still a gap in the availability of AI products to the common public with high component dependency. Objective and Methodology: To address the issue of additional component dependency on AI products, we propose a model that can use available Smartphone resources to perceive real-world huddles and assist ordinary people with their daily needs. The proposed AI model is to predict the user’s indoor position (Node) at the computer science and engineering block of CMR Institute of Technology (CMRIT) by using Smartphone sensors and wireless signals. We used SVR to predict the regular walk steps needed between two Nodes and Pedestrian Dead Reckoning (PDR) to predict the walk steps needed while the signal was lost in the indoor environment. Results: The Support vector regression (SVR) models make the locations to be available within the specified building boundaries for proper guidance. The PDR approach supports the user while signal loss between two Received Signal Strength Indicators (RSSI). The Pedestrian dead reckoning - Support Vector Regression (PD-SVR) results are showing 98% accuracy in NODE predictions with routing tables. The indoor positioning is 100% accurate with dynamic crowd-sourcing Node preparation. Conclusion: The results are compared with other indoor navigation models K-nearest neighbor (KNN) and DF-SVM are given 95% accurate NODE estimation with minimal need for network components.