Gesture Embedding and Classification

To recognize gestures:

def lm_postprocess_with_gesture_classification(current_region, inference, previous_frame_regions, 
                                              gesture_model, classifier_model, 
                                              alpha=0.6, iou_threshold=0.4, crop_width=224, crop_height=224):
    """Process landmark model output with gesture embedding and classification"""
    # First, process landmarks as before
    lm_postprocess(current_region, inference, previous_frame_regions, alpha, iou_threshold, crop_width, crop_height)
    # Then extract gesture embedding and classify
    if hasattr(current_region, 'landmarks') and current_region.lm_score > 0.5:
        embedding = gesture_model([
            np.array(current_region.landmarks, dtype=np.float32).reshape(1, 21, 3),
            np.array([[current_region.handedness]], dtype=np.float32),
            None # world_landmarks if available
        ])
        gesture_name, confidence = classify_gesture(embedding, classifier_model, gesture_labels)
        current_region.gesture_name = gesture_name
        current_region.gesture_confidence = confidence
    return current_region