detector.py

# -*- coding: utf-8 -*-
"""
Created on Mon May 11 12:23:48 2020

@author: Mehak
"""

import numpy as np
import cv2
import matplotlib.pyplot as plt

class Detectors(object):

    def __init__ (self):

        #To create the object for backgrounf subtraction from individual frames
        self.fgbg = cv2.createBackgroundSubtractorMOG2(detectShadows = False)

    def detectBall(self, img):

        #Convert RGB to Gray and apply background subtraction
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        fgmask = self.fgbg.apply(gray)

        # Detect all contours in the image
        contours, hierarchy = cv2.findContours(fgmask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)

        #Find valid contours (radius greater than 10) and their minenclosing circles
        circles = []
        validContours = []
        for contour in contours:
            try:
                (x, y), radius = cv2.minEnclosingCircle(contour)
                if(radius > 10):
                    circles.append([x,y,radius])
                    validContours.append(contour)
            except ZeroDivisionError:
                pass

        #Find convex hull of sets of neighbouring contours
        l = len(circles)
        sets = np.arange(1,l+1,1)
        circles = np.array(circles)
        validContours = np.array(validContours)
        thresh = 25   #Distance threshold
        for i in range(0,l-1,1):
            for j in range(i+1, l,1):
                dist = np.linalg.norm(circles[i][0:2]-circles[j][0:2])
                if(dist < thresh):
                    sets[j] = sets[i]

        convHull = []
        for i in np.unique(sets):
            idx = np.where(sets == i)
            cont = np.vstack(validContours[i] for i in idx[0])
            convHull.append(cv2.convexHull(cont))

        similarity = []
        centers = []
        for ch in convHull:
            try:
                (x,y), radius = cv2.minEnclosingCircle(ch)
                theta = np.arange(0, 2*np.pi, 2*np.pi/len(ch))
                theta = theta.reshape(len(theta),1)
                xa = np.int32(x+ radius*np.cos(theta))
                ya = np.int32(y+ radius*np.sin(theta))
                circle = np.array([xa.T,ya.T]).T
                centers.append([x,y,radius])
                similarity.append(cv2.matchShapes(ch, circle,1,0.0)/len(ch))
            except ZeroDivisionError:
                pass

        #find the most similar convex hull and min enclosing circles
        try:
            index = np.argsort(similarity)[0]
            #cv2.circle(img, (int(centers[index][0]), int(centers[index][1])), int(centers[index][2]), (255,255,255), 3)
            #plt.imshow(img)
            #plt.show()
            return centers[index]
        except IndexError:
            return None