Neural Network Geo-Location Estimation With Low-Power Pose Sensors

The system segments the geo-location determination process into two parts: GPS provides initial absolute position, while low-power sensors (accelerometer, gyroscope, magnetometer) handle continuous position tracking. This segmentation allows the high-accuracy GPS to be used sparingly for initialization, while power-hungry operations are replaced by low-power sensor-based inertial navigation.

A neural network acts as an intermediary that processes sensor data from multiple sources (accelerometer, gyroscope, magnetometer, barometer) and fuses this information to estimate current position. The neural network translates raw sensor measurements into accurate geo-location estimates without requiring continuous GPS reception, thereby reducing power consumption while maintaining accuracy.

Instead of continuous high-frequency GPS polling, the system uses periodic GPS updates combined with continuous low-power sensor monitoring. The neural network processes sensor data continuously to track position changes, while GPS provides periodic absolute position corrections. This periodic action pattern maintains accuracy while dramatically reducing the power consumption associated with frequent GPS satellite communications.

The system performs preliminary position determination using low-power sensors before triggering a GPS update. The neural network analyzes sensor data to detect when position changes exceed a threshold, and only then activates GPS for absolute position verification. This preliminary filtering action prevents unnecessary GPS polling, reducing power consumption while maintaining update accuracy when actually needed.