mmWave Radar Gesture Recognition Non-Gesture Rejection

The patent segments the activity detection process into multiple stages: initial activity detection module (ADM), post-ADM gating module, and gesture-specific gating module. Each stage filters out different types of non-gestures, progressively reducing false alarms while maintaining gesture detection coverage. This multi-stage segmentation allows the system to handle the infinite variety of non-gesture activities without compromising gesture recognition.

The patent applies preliminary gating actions before final gesture classification. The post-ADM gating module performs preliminary filtering by checking basic conditions (hand presence, motion magnitude, motion type) before activities are classified as gestures. This preliminary action eliminates obvious non-gestures early in the process, reducing the burden on subsequent classification stages and lowering false alarm rates.

The complex rejection process is segmented into distinct modular components: post-ADM gating module with basic condition checks, and gesture-specific gating module with deep learning-based classification. Each module has a specific function and operates independently, making the overall complex system manageable and maintainable while achieving high accuracy through cumulative filtering.

The patent applies multiple layers of gating conditions, some of which may be redundant or overly strict for individual cases, but collectively provide robust false alarm rejection. The gesture-specific gating module uses deep learning models that evaluate multiple features and conditions, applying partial actions (individual gating checks) that collectively achieve high accuracy despite individual complexity.