Python cv2.HOUGH_GRADIENT Examples

def findPiccircle(frame, color):

	hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)	
	color_dict = color_list.getColorList()
	mask = cv2.inRange(hsv, color_dict[color][0], color_dict[color][1])
	dilated = cv2.dilate(mask, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3)), iterations=2)
	## 需要修改minRadius以及maxRadius，用来限制识别圆的大小，排除其他的干扰
	circles = cv2.HoughCircles(dilated, cv2.HOUGH_GRADIENT, 1, 1000, param1=15, param2=10, minRadius=15, maxRadius=50)
	
	center = None
	if circles is not None:
		x, y, radius = circles[0][0]
		center = (x, y)
		cv2.circle(frame, center, radius, (0, 255, 0), 2)
		cv2.circle(frame, center, 2, (0,255,0), -1, 8, 0 );
		print('圆心：{}, {}'.format(x, y))
		
	cv2.imshow('result', frame)	
	
	if center != None:
		return center 

def hough_circles(self):
        src = self.cv_read_img(self.src_file)
        if src is None:
            return

        dst = cv.pyrMeanShiftFiltering(src, 10, 100)
        cimage = cv.cvtColor(dst, cv.COLOR_BGR2GRAY)
        circles = cv.HoughCircles(cimage, cv.HOUGH_GRADIENT, 1, 20, param1=50, param2=30, minRadius=0, maxRadius=0)
        circles = np.uint16(np.around(circles))
        for i in circles[0, :]:
            cv.circle(src, (i[0], i[1]), i[2], (0, 0, 255), 2)
            cv.circle(src, (i[0], i[1]), 2, (255, 0, 255), 2)

        self.decode_and_show_dst(src)

    # 轮廓发现 

def capture_white_circles(self):
        self.prep_for_white_circles()
        img = cv2.cvtColor(self.white_query, cv2.COLOR_BGR2GRAY)
        img = cv2.medianBlur(img, 1)
        cimg = cv2.cvtColor(self.query, cv2.COLOR_BGR2RGB)
        circles = cv2.HoughCircles(img, cv2.HOUGH_GRADIENT, 1, img.shape[0] / 15,
                                   param1=50, param2=22, minRadius=5, maxRadius=60)
        if circles is None:
            return
        circles = np.uint16(np.around(circles))
        new_circles = []
        for i in circles[0, :]:
            if self.in_box(i[0], i[1]) and not self.in_blacklist(i[0], i[1]):
                self.circlePoints.append((i[0], i[1]))
                new_circles.append(i)
        if self._debug:
            # self.draw_circles(circles, cimg)
            if len(new_circles) > 0:
                self.draw_circles(np.array([new_circles]), cimg) 

def read_captured_circles(self):
        img = cv2.cvtColor(self.query, cv2.COLOR_BGR2GRAY)
        img = cv2.medianBlur(img, 7)
        cimg = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)

        circles = cv2.HoughCircles(img, cv2.HOUGH_GRADIENT, 1, 30,
                                   param1=50, param2=30, minRadius=20, maxRadius=50)
        if circles is None:
            return
        circles = np.uint16(np.around(circles))
        for i in circles[0, :]:
            if i[1] < 400:
                continue
            self.circlePoints.append((i[0], i[1]))
        if self._debug:
            self.draw_circles(circles, cimg) 

def FindWand():
    global rval,old_frame,old_gray,p0,mask,color,ig,img,frame
    try:
        rval, old_frame = cam.read()
	cv2.flip(old_frame,1,old_frame)
        old_gray = cv2.cvtColor(old_frame,cv2.COLOR_BGR2GRAY)
        equalizeHist(old_gray)
	old_gray = GaussianBlur(old_gray,(9,9),1.5)
        dilate_kernel = np.ones(dilation_params, np.uint8)
        old_gray = cv2.dilate(old_gray, dilate_kernel, iterations=1)
        clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8,8))
        old_gray = clahe.apply(old_gray)
        #TODO: trained image recognition
        p0 = cv2.HoughCircles(old_gray,cv2.HOUGH_GRADIENT,3,50,param1=240,param2=8,minRadius=4,maxRadius=15)
	if p0 is not None:
            p0.shape = (p0.shape[1], 1, p0.shape[2])
            p0 = p0[:,:,0:2] 
            mask = np.zeros_like(old_frame)
            ig = [[0] for x in range(20)]
        print "finding..."
        threading.Timer(3, FindWand).start()
    except:
        e = sys.exc_info()[1]
        print "Error: %s" % e 
        exit 

def CountCoins(img, cimg):
    circles = []
    try:
        circles = cv2.HoughCircles(img,cv2.HOUGH_GRADIENT,1,20,
                            param1=50,param2=30,minRadius=0,maxRadius=0)
    except:
        circles = cv2.HoughCircles(img, cv.CV_HOUGH_GRADIENT,1,20,param1=50,param2=30,minRadius=0,maxRadius=0)

    if(circles is None):
        return

    circles = np.uint16(np.around(circles))
    for i in circles[0, :]:
        # draw the outer circle
        cv2.circle(cimg, (i[0], i[1]), i[2], (0, 255, 0), 2)
        # draw the center of the circle
        cv2.circle(cimg, (i[0], i[1]), 2, (0, 0, 255), 3) 

def CountCoins(img, cimg):
    circles = []
    try:
        circles = cv2.HoughCircles(img,cv2.HOUGH_GRADIENT,1,20,
                            param1=50,param2=30,minRadius=0,maxRadius=0)
    except:
        circles = cv2.HoughCircles(img, cv.CV_HOUGH_GRADIENT,1,20,param1=50,param2=30,minRadius=0,maxRadius=0)

    if(circles is None):
        return

    circles = np.uint16(np.around(circles))
    for i in circles[0, :]:
        # draw the outer circle
        cv2.circle(cimg, (i[0], i[1]), i[2], (0, 255, 0), 2)
        # draw the center of the circle
        cv2.circle(cimg, (i[0], i[1]), 2, (0, 0, 255), 3) 

def detect_circles(self,gray,img):
        circles = cv2.HoughCircles(gray, cv2.HOUGH_GRADIENT, 1, 100, param2=150, minRadius=160)
        circles = np.uint16(np.around(circles))  # 把circles包含的圆心和半径的值变成整数
        cir = img.copy()

        for i in circles[0, :]:
            cv2.circle(cir, (i[0], i[1]), i[2], (0, 255, 0), 2, cv2.LINE_AA)  # 画圆
            cv2.circle(cir, (i[0], i[1]), 2, (0, 255, 0), 2, cv2.LINE_AA)  # 画圆心
        cv2.imshow("circles", cir)
        return cir

    # 霍夫直线变换：检测指针 

def _get_circles(image_segment, params):

    temp = cv2.HoughCircles(image_segment.segmented_image, cv2.HOUGH_GRADIENT, params.dp, params.minDist,
                            param1=params.param1, param2=params.param2,
                            minRadius=params.minRadius, maxRadius=params.maxRadius)
    if temp is None:
        return []

    circle_tuples = temp[0]
    if len(circle_tuples) > params.max_num:
        circle_tuples = circle_tuples[:params.max_num]

    circles = [instantiators['circle'](instantiators['point'](x, y), radius)
               for x, y, radius in circle_tuples]
    return circles 

def capture_white_circles(self, x_limit=480, y_limit=670):
        self.prep_for_white_circles()
        img = cv2.cvtColor(self.white_query, cv2.COLOR_BGR2GRAY)
        cimg = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
        circles = cv2.HoughCircles(img, cv2.HOUGH_GRADIENT, 1, 40,
                                   param1=50, param2=30, minRadius=5, maxRadius=60)
        if circles is None:
            return
        circles = np.uint16(np.around(circles))
        for i in circles[0, :]:
            if i[0] <= x_limit and i[1] <= y_limit:
                self.circlePoints.append((i[0], i[1]))
        if self._debug:
            self.draw_circles(circles, cimg) 

def FindNewPoints():
    global old_frame,old_gray,p0,mask,color,ig,img,frame
    try:
        try:
            old_frame = cam.capture(stream, format='jpeg')
        except:
            print("resetting points")
        data = np.fromstring(stream.getvalue(), dtype=np.uint8)
        old_frame = cv2.imdecode(data, 1)
        cv2.flip(old_frame,1,old_frame)
        old_gray = cv2.cvtColor(old_frame,cv2.COLOR_BGR2GRAY)
        #cv2.equalizeHist(old_gray,old_gray)
	    #old_gray = cv2.GaussianBlur(old_gray,(9,9),1.5)
        #dilate_kernel = np.ones(dilation_params, np.uint8)
        #old_gray = cv2.dilate(old_gray, dilate_kernel, iterations=1)

        #TODO: trained image recognition
        p0 = cv2.HoughCircles(old_gray,cv2.HOUGH_GRADIENT,3,100,param1=100,param2=30,minRadius=4,maxRadius=15)
        p0.shape = (p0.shape[1], 1, p0.shape[2])
        p0 = p0[:,:,0:2]
        mask = np.zeros_like(old_frame)
        ig = [[0] for x in range(20)]
        print("finding...")
        TrackWand()
	    #This resets the scene every three seconds
        threading.Timer(3, FindNewPoints).start()
    except:
        e = sys.exc_info()[1]
        print("FindWand Error: %s" % e )
        End()
        exit 

def cropImage(self, image, raidNo, radius):
        gray=cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
        gray=cv2.GaussianBlur(gray, (7, 7), 2)
        output = image.copy()
        height, width, channel = output.shape
        output = output[0:height*2/3,0:width]
        image_cols, image_rows, _ = image.shape
        circles = cv2.HoughCircles(gray, cv2.HOUGH_GRADIENT, 1, 20, param1=50,param2=30, minRadius=radius, maxRadius=radius)
        if circles is not None:
            circles = np.round(circles[0, :]).astype("int")
            for (x, y, r) in circles:
                        log.debug('[Crop: ' + str(raidNo) + ' (' + str(self.uniqueHash) +') ] ' + 'cropImage: Detect crop coordinates x: ' + str(x) +' y: ' + str(y) +' with radius: ' + str(r))
                        new_crop = output[y-r:y+r, x-r:x+r]
                        return new_crop
        return False 

def cropImage(self, screenshot, captureTime, captureLat, captureLng, src_path):
        p = None
        raidNo = 0
        processes = []
        
        hash = str(time.time())
        orgScreen = screenshot
        height, width, channel = screenshot.shape
        gray=cv2.cvtColor(screenshot,cv2.COLOR_BGR2GRAY)
        gray=cv2.GaussianBlur(gray, (7, 7), 2)

        minRadius = int(((width / 4.736)) / 2) 
        maxRadius = int(((width / 4.736)) / 2)
        log.debug('Searching for Raid Circles with Radius from %s to %s px' % (str(minRadius), str(maxRadius)))
        
        circles = cv2.HoughCircles(gray, cv2.HOUGH_GRADIENT, 1, 20, param1=50,param2=30, minRadius=minRadius, maxRadius=maxRadius)
        
        if circles is not None:
            circles = np.round(circles[0, :]).astype("int")
            for (x, y, r) in circles:
                log.debug('Found Circle with x:%s, y:%s, r:%s' % (str(x), str(y), str(r)))
                raidNo += 1
                raidCropFilepath = os.path.join(args.temp_path, str(hash) + "_raidcrop" + str(raidNo) +".jpg")
                new_crop = orgScreen[y-r-int((r*2*0.03)):y+r+int((r*2*0.75)), x-r-int((r*2*0.03)):x+r+int((r*2*0.3))]
                cv2.imwrite(raidCropFilepath, new_crop)
                if args.ocr_multitask:
                    p = multiprocessing.Process(target=RaidScan.process, name='OCR-crop-analysis-' + str(raidNo), args=(raidCropFilepath, hash, raidNo, captureTime, captureLat, captureLng, src_path, r))
                else:
                    p = Thread(target=RaidScan.process, name='OCR-processing', args=(raidCropFilepath, hash, raidNo, captureTime, captureLat, captureLng, src_path, r))
                processes.append(p)          
                p.daemon = True
                p.start() 

def find_circular_wells(self):
        """Simply use Hough transform to find circles in MultiWell Plate rgb image.
        The parameters used are optimised for 24 or 48WP"""


        dwnscl_factor = self.img_shape[0]/self.blur_im.shape[0]

        # find circles
        # parameters in downscaled units
        circle_goodness = 70;
        highest_canny_thresh = 10;
        min_well_dist = self.blur_im.shape[1]/3;    # max 3 wells along short side. bank on FOV not taking in all the entirety of the well
        min_well_radius = self.blur_im.shape[1]//7; # if 48WP 3 wells on short side ==> radius <= side/6
        max_well_radius = self.blur_im.shape[1]//4; # if 24WP 2 wells on short side. COnsidering intrawells space, radius <= side/4
        # find circles
        _circles = cv2.HoughCircles(self.blur_im,
                                   cv2.HOUGH_GRADIENT,
                                   dp=1,
                                   minDist=min_well_dist,
                                   param1=highest_canny_thresh,
                                   param2=circle_goodness,
                                   minRadius=min_well_radius,
                                   maxRadius=max_well_radius)
        _circles = np.squeeze(_circles); # because why the hell is there an empty dimension at the beginning?

        # convert back to pixels
        _circles *= dwnscl_factor;

        # output back into class property
        self.wells['x'] = _circles[:,0].astype(int)
        self.wells['y'] = _circles[:,1].astype(int)
        self.wells['r'] = _circles[:,2].astype(int)
        return 

def detect_blob(self, img, filters):
        """
            "filters" must be something similar to:
            filters = {
                'R': (150, 255), # (min, max)
                'S': (150, 255),
            }

        """
        acc_mask = ones(img.shape[:2], dtype=uint8) * 255

        rgb = img.copy()
        hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)

        for c, (min, max) in filters.items():
            img = rgb if c in 'RGB' else hsv

            mask = img[:, :, self.channels[c]]
            mask[mask < min] = 0
            mask[mask > max] = 0

            acc_mask &= mask

        kernel = ones((5, 5), uint8)
        acc_mask = cv2.dilate(cv2.erode(acc_mask, kernel), kernel)

        circles = cv2.HoughCircles(acc_mask, cv2.HOUGH_GRADIENT, 3, img.shape[0] / 5.)
        return circles.reshape(-1, 3) if circles is not None else [] 

def __readCircleCords(self,filename,hash,ratio, crop = False, canny=False):
        log.debug("__readCircleCords: Reading circlescords")

        try:
            screenshotRead = cv2.imread(filename)
        except:
            log.error("Screenshot corrupted :(")
            return False

        if screenshotRead is None:
            log.error("Screenshot corrupted :(")
            return False

        height, width, _ = screenshotRead.shape
        
        if crop:
            screenshotRead = screenshotRead[int(height)-int(height/5):int(height),int(width)/2-int(width)/8:int(width)/2+int(width)/8]

        log.debug("__readCircleCords: Determined screenshot scale: " + str(height) + " x " + str(width))
        gray = cv2.cvtColor(screenshotRead, cv2.COLOR_BGR2GRAY)
        # detect circles in the image

        radMin = int((width / float(ratio) - 3) / 2)
        radMax = int((width / float(ratio) + 3) / 2)
        
        if canny:
            gray = cv2.GaussianBlur(gray, (3, 3), 0)
            gray = cv2.Canny(gray, 100, 50, apertureSize=3)
        
        log.debug("__readCircleCords: Detect radius of circle: Min " + str(radMin) + " Max " + str(radMax))
        circles = cv2.HoughCircles(gray, cv2.HOUGH_GRADIENT, 1, width / 8, param1=100, param2=15, minRadius=radMin,
                                   maxRadius=radMax)
        circle = 0
        # ensure at least some circles were found
        if circles is not None:
            # convert the (x, y) coordinates and radius of the circles to integers
            circles = np.round(circles[0, :]).astype("int")
            # loop over the (x, y) coordinates and radius of the circles
            for (x, y, r) in circles:
                log.debug("__readCircleCords: Found Circle x: %s y: %s" % (str(width/2), str((int(height)-int(height/5))+y)))
                return True, width/2, (int(height)-int(height/5))+y, height, width
        else:
            log.debug("__readCircleCords: Found no Circle")
            return False, 0, 0, 0, 0 

def __readCircleCount(self,filename,hash,ratio, xcord = False, crop = False, click = False, canny=False):
        log.debug("__readCircleCount: Reading circles")

        try:
            screenshotRead = cv2.imread(filename)
        except:
            log.error("Screenshot corrupted :(")
            return -1

        if screenshotRead is None:
            log.error("Screenshot corrupted :(")
            return -1

        height, width, _ = screenshotRead.shape
        
        if crop:
            screenshotRead = screenshotRead[int(height)-int(height/4.5):int(height),int(width)/2-int(width)/8:int(width)/2+int(width)/8]

        log.debug("__readCircleCount: Determined screenshot scale: " + str(height) + " x " + str(width))
        gray = cv2.cvtColor(screenshotRead, cv2.COLOR_BGR2GRAY)
        # detect circles in the image

        radMin = int((width / float(ratio) - 3) / 2)
        radMax = int((width / float(ratio) + 3) / 2)
        
        if canny:
            gray = cv2.GaussianBlur(gray, (3, 3), 0)
            gray = cv2.Canny(gray, 100, 50, apertureSize=3)
        
        log.debug("__readCircleCount: Detect radius of circle: Min " + str(radMin) + " Max " + str(radMax))
        circles = cv2.HoughCircles(gray, cv2.HOUGH_GRADIENT, 1, width / 8, param1=100, param2=15, minRadius=radMin,
                                   maxRadius=radMax)
        circle = 0
        # ensure at least some circles were found
        if circles is not None:
            # convert the (x, y) coordinates and radius of the circles to integers
            circles = np.round(circles[0, :]).astype("int")
            # loop over the (x, y) coordinates and radius of the circles
            for (x, y, r) in circles:

                if not xcord:
                    circle += 1
                    if click:
                        log.debug('__readCircleCount: found Circle - click it')
                        self.screenWrapper.click(width/2, ((int(height)-int(height/4.5)))+y)
                        time.sleep(2)
                else:
                    if x >= (width / 2) - 100 and x <= (width / 2) + 100 and y >= (height - (height / 3)):
                        circle += 1
                        if click:
                            log.debug('__readCircleCount: found Circle - click on: it' )
                            self.screenWrapper.click(width/2, ((int(height)-int(height/4.5)))+y)
                            time.sleep(2)

            log.debug("__readCircleCount: Determined screenshot to have " + str(circle) + " Circle.")
            return circle
        else:
            log.debug("__readCircleCount: Determined screenshot to have 0 Circle")
            return -1 

def TennisDetect(self, img, VideoReturn):   # 检测网球（利用霍夫圆检测和HSV色彩检测）
        x_pos = 0  # initialize the tennis's position
        y_pos = 0
        radius = 0
        img_out = copy.copy(img)
        img = img[self.cut_edge:479, :, :]
        img = cv2.blur(img, (5,5))  # denoising
        hsv_img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)  # rgb to HSV

        # img_out = copy.copy(hsv_img[:, :, 0])
        # img_out = cv2.cvtColor(img_out, cv2.COLOR_GRAY2BGR) 

        gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)  # rgb to gray

        circles = cv2.HoughCircles(gray_img, cv2.HOUGH_GRADIENT, 1, 60, param1=100, param2=20, minRadius=15, maxRadius=60)  # HOUGH circle detection

        if circles is not None:  
            x = circles[0][:, 0].astype(int)  # extract the x, y, r of all detected circles
            y = circles[0][:, 1].astype(int)
            r = circles[0][:, 2].astype(int)
            s_r = (r/1.5).astype(int)  

            num = circles[0].shape[0]
            rate = np.zeros(num)

            for i in range(num):  # traverse all detected circles
                detect_area = (hsv_img[y[i]-s_r[i]: y[i]+s_r[i], x[i]-s_r[i]:x[i]+s_r[i]])  # A square in the detected circle
                height, width, channel = detect_area.shape

                if height !=0 and width !=0:
                    tennis_color_mask = cv2.inRange(detect_area, self.lower_range, self.higher_range)
                    num_point = np.sum(tennis_color_mask / 255)
                    rate[i] = num_point / (height * width)              
                    img_out = cv2.circle(img_out, (x[i], y[i]+self.cut_edge), r[i], (0,255,0), thickness=2)
                    
            i = np.argmax(rate)  # select the circle with the maximum rate as the detected tennis
            if rate[i] > 0.4:   # if the percent of tennis_color_point in detect_area > 50%, then regard it as the tennis
                x_pos = x[i]
                y_pos = y[i]
                radius = r[i]
            print('x: ', x[i], '  y: ', y[i], '  r: ', r[i], '   rate: ', rate[i])

        if VideoReturn:  # if it needs to return the frame with the detected tennis
            img_out = cv2.circle(img_out, (x_pos, y_pos+self.cut_edge), radius, (0,0,255), thickness=10)
            return img_out, x_pos, y_pos, radius
        else:  # if it only needs to return the position of the detected tennis
            return x_pos, y_pos, radius 

def TennisDetect(self, img, VideoReturn):   # 检测网球（利用霍夫圆检测和HSV色彩检测）
        x_pos = 0  # initialize the tennis's position
        y_pos = 0
        radius = 0

        img = cv2.blur(img, (5,5))  # denoising
        hsv_img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)  # rgb to HSV
        (h_img, s_img, v_img) = cv2.split(hsv_img)

        gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)  # rgb to gray
        # gray_img = cv2.equalizeHist(gray_img)  # hist equalization

        circles = cv2.HoughCircles(gray_img, cv2.HOUGH_GRADIENT, 1, 120, param1=100, param2=20, minRadius=15, maxRadius=60)  # HOUGH circle detection

        if circles is not None:  
            x = circles[0][:, 0].astype(int)  # extract the x, y, r of all detected circles
            y = circles[0][:, 1].astype(int)
            r = circles[0][:, 2].astype(int)
            s_r = (r/1.5).astype(int)  

            num = circles[0].shape[0]
            distance = np.ones(num) * 100
            mean_h = np.zeros(num)
            mean_s = np.zeros(num)

            for i in range(num):  # traverse all detected circles:
                detect_area_h = (h_img[y[i]-s_r[i]: y[i]+s_r[i], x[i]-s_r[i]:x[i]+s_r[i]])  # A square in the detected circle (H)
                detect_area_s = (s_img[y[i]-s_r[i]: y[i]+s_r[i], x[i]-s_r[i]:x[i]+s_r[i]])  # A square in the detected circle (S)

                # Through 33 photos of the tennis captured by Raspberry Camera, we find the average H of tennis areas is 36, the average S of tennis areas is 163, and the var of H is much smaller than that of S.
                if len(detect_area_h):
                    num_point =  detect_area_h[detect_area_h > 30 and detect_area_h < 40]
                    # height, width = detect_area_h.shape
                    # rate = num_point / (height*width)
                    # print('rate',  rate)
                    mean_h[i] = np.mean(detect_area_h)
                    mean_s[i] = np.mean(detect_area_s)
                    distance[i] = np.sqrt(0.98*(mean_h[i] - 35)**2 + 0.02*(mean_s[i] - 163)**2) 
                
            i = np.argmin(distance)  # select the circle with the minimum distance as the detected tennis
            if distance[i] < 20:   # if distance > 15, the selected circle cannot be a tennis
                x_pos = x[i]
                y_pos = y[i]
                radius = r[i]
            print('x: ', x[i], '  y: ', y[i], '  r: ', r[i], '   distance: ', distance[i], '  mean_h:', mean_h[i], '  mean_s:', mean_s[i])


        if VideoReturn:  # if it needs to return the frame with the detected tennis
            img_out = copy.copy(img)
            img_out = cv2.circle(img_out, (x_pos, y_pos), radius, (0,0,255), thickness=10)
            return img_out, x_pos, y_pos, radius
        else:  # if it only needs to return the position of the detected tennis
            return x_pos, y_pos, radius 

def find_eye_center(image):
    """
    Find center of eye using Fabian's algorithm
    :param image: Gray scale image of eye
    :return: row, col identified as center
    """
    # print image.shape
    global showImage

    scaled_image = utils.image_resize(image.copy(), width=EYE_ROI_WIDTH)
    gradient_energy_x = cv2.Sobel(scaled_image, cv2.CV_64F, 1, 0, ksize=3)
    gradient_energy_y = cv2.Sobel(scaled_image, cv2.CV_64F, 0, 1, ksize=3)
    gradient_magnitude = (gradient_energy_x ** 2 + gradient_energy_y ** 2) ** 0.5
    threshold = np.mean(gradient_magnitude) + np.std(gradient_magnitude) * 3
    gradient_energy_x /= gradient_magnitude
    gradient_energy_y /= gradient_magnitude
    mask = gradient_magnitude < threshold
    gradient_energy_x[mask] = 0
    gradient_energy_y[mask] = 0
    scaled_image = cv2.GaussianBlur(scaled_image, (5, 5), 0, 0)

    inverted_image = 255 * np.ones_like(scaled_image) - scaled_image
    if showImage:
        # cv2.HoughCircles(scaled_image, cv2.HOUGH_GRADIENT, 2, 12.0)
        cv2.imshow("EyeDebug", inverted_image)
        if (cv2.waitKey() & 0xFF) == ord('s'):
            showImage = False
            cv2.destroyWindow('EyeDebug')

    indices = np.indices(inverted_image.shape).astype(np.float32)
    indices += 1e-8
    output_sum = np.zeros_like(inverted_image).astype(np.float32)
    for row in range(output_sum.shape[0]):
        for col in range(output_sum.shape[1]):
            val1 = (indices[0] - row) * gradient_energy_y
            val2 = (indices[1] - col) * gradient_energy_x
            val = (val1 + val2)
            output_sum += inverted_image * (val - val.mean()) / val.std()
            # compute_location_weight(row, col, inverted_image, gradient_energy_x, gradient_energy_y)

    index = np.unravel_index(np.argmax(output_sum), output_sum.shape)
    rescaled_index = (
        index[0] * image.shape[0] / scaled_image.shape[0], index[1] * image.shape[1] / scaled_image.shape[1])
    return rescaled_index 

def detect_sushi(img):
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    blur = cv2.medianBlur(gray, 5)
    circles = cv2.HoughCircles(blur, cv2.HOUGH_GRADIENT,
                                    dp=1, minDist=100, param1=50, param2=30,
                                    minRadius=170, maxRadius=400)

    if circles is None:
        return None

    obj_width = 0
    obj_height = 0
    circles = np.uint16(np.around(circles))
    for (x, y, r) in circles[0]:
        x, y, r = int(x), int(y), int(r)
        obj_top = int(y - r - 10)
        if obj_top < 0:
            obj_top = 0

        obj_left = int(x - r - 10)
        if obj_left < 0:
            obj_left = 0

        obj_width = int(r*2+20)
        obj_right = obj_left + obj_width
        if obj_right > WIDTH:
            obj_right = WIDTH
            obj_width = WIDTH - obj_left

        obj_height = int(r*2+20)
        obj_bottom =  obj_top + obj_height
        if obj_bottom > HEIGHT:
            obj_bottom = HEIGHT
            obj_height = HEIGHT - obj_top
        break

    if obj_width < 380 or obj_height < 380 or obj_width > 420 or obj_height > 420:
        # skip if circle is small or too large
        return None

    # return the detected rectangle
    return (obj_top, obj_bottom, obj_left, obj_right)

# infinite loop to detect sushi 

def find_blob() :
    radius = 0
    # Load input image
    _, bgr_image = img.read()

    orig_image = bgr_image

    bgr_image = cv2.medianBlur(bgr_image, 3)

    # Convert input image to HSV
    hsv_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2HSV)

    # Threshold the HSV image, keep only the red pixels
    lower_red_hue_range = cv2.inRange(hsv_image, (0, 100, 100), (10, 255, 255))
    upper_red_hue_range = cv2.inRange(hsv_image, (160, 100, 100), (179, 255, 255))
    # Combine the above two images
    red_hue_image = cv2.addWeighted(lower_red_hue_range, 1.0, upper_red_hue_range, 1.0, 0.0)

    red_hue_image = cv2.GaussianBlur(red_hue_image, (9, 9), 2, 2)

    # Use the Hough transform to detect circles in the combined threshold image
    circles = cv2.HoughCircles(red_hue_image, cv2.HOUGH_GRADIENT, 1, 120, 100, 20, 10, 0)
    circles = np.uint16(np.around(circles))
    # Loop over all detected circles and outline them on the original image
    all_r = np.array([])
    # print("circles: %s"%circles)
    if circles is not None:
        try:
            for i in circles[0,:]:
                # print("i: %s"%i)
                all_r = np.append(all_r, int(round(i[2])))
            closest_ball = all_r.argmax()
            center=(int(round(circles[0][closest_ball][0])), int(round(circles[0][closest_ball][1])))
            radius=int(round(circles[0][closest_ball][2]))
            if draw_circle_enable:
                cv2.circle(orig_image, center, radius, (0, 255, 0), 5)
        except IndexError:
            pass
            # print("circles: %s"%circles)

    # Show images
    if show_image_enable:
        cv2.namedWindow("Threshold lower image", cv2.WINDOW_AUTOSIZE)
        cv2.imshow("Threshold lower image", lower_red_hue_range)
        cv2.namedWindow("Threshold upper image", cv2.WINDOW_AUTOSIZE)
        cv2.imshow("Threshold upper image", upper_red_hue_range)
        cv2.namedWindow("Combined threshold images", cv2.WINDOW_AUTOSIZE)
        cv2.imshow("Combined threshold images", red_hue_image)
        cv2.namedWindow("Detected red circles on the input image", cv2.WINDOW_AUTOSIZE)
        cv2.imshow("Detected red circles on the input image", orig_image)

    k = cv2.waitKey(5) & 0xFF
    if k == 27:
        return (0, 0), 0
    if radius > 3:
        return center, radius
    else:
        return (0, 0), 0 

def houghCircles(path, counter):
    img = cv2.imread(path, 0)
    # img = cv2.medianBlur(img, 5)

    x = cv2.Sobel(img, -1, 1, 0, ksize=3)
    y = cv2.Sobel(img, -1, 0, 1, ksize=3)
    absx = cv2.convertScaleAbs(x)
    absy = cv2.convertScaleAbs(y)
    img = cv2.addWeighted(absx, 0.5, absy, 0.5, 0)

    # ycrcb = cv2.cvtColor(img, cv2.COLOR_BGR2YCR_CB)
    # channels = cv2.split(ycrcb)
    # cv2.equalizeHist(channels[0], channels[0])  #输入通道、输出通道矩阵
    # cv2.merge(channels, ycrcb)  #合并结果通道
    # cv2.cvtColor(ycrcb, cv2.COLOR_YCR_CB2BGR, img)

    # img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

    cimg = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)

    # cv2.imshow("img2", img)
    # cv2.imshow("grayimg", grayimg)

    circles = cv2.HoughCircles(
        img,
        cv2.HOUGH_GRADIENT,
        1,
        50,
        param1=50,
        param2=10,
        minRadius=2,
        maxRadius=0)

    circles = np.uint16(np.around(circles))
    for i in circles[0, :]:
        # draw the outer circle
        # cv2.circle(cimg, (i[0], i[1]), i[2], (0, 255, 0), 1)
        # draw the center of the circle
        cv2.circle(cimg, (i[0], i[1]), 2, (0, 0, 255), 2)
    # cv2.imshow("img" + str(counter), cimg)
    return (i[0] + 3, i[1] + 3)


#彩色直方图均衡化