app.py

import logging
import queue
from typing import List, NamedTuple
import av
import cv2
import numpy as np
import streamlit as st
from streamlit_webrtc import WebRtcMode, webrtc_streamer
from sample_utils.turn import get_ice_servers
import mediapipe as mp
import os
import time

# Logger Setup
logger = logging.getLogger(__name__)

# Streamlit settings
st.set_page_config(page_title="Virtual Keyboard", page_icon="🏋️")
st.title("Interactive Virtual Keyboard")
st.subheader('''Turn on the webcam and use hand gestures to interact with the virtual keyboard.
Use 'a' and 'd' from the keyboard to change the background.''')

# Initialize MediaPipe Hand Detector
mp_hands = mp.solutions.hands
hands = mp_hands.Hands(static_image_mode=False, max_num_hands=2, min_detection_confidence=0.7)
mp_drawing = mp.solutions.drawing_utils

# Define virtual keyboard layout
keys = [["Q", "W", "E", "R", "T", "Y", "U", "I", "O", "P"],
        ["A", "S", "D", "F", "G", "H", "J", "K", "L", ";"],
        ["Z", "X", "C", "V", "B", "N", "M", ",", ".", "/"]]

class Detection(NamedTuple):
    label: str
    score: float
    box: np.ndarray

result_queue: "queue.Queue[List[Detection]]" = queue.Queue()

# Load background images
listImg = os.listdir('model/street') if os.path.exists('model/street') else []
if not listImg:
    st.error("Error: 'street' directory is missing or empty. Please add background images.")
    st.stop()
else:
    imgList = [cv2.imread(f'model/street/{imgPath}') for imgPath in listImg]
    imgList = [img for img in imgList if img is not None]

indexImg = 0
output_text = ""

if "output_text" not in st.session_state:
    st.session_state["output_text"] = ""

# Video Frame Callback
def video_frame_callback(frame: av.VideoFrame) -> av.VideoFrame:
    global indexImg, output_text

    img = frame.to_ndarray(format="bgr24")
    img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

    # Process the frame with MediaPipe
    result = hands.process(img_rgb)

    detections = []
    if result.multi_hand_landmarks:
        for hand_landmarks in result.multi_hand_landmarks:
            mp_drawing.draw_landmarks(
                img, hand_landmarks, mp_hands.HAND_CONNECTIONS,
                mp_drawing.DrawingSpec(color=(0, 255, 0), thickness=2, circle_radius=4),
                mp_drawing.DrawingSpec(color=(0, 0, 255), thickness=2)
            )
            # Extract bounding box for detection info
            x_min, y_min = 1.0, 1.0
            x_max, y_max = 0.0, 0.0
            for lm in hand_landmarks.landmark:
                x_min = min(x_min, lm.x)
                y_min = min(y_min, lm.y)
                x_max = max(x_max, lm.x)
                y_max = max(y_max, lm.y)
            
            h, w, _ = img.shape
            bbox = np.array([int(x_min * w), int(y_min * h), int((x_max - x_min) * w), int((y_max - y_min) * h)])
            detections.append(Detection(label="Hand", score=1.0, box=bbox))
            
        logger.info(f"Detected {len(detections)} hand(s).")
    else:
        logger.info("No hands detected.")

    result_queue.put(detections)
    st.session_state["output_text"] = output_text
    return av.VideoFrame.from_ndarray(img, format="bgr24")

# WebRTC Streamer
webrtc_streamer(
    key="virtual-keyboard",
    mode=WebRtcMode.SENDRECV,
    rtc_configuration={"iceServers": get_ice_servers(), "iceTransportPolicy": "relay"},
    media_stream_constraints={"video": True, "audio": False},
    video_frame_callback=video_frame_callback,
    async_processing=True,
)