recognise.py

import cv2
import numpy as np
from keras.models import load_model
import math
import copy
import pyttsx3
import subprocess
import sys
import os


    


model = load_model('transferlearning.h5') 

#Mapping the classes with gestures
gestures = {
    0:'No',1:'1',2: '3',3: '4', 4:'5',5: '7',6: '8',7: '9', 8:'A',9: 'B',10: 'Baby',11: 'Brother',12: 'C',13: 'D', 14:"Don't like", 15:'E',16: 'F',17: 'Friend',18: 'G',19: 'H', 20:'Help',21: 'House', 22:'I',23: 'J',24: 'K',25: 'L', 26:'Like',27: 'Love',28: 'M', 29:'Make',30: 'More',31: 'N', 32:'Name',33: 'No', 34:'O_OR_0',35: 'P',36: 'Pay', 37:'Play', 38:'Q', 39:'R', 40:'S',41: 'Stop', 42:'T',43: 'U', 44:'V_OR_2',45: 'W_OR_6', 46:'With', 47:'X', 48:'Y',49: 'Yes', 50:'Z',51: 'nothing'
}

def predict(gesture):   #Method for predicting the gesture
    img = cv2.resize(gesture, (200,200))
    img = img.reshape(-1,200,200,3)
    img = img/255.0
    prd = model.predict(img)
    index = prd.argmax()    #Selecting Best Estimate
    return gestures[index]

capture= cv2.VideoCapture(0)
rval, frame = capture.read()
pred_text = ''
count_frames = 0
flag = False

while True:
    
    if frame is not None: 
        
        frame = cv2.flip(frame, 1)
        frame = cv2.resize( frame, (400,400) )
        cv2.rectangle(frame, (300,300), (100,100), (0,255,0), 2) #Defining ROI
        
        #The process for thresholding the desired input frame and contour detection
        crop_img = frame[100:300, 100:300]
        grey = cv2.cvtColor(crop_img, cv2.COLOR_BGR2GRAY)
        blur = cv2.GaussianBlur(grey, (7, 7), 0)
        thresh = cv2.threshold(blur,210,255,cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)[1]
        contours, hierarchy = cv2.findContours(thresh.copy(), cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
        length = len(contours)
        maxArea = -1
        if length > 0:
            for i in range(length):  # find the biggest contour (according to area)
                temp = contours[i]
                area = cv2.contourArea(temp)
                if area > maxArea:
                    maxArea = area
                    area_index= i

            result = contours[area_index]
            hull = cv2.convexHull(result) 

            track_hand= np.zeros(crop_img.shape, np.uint8)    #Creates black frame for displaying the detected contours
            cv2.drawContours(track_hand, [result], 0, (0, 255, 0), 2)
            cv2.drawContours(track_hand, [hull], 0, (0, 0, 255), 3)
        cv2.imshow('output', track_hand)
        blackboard = np.zeros(frame.shape, dtype=np.uint8)
        
        if flag == True:
            wait=0
            pred_text = predict(crop_img)
            count_frames = 0
            cv2.putText(blackboard, pred_text, (100, 180), cv2.FONT_HERSHEY_SCRIPT_SIMPLEX, 4, (255, 255, 255))
            wait+=1
            if wait==1:
                file=open("testing_file.txt","w")
                file.write(pred_text)
                file.close()
             #   engine = pyttsx3.init()
              #  engine.say(pred_text)
               # engine.setProperty('rate',120)  
               # engine.setProperty('volume',0.9) 
               # wait=0
            #engine.runAndWait()

        result= np.hstack((frame, blackboard)) #Concatening both frames
        
        cv2.imshow("Frame", result)
        cv2.imshow("Thresholded", thresh)
        
        
    rval, frame = capture.read()
    keypress = cv2.waitKey(1)
    if keypress == ord('c') or keypress==ord('C'):  #Press C/c for enabling translation mode
        flag = True

 
    if keypress == ord('q') or keypress==ord('Q'):    #Press q to exit
        subprocess.run("pkill -f sound.py",shell=True)
        break

capture.release()
cv2.destroyAllWindows()