Deep Learning Pedestrian Dead Reckoning for 6DOF Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional mixed-reality systems face challenges in accurately tracking the trajectory of mobile computing devices when exteroceptive sensing capabilities are hindered, unreliable, or unavailable, due to issues like drift, high computational cost, and failure to account for user and device-specific characteristics.
Innovation Solution
The implementation of deep learning-based pedestrian dead reckoning systems that utilize inertial tracking data from accelerometers, gyroscopes, and compasses to estimate six degree of freedom (6DOF) positioning, fusing this data with predictive models trained on both inertial and exteroceptive sensor data to enhance accuracy and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional dead reckoning methods use only interoceptive sensor data to estimate trajectory, then the system can operate without exteroceptive sensors, but the accuracy deteriorates due to drift and failure to account for user-specific characteristics
Solution Approach 1:
The system performs preliminary training offline using both interoceptive and exteroceptive sensor data to build a predictive model that captures user-specific characteristics. This pre-computed model is then deployed for real-time trajectory estimation, allowing the system to achieve high accuracy without relying on exteroceptive sensors during operation.
Solution Approach 2:
The predictive model creates a virtual copy of the relationship between interoceptive sensor data and ground truth trajectory by training on paired sensor data. This copied relationship allows the system to estimate accurate trajectory from interoceptive data alone, compensating for the absence of exteroceptive sensors.
2Speed
If conventional dead reckoning processes continuously process sensor data to maintain positioning, then the positioning can be maintained in real-time, but the computational cost increases
Solution Approach 1:
The computationally intensive work of learning the relationship between sensor data and trajectory is performed in advance during an offline training phase. During real-time operation, the system only needs to inference from the pre-trained model, dramatically reducing computational energy consumption while maintaining real-time positioning speed.
3Measurement precision
If conventional systems use exteroceptive sensors for trajectory tracking, then the positioning accuracy is improved, but the system fails when exteroceptive sensing capabilities are hindered or unavailable
Solution Approach 1:
The predictive model serves as an intermediary that translates interoceptive sensor data into accurate trajectory estimates by learning the complex relationship between body movement and device trajectory. This intermediary enables the system to achieve exteroceptive-level accuracy using only interoceptive sensors, ensuring reliability in environments where exteroceptive sensing is denied or unavailable.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems are provided for estimating 6DOF positioning of a computing device while in a pedestrian dead reckoning mode. The systems obtain a set of inertial tracking data from the set of one or more inertial tracking components while the system is in a pedestrian dead reckoning mode. Then, the systems obtain an estimated 3DOF velocity of the system based inertial tracking data, using a predictive model trained on a set of observed exteroceptive sensor data and observed inertial tracking data. The systems also obtain estimated 6DOF positioning of the systems based on the estimated 3DOF velocity.