Classification of User Behavior Patterns for Indoor Navigation Problem.

Area of Science:

• Computer Science
• Robotics
• Human-Computer Interaction

Background:

• Indoor navigation is challenging due to limitations of traditional Global Positioning System (GPS).
• Existing indoor positioning systems often require significant infrastructure.
• There is a need for cost-effective and scalable indoor navigation solutions.

Purpose of the Study:

• To propose a novel behavior-based localization approach for indoor navigation.
• To develop a system that estimates user position by matching observed behavior to a predefined route.
• To validate the effectiveness of real-time activity recognition for navigation.

Main Methods:

• Utilized smartphone sensors to collect velocity data.
• Developed a Long Short-Term Memory (LSTM) based model for user behavior classification (standing, walking, stairs, elevators).
• Classified activities within one-second time windows and full sequences for navigation estimation.

Main Results:

• Achieved 75% accuracy in classifying individual user activities within one-second intervals.
• Reached 98.6% accuracy for full-sequence activity classification using majority voting.
• Demonstrated the viability of behavior-based localization for indoor navigation.

Conclusions:

• Real-time activity recognition is a feasible foundation for indoor navigation systems.
• Behavior-based localization offers a cost-effective alternative to infrastructure-dependent systems.
• The proposed method is suitable for specific scenarios like guiding personnel in complex buildings.