FP-RSBP BISINDO Sign Language Translator

Group Member
General
Training Model
Training Result
- Confussion Matrix
- Result Curve
Create Flask Web Application
Integration in HTML, CSS, and Javascript

Group Member

NRP	Name	Github Account
5025211015	Muhammad Daffa Ashdaqfillah	daf2a
5025211232	Ulima Kultsum	ulimakrh
5025211252	Nur Azizah	dancingaroundthelies

Description

This project is a final project for "Rekayasa Sistem Berbasis Pengetahuan (RSBP)". This project is a web-based application that can translate BISINDO sign language to Indonesian language. This application is made to help people who are not familiar with BISINDO sign language to communicate with people who are deaf.

Appoarch Method

This project employs the Object Detection method to identify hand gestures, utilizing a customized YOLO v8 dataset model. This model is designed to detect various hand gestures and provide the bounding box coordinates delineating these gestures. Once the bounding box is identified, the corresponding region of interest (ROI) containing the hand gesture is cropped. Subsequently, this cropped image undergoes further processing through a Convolutional Neural Network (CNN) model for precise classification of the detected hand gesture. The CNN model's output yields the specific label corresponding to the gesture identified. To enhance user experience, these labels are translated into Indonesian using a Rule-Based method before being displayed on the screen interface.

In terms of visualization, the project harnesses MediaPipe, a tool enabling the drawing of hand gesture bounding boxes and their associated labels directly on the screen. This integration ensures users can visually comprehend the detected gestures in real-time with graphical representations.

Moreover, the project is built as a web-based application using Flask, allowing seamless accessibility and interaction through a web interface. This framework facilitates the deployment of the hand gesture detection system within a browser, ensuring ease of use across various devices without necessitating complex setups or installations.

Visual Demo

Dashboard

Stream Youtube Detection

Camera Detection

Step by Step Method

Train the model using YOLOv8 with custom dataset
Create a CNN model to classify the cropped image
Create a Flask app to run the model
Create a web interface using HTML and CSS
Create feature interaction using Javascript

Features

Detect hand gesture from Camera
Detect hand gesture from Youtube video
Switch on/off the detection
Switch on/off the landmark
Flip the video
Modify Confidence Threshold
The result accumulates the detected words over 10 sequential frames

How to Run

Clone this repository
Install all the requirements
Run the app.py file
Open the localhost link in your browser

Training Model

Check GPU

!nvidia-smi

Import OS

import os
HOME = os.getcwd()
print(HOME)

Yolo Instalation

%pip install ultralytics==8.0.20

from IPython import display
display.clear_output()

import ultralytics
ultralytics.checks()

from ultralytics import YOLO

from IPython.display import display, Image

Import Dataset Form Roboflow

%mkdir {HOME}/datasets
%cd {HOME}/datasets

%pip install roboflow --quiet

from roboflow import Roboflow

from roboflow import Roboflow
rf = Roboflow(api_key="moOAxzoPZOtIzhyyco0r")
project = rf.workspace("bisindo-qndjb").project("bisindov2")
dataset = project.version(1).download("yolov8")

Training

%cd {HOME}

yolo task=detect mode=train model=yolov8s.pt data={dataset.location}/data.yaml epochs=25 imgsz=800 plots=True

%ls {HOME}/runs/detect/train/

%cd {HOME}
Image(filename=f'{HOME}/runs/detect/train/confusion_matrix.png', width=600)

%cd {HOME}
Image(filename=f'{HOME}/runs/detect/train/results.png', width=600)

Validate Model

%cd {HOME}

!yolo task=detect mode=val model={HOME}/runs/detect/train/weights/best.pt data={dataset.location}/data.yaml

Inference Model

%cd {HOME}
!yolo task=detect mode=predict model={HOME}/runs/detect/train/weights/best.pt conf=0.25 source={dataset.location}/test/images save=True

Run Model on CAMERA

%cd {HOME}
!yolo task=detect mode=predict model=D:/Kuliah/\(2023\)S5-RSBP/FP/runs/detect/train/weights/best.pt source=0 show=true

Training Result

Confussion Matrix

Result Curve

Create Flask Web Application

Install Requirements Dependencies

To install the Python requirements from the requirements.txt file, run the following command in your terminal or command prompt:

pip install -r requirements.txt

Create new app.py file for main program

Import libray that needed for this project

from ultralytics import YOLO
import time
import numpy as np
import mediapipe as mp


import cv2
from flask import Flask, render_template, request, Response, session, redirect, url_for

from flask_socketio import SocketIO
import yt_dlp as youtube_dl

Explanation for each library:

ultralytics.YOLO: Used for real-time object detection.
time: Handles time-related functions.
numpy as np: Supports numerical computations and arrays.
mediapipe as mp: Facilitates various media processing tasks like object detection and hand tracking.
cv2: Offers tools for computer vision, image, and video processing.
Flask: A lightweight web framework for building web applications.
Flask_socketio and SocketIO: Enables WebSocket support for real-time communication in Flask.
yt_dlp as youtube_dl: Used to stream media content from various streaming sites, like YouTube.

Initialize Model

model_object_detection = YOLO("bisindo.pt")

Create Function for Detection

def show(self, url):
        print(url)
        self._preview = False
        self._flipH = False
        self._detect = False
        self._mediaPipe = False

        self._confidence = 75.0
        ydl_opts = {
            "quiet": True,
            "no_warnings": True,
            "format": "best",
            "forceurl": True,
        }

        if url == '0':
            cap = cv2.VideoCapture(0)
        else:
            
            ydl = youtube_dl.YoutubeDL(ydl_opts)

            info = ydl.extract_info(url, download=False)
            url = info["url"]

            cap = cv2.VideoCapture(url)

        while True:
            if self._preview:
                if stop_flag:
                    print("Process Stopped")
                    return

                grabbed, frame = cap.read()
                if not grabbed:
                    break
                if self.flipH:
                    frame = cv2.flip(frame, 1)

                if self.detect:
                    frame_yolo = frame.copy()
                    results_yolo = model_object_detection.predict(frame_yolo, conf=self._confidence / 100)

                    frame_yolo, labels = results_yolo[0].plot()
                    list_labels = []
                    # labels_confidences
                    for label in labels:
                        confidence = label.split(" ")[-1]
                        label = (label.split(" "))[:-1]
                        label = " ".join(label)
                        list_labels.append(label)
                        list_labels.append(confidence)
                        socketio.emit('label', list_labels)

                if self.mediaPipe:
                    # Convert the image to RGB for processing with MediaPipe
                    image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
                    results = self.hands.process(image)
                    
                    if results.multi_hand_landmarks:
                        for hand_landmarks in results.multi_hand_landmarks:
                            mp.solutions.drawing_utils.draw_landmarks(
                                frame,
                                hand_landmarks,
                                self.mp_hands.HAND_CONNECTIONS,
                                landmark_drawing_spec=mp.solutions.drawing_utils.DrawingSpec(color=(255, 0, 0), thickness=4, circle_radius=3),
                                connection_drawing_spec=mp.solutions.drawing_utils.DrawingSpec(color=(255, 255, 255), thickness=2, circle_radius=2), 
                            )

                frame = cv2.imencode(".jpg", frame)[1].tobytes()
                yield ( 
                    b'--frame\r\n'
                    b'Content-Type: image/jpeg\r\n\r\n' + frame + b'\r\n'
                )
            else:
                snap = np.zeros((
                    1000,
                    1000
                ), np.uint8)
                label = "Streaming Off"
                H, W = snap.shape
                font = cv2.FONT_HERSHEY_PLAIN
                color = (255, 255, 255)
                cv2.putText(snap, label, (W//2 - 100, H//2),
                            font, 2, color, 2)
                frame = cv2.imencode(".jpg", snap)[1].tobytes()
                yield (b'--frame\r\n'
                       b'Content-Type: image/jpeg\r\n\r\n' + frame + b'\r\n')

Explanation for each object:

preview(): Displays the video stream on the web interface.
flipH(): Flips the video horizontally.
detect(): Detects using Yolo model the hand gesture from the video stream.
mediaPipe(): Draws the hand gesture landmark on the video stream.

Explanation for flow of the program:

If user input is "camera", the program will run the camera detection.
If user input is "url youtube video", the program will run the youtube detection.
If user activate the preview, the program will run the video stream/camera.
If user activate the detection, the program will run the detection.
If user activate the landmark, the program will run the landmark.
If user activate the flip, video will be flipped.
Threshold is used to set the minimum confidence threshold of the detection.
If the preview is not activated, the program will show streaming off label.

Integration in HTML, CSS, and Javascript

In CSS file, create a style for the stream and output

/* * Local selectors */
#container {
    width: 100%;
    height: 586px;
    border: 8px #2c374a solid;
    background-color: #0F172A;
    border-radius: 5px;
}

#videoElement {
    height: 570px;
    width: 100%;
    background-color: #0F172A;

    display: block;
    margin-left: auto;
    margin-right: auto;
}

#terminal {
    border-radius: 5px;
    border: 5px #1C2637 solid;
    font-family: monospace;
    font-size: 12px;
    background-color: #0F172A;
    height: 490px;
    overflow-y: scroll;
}

In Javascript, create a function needed for the web interface

For Camera or Video Button

function startCamera() {
    var url = '0';
    $('#urlForm').attr('action', '/index'); 
    $('#urlForm').attr('method', 'POST'); 
    $('#urlForm').find('#url').val(url);
    $('#urlForm').submit();
}

function startVideo() {
    var url = $('#url').val();
    $('#urlForm').attr('action', '/index'); 
    $('#urlForm').attr('method', 'POST'); 
    $('#urlForm').find('#url').val(url);
    $('#urlForm').submit();
}

For terminal output, socket, and final output

var socket = io.connect('http://127.0.0.1:5000/');

let consecutiveWords = [];
let finalSentence = "";
let wordCounter = 0;

function appendToTerminal(message) {
    var terminal = document.getElementById("terminal");
    var p = document.createElement("p");
    p.innerHTML = `<table class="table table-striped text-center" style="border: none;">
                    <tr class="row">
                        <td class="col-md-6" style="color: #01ECEC; border: none;">${message[0]}</td>
                        <td class="col-md-6" style="color: #01ECEC; border: none;">${message[1]}</td>
                    </tr>
                </table>`;
    terminal.appendChild(p);
    terminal.scrollTop = terminal.scrollHeight;

    if (consecutiveWords.length === 0 || consecutiveWords[consecutiveWords.length - 1] === message[0]) {
        consecutiveWords.push(message[0]);
        wordCounter++; 
    } else {
        consecutiveWords = [message[0]];
        wordCounter = 1;
    }

    if (wordCounter >= 10) {
        finalSentence += (finalSentence.length > 0 ? " " : "") + consecutiveWords[0];
        document.getElementById("finalSentencePara").innerText = finalSentence;
        consecutiveWords = [];
        wordCounter = 0;
    }
}

socket.on("label", (data) => {
    appendToTerminal(data);
});

Integration with SocketIO:

Listens for data labeled as "label" from a SocketIO connection.
Calls appendToTerminal() to display the received data in the terminal and potentially update an advertisement based on the data.

Function appendToTerminal(message):

Takes a message as input.
Adds a table with two columns to the terminal for displaying the message.
Keeps track of consecutive words and their counts.
Constructs a final sentence if a word appears more than ten times consecutively.

For Toggle Button

function toggleHSwitch() {
    var switchElement = $("#flip-horizontal");
    var switchIsOn = switchElement.is(":checked");

    if (switchIsOn) {
        console.log("SWITCH ON")
        $.getJSON("/request_flipH_switch", function (data) {
            console.log("Switch on request sent.");
        });
    } else {
        console.log("SWITCH OFF")
        $.getJSON("/request_flipH_switch", function (data) {
            console.log("Switch off request sent.");
        });
    }
}

function toggleMediaPipeSwitch() {
    var switchElement = $("#mediapipe");
    var switchIsOn = switchElement.is(":checked");

    if (switchIsOn) {
        console.log("SWITCH ON")
        $.getJSON("/request_mediapipe_switch", function (data) {
            console.log("Switch on request sent.");
        });
    } else {
        console.log("SWITCH OFF")
        $.getJSON("/request_mediapipe_switch", function (data) {
            console.log("Switch off request sent.");
        });
    }
}

function toggleDetSwitch() {

    var switchElement = $("#run_detection");
    var switchIsOn = switchElement.is(":checked");

    if (switchIsOn) {
        console.log("SWITCH ON")
        $.getJSON("/request_run_model_switch", function (data) {
            console.log("Switch on request sent.");
        });
    } else {
        console.log("SWITCH OFF")
        $.getJSON("/request_run_model_switch", function (data) {
            console.log("Switch off request sent.");
        });
    }
}

function toggleOffSwitch() {
    var switchElement = $("#turn_off");
    var switchIsOn = switchElement.is(":checked");

    if (switchIsOn) {
        console.log("Camera ON")
        $.getJSON("/request_preview_switch", function (data) {
            console.log("Switch on request sent.");
        });
    } else {
        console.log("Camera OFF")
        $.getJSON("/request_preview_switch", function (data) {
            console.log("Switch off request sent.");
        });
    }
}

For HTML, integrate the Javascript function

For Camera and Terminal Output

<!-- Video -->
<div class="col-span-8 mx-4 mt-3">
    <div id="container">
        <img class="center" src="/video_feed" id="videoElement">
    </div>
</div>

<!-- Terminal -->
<div class="col-span-2 mr-4">
    <h2 class="border-b border-slate-800 py-4 mb-4 text-3xl flex justify-end font-bold leading-none tracking-tight md:text-4xl lg:text-4xl text-cyan-100 ">Output</h1>
    <div id="terminal" class="w-full"></div>
</div>

For toggle switch

<div class="flex gap-3 mb-4">
    <label class="switch">
        <input id="turn_off" value="1" name="turn_off" type="checkbox" onclick="toggleOffSwitch()"/>
        <span class="slider round"></span>
    </label>
    <label for="turn_off" class="form-label text-cyan-500">Show Video</label><br>
</div>
<div class="flex gap-3 mb-4">
    <label class="switch">
        <input id="run_detection" value="0" name="run_detection" type="checkbox"
               onclick="toggleDetSwitch()"/>
        <span class="slider round"></span>
    </label>
    <label for="run_detection" class="form-label text-cyan-500">Run Detection</label><br>
</div>
<div class="flex gap-3 mb-4">
    <label class="switch">
        <input id="mediapipe" value="0" name="mediapipe" type="checkbox"
               onclick="toggleMediaPipeSwitch()"/>
        <span class="slider round"></span>
    </label>
    <label for="mediapipe" class="form-label text-cyan-500">Show Landmark</label><br>
</div>
<div class="flex gap-3 mb-4">
    <label class="switch">
        <input id="flip-horizontal" value="0" name="flip-horizontal" type="checkbox"
               onclick="toggleHSwitch()"/>
        <span class="slider round"></span>
    </label>
    <label for="flip-horizontal" class="form-label text-cyan-500">Flip Video</label><br>
</div>

<div class="gap-3 py-4 text-center border-b border-slate-800 mb-5">
    <form action="/" method="POST" style="text-align: center;" class="mb-4" >
        <label for="slider" class="form-label text-cyan-500">Confidence Threshold</label>
        <input type="range" id="slider" name="slider" min="1" max="100">
    </form>
    <input type="hidden" id="sliderValue" name="sliderValue" value="75">
    <span class="rounded-lg py-2 px-3 bg-slate-800 text-cyan-500" id="conf_display">75</span>
</div>

For Final Output

<div>
    <p id="finalSentencePara" class="text-cyan-200 mt-4 text-center">
    </p>
</div>

Best Flask open-source libraries and packages

Sign_Language_Translator_Web_App_Yolo_Flask