US20250362741
2025-11-27
Physics
G06F3/012
The disclosure outlines a method for enhancing user localization within a metaverse environment. It involves detecting movements through wearable headgear equipped with an inertial sensor and camera, generating corresponding sensor and visual data. This data is mapped to a virtual world to accurately localize the user. Machine Learning (ML) models are utilized to process and refine this data, distinguishing between stable and dynamic key points to improve localization accuracy.
Traditional systems face challenges in user localization due to unreliable visual tracking, especially in environments with poor or repetitive textures, motion blur, or dynamic objects. These issues lead to unrealistic user movements within the metaverse. Previous solutions often fail to integrate user dynamism effectively, relying heavily on visual data without adequately addressing sensor biases or mapping distortions.
The method integrates visual and inertial data using ML models to enhance localization. Key points are extracted from visual data, with stable points retained and dynamic points removed using AI filters. The system prioritizes sensor data over low-weight visual impacts, ensuring more reliable user positioning within the virtual world. This approach also involves preprocessing visual data by tracking features across frames.
The system comprises memory and processors that execute instructions for processing visual and sensor data through ML models. It determines the quality of key points by computing weight parameters, integrates data from both sources, and maps it using pre-learned weights. The processors also handle loop closure and pose optimization, further refining localization accuracy.
The proposed method is supported by a non-transitory computer-readable storage medium containing a program executable by processors. These processors may include general-purpose CPUs, GPUs, VPUs, or AI-dedicated NPUs. The AI models are trained using predefined rules stored in memory, either directly on the device or via a separate server/system, ensuring robust user localization in the metaverse.