Building Classification via Elevator and Escalator Mobility Patterns

The patent introduces map data and geographic information as an intermediary layer between raw sensor data and building classification results. The system collects sensor data from multiple mobile devices, processes it through machine learning models, and integrates the results with existing map data to determine transport modes and classify buildings. This intermediary approach simplifies the overall system complexity by leveraging existing geographic information infrastructure.

The system performs multiple functions using the same sensor data collection infrastructure: it determines transport modes (elevators, escalators, stairs), classifies building types, and analyzes mobility patterns. This multi-functional approach reduces implementation complexity by consolidating data collection and processing mechanisms rather than building separate systems for each function.

The patent employs pre-trained machine learning models that have been trained offline on large datasets. The models are trained in advance to recognize patterns associated with different transport modes and building types, so that during actual operation, the system only needs to input sensor data and receive rapid predictions. This preliminary training action separates the computationally intensive training phase from the real-time detection phase, reducing processing time during deployment.

The system processes sensor data from multiple mobile devices simultaneously and aggregates results to determine transport modes and building types. By collecting data from multiple sources rather than relying on a single device or limited data samples, the system achieves higher detection precision while distributing the computational load across multiple data contributions rather than requiring intensive processing of a single comprehensive dataset.

The patent extracts only the essential features and patterns from the collected sensor data that are relevant to determining transport modes and building types. Rather than processing and storing all raw sensor data, the system identifies and extracts key mobility patterns, movement characteristics, and usage behaviors that are sufficient for reliable building classification. This extraction approach maintains determination reliability while significantly reducing the volume of data that needs to be processed and stored.

The system merges and aggregates sensor data from multiple mobile devices to determine building types and transport modes. By combining data from multiple sources, the system achieves more reliable determinations through data validation and consensus, while the aggregation process itself reduces redundancy and eliminates duplicate information that would otherwise increase data volume. The merged data provides more reliable results than individual device data while being more compact than the sum of all individual datasets.