Esempio n. 1
0
def interpolate_dominate_lines(signals: dict,
                               interpolations: dict,
                               mean_slope: float,
                               lower_bound: int,
                               upper_bound: int,
                               horizontal_limit: int,
                               slope_variance=SLOPE_VARIANCE):
    '''
    Interpolates lines based on ROI mask and mean_slope
    :param signals: dominate signals in image
    :param mean_slope: mean of dominate signals
    :param lower_bound: int lower y value in image
    :param upper_bound: int upper y value in image
    :param horizontal_limit int maximum possible x-value
    :param slope_variance: acceptable slope variance
    '''
    try:
        for _line in signals:
            _slope = signals[_line]['slope']
            _offset = signals[_line]['offset']
            x1, y1 = signals[_line]['p1']
            x2, y2 = signals[_line]['p2']
            if not _slope:
                continue
            abs_slope = abs(_slope)
            delta = abs(abs_slope - mean_slope)
            if delta < slope_variance:
                logger.debug('%s\t %s', (x1, y1), (x2, y2))
                assert _slope != 0, 'slope is zero in signals'
                new_p1 = (int((lower_bound - _offset) / _slope), lower_bound)
                new_p2 = (int((upper_bound - _offset) / _slope), upper_bound)
                is_valid = True
                for _x, _y in [new_p1, new_p2]:
                    if _x < 0:
                        logger.warning(
                            'line extends too far left, throwing out %s',
                            (_x, _y))
                        is_valid = False
                    elif _x > horizontal_limit:
                        logger.warning(
                            'line extends too far right, throwing out %s',
                            (_x, _y))
                        is_valid = False
                if is_valid:
                    interpolations[_line] = {
                        'slope': _slope,
                        'offset': _offset,
                        'p1': new_p1,
                        'p2': new_p2
                    }
    except Exception as err:
        logger.error('interpolation error: %s', err)
Esempio n. 2
0
 def draw_lines(self,
                lines: ndarray,
                image=None,
                color=None,
                thickness=2) -> ndarray:
     '''
     Lines are drawn on the image inplace.
     """
     :param image:     image to apply lines
     :param lines:     numpy.ndarray of tuple (x1,y1,x2,y2)
     :param color:     tuple (r, g, b) uint8
     :param thickness: pixel width of highlight
     :return <numpy.ndarray>
     '''
     if image is not None:
         assert image.shape == self.image.shape, 'images must be same shape, to draw lines'
         self.image_tf = image
     y_height, x_width, channels = self.image.shape
     self.image_tf = zeros((y_height, x_width, channels), dtype=uint8)
     # assign default color
     if color is None:
         color = [255, 0, 0]
     # draw lines
     logger.debug(
         '-------------------------------------------------------------')
     # use accumulated signals
     region_mask = array(self.get_roi_mask(), dtype=bool)
     upper_bound = int(y_height - 1)
     lower_bound = int(upper_bound / 2 + self.Y_OFFSET)
     horizontal_limit = (x_width - 1)
     # quantify signals
     mean_slope = find_dominate_signals(lines, self.roi_filter_lines,
                                        region_mask)
     mean_slope = find_mean_slope(self.roi_filter_lines, mean_slope)
     # extend lines into lanes
     interpolate_dominate_lines(self.roi_filter_lines,
                                self.slope_filter_lines, mean_slope,
                                lower_bound, upper_bound, horizontal_limit)
     self.right_lane, self.left_lane = convert_lane_edges_to_polygons(
         self.slope_filter_lines, lower_bound, upper_bound)
     # fill lane polygons on images
     if self.right_lane.any():
         fillPoly(self.image_tf, int32([self.right_lane]), color)
     else:
         logger.error('did not create right-side lane')
     if self.left_lane.any():
         fillPoly(self.image_tf, int32([self.left_lane]), color)
     else:
         logger.error('did not create left-side lane')
     return self.image_tf
Esempio n. 3
0
def convert_lane_edges_to_polygons(
        edges: dict,
        lower_bound: int,
        upper_bound: int,
        lower_x_offset=LOWER_X_OFFSET,
        upper_x_offset=UPPER_X_OFFSET) -> (array, array):
    '''
    Converts interpolated lines lanes into lane polygons
    :param edges: dict of line info
    :param lower_bound: lower y value in image
    :param upper_bound: upper y value in image
    :return: <tuple> (left_lane: array, right_lane: array)
    '''
    left = {}
    right = {}

    for _line in edges:
        _slope = edges[_line]['slope']
        if _slope < 0:
            right[_line] = edges[_line]
        else:
            left[_line] = edges[_line]

    if left:
        mean_left_lower, mean_left_upper = get_point_stats(left, lower_bound)
        poly_left = array([((mean_left_lower - lower_x_offset), lower_bound),
                           ((mean_left_upper - upper_x_offset), upper_bound),
                           ((mean_left_upper + upper_x_offset), upper_bound),
                           ((mean_left_lower + lower_x_offset), lower_bound)])
    else:
        poly_left = array(None)

    if right:
        mean_right_lower, mean_right_upper = get_point_stats(
            right, lower_bound)
        poly_right = array([((mean_right_lower - lower_x_offset), lower_bound),
                            ((mean_right_upper - upper_x_offset), upper_bound),
                            ((mean_right_upper + upper_x_offset), upper_bound),
                            ((mean_right_lower + lower_x_offset), lower_bound)
                            ])
    else:
        poly_right = array(None)

    logger.debug('polygons (left-lane, right-lane):\n(array(%s),\narray(%s))',
                 poly_left, poly_right)
    return poly_left, poly_right
Esempio n. 4
0
def find_dominate_signals(lines: ndarray,
                          signals: dict,
                          region_mask: array,
                          slope_max_cutoff=SLOPE_MAX_CUTOFF,
                          slope_thresh=SLOPE_THRESHOLD,
                          magnitude_thresh=MAGNITUDE_THRESH) -> float:
    '''
    Filters subset of dominate signals in line segments and returns mean slope
    :param lines: <numpy.ndarray> line segments
    :param signals dict filtered by one point valid in ROI
    :param region_mask: ROI mask shape
    :param slope_thresh: filters by slope general lane pitch
    :param magnitude_thresh: filters lines by dominant signal length
    :return: mean_slope: int of dominant signals
    '''
    i = 0
    max_slope = 0.0
    min_slope = 0.0
    max_signal = 0
    x_valid = 0
    try:
        for _line in lines:
            for x1, y1, x2, y2 in _line:
                if x1 == x2 or y1 == y2:
                    logger.warning('disregarding: %s, %s', (x1, y1), (x2, y2))
                    continue
                if not valid_within_fov([(x1, y1), (x2, y2)], region_mask):
                    logger.debug('invalid in FOV, thowing out %s',
                                 (x1, y1, x2, y2))
                    continue
                slope = (y2 - y1) / (x2 - x1)
                assert slope != 0, 'slope is zero {0}'.format(_line)
                offset = y1 - (x1 * slope)
                magnitude = sqrt((square(x2 - x1) + square(y2 - y1)))
                abs_slope = abs(slope)
                if magnitude > magnitude_thresh and abs_slope > slope_thresh and abs_slope < slope_max_cutoff:
                    if min_slope == 0:
                        min_slope = abs_slope
                    else:
                        min_slope = min(abs_slope, min_slope)
                    logger.debug(
                        'points: %s\t slope: %6.3f\t magnitude: %6.3f',
                        (x1, y1, x2, y2), slope, magnitude)
                    if magnitude > max_signal:
                        max_slope = abs(slope)
                        max_signal = magnitude
                signals[i] = {
                    'slope': slope,
                    'offset': offset,
                    'magnitude': magnitude,
                    'p1': (x1, y1),
                    'p2': (x2, y2)
                }
                i = i + 1
    except Exception as err:
        logger.error('bad signal: %s in %s', err, signals)
    mean_slope = (max_slope + min_slope) / 2
    logger.debug('mean_slope: %s, max: %s, min: %s, max_signal: %s',
                 mean_slope, max_slope, min_slope, max_signal)
    return mean_slope
Esempio n. 5
0
def get_slope_stats(slopes: list, threshold=SLOPE_THRESHOLD) -> dict:
    '''
    Gives statistics on slope variances
    :param slopes: list of all slopes
    :return: <list> [lane_label, min, max, mean, std]
    '''
    _ret = {}
    try:
        for _slopes in slopes:
            _min = -threshold
            _max = threshold
            _lane = '?'
            _stdev = []
            if _slopes:
                for _slope in _slopes:
                    if not _slope:
                        continue
                    elif _slope < 0:
                        _lane = 'right'
                    else:
                        _lane = 'left'
                    _min = min(_slope, _min)
                    _max = max(_slope, _max)
                assert len(_slopes) != 0, 'no slopes found'
                _mean = sum(_slopes) / len(_slopes)
                for _slope in _slopes:
                    _stdev.append(square(_slope - _mean))
                _std = sum(_stdev) / len(_slopes)
                _ret[_lane] = {
                    'min': _min,
                    'max': _max,
                    'mean': _mean,
                    'std': _std
                }
            else:
                logger.debug('no slopes in input')
    except Exception as err:
        logger.error('bad stats: %s in %s', err, slopes)
    return _ret