Classify Gesture Embedding

This function takes the gesture embedding and classifies it into one of the predefined static gestures.

Key Implementation

Function: classify_gesture in hand_landmark.py

def classify_gesture(gesture_embedding, gesture_classifier_model, gesture_labels, confidence_threshold=0.5):
    """Classify gesture from embedding"""
    try:
        # Run gesture classification
        classification_results = gesture_classifier_model([gesture_embedding])
        # Get raw probabilities for each gesture class
        raw_probabilities = list(classification_results.values())[0][0]  # Remove batch dimension
        # Apply sigmoid activation
        final_probabilities = [1.0 / (1.0 + np.exp(-score)) for score in raw_probabilities]
        # Find the gesture with highest probability from sigmoid-activated scores
        predicted_class = np.argmax(final_probabilities)
        confidence = final_probabilities[predicted_class]
        # Only return gesture if confidence is above threshold
        if confidence > confidence_threshold:
            return gesture_labels[predicted_class], confidence
        else:
            return "None", 0.0
    except Exception as e:
        print(f"Error in gesture classification: {e}")
        return "None", 0.0