# Copyright 2016 The Android Open Source Project # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """Image processing utilities using openCV.""" import logging import math import os import pathlib import cv2 import numpy import scipy.spatial import camera_properties_utils import capture_request_utils import error_util import image_processing_utils AE_AWB_METER_WEIGHT = 1000 # 1 - 1000 with 1000 the highest ANGLE_CHECK_TOL = 1 # degrees ANGLE_NUM_MIN = 10 # Minimum number of angles for find_angle() to be valid ARUCO_CORNER_COUNT = 4 # total of 4 corners to a aruco marker TEST_IMG_DIR = os.path.join(os.environ['CAMERA_ITS_TOP'], 'test_images') CH_FULL_SCALE = 255 CHART_FILE = os.path.join(TEST_IMG_DIR, 'ISO12233.png') CHART_HEIGHT_31CM = 13.5 # cm height of chart for 31cm distance chart CHART_HEIGHT_22CM = 9.5 # cm height of chart for 22cm distance chart CHART_DISTANCE_90CM = 90.0 # cm CHART_DISTANCE_31CM = 31.0 # cm CHART_DISTANCE_22CM = 22.0 # cm CHART_SCALE_RTOL = 0.1 CHART_SCALE_START = 0.65 CHART_SCALE_STOP = 1.35 CHART_SCALE_STEP = 0.025 CIRCLE_AR_ATOL = 0.1 # circle aspect ratio tolerance CIRCLISH_ATOL = 0.10 # contour area vs ideal circle area & aspect ratio TOL CIRCLISH_LOW_RES_ATOL = 0.15 # loosen for low res images CIRCLE_MIN_PTS = 20 CIRCLE_RADIUS_NUMPTS_THRESH = 2 # contour num_pts/radius: empirically ~3x CIRCLE_COLOR_ATOL = 0.05 # circle color fill tolerance CIRCLE_LOCATION_VARIATION_RTOL = 0.05 # tolerance to remove similar circles CV2_CONTRAST_ALPHA = 1.25 # contrast CV2_CONTRAST_BETA = 0 # brightness CV2_THESHOLD_LOWER_BLACK = 0 CV2_LINE_THICKNESS = 3 # line thickness for drawing on images CV2_BLACK = (0, 0, 0) CV2_BLUE = (0, 0, 255) CV2_RED = (255, 0, 0) # color in cv2 to draw lines CV2_RED_NORM = tuple(numpy.array(CV2_RED) / 255) CV2_GREEN = (0, 255, 0) CV2_GREEN_NORM = tuple(numpy.array(CV2_GREEN) / 255) CV2_WHITE = (255, 255, 255) CV2_YELLOW = (255, 255, 0) CV2_THRESHOLD_BLOCK_SIZE = 11 CV2_THRESHOLD_CONSTANT = 2 CV2_HOME_DIRECTORY = os.path.dirname(cv2.__file__) CV2_ALTERNATE_DIRECTORY = pathlib.Path(CV2_HOME_DIRECTORY).parents[3] HAARCASCADE_FILE_NAME = 'haarcascade_frontalface_default.xml' FACES_ALIGNED_MIN_NUM = 2 FACE_CENTER_MATCH_TOL_X = 10 # 10 pixels or ~1.5% in 640x480 image FACE_CENTER_MATCH_TOL_Y = 20 # 20 pixels or ~4% in 640x480 image FACE_CENTER_MIN_LOGGING_DIST = 50 FACE_MIN_CENTER_DELTA = 15 FOV_THRESH_TELE25 = 25 FOV_THRESH_TELE40 = 40 FOV_THRESH_TELE = 60 FOV_THRESH_UW = 90 IMAGE_ROTATION_THRESHOLD = 40 # rotation by 20 pixels LOW_RES_IMG_THRESH = 320 * 240 NUM_AE_AWB_REGIONS = 4 SCALE_CHART_33_PERCENT = 0.33 SCALE_CHART_67_PERCENT = 0.67 SCALE_WIDE_IN_22CM_RIG = 0.67 SCALE_TELE_IN_22CM_RIG = 0.5 SCALE_TELE_IN_31CM_RIG = 0.67 SCALE_TELE40_IN_22CM_RIG = 0.33 SCALE_TELE40_IN_31CM_RIG = 0.5 SCALE_TELE25_IN_31CM_RIG = 0.33 SQUARE_AREA_MIN_REL = 0.05 # Minimum size for square relative to image area SQUARE_CROP_MARGIN = 0 # Set to aid detection of QR codes SQUARE_TOL = 0.05 # Square W vs H mismatch RTOL SQUARISH_RTOL = 0.10 SQUARISH_AR_RTOL = 0.10 VGA_HEIGHT = 480 VGA_WIDTH = 640 def convert_to_y(img, color_order='RGB'): """Returns a Y image from a uint8 RGB or BGR ordered image. Args: img: a uint8 openCV image. color_order: str; 'RGB' or 'BGR' to signify color plane order. Returns: The Y plane of the input img. """ if img.dtype != 'uint8': raise AssertionError(f'Incorrect input type: {img.dtype}! Expected: uint8') if color_order == 'RGB': y, _, _ = cv2.split(cv2.cvtColor(img, cv2.COLOR_RGB2YUV)) elif color_order == 'BGR': y, _, _ = cv2.split(cv2.cvtColor(img, cv2.COLOR_BGR2YUV)) else: raise AssertionError(f'Undefined color order: {color_order}!') return y def binarize_image(img_gray): """Returns a binarized image based on cv2 thresholds. Args: img_gray: A grayscale openCV image. Returns: An openCV image binarized to 0 (black) and 255 (white). """ _, img_bw = cv2.threshold(numpy.uint8(img_gray), 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU) return img_bw def _load_opencv_haarcascade_file(): """Return Haar Cascade file for face detection.""" for cv2_directory in (CV2_HOME_DIRECTORY, CV2_ALTERNATE_DIRECTORY,): for path, _, files in os.walk(cv2_directory): if HAARCASCADE_FILE_NAME in files: haarcascade_file = os.path.join(path, HAARCASCADE_FILE_NAME) logging.debug('Haar Cascade file location: %s', haarcascade_file) return haarcascade_file raise error_util.CameraItsError('haarcascade_frontalface_default.xml was ' f'not found in {CV2_HOME_DIRECTORY} ' f'or {CV2_ALTERNATE_DIRECTORY}') def find_opencv_faces(img, scale_factor, min_neighbors): """Finds face rectangles with openCV. Args: img: numpy array; 3-D RBG image with [0,1] values scale_factor: float, specifies how much image size is reduced at each scale min_neighbors: int, specifies minimum number of neighbors to keep rectangle Returns: List of rectangles with faces """ # prep opencv opencv_haarcascade_file = _load_opencv_haarcascade_file() face_cascade = cv2.CascadeClassifier(opencv_haarcascade_file) img_uint8 = image_processing_utils.convert_image_to_uint8(img) img_gray = cv2.cvtColor(img_uint8, cv2.COLOR_RGB2GRAY) # find face rectangles with opencv faces_opencv = face_cascade.detectMultiScale( img_gray, scale_factor, min_neighbors) logging.debug('%s', str(faces_opencv)) return faces_opencv def find_all_contours(img): cv2_version = cv2.__version__ if cv2_version.startswith('3.'): # OpenCV 3.x _, contours, _ = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) else: # OpenCV 2.x and 4.x contours, _ = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) return contours def calc_chart_scaling(chart_distance, camera_fov): """Returns charts scaling factor. Args: chart_distance: float; distance in cm from camera of displayed chart camera_fov: float; camera field of view. Returns: chart_scaling: float; scaling factor for chart """ chart_scaling = 1.0 fov = float(camera_fov) is_chart_distance_22cm = math.isclose( chart_distance, CHART_DISTANCE_22CM, rel_tol=CHART_SCALE_RTOL) is_chart_distance_31cm = math.isclose( chart_distance, CHART_DISTANCE_31CM, rel_tol=CHART_SCALE_RTOL) is_chart_distance_90cm = math.isclose( chart_distance, CHART_DISTANCE_90CM, rel_tol=CHART_SCALE_RTOL) if FOV_THRESH_TELE < fov < FOV_THRESH_UW and is_chart_distance_22cm: chart_scaling = SCALE_WIDE_IN_22CM_RIG elif FOV_THRESH_TELE40 < fov <= FOV_THRESH_TELE and is_chart_distance_22cm: chart_scaling = SCALE_TELE_IN_22CM_RIG elif fov <= FOV_THRESH_TELE40 and is_chart_distance_22cm: chart_scaling = SCALE_TELE40_IN_22CM_RIG elif fov <= FOV_THRESH_TELE25 and is_chart_distance_31cm: chart_scaling = SCALE_TELE25_IN_31CM_RIG elif fov <= FOV_THRESH_TELE40 and is_chart_distance_31cm: chart_scaling = SCALE_TELE40_IN_31CM_RIG elif fov <= FOV_THRESH_TELE40 and is_chart_distance_90cm: chart_scaling = SCALE_CHART_67_PERCENT elif fov <= FOV_THRESH_TELE and is_chart_distance_31cm: chart_scaling = SCALE_TELE_IN_31CM_RIG elif chart_distance > CHART_DISTANCE_31CM: chart_scaling = SCALE_CHART_33_PERCENT return chart_scaling def scale_img(img, scale=1.0): """Scale image based on a real number scale factor.""" dim = (int(img.shape[1] * scale), int(img.shape[0] * scale)) return cv2.resize(img.copy(), dim, interpolation=cv2.INTER_AREA) class Chart(object): """Definition for chart object. Defines PNG reference file, chart, size, distance and scaling range. """ def __init__( self, cam, props, log_path, chart_file=None, height=None, distance=None, scale_start=None, scale_stop=None, scale_step=None, rotation=None): """Initial constructor for class. Args: cam: open ITS session props: camera properties object log_path: log path to store the captured images. chart_file: str; absolute path to png file of chart height: float; height in cm of displayed chart distance: float; distance in cm from camera of displayed chart scale_start: float; start value for scaling for chart search scale_stop: float; stop value for scaling for chart search scale_step: float; step value for scaling for chart search rotation: clockwise rotation in degrees (multiple of 90) or None """ self._file = chart_file or CHART_FILE if math.isclose( distance, CHART_DISTANCE_31CM, rel_tol=CHART_SCALE_RTOL): self._height = height or CHART_HEIGHT_31CM self._distance = distance else: self._height = height or CHART_HEIGHT_22CM self._distance = CHART_DISTANCE_22CM self._scale_start = scale_start or CHART_SCALE_START self._scale_stop = scale_stop or CHART_SCALE_STOP self._scale_step = scale_step or CHART_SCALE_STEP self.opt_val = None self.locate(cam, props, log_path, rotation) def _set_scale_factors_to_one(self): """Set scale factors to 1.0 for skipped tests.""" self.wnorm = 1.0 self.hnorm = 1.0 self.xnorm = 0.0 self.ynorm = 0.0 self.scale = 1.0 def _calc_scale_factors(self, cam, props, fmt, log_path, rotation): """Take an image with s, e, & fd to find the chart location. Args: cam: An open its session. props: Properties of cam fmt: Image format for the capture log_path: log path to save the captured images. rotation: clockwise rotation of template in degrees (multiple of 90) or None Returns: template: numpy array; chart template for locator img_3a: numpy array; RGB image for chart location scale_factor: float; scaling factor for chart search """ req = capture_request_utils.auto_capture_request() cap_chart = capture_request_utils.stationary_lens_capture(cam, req, fmt) img_3a = image_processing_utils.convert_capture_to_rgb_image( cap_chart, props) img_3a = image_processing_utils.rotate_img_per_argv(img_3a) af_scene_name = os.path.join(log_path, 'af_scene.jpg') image_processing_utils.write_image(img_3a, af_scene_name) template = cv2.imread(self._file, cv2.IMREAD_ANYDEPTH) if rotation is not None: logging.debug('Rotating template by %d degrees', rotation) template = numpy.rot90(template, k=rotation / 90) focal_l = cap_chart['metadata']['android.lens.focalLength'] pixel_pitch = ( props['android.sensor.info.physicalSize']['height'] / img_3a.shape[0]) logging.debug('Chart distance: %.2fcm', self._distance) logging.debug('Chart height: %.2fcm', self._height) logging.debug('Focal length: %.2fmm', focal_l) logging.debug('Pixel pitch: %.2fum', pixel_pitch * 1E3) logging.debug('Template width: %dpixels', template.shape[1]) logging.debug('Template height: %dpixels', template.shape[0]) chart_pixel_h = self._height * focal_l / (self._distance * pixel_pitch) scale_factor = template.shape[0] / chart_pixel_h if rotation == 90 or rotation == 270: # With the landscape to portrait override turned on, the width and height # of the active array, normally w x h, will be h x (w * (h/w)^2). Reduce # the applied scaling by the same factor to compensate for this, because # the chart will take up more of the scene. Assume w > h, since this is # meant for landscape sensors. rotate_physical_aspect = ( props['android.sensor.info.physicalSize']['height'] / props['android.sensor.info.physicalSize']['width']) scale_factor *= rotate_physical_aspect ** 2 logging.debug('Chart/image scale factor = %.2f', scale_factor) return template, img_3a, scale_factor def locate(self, cam, props, log_path, rotation): """Find the chart in the image, and append location to chart object. Args: cam: Open its session. props: Camera properties object. log_path: log path to store the captured images. rotation: clockwise rotation of template in degrees (multiple of 90) or None The values appended are: xnorm: float; [0, 1] left loc of chart in scene ynorm: float; [0, 1] top loc of chart in scene wnorm: float; [0, 1] width of chart in scene hnorm: float; [0, 1] height of chart in scene scale: float; scale factor to extract chart opt_val: float; The normalized match optimization value [0, 1] """ fmt = {'format': 'yuv', 'width': VGA_WIDTH, 'height': VGA_HEIGHT} cam.do_3a() chart, scene, s_factor = self._calc_scale_factors(cam, props, fmt, log_path, rotation) scale_start = self._scale_start * s_factor scale_stop = self._scale_stop * s_factor scale_step = self._scale_step * s_factor offset = scale_step / 2 self.scale = s_factor logging.debug('scale start: %.3f, stop: %.3f, step: %.3f', scale_start, scale_stop, scale_step) logging.debug('Used offset of %.3f to include stop value.', offset) max_match = [] # convert [0.0, 1.0] image to [0, 255] and then grayscale scene_uint8 = image_processing_utils.convert_image_to_uint8(scene) scene_gray = image_processing_utils.convert_rgb_to_grayscale(scene_uint8) # find scene logging.debug('Finding chart in scene...') for scale in numpy.arange(scale_start, scale_stop + offset, scale_step): scene_scaled = scale_img(scene_gray, scale) if (scene_scaled.shape[0] < chart.shape[0] or scene_scaled.shape[1] < chart.shape[1]): logging.debug( 'Skipped scale %.3f. scene_scaled shape: %s, chart shape: %s', scale, scene_scaled.shape, chart.shape) continue result = cv2.matchTemplate(scene_scaled, chart, cv2.TM_CCOEFF_NORMED) _, opt_val, _, top_left_scaled = cv2.minMaxLoc(result) logging.debug(' scale factor: %.3f, opt val: %.3f', scale, opt_val) max_match.append((opt_val, scale, top_left_scaled)) # determine if optimization results are valid opt_values = [x[0] for x in max_match] if not opt_values or (2.0 * min(opt_values) > max(opt_values)): estring = ('Warning: unable to find chart in scene!\n' 'Check camera distance and self-reported ' 'pixel pitch, focal length and hyperfocal distance.') logging.warning(estring) self._set_scale_factors_to_one() else: if (max(opt_values) == opt_values[0] or max(opt_values) == opt_values[len(opt_values) - 1]): estring = ('Warning: Chart is at extreme range of locator.') logging.warning(estring) # find max and draw bbox matched_scale_and_loc = max(max_match, key=lambda x: x[0]) self.opt_val = matched_scale_and_loc[0] self.scale = matched_scale_and_loc[1] logging.debug('Optimum scale factor: %.3f', self.scale) logging.debug('Opt val: %.3f', self.opt_val) top_left_scaled = matched_scale_and_loc[2] logging.debug('top_left_scaled: %d, %d', top_left_scaled[0], top_left_scaled[1]) h, w = chart.shape bottom_right_scaled = (top_left_scaled[0] + w, top_left_scaled[1] + h) logging.debug('bottom_right_scaled: %d, %d', bottom_right_scaled[0], bottom_right_scaled[1]) top_left = ((top_left_scaled[0] // self.scale), (top_left_scaled[1] // self.scale)) bottom_right = ((bottom_right_scaled[0] // self.scale), (bottom_right_scaled[1] // self.scale)) self.wnorm = ((bottom_right[0]) - top_left[0]) / scene.shape[1] self.hnorm = ((bottom_right[1]) - top_left[1]) / scene.shape[0] self.xnorm = (top_left[0]) / scene.shape[1] self.ynorm = (top_left[1]) / scene.shape[0] patch = image_processing_utils.get_image_patch( scene_uint8, self.xnorm, self.ynorm, self.wnorm, self.hnorm) / 255 image_processing_utils.write_image( patch, os.path.join(log_path, 'template_scene.jpg')) def component_shape(contour): """Measure the shape of a connected component. Args: contour: return from cv2.findContours. A list of pixel coordinates of the contour. Returns: The most left, right, top, bottom pixel location, height, width, and the center pixel location of the contour. """ shape = {'left': numpy.inf, 'right': 0, 'top': numpy.inf, 'bottom': 0, 'width': 0, 'height': 0, 'ctx': 0, 'cty': 0} for pt in contour: if pt[0][0] < shape['left']: shape['left'] = pt[0][0] if pt[0][0] > shape['right']: shape['right'] = pt[0][0] if pt[0][1] < shape['top']: shape['top'] = pt[0][1] if pt[0][1] > shape['bottom']: shape['bottom'] = pt[0][1] shape['width'] = shape['right'] - shape['left'] + 1 shape['height'] = shape['bottom'] - shape['top'] + 1 shape['ctx'] = (shape['left'] + shape['right']) // 2 shape['cty'] = (shape['top'] + shape['bottom']) // 2 return shape def find_circle_fill_metric(shape, img_bw, color): """Find the proportion of points matching a desired color on a shape's axes. Args: shape: dictionary returned by component_shape(...) img_bw: binarized numpy image array color: int of [0 or 255] 0 is black, 255 is white Returns: float: number of x, y axis points matching color / total x, y axis points """ matching = 0 total = 0 for y in range(shape['top'], shape['bottom']): total += 1 matching += 1 if img_bw[y][shape['ctx']] == color else 0 for x in range(shape['left'], shape['right']): total += 1 matching += 1 if img_bw[shape['cty']][x] == color else 0 logging.debug('Found %d matching points out of %d', matching, total) return matching / total def find_circle(img, img_name, min_area, color, use_adaptive_threshold=False): """Find the circle in the test image. Args: img: numpy image array in RGB, with pixel values in [0,255]. img_name: string with image info of format and size. min_area: float of minimum area of circle to find color: int of [0 or 255] 0 is black, 255 is white use_adaptive_threshold: True if binarization should use adaptive threshold. Returns: circle = {'x', 'y', 'r', 'w', 'h', 'x_offset', 'y_offset'} """ circle = {} img_size = img.shape if img_size[0]*img_size[1] >= LOW_RES_IMG_THRESH: circlish_atol = CIRCLISH_ATOL else: circlish_atol = CIRCLISH_LOW_RES_ATOL # convert to gray-scale image and binarize using adaptive/global threshold if use_adaptive_threshold: img_gray = cv2.cvtColor(img.astype(numpy.uint8), cv2.COLOR_BGR2GRAY) img_bw = cv2.adaptiveThreshold(img_gray, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, CV2_THRESHOLD_BLOCK_SIZE, CV2_THRESHOLD_CONSTANT) else: img_gray = image_processing_utils.convert_rgb_to_grayscale(img) img_bw = binarize_image(img_gray) # find contours contours = find_all_contours(255-img_bw) # Check each contour and find the circle bigger than min_area num_circles = 0 circle_contours = [] logging.debug('Initial number of contours: %d', len(contours)) min_circle_area = min_area * img_size[0] * img_size[1] logging.debug('Screening out circles w/ radius < %.1f (pixels) or %d pts.', math.sqrt(min_circle_area / math.pi), CIRCLE_MIN_PTS) for contour in contours: area = cv2.contourArea(contour) num_pts = len(contour) if (area > min_circle_area and num_pts >= CIRCLE_MIN_PTS): shape = component_shape(contour) radius = (shape['width'] + shape['height']) / 4 colour = img_bw[shape['cty']][shape['ctx']] circlish = (math.pi * radius**2) / area aspect_ratio = shape['width'] / shape['height'] fill = find_circle_fill_metric(shape, img_bw, color) logging.debug('Potential circle found. radius: %.2f, color: %d, ' 'circlish: %.3f, ar: %.3f, pts: %d, fill metric: %.3f', radius, colour, circlish, aspect_ratio, num_pts, fill) if (colour == color and math.isclose(1.0, circlish, abs_tol=circlish_atol) and math.isclose(1.0, aspect_ratio, abs_tol=CIRCLE_AR_ATOL) and num_pts/radius >= CIRCLE_RADIUS_NUMPTS_THRESH and math.isclose(1.0, fill, abs_tol=CIRCLE_COLOR_ATOL)): radii = [ image_processing_utils.distance( (shape['ctx'], shape['cty']), numpy.squeeze(point)) for point in contour ] minimum_radius, maximum_radius = min(radii), max(radii) logging.debug('Minimum radius: %.2f, maximum radius: %.2f', minimum_radius, maximum_radius) if circle: old_circle_center = (circle['x'], circle['y']) new_circle_center = (shape['ctx'], shape['cty']) # Based on image height center_distance_atol = img_size[0]*CIRCLE_LOCATION_VARIATION_RTOL if math.isclose( image_processing_utils.distance( old_circle_center, new_circle_center), 0, abs_tol=center_distance_atol ) and maximum_radius - minimum_radius < circle['radius_spread']: logging.debug('Replacing the previously found circle. ' 'Circle located at %s has a smaller radius spread ' 'than the previously found circle at %s. ' 'Current radius spread: %.2f, ' 'previous radius spread: %.2f', new_circle_center, old_circle_center, maximum_radius - minimum_radius, circle['radius_spread']) circle_contours.pop() circle = {} num_circles -= 1 circle_contours.append(contour) # Populate circle dictionary circle['x'] = shape['ctx'] circle['y'] = shape['cty'] circle['r'] = (shape['width'] + shape['height']) / 4 circle['w'] = float(shape['width']) circle['h'] = float(shape['height']) circle['x_offset'] = (shape['ctx'] - img_size[1]//2) / circle['w'] circle['y_offset'] = (shape['cty'] - img_size[0]//2) / circle['h'] circle['radius_spread'] = maximum_radius - minimum_radius logging.debug('Num pts: %d', num_pts) logging.debug('Aspect ratio: %.3f', aspect_ratio) logging.debug('Circlish value: %.3f', circlish) logging.debug('Location: %.1f x %.1f', circle['x'], circle['y']) logging.debug('Radius: %.3f', circle['r']) logging.debug('Circle center position wrt to image center: %.3fx%.3f', circle['x_offset'], circle['y_offset']) num_circles += 1 # if more than one circle found, break if num_circles == 2: break if num_circles == 0: image_processing_utils.write_image(img/255, img_name, True) if not use_adaptive_threshold: return find_circle( img, img_name, min_area, color, use_adaptive_threshold=True) else: raise AssertionError('No circle detected. ' 'Please take pictures according to instructions.') if num_circles > 1: image_processing_utils.write_image(img/255, img_name, True) cv2.drawContours(img, circle_contours, -1, CV2_RED, CV2_LINE_THICKNESS) img_name_parts = img_name.split('.') image_processing_utils.write_image( img/255, f'{img_name_parts[0]}_contours.{img_name_parts[1]}', True) if not use_adaptive_threshold: return find_circle( img, img_name, min_area, color, use_adaptive_threshold=True) raise AssertionError('More than 1 circle detected. ' 'Background of scene may be too complex.') return circle def append_circle_center_to_img(circle, img, img_name, save_img=True): """Append circle center and image center to image and save image. Draws line from circle center to image center and then labels end-points. Adjusts text positioning depending on circle center wrt image center. Moves text position left/right half of up/down movement for visual aesthetics. Args: circle: dict with circle location vals. img: numpy float image array in RGB, with pixel values in [0,255]. img_name: string with image info of format and size. save_img: optional boolean to not save image """ line_width_scaling_factor = 500 text_move_scaling_factor = 3 img_size = img.shape img_center_x = img_size[1]//2 img_center_y = img_size[0]//2 # draw line from circle to image center line_width = int(max(1, max(img_size)//line_width_scaling_factor)) font_size = line_width // 2 move_text_dist = line_width * text_move_scaling_factor cv2.line(img, (circle['x'], circle['y']), (img_center_x, img_center_y), CV2_RED, line_width) # adjust text location move_text_right_circle = -1 move_text_right_image = 2 if circle['x'] > img_center_x: move_text_right_circle = 2 move_text_right_image = -1 move_text_down_circle = -1 move_text_down_image = 4 if circle['y'] > img_center_y: move_text_down_circle = 4 move_text_down_image = -1 # add circles to end points and label radius_pt = line_width * 2 # makes a dot 2x line width filled_pt = -1 # cv2 value for a filled circle # circle center cv2.circle(img, (circle['x'], circle['y']), radius_pt, CV2_RED, filled_pt) text_circle_x = move_text_dist * move_text_right_circle + circle['x'] text_circle_y = move_text_dist * move_text_down_circle + circle['y'] cv2.putText(img, 'circle center', (text_circle_x, text_circle_y), cv2.FONT_HERSHEY_SIMPLEX, font_size, CV2_RED, line_width) # image center cv2.circle(img, (img_center_x, img_center_y), radius_pt, CV2_RED, filled_pt) text_imgct_x = move_text_dist * move_text_right_image + img_center_x text_imgct_y = move_text_dist * move_text_down_image + img_center_y cv2.putText(img, 'image center', (text_imgct_x, text_imgct_y), cv2.FONT_HERSHEY_SIMPLEX, font_size, CV2_RED, line_width) if save_img: image_processing_utils.write_image(img/255, img_name, True) # [0, 1] values def is_circle_cropped(circle, size): """Determine if a circle is cropped by edge of image. Args: circle: list [x, y, radius] of circle size: tuple (x, y) of size of img Returns: Boolean True if selected circle is cropped """ cropped = False circle_x, circle_y = circle[0], circle[1] circle_r = circle[2] x_min, x_max = circle_x - circle_r, circle_x + circle_r y_min, y_max = circle_y - circle_r, circle_y + circle_r if x_min < 0 or y_min < 0 or x_max > size[0] or y_max > size[1]: cropped = True return cropped def find_white_square(img, min_area): """Find the white square in the test image. Args: img: numpy image array in RGB, with pixel values in [0,255]. min_area: float of minimum area of circle to find Returns: square = {'left', 'right', 'top', 'bottom', 'width', 'height'} """ square = {} num_squares = 0 img_size = img.shape # convert to gray-scale image img_gray = image_processing_utils.convert_rgb_to_grayscale(img) # otsu threshold to binarize the image img_bw = binarize_image(img_gray) # find contours contours = find_all_contours(img_bw) # Check each contour and find the square bigger than min_area logging.debug('Initial number of contours: %d', len(contours)) min_area = img_size[0]*img_size[1]*min_area logging.debug('min_area: %.3f', min_area) for contour in contours: area = cv2.contourArea(contour) num_pts = len(contour) if (area > min_area and num_pts >= 4): shape = component_shape(contour) squarish = (shape['width'] * shape['height']) / area aspect_ratio = shape['width'] / shape['height'] logging.debug('Potential square found. squarish: %.3f, ar: %.3f, pts: %d', squarish, aspect_ratio, num_pts) if (math.isclose(1.0, squarish, abs_tol=SQUARISH_RTOL) and math.isclose(1.0, aspect_ratio, abs_tol=SQUARISH_AR_RTOL)): # Populate square dictionary angle = cv2.minAreaRect(contour)[-1] if angle < -45: angle += 90 square['angle'] = angle square['left'] = shape['left'] - SQUARE_CROP_MARGIN square['right'] = shape['right'] + SQUARE_CROP_MARGIN square['top'] = shape['top'] - SQUARE_CROP_MARGIN square['bottom'] = shape['bottom'] + SQUARE_CROP_MARGIN square['w'] = shape['width'] + 2*SQUARE_CROP_MARGIN square['h'] = shape['height'] + 2*SQUARE_CROP_MARGIN num_squares += 1 if num_squares == 0: raise AssertionError('No white square detected. ' 'Please take pictures according to instructions.') if num_squares > 1: raise AssertionError('More than 1 white square detected. ' 'Background of scene may be too complex.') return square def get_angle(input_img): """Computes anglular inclination of chessboard in input_img. Args: input_img (2D numpy.ndarray): Grayscale image stored as a 2D numpy array. Returns: Median angle of squares in degrees identified in the image. Angle estimation algorithm description: Input: 2D grayscale image of chessboard. Output: Angle of rotation of chessboard perpendicular to chessboard. Assumes chessboard and camera are parallel to each other. 1) Use adaptive threshold to make image binary 2) Find countours 3) Filter out small contours 4) Filter out all non-square contours 5) Compute most common square shape. The assumption here is that the most common square instances are the chessboard squares. We've shown that with our current tuning, we can robustly identify the squares on the sensor fusion chessboard. 6) Return median angle of most common square shape. USAGE NOTE: This function has been tuned to work for the chessboard used in the sensor_fusion tests. See images in test_images/rotated_chessboard/ for sample captures. If this function is used with other chessboards, it may not work as expected. """ # Tuning parameters square_area_min = (float)(input_img.shape[1] * SQUARE_AREA_MIN_REL) # Creates copy of image to avoid modifying original. img = numpy.array(input_img, copy=True) # Scale pixel values from 0-1 to 0-255 img_uint8 = image_processing_utils.convert_image_to_uint8(img) img_thresh = cv2.adaptiveThreshold( img_uint8, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 201, 2) # Find all contours. contours = find_all_contours(img_thresh) # Filter contours to squares only. square_contours = [] for contour in contours: rect = cv2.minAreaRect(contour) _, (width, height), angle = rect # Skip non-squares if not math.isclose(width, height, rel_tol=SQUARE_TOL): continue # Remove very small contours: usually just tiny dots due to noise. area = cv2.contourArea(contour) if area < square_area_min: continue square_contours.append(contour) areas = [] for contour in square_contours: area = cv2.contourArea(contour) areas.append(area) median_area = numpy.median(areas) filtered_squares = [] filtered_angles = [] for square in square_contours: area = cv2.contourArea(square) if not math.isclose(area, median_area, rel_tol=SQUARE_TOL): continue filtered_squares.append(square) _, (width, height), angle = cv2.minAreaRect(square) filtered_angles.append(angle) if len(filtered_angles) < ANGLE_NUM_MIN: logging.debug( 'A frame had too few angles to be processed. ' 'Num of angles: %d, MIN: %d', len(filtered_angles), ANGLE_NUM_MIN) return None return numpy.median(filtered_angles) def correct_faces_for_crop(faces, img, crop): """Correct face rectangles for sensor crop. Args: faces: list of dicts with face information relative to sensor's aspect ratio img: np image array crop: dict of crop region size with 'top', 'right', 'left', 'bottom' as keys to desired region of the sensor to read out Returns: list of face locations (left, right, top, bottom) corrected """ faces_corrected = [] crop_w = crop['right'] - crop['left'] crop_h = crop['bottom'] - crop['top'] logging.debug('crop region: %s', str(crop)) img_w, img_h = img.shape[1], img.shape[0] crop_aspect_ratio = crop_w / crop_h img_aspect_ratio = img_w / img_h for rect in [face['bounds'] for face in faces]: logging.debug('rect: %s', str(rect)) if crop_aspect_ratio >= img_aspect_ratio: # Sensor width is being cropped, so we need to adjust the horizontal # coordinates of the face rectangles to account for the crop. # Since we are converting from sensor coordinates to image coordinates img_crop_h_ratio = img_h / crop_h scaled_crop_w = crop_w * img_crop_h_ratio excess_w = (img_w - scaled_crop_w) / 2 left = int( round((rect['left'] - crop['left']) * img_crop_h_ratio + excess_w)) right = int( round((rect['right'] - crop['left']) * img_crop_h_ratio + excess_w)) top = int(round((rect['top'] - crop['top']) * img_crop_h_ratio)) bottom = int(round((rect['bottom'] - crop['top']) * img_crop_h_ratio)) else: # Sensor height is being cropped, so we need to adjust the vertical # coordinates of the face rectangles to account for the crop. img_crop_w_ratio = img_w / crop_w scaled_crop_h = crop_h * img_crop_w_ratio excess_w = (img_h - scaled_crop_h) / 2 left = int(round((rect['left'] - crop['left']) * img_crop_w_ratio)) right = int(round((rect['right'] - crop['left']) * img_crop_w_ratio)) top = int( round((rect['top'] - crop['top']) * img_crop_w_ratio + excess_w)) bottom = int( round((rect['bottom'] - crop['top']) * img_crop_w_ratio + excess_w)) faces_corrected.append([left, right, top, bottom]) logging.debug('faces_corrected: %s', str(faces_corrected)) return faces_corrected def eliminate_duplicate_centers(coordinates_list): """Checks center coordinates of OpenCV's face rectangles. Method makes sure that the list of face rectangles' centers do not contain duplicates from the same face Args: coordinates_list: list; coordinates of face rectangles' centers Returns: non_duplicate_list: list; coordinates of face rectangles' centers without duplicates on the same face """ output = set() for _, xy1 in enumerate(coordinates_list): for _, xy2 in enumerate(coordinates_list): if scipy.spatial.distance.euclidean(xy1, xy2) < FACE_MIN_CENTER_DELTA: continue if xy1 not in output: output.add(xy1) else: output.add(xy2) return list(output) def match_face_locations(faces_cropped, faces_opencv, img, img_name): """Assert face locations between two methods. Method determines if center of opencv face boxes is within face detection face boxes. Using math.hypot to measure the distance between the centers, as math.dist is not available for python versions before 3.8. Args: faces_cropped: list of lists with (l, r, t, b) for each face. faces_opencv: list of lists with (x, y, w, h) for each face. img: numpy [0, 1] image array img_name: text string with path to image file """ # turn faces_opencv into list of center locations faces_opencv_center = [(x+w//2, y+h//2) for (x, y, w, h) in faces_opencv] cropped_faces_centers = [ ((l+r)//2, (t+b)//2) for (l, r, t, b) in faces_cropped] faces_opencv_center.sort(key=lambda t: [t[1], t[0]]) cropped_faces_centers.sort(key=lambda t: [t[1], t[0]]) logging.debug('cropped face centers: %s', str(cropped_faces_centers)) logging.debug('opencv face center: %s', str(faces_opencv_center)) faces_opencv_centers = [] num_centers_aligned = 0 # eliminate duplicate openCV face rectangles' centers the same face faces_opencv_centers = eliminate_duplicate_centers(faces_opencv_center) logging.debug('opencv face centers: %s', str(faces_opencv_centers)) for (x, y) in faces_opencv_centers: for (x1, y1) in cropped_faces_centers: centers_dist = math.hypot(x-x1, y-y1) if centers_dist < FACE_CENTER_MIN_LOGGING_DIST: logging.debug('centers_dist: %.3f', centers_dist) if (abs(x-x1) < FACE_CENTER_MATCH_TOL_X and abs(y-y1) < FACE_CENTER_MATCH_TOL_Y): num_centers_aligned += 1 # If test failed, save image with green AND OpenCV red rectangles image_processing_utils.write_image(img, img_name) if num_centers_aligned < FACES_ALIGNED_MIN_NUM: for (x, y, w, h) in faces_opencv: cv2.rectangle(img, (x, y), (x+w, y+h), CV2_RED_NORM, 2) image_processing_utils.write_image(img, img_name) logging.debug('centered: %s', str(num_centers_aligned)) raise AssertionError(f'Face rectangles in wrong location(s)!. ' f'Found {num_centers_aligned} rectangles near cropped ' f'face centers, expected {FACES_ALIGNED_MIN_NUM}') def draw_green_boxes_around_faces(img, faces_cropped, img_name): """Correct face rectangles for sensor crop. Args: img: numpy [0, 1] image array faces_cropped: list of lists with (l, r, t, b) for each face img_name: text string with path to image file Returns: image with green rectangles """ # draw boxes around faces in green and save image for (l, r, t, b) in faces_cropped: cv2.rectangle(img, (l, t), (r, b), CV2_GREEN_NORM, 2) image_processing_utils.write_image(img, img_name) def find_aruco_markers(input_img, output_img_path): """Detects ArUco markers in the input_img. Finds ArUco markers in the input_img and draws the contours around them. Args: input_img: input img in numpy array with ArUco markers to be detected output_img_path: path of the image to be saved with contours around the markers detected Returns: corners: list of detected corners ids: list of int ids for each ArUco markers in the input_img rejected_params: list of rejected corners """ parameters = cv2.aruco.DetectorParameters_create() # ArUco markers used are 4x4 aruco_dict = cv2.aruco.getPredefinedDictionary(cv2.aruco.DICT_4X4_100) corners, ids, rejected_params = cv2.aruco.detectMarkers( input_img, aruco_dict, parameters=parameters) # Early return if sufficient markers found if ids is not None and len(ids) >= ARUCO_CORNER_COUNT: logging.debug('All ArUco markers detected.') cv2.aruco.drawDetectedMarkers(input_img, corners, ids) image_processing_utils.write_image(input_img / 255, output_img_path) return corners, ids, rejected_params # Try with high-contrast greyscale if needed logging.debug('Trying ArUco marker detection with greyscale image.') bw_img = convert_image_to_high_contrast_black_white(input_img) corners, ids, rejected_params = cv2.aruco.detectMarkers( bw_img, aruco_dict, parameters=parameters) if ids is not None and len(ids) >= ARUCO_CORNER_COUNT: logging.debug('All ArUco markers detected with greyscale image.') # Handle case where no markers are found if ids is None: image_processing_utils.write_image(input_img/255, output_img_path) raise AssertionError('ArUco markers not detected.') # Log and save results logging.debug('Number of ArUco markers detected w/ greyscale: %d', len(ids)) logging.debug('IDs of the ArUco markers detected: %s', ids) logging.debug('Corners of the ArUco markers detected: %s', corners) cv2.aruco.drawDetectedMarkers(bw_img, corners, ids) image_processing_utils.write_image(bw_img / 255, output_img_path) return corners, ids, rejected_params def get_patch_from_aruco_markers( input_img, aruco_marker_corners, aruco_marker_ids): """Returns the rectangle patch from the aruco marker corners. Note: Refer to image used in scene7 for ArUco markers location. Args: input_img: input img in numpy array with ArUco markers to be detected aruco_marker_corners: array of aruco marker corner coordinates detected by opencv_processing_utils.find_aruco_markers aruco_marker_ids: array of ids of aruco markers detected by opencv_processing_utils.find_aruco_markers Returns: Numpy float image array of the rectangle patch """ outer_rect_coordinates = {} for corner, marker_id in zip(aruco_marker_corners, aruco_marker_ids): corner = corner.reshape(4, 2) # opencv returns 3D array index = marker_id[0] # Roll the array 4x to align with the coordinates of the corner adjacent # to the corner of the rectangle # Marker id: 0 => index 2 coordinates # Marker id: 1 => index 3 coordinates # Marker id: 2 => index 0 coordinates # Marker id: 3 => index 1 coordinates corner = numpy.roll(corner, 4) outer_rect_coordinates[index] = tuple(corner[index]) red_corner = tuple(map(int, outer_rect_coordinates[0])) green_corner = tuple(map(int, outer_rect_coordinates[1])) gray_corner = tuple(map(int, outer_rect_coordinates[2])) blue_corner = tuple(map(int, outer_rect_coordinates[3])) logging.debug('red_corner: %s', red_corner) logging.debug('blue_corner: %s', blue_corner) logging.debug('green_corner: %s', green_corner) logging.debug('gray_corner: %s', gray_corner) # Ensure that the image is not rotated blue_gray_y_diff = abs(gray_corner[1] - blue_corner[1]) red_green_y_diff = abs(green_corner[1] - red_corner[1]) if ((blue_gray_y_diff > IMAGE_ROTATION_THRESHOLD) or (red_green_y_diff > IMAGE_ROTATION_THRESHOLD)): raise AssertionError('Image rotation is not within the threshold. ' f'Actual blue_gray_y_diff: {blue_gray_y_diff}, ' f'red_green_y_diff: {red_green_y_diff} ' f'Expected {IMAGE_ROTATION_THRESHOLD}') cv2.rectangle(input_img, red_corner, gray_corner, CV2_RED_NORM, CV2_LINE_THICKNESS) return input_img[red_corner[1]:gray_corner[1], red_corner[0]:gray_corner[0]].copy() def get_chart_boundary_from_aruco_markers( aruco_marker_corners, aruco_marker_ids, input_img, output_img_path): """Returns top left and bottom right coordinates from the aruco markers. Note: Refer to image used in scene8 for ArUco markers location. Args: aruco_marker_corners: array of aruco marker corner coordinates detected by opencv_processing_utils.find_aruco_markers. aruco_marker_ids: array of ids of aruco markers detected by opencv_processing_utils.find_aruco_markers. input_img: 3D RGB numpy [0, 255] uint8; input image. output_img_path: string; output image path. Returns: top_left: tuple; aruco marker corner coordinates in pixel. top_right: tuple; aruco marker corner coordinates in pixel. bottom_right: tuple; aruco marker corner coordinates in pixel. bottom_left: tuple; aruco marker corner coordinates in pixel. """ outer_rect_coordinates = {} for corner, marker_id in zip(aruco_marker_corners, aruco_marker_ids): corner = corner.reshape(4, 2) # reshape opencv 3D array to 4x2 index = marker_id[0] corner = numpy.roll(corner, ARUCO_CORNER_COUNT) outer_rect_coordinates[index] = tuple(corner[index]) logging.debug('Corners: %s', corner) logging.debug('Index: %s', index) logging.debug('Outer rect coordinates: %s', outer_rect_coordinates[index]) top_left = tuple(map(int, outer_rect_coordinates[0])) top_right = tuple(map(int, outer_rect_coordinates[1])) bottom_right = tuple(map(int, outer_rect_coordinates[2])) bottom_left = tuple(map(int, outer_rect_coordinates[3])) # Outline metering rectangles with corresponding colors rect_w = round((bottom_right[0] - top_left[0])/NUM_AE_AWB_REGIONS) top_x, top_y = top_left[0], top_left[1] bottom_x, bottom_y = bottom_left[0], bottom_left[1] cv2.rectangle( input_img, (top_x, top_y), (bottom_x + rect_w, bottom_y), CV2_BLUE, CV2_LINE_THICKNESS) cv2.rectangle( input_img, (top_x + rect_w, top_y), (bottom_x + rect_w * 2, bottom_y), CV2_WHITE, CV2_LINE_THICKNESS) cv2.rectangle( input_img, (top_x + rect_w * 2, top_y), (bottom_x + rect_w * 3, bottom_y), CV2_BLACK, CV2_LINE_THICKNESS) cv2.rectangle( input_img, (top_x + rect_w * 3, top_y), bottom_right, CV2_YELLOW, CV2_LINE_THICKNESS) image_processing_utils.write_image(input_img/255, output_img_path) logging.debug('ArUco marker top_left: %s', top_left) logging.debug('ArUco marker bottom_right: %s', bottom_right) return top_left, top_right, bottom_right, bottom_left def define_metering_rectangle_values( props, top_left, top_right, bottom_right, bottom_left, w, h): """Find normalized values of coordinates and return 4 metering rects. Args: props: dict; camera properties object. top_left: coordinates; defined by aruco markers for targeted image. top_right: coordinates; defined by aruco markers for targeted image. bottom_right: coordinates; defined by aruco markers for targeted image. bottom_left: coordinates; defined by aruco markers for targeted image. w: int; active array width in pixels. h: int; active array height in pixels. Returns: meter_rects: 4 metering rectangles made of (x, y, width, height, weight). x, y are the top left coordinate of the metering rectangle. """ # If testing front camera, mirror coordinates either left/right or up/down # Preview are flipped on device's natural orientation # For sensor orientation 90 or 270, it is up or down # For sensor orientation 0 or 180, it is left or right if (props['android.lens.facing'] == camera_properties_utils.LENS_FACING['FRONT']): if props['android.sensor.orientation'] in (90, 270): tl_coordinates = (bottom_left[0], h - bottom_left[1]) br_coordinates = (top_right[0], h - top_right[1]) logging.debug('Found sensor orientation %d, flipping up down', props['android.sensor.orientation']) else: tl_coordinates = (w - top_right[0], top_right[1]) br_coordinates = (w - bottom_left[0], bottom_left[1]) logging.debug('Found sensor orientation %d, flipping left right', props['android.sensor.orientation']) logging.debug('Mirrored top-left coordinates: %s', tl_coordinates) logging.debug('Mirrored bottom-right coordinates: %s', br_coordinates) else: tl_coordinates, br_coordinates = top_left, bottom_right # Normalize coordinates' values to construct metering rectangles meter_rects = [] tl_normalized_x = tl_coordinates[0] / w tl_normalized_y = tl_coordinates[1] / h br_normalized_x = br_coordinates[0] / w br_normalized_y = br_coordinates[1] / h rect_w = round((br_normalized_x - tl_normalized_x) / NUM_AE_AWB_REGIONS, 2) rect_h = round(br_normalized_y - tl_normalized_y, 2) for i in range(NUM_AE_AWB_REGIONS): x = round(tl_normalized_x + (rect_w * i), 2) y = round(tl_normalized_y, 2) meter_rect = [x, y, rect_w, rect_h, AE_AWB_METER_WEIGHT] meter_rects.append(meter_rect) logging.debug('metering rects: %s', meter_rects) return meter_rects def convert_image_to_high_contrast_black_white( img, contrast=CV2_CONTRAST_ALPHA, brightness=CV2_CONTRAST_BETA): """Convert capture to high contrast black and white image. Args: img: numpy array of image. contrast: gain parameter between the value of 0 to 3. brightness: bias parameter between the value of 1 to 100. Returns: high_contrast_img: high contrast black and white image. """ copy_img = numpy.ndarray.copy(img) uint8_img = image_processing_utils.convert_image_to_uint8(copy_img) gray_img = convert_to_y(uint8_img) img_bw = cv2.convertScaleAbs( gray_img, alpha=contrast, beta=brightness) _, high_contrast_img = cv2.threshold( numpy.uint8(img_bw), CV2_THESHOLD_LOWER_BLACK, CH_FULL_SCALE, cv2.THRESH_BINARY + cv2.THRESH_OTSU ) high_contrast_img = numpy.expand_dims( (CH_FULL_SCALE - high_contrast_img), axis=2) return high_contrast_img