Sign_create_dataset.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Création du Dataset\n",
    "\n",
    "Dans cette première partie, nous constituons un dataset avec les différents signes de l'alphabet. Nous avons choisi de collecter des images traitées en amont via un \"filtre\" (Thresholded) qui permettra de différencier au mieux chaque signe effectué avec la main.\n",
    "\n",
    "Durant cette aquisition de donnée, nous chargeons préalablement le background, qui ne sera pas pris en compte lorsque nous positionnerons la main. Une fois fait, 300 images seront enregistrées dans un dossier conçu à cet effet. L'opération sera reproduite pour chaque catégorie (chaque signe de l'alphabet).\n",
    "\n",
    "Le dataset constitué (près de 8100 images) sera compressé pour l'importer sur *Colab* afin d'entrainer notre modèle."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "import cv2\n",
    "import os\n",
    "import time\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [],
   "source": [
    "background = None\n",
    "accumulated_weight = 0.9\n",
    "\n",
    "# Création des dimensions du ROI\n",
    "\n",
    "ROI_top = 50\n",
    "ROI_bottom = 300\n",
    "ROI_right = 50\n",
    "ROI_left = 300\n",
    "\n",
    "# Project: gesture-recognition, License: MIT License, OpenSource\n",
    "\n",
    "def cal_accum_avg(frame, accumulated_weight):\n",
    "    global background\n",
    "    \n",
    "    if background is None:\n",
    "        background = frame.copy().astype(\"float\")\n",
    "        return None\n",
    "    \n",
    "    cv2.accumulateWeighted(frame, background, accumulated_weight)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Fonction pour l'application du filtre (Thresholded)\n",
    "def segment_hand(frame, threshold=20):\n",
    "    global background\n",
    "    \n",
    "    diff = cv2.absdiff(background.astype(\"uint8\"), frame)\n",
    "    thresholded = cv2.adaptiveThreshold(diff,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY,11,2)\n",
    "    \n",
    "    return (thresholded)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Ouverture de la caméra\n",
    "cam = cv2.VideoCapture(0)\n",
    "\n",
    "# Définition des variables (nombre de frame, dossier d'images)\n",
    "num_frames = 0\n",
    "element = '_SPACE'\n",
    "\n",
    "num_imgs_taken = 0\n",
    "\n",
    "while True:\n",
    "    ret, frame = cam.read()\n",
    "    \n",
    "    # Eviter l'inversement de l'image\n",
    "    frame = cv2.flip(frame, 1)\n",
    "    frame_copy = frame.copy()\n",
    "    roi = frame[ROI_top:ROI_bottom, ROI_right:ROI_left]\n",
    "    \n",
    "    # Traitement\n",
    "    gray_frame = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)\n",
    "    gray_frame = cv2.GaussianBlur(gray_frame, (5, 5), 0)\n",
    "    \n",
    "    if num_frames < 100:\n",
    "        cal_accum_avg(gray_frame, accumulated_weight)\n",
    "        \n",
    "        # Détection du Background\n",
    "        if num_frames <= 299:\n",
    "            cv2.putText(frame_copy, \"Chargement..\", (80, 400), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0,0,255), 2)\n",
    "            \n",
    "    # Configuration (Nous pouvons mettre notre main et la positionner correctement avant enregistrement)\n",
    "    elif num_frames <= 300: \n",
    "        hand = segment_hand(gray_frame)\n",
    "        cv2.putText(frame_copy, \"Configuration..\" + str(element), (200, 400), cv2.FONT_HERSHEY_SIMPLEX, 1, (255,0,0),2)\n",
    "        \n",
    "        # Voir si la main est bien détectée\n",
    "        if hand is not None:\n",
    "            thresholded = hand\n",
    "            \n",
    "            # Dessiner les contours de la main\n",
    "            cv2.putText(frame_copy, str(num_frames)+\" Pour \"+ str(element),(70, 45), cv2.FONT_HERSHEY_SIMPLEX, 1, (255,0,0), 2)\n",
    "            \n",
    "            # Affiche l'image prise\n",
    "            cv2.imshow(\"Image traitée\", thresholded)\n",
    "    \n",
    "    else: \n",
    "        # Filtre sur la main\n",
    "        hand = segment_hand(gray_frame)\n",
    "        \n",
    "        # Repérer si la main est là\n",
    "        if hand is not None:\n",
    "            \n",
    "            thresholded = hand\n",
    "            \n",
    "            # Détection de la main (avec le filtre)\n",
    "            cv2.putText(frame_copy, str(num_frames), (70, 45), cv2.FONT_HERSHEY_SIMPLEX, 1, (0,255,0), 2)\n",
    "            cv2.putText(frame_copy, str(num_imgs_taken) + 'Enregistrement pour ' + str(element), (200, 400), cv2.FONT_HERSHEY_SIMPLEX, 1, (0,255,0), 2)\n",
    "            \n",
    "            # Enregistrement des images de la main traitée dans la frame Filtre\n",
    "            cv2.imshow(\"Image traitée\", thresholded)\n",
    "            if num_imgs_taken <= 299:\n",
    "                cv2.imwrite(r\"C:\\\\Users\\\\utilisateur\\\\Documents\\\\dataset2\\\\\"+str(element)+\"\\\\\"+str(num_imgs_taken+1) +'.jpg', thresholded)\n",
    "                \n",
    "            else:\n",
    "                break\n",
    "            num_imgs_taken +=1\n",
    "            \n",
    "        else:\n",
    "            cv2.putText(frame_copy, 'Pas de main détectée !', (200, 400), cv2.FONT_HERSHEY_SIMPLEX, 1, (0,0,255), 2)\n",
    "            \n",
    "    # Dessine une ROI de la frame Filtre\n",
    "    cv2.rectangle(frame_copy, (ROI_left, ROI_top), (ROI_right,ROI_bottom), (255,128,0), 3)\n",
    "    \n",
    "    # Incrementation\n",
    "    num_frames += 1\n",
    "    # Montre la frame Filtre\n",
    "    cv2.imshow(\"Filtre\", frame_copy)\n",
    "    # Ferme l'application\n",
    "    if cv2.waitKey(1) & 0xFF == ord('q'):\n",
    "        break\n",
    "        \n",
    "cv2.destroyAllWindows()\n",
    "cam.release()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}