Augmented Reality Object Recognition with Dynamic AI Offloading

The patent divides the object recognition system into two segments: a lightweight model running on the AR device for basic functionality, and a heavy model running on an external electronic device for enhanced capabilities. This segmentation allows the AR device to maintain small size while still accessing high-end processing power when needed through cloud connectivity.

The patent introduces an external electronic device as an intermediary between the AR device and high-end neural processing. The external device acts as a mediator that provides advanced computing power without being physically integrated into the AR device, thus maintaining the AR device's compact form factor while enabling access to enhanced functionality.

The patent implements a dynamic model selection mechanism that adjusts between lightweight and heavy AI models based on real-time conditions such as network availability, device performance, and recognition requirements. This dynamic approach allows the system to optimize processing speed by selecting appropriate models without permanently increasing device complexity.

The patent changes the parameter of model weight/size dynamically. The system selects between different model configurations (lightweight vs. heavy) based on operational conditions, effectively changing the computational parameters to match current needs without requiring the device to permanently accommodate the most complex model.

The patent segments the AI model portfolio into lightweight models for basic recognition tasks and heavy models for high-accuracy requirements. The system automatically selects the appropriate segment based on the specific recognition task, ensuring that accuracy is maintained when needed while keeping the device size manageable through selective model deployment.

The patent creates a universal object recognition system that can handle both simple and complex recognition tasks by incorporating multiple model types. The system universally supports both lightweight and heavy models, allowing it to adapt to different accuracy requirements without requiring separate dedicated systems for each task type.

The patent applies partial action by processing only the necessary portions of images at full resolution. The system uses gaze tracking to identify regions of interest and processes those areas with high accuracy while using lighter processing for other areas, thereby maintaining recognition accuracy where needed while reducing overall energy consumption.

The patent implements local quality by applying different processing intensities to different regions of the image. High-quality full-resolution processing is applied locally to gaze-focused regions where accuracy is critical, while other regions receive reduced processing, optimizing the balance between recognition accuracy and energy consumption.