Inertial Sensing for Predictive Field of View Optimization in Smartphone Augmented Reality
Faris Abuhashish, Ismahafezi Ismail · 2025
Smartphone IMU data processing through Bidirectional LSTM networks results in Motion Trace as a new approach for predicting augmented reality field of view. The system functions with minimal power usage and delivers dependable performance mainly during dark conditions. The model yielded successful results in multiple dataset evaluations that resulted in short-term FOV predictions reaching mean squared error (MSE) levels of 0.84 mm as it improved its accuracy with time. The research establishes Motion Trace as an optimal tool for AR system optimization because it reduces FOV estimation time and minimizes energy consumption essential for mobile and resource-limited devices in real-time scenarios. for 28 seconds. The researchers present Motion Trace as a system which combines smartphone IMU information with a Bidirectional LSTM network to estimate field of view (FOV) in augmented reality (AR). The research approach generates datasets from simulated and actual information before validating a model architecture through cross-dataset evaluation testing. The results reported minimum MSE value at 0.84 mm during short-term predictions and proved beneficial for low power usage and effective performance under low-light environments. The model demonstrates consistent accuracy increases during its use period which indicates it will work properly in limited resource AR environments. The study demonstrates a real-time FOV prediction system that saves energy between data points in the conclusion before suggesting future research which integrates sensor data with improved neural designs to enhance system performance. The validation methodology uses multiple extensive measures to test method consistency across different datasets.