Real-Time Radio Environment Mapping for Multi-Class Occupancy Detection
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Accurate indoor occupancy detection is essential for smart environments, yet existing methods often overlook the combined benefits of fine-grained spatial mapping and robust deep learning classification. This work presents a hybrid framework that integrates Radio Environment Mapping (REM) with convolutional neural networks (CNNs) for multistate occupancy detection using WiFi received signal strength indicator (RSSI) data. A TurtleBot3 mobile robot is employed to collect synchronized RSSI and odometry measurements in an indoor environment. Three occupancy scenarios are considered: no-occupancy, singleoccupancy, and double-occupancy. For each scenario, 500 REMs are generated to capture spatial signal variations. Extra Trees Regression constructs high-resolution REMs, which are subsequently classified by a CNN model. Experimental evaluation shows consistently high classification accuracy across all scenarios, effectively distinguishing between occupancy states while preserving detailed spatial features. The method demonstrates robustness to environmental changes and variability in human presence. The suggested method combines deep learning classification with regression-based mapping to achieve both classification reliability and spatial accuracy.
Fields of Study & Subjects
Indoor and Outdoor Localization Technologies
Robotics and Sensor-Based Localization
Cognitive Radio Networks and Spectrum Sensing
Publication details
- Journal / source
- International Conference on Smart City and Information System (ICSCIS)
- Pages
- 113–116
- Publication date
- 30 January 2026
- Language
- EN