inference.py

"""
Sample inference script for ASL Sign Language Detection
"""
import numpy as np
import cv2
import mediapipe as mp
from tensorflow.keras.models import load_model

# Load model and class names
model = load_model('asl_lstm_model.h5')
class_names = np.load('class_names.npy')

# Initialize MediaPipe
mp_hands = mp.solutions.hands
hands = mp_hands.Hands(static_image_mode=True, max_num_hands=1, min_detection_confidence=0.5)

def extract_landmarks(image_path):
    """Extract hand landmarks from image"""
    image = cv2.imread(image_path)
    if image is None:
        return None
    image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    results = hands.process(image_rgb)
    if results.multi_hand_landmarks:
        from google.protobuf.json_format import MessageToJson
        import json
        jsonObj = MessageToJson(results.multi_hand_landmarks[0])
        lmk = json.loads(jsonObj)['landmark']
        coords = []
        for point in lmk:
            coords.extend([point['x'], point['y']])
        return np.array(coords)
    return None

def distance_between(p1_idx, p2_idx, landmarks):
    """Calculate distance between two landmark points"""
    p1 = np.array([landmarks[p1_idx * 2], landmarks[p1_idx * 2 + 1]])
    p2 = np.array([landmarks[p2_idx * 2], landmarks[p2_idx * 2 + 1]])
    return np.sqrt(np.sum((p1 - p2) ** 2))

def landmark_to_dist_emb(landmarks):
    """Convert landmarks to distance features"""
    emb = np.array([
        distance_between(4, 8, landmarks), distance_between(4, 12, landmarks),
        distance_between(4, 16, landmarks), distance_between(4, 20, landmarks),
        distance_between(0, 4, landmarks), distance_between(0, 8, landmarks),
        distance_between(0, 12, landmarks), distance_between(0, 16, landmarks),
        distance_between(0, 20, landmarks), distance_between(8, 12, landmarks),
        distance_between(12, 16, landmarks), distance_between(1, 4, landmarks),
        distance_between(5, 8, landmarks), distance_between(9, 12, landmarks),
        distance_between(13, 16, landmarks), distance_between(17, 20, landmarks),
        distance_between(2, 8, landmarks), distance_between(2, 12, landmarks),
        distance_between(2, 16, landmarks), distance_between(2, 20, landmarks)
    ])
    return emb / np.linalg.norm(emb)

def predict_sign(image_path):
    """Predict sign language gesture from image"""
    landmarks = extract_landmarks(image_path)
    if landmarks is None:
        return None, None
    
    features = landmark_to_dist_emb(landmarks)
    # Repeat to create sequence of 50 timesteps
    sequence = np.repeat([features], 50, axis=0).reshape(1, 50, 20)
    
    prediction = model.predict(sequence, verbose=0)
    class_idx = np.argmax(prediction)
    confidence = prediction[0][class_idx]
    
    return class_names[class_idx], confidence

# Example usage
if __name__ == "__main__":
    result, confidence = predict_sign("path/to/your/image.jpg")
    if result:
        print(f"Predicted sign: {result} (confidence: {confidence:.2%})")
    else:
        print("No hand detected in image")