MovingWindow.py

import CreateData as d

import imutils, time
import cv2
from keras.models import load_model
import numpy as np
from PIL import Image

class SlidingWindow:
    def __init__(self, img_path):
        self.img = cv2.imread(img_path)
        self.img_path = img_path
        (self.win_w, self.win_h) = (64, 64)  # Window size
        self.step_size = 32  # Pixels to move the window
        self.scale = 1.5  # Scale to resize the image after each iteration
        self.min_size = (200, 200)  # Minimum size of the image to stop resizing

    def resizeImage(self, ):
        # compute the new dimensions of the image and resize it
        if self.img.size > 2700000:
            w = int(self.img.shape[1] / 2)
            self.img = imutils.resize(self.img, width=w)
            self.step_size = int(self.step_size / 2)

        yield self.img

        while True:
            w = int(self.img.shape[1] / self.scale)
            self.img = imutils.resize(self.img, width=w)

            if self.img.shape[0] < self.min_size[1] or self.img.shape[1] < self.min_size[0]:
                break

            self.step_size = int(self.step_size / 2)

            yield self.img

    def moveWindow(self, ):
        # slide a window across the image
        for x in range(0, self.img.shape[0], self.step_size):
            for y in range(0, self.img.shape[1], self.step_size):
                yield (x, y, self.img[x:x + self.win_w, y:y + self.win_h])

    def shapeWindow(self, im):
        im = im.tolist()


        for x in range(len(im)):
            for y in range(len(im[x])):
                im[x][y] = [int(sum(im[0][0]) / 3)]

        im = np.array(im).reshape(64, 64)  # Convert to numpy array
        im = Image.fromarray(im, 'L')  # Convert to PIL image and make grayscale
        im = np.array(im).reshape(1, 64, 64, 1)  # Convert back to numpy array and add dimensions

        return im


    def loopWindow(self, ):
        # loop over the sliding window for each resized image
        for resized in self.resizeImage():
            for (x, y, window) in self.moveWindow():
                if window.shape[0] != self.win_h or window.shape[1] != self.win_h:
                    continue

                window = self.shapeWindow(window)

                pred = model.predict(window)[0][0]

                print(pred)

                if pred == 1.0:
                    cv2.rectangle(resized, (x, y), (x + self.win_w, y + self.win_h), (255, 0, 0), 2)

                clone = resized.copy()
                cv2.rectangle(clone, (x, y), (x + self.win_w, y + self.win_h), (0, 255, 0), 2)
                cv2.imshow("Window", clone)
                cv2.waitKey(1)
                time.sleep(0.025)


if __name__ == '__main__':
    d = d.Data()

    model = load_model('models/m1')

    s = SlidingWindow('data/validation/trafficjam.jpeg')
    s.loopWindow()