DE4108961A1 - Method and device for detecting a web in an image - Google Patents

Abstract

To detect a path in an aerial image (visible image or infrared image) or in a radar image, the following steps are carried out: contour lines, which are characteristic of the web to be detected and its surroundings, are assigned to straight line segments which are designated by individual and relative attributes, which results from description data for the respective location that an operator issues to create a file of segment data Fd; Processing of the image in which the web is to be recognized in order to extract straight line segments from which it approximates its contour lines, which are denoted by individual and relative attributes forming a file of segment measurands Fm; and - comparing the set of elements of the file Fd with the elements of the file Fm to match the searched path in the image if they match.

Description

The This invention relates generally to the field of image processing, in particular Detecting a web in an aerial view (visible or infrared) or a radar image.

Of the Invention is based on the object, the detection of a web in an aerial photograph in such a way that they relatively little by detail changes of the environment one Railway is impaired, with a slight application is sought with relatively straightforward processing means.

• – Zuordnung von Konturlinien, die kennzeichnend für die zu erkennende Bahn und ihre Umgebung sind, zu Geradensegmenten, welche durch individuelle Attribute wie ihre Länge und die Lage ihrer Endpunkte bezeichnet sind und die relativ sind, wie ihre relativen Orientierungen und Abstände, die aus Beschreibungsdaten für den Ort resultieren, die von einer Bedienungsperson ausgegeben werden und eine Datei von Segmentdaten bilden; diese Beschreibung muß nicht erschöpfend sein, um die Identifizierung der Bahn zu ermöglichen. Für bestimmte Orte kann eine summarische Beschreibung ausreichen, die beispielsweise nur die Breite der Bahn beinhaltet;
• – Verarbeitung des Bildes, in welchem die Bahn erkannt werden soll, um daraus eine Datei von Segmentmeßwerten zu gewinnen, mit folgenden Schritten:
• – die Konturlinien des Bildes werden extrahiert;
• – die Hauptrichtungen der Konturlinien werden bestimmt durch Anwendung der Hough-Transformierten und Auswahl von Akkumulationspunkten, die in der Transformierten der Bildkonturlinien erscheinen;
• – eine Approximierung der Konturlinien mit geraden Segmenten entlang den Hauptrichtungen; und
• – Erzeugung einer Datei von Segment-Meßwerten, worin die Segmente erfaßt sind, welche die Bildkonturlinien approximieren, die durch individuelle und relative Attribute gleicher Art wie die der Datei von Segmentdaten definiert sind; und
• – Vergleichen der Menge von Elementen der Datei von Segmentdaten mit den Elementen der Datei von Segment-Meßwerten, um bei Übereinstimmung zur Erkennung der gesuchten Bahn im Bild zu gelangen.

The invention relates to a method for detecting a web in an image, with the following successive steps:

• Alignment of contour lines which are characteristic of the track and its surroundings to be recognized, to straight line segments which are designated by individual attributes such as their length and the location of their endpoints and which are relative, such as their relative orientations and distances, from description data for result in the location being output by an operator and forming a file of segment data; this description need not be exhaustive to allow identification of the web. For certain locations, a summary description may suffice, including, for example, only the width of the web;
• Processing of the image in which the web is to be recognized in order to obtain a file of segment measured values, comprising the following steps:
• The contour lines of the image are extracted;
• The principal directions of the contour lines are determined by applying the Hough transform and selecting accumulation points that appear in the transform of the image contour lines;
• An approximation of the contour lines with straight segments along the main directions; and
• - generating a file of segment measurements, wherein the segments are detected which approximate the image contour lines defined by individual and relative attributes of the same kind as the file of segment data; and
• Comparing the set of elements of the file of segment data with the elements of the file of segment measured values, in order to arrive at the match in recognition of the searched trajectory in the image.

object The invention is further an apparatus for implementation this method of recognizing a web in an image.

Further Features and advantages of the invention will become apparent from the following Description and from the drawing, to which reference is made. In the drawing show:

1 a flow chart of the method according to the invention;

2 an overview diagram for illustrating the method for segmenting an image, for approximating the image contour lines by straight segments in the course of 1 illustrated method for detecting a web;

3 a scheme showing the relationships between the polar coordinates and the Cartesian coordinates of a straight line;

4 representing a point on a contour line having an orientation gradient that is within a certain angular interval, as seen by the Hough transform; and

5 a graph formed of nodes and arcs describing straight line segments approximating the image contour lines and the relative positions of these segments.

The Invention deals with. a method of detection a trajectory in a picture based on parameters, by the dimensions and relative positions of straight segments are formed, which approximate the contour lines that characterizing for the searched railway and their environment are.

The 1 is a general functional diagram for explaining the recognition method. Conventionally, the data and / or results are indicated in trapezoids, while the phases of the process, ie the various processing steps, are represented by rectangles.

The method essentially comprises three steps:
In a first step, the operator provides description data for the web to be recognized in the form of a more or less complete file of data Fd, including attributes defining straight line segments which approximate the contour lines of the web to be recognized and its surroundings. These attributes, which will be discussed in more detail below, may for a straight line segment be the object type to which it relates: the orbit, or its environment, its length, its angular orientation, its distances and its relative orientations with respect to other segments, etc.

In a second step, the image Im is subjected to segmentation processing which through a rectangle 100 is symbolized, with the aim of extracting therefrom the contour lines and to approximate these contour lines by straight line segments, denoted by a file of measured data Fm. which contain their attributes that are of the same kind as those of the file of data Fd, except for the one concerning the object type and missing.

In a third step, the data file Fd is compared with the measurement data file Fm, which is a rectangle 200 is symbolized in order to match the data from the file Fd with the elements of the file Fm to identify Is the straight segments belonging in the image to the track to be detected, so as to recognize the lane in the image.

• – wie durch ein Rechteck 110 symbolisiert ist, werden die sogenannten primitiven Bildkonturlinien extrahiert;
• – gemäß Rechteck 120 werden die Hauptrichtungen der Konturlinien durch Anwendung der Hough-Transformierten bestimmt;
• – gemäß Rechteck 130 werden die Bildkonturlinien durch Geradensegmente entlang den Hauptrichtungen approximiert;
• – gemäß Rechteck 140 wird eine Überbrückung zwischen den miteinander fluchtenden Geradensegmenten vorgenommen, um die Lücken auszufüllen, die in den Konturlinien vorhanden sind; und
• – gemäß dem Rechteck 150 wird eine Datei von Segment-Meßgrößen erzeugt, worin die Gesamtheit von Geradensegmenten erfaßt ist, welche die Bildkonturlinien approximieren.

The in 1 shown segmentation step 100 is in detail in 2 shown. It contains the following phases:

• - like a rectangle 110 is symbolized, the so-called primitive image contour lines are extracted;
• - according to rectangle 120 the principal directions of the contour lines are determined by applying the Hough transform;
• - according to rectangle 130 the image contour lines are approximated by straight line segments along the main directions;
• - according to rectangle 140 a bridge is made between the aligned straight line segments to fill in the gaps present in the contour lines; and
• - according to the rectangle 150 a file of segment measured quantities is generated in which the totality of straight line segments which approximate the image contour lines is detected.

The phase of extraction of primitive contour lines according to 110 in 2 makes it possible to arrive at a binary image in which the contour lines appear, which belong to interruptions of the brightness in the image. It begins with an extraction of the contour zones or calculation of the gradient image to be achieved by making spatial folds of the light image with two binary masks of variable size N and the orientation 0 or 90 °.

For example, for N = 5, the 0 ° orientation mask, referred to as the horizontal gradient mask, has the following composition: -1 -1 0 1 1 -1 -1 0 1 1 -1 -1 0 1 1 -1 -1 0 1 1 -1 -1 0 1 1 while the mask with the orientation of 90 °, which is referred to as a vertical gradient mask, has the following composition: -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1

Let G _H and G _{V be} the results of the spatial convolutions of the image with these masks; one then calculates for each pixel the amount or the amplitude of the gradient using the following formula:

Furthermore, the orientation of the gradient or the direction of maximum brightness change is calculated using the following formula:

These provides an orientation angle related to the horizontal is.

The Gradient amplitudes which are below a certain fixed threshold are set to zero to suppress fluctuations that be generated by noise, and only the large values to take account of by contrast jumps caused and correspond to the points of the picture contour zones.

To Extraction of the contour zones is done in a scaffolding such that the contour zones are diluted and converted to contour lines, along the apex lines the amplitude image of the gradient. For this purpose, the amount of the Gradients at each pixel, where it is different from zero, with compared to the magnitude of the gradient of those two pixels, which are closest in the direction of the gradient, which in this case is set to a multiple of 45 ° is, and only that pixel is kept, in which the Gradient has the largest amount. By iterative In a picture contour zone, the procedure only becomes a line of dots stored, which correspond to the maximum values of the amount of the gradient. You then have a binary image, which consists of contour lines.

The phase of determining the main directions of the contour lines, block 120 in 2 , is based on the application of Hough transforms.

This Hough transform is based on the polar equation for a straight line in Cartesian coordinates: ρ = x cosθ + ysin θ

Where x and y are the Cartesian coordinates for the points of the line, ρ is the distance of the line from the origin and θ is the angle that the normal of the line forms with the axis Ox, ie the angle of the gradient when the line is considered a contour line , 3 shows the relations between the Cartesian coordinates and the polar coordinates for a straight line.

The Hough transform makes it possible to move from the space of the Cartesian parameters (x, y) to the space of the polar coordinates (ρ, θ). It generates a correspondence between a straight line and a point with the coordinates (ρ, θ) and to a correspondence between a point (x, y) and a sine function of the following equation: ρ = x cosθ + ysin θ

This sine function can also be regarded as a transform of the totality of straight lines passing through the coordinate point (x, y). For further details regarding the Hough Transform, reference is made to the following reference: J. ILLINGWORTH and J. KITTLER in "A survey of the Hough Transform", published in Computer Vision, Graphics and Processing 44, pp. 87-166 (1988). ,

For searching the main directions, in each point of a contour line, one is only interested in those straight lines whose normal forms an angle with the abscissa axis which corresponds to the direction of the gradient which was determined with a specific accuracy tolerance + δθ, here 15 °. This limit runs as in 4 In addition, in the Hough transform, each point of an image contour line is assigned to a portion of the sine function that is restricted to a range of values (θ-δθ, θ + δθ) centered on the angle θo of orientation of the gradient is. When a contour line of the image comes close to a straight line in which the direction of the gradient corresponds to the angle θo and whose distance from the origin is ρo, all parts of the sine functions representing their points in the Hough transform intersect at a coordinate point (ρo, θo). The position of the crossing points of this type in the Hough transform of the image contour lines is thus characteristic of the main direction.

• – Für jeden Bildpunkt mit den Koordinaten (xi, yi), der auf einer Konturlinie liegt und dessen Gradient die Orientierung θi aufweist, berechnet man folgenden Ausdruck: ρ = xicosθ + yisinθ mit θε[θi – 15°, θi + 15°]
• – Man bildet eine Tabelle von Werten (ρj, θj), deren Felder jedesmal dann inkrementiert werden, wenn ein Teil einer Sinusfunktion hindurchgeht, der einem Punkt einer Bildkonturlinie entspricht. Diese Tabelle, die unter der Bezeichnung "nappe de Hough" bekannt ist, kann als zweidimensionales Histogramm angesehen werden;
• – schließlich werden die Maxima dieser Tabelle ausgewählt, die zu den Akkumulationspunkten gehören, welche den Hauptrichtungen entsprechen, denen die Bildkonturlinien folgen, und wofür die Koordinaten (ρ, O) die zugeordneten Gleichungen definieren.

The main directions are thus searched for according to the following procedure:

• For each pixel with the coordinates (xi, yi) lying on a contour line and whose gradient has the orientation θi, the following expression is calculated: ρ = xicosθ + yisinθ with θε [θi - 15 °, θi + 15 °]
• - Form a table of values (ρj, θj) whose fields are incremented each time a part of a sine function corresponding to one point of an image contour line passes. This table, known as "nappe de Hough", can be considered as a two-dimensional histogram;
• Finally, the maximums of this table belonging to the accumulation points corresponding to the principal directions followed by the image contour lines are selected and for which the coordinates (ρ, O) define the associated equations.

The phase 130 in 2 in which the line segments which approximate the image contour lines are determined, is that the main directions which follow the contour lines quite accurately are limited. For this purpose, a narrow-width rectangular window is opened which is centered about each main direction, and of a main direction only those segments are kept which approximate the contour lines contained in the associated window and which have a compatible orientation gradient.

The bridging phase 140 to 2 This is to eliminate any gaps between aligned straight line segments due to local fluctuations due to noise. It may also serve to fill in larger gaps between aligned straight line segments resulting from the masking or crossing of webs when their suppression is desired to facilitate eventual tracking of the web. It consists, for example, in that the aligned straight segments are connected, which are separated only by a distance which is smaller than their average length.

In the in 2 shown phase 150 To generate a file Fm of segment measured quantities, those straight line segments which arise as a result of the bridging phase are detected.

The straight line segments and their relative positions in the image are described by a graphic type structure consisting of nodes connected by arcs, as in FIG 5 shown.

The nodes correspond to the segments. They are defined by individual attributes, which are the characteristics of each segment defined with respect to an image, such as the length of the considered segment, its orientation in the image, the location of its endpoints, the order in which it appears in the image, as it progresses Time develops and is part of, for example, a series of images of an overflown terrain.

The Arcs correspond to the relationships between segments and are defined by relative attributes which are the characteristics are any segment considered with respect to the other segments, like the parallelism or intersection relationship, the relative Orientation, the relative distance.

The individual and relative attributes of each segment are derived determined by the positions of the line segments in the image and in a file Fm summarized the segment measured variables.

In In the initial phase of the process, an operator creates one analog file called file Fd of segment data starting from a reconstruction of the graphics for positioning of segments based on description information for the contours of the track to be recognized and its surroundings after approximation done by straight segments. However, this file contains Fd an additional attribute, which is valid for each line segment the object type it concerns, d. H. the train or her Surroundings.

In the final phase 200 in 1 in which the files Fd and Fm are compared, the elements of the file Fd are compared with the elements of the file Fm until a more or less complete match is found, which makes it possible to identify those segments in the course Fm of measured variables, which correspond to the edges of the searched runway or track so that it is recognized.

This Comparison does not require that the Fd file or the Reference graphic, from which it results, the place of recognition Describes the course exhaustively. She does not have to Attributes of all straight segments containing the web contours and approximate their environment, but only a certain number of them, which are characteristic for the place of the course. The absence of the attributes of certain line segments or even that Lack of certain segments in the file of data, however contained in the file of measured quantities do not question the success of the ID. The minimal information to be provided, which depends on the location, must allow the identification of the web and eliminate any ambiguity.

QUOTES INCLUDE IN THE DESCRIPTION

This list The documents listed by the applicant have been automated generated and is solely for better information recorded by the reader. The list is not part of the German Patent or utility model application. The DPMA takes over no liability for any errors or omissions.

Cited non-patent literature

• - J. ILLINGWORTH and J. KITTLER in "A survey of the Hough Transform", published in Computer Vision, Graphics and Processing 44, pp. 87 to 116 - (1988) [0028]

Claims (5)

Method of recognizing a web in an image, characterized in that it comprises the following successive steps: - assigning contour lines, which are characteristic of the web to be recognized and its surroundings, to straight-line segments, which are designated by individual attributes, such as their length, the position of their ends, and relative attributes, such as their orientations and relative distances, as a result of location descriptor data issued by an operator to create a file of segment data Fd, processing of the image in which the web is to be recognized, in order to obtain a file of straight-line segment measured quantities, comprising the following phases: - extraction of the image contour lines; Determining the principal directions of the image contour lines using the Hough transforms and selecting the accumulation points that appear in the transform of the image contour lines; - Approximation of the contour lines by straight segments along the main directions; and - generating a file of segment measurands Fm, wherein the line segments which approximate the image contour lines defined by individual and relative attributes of the same kind as the file Fd of segment data are detected; and - comparing the set of elements of the file Fd with the elements of the file Fm in order to arrive at the match for recognizing the searched trajectory in the image. Method according to claim 1, characterized in that that image processing to extract a file from Straight line measured quantities also before Phase of file creation a phase of bridging contains, through which the gaps between aligned Straight line segments are filled. Method according to claim 2, characterized in that that the bridging phase is that the aligned straight segments are connected, which are separated only by a distance smaller than theirs average length is. Method according to claim 1, characterized in that the determination of the principal directions of the contour lines in the image by applying the Hough transforms consists in: - a correspondence between each point of a contour line with the coordinates (xi, yi) and with the gradient of the orientation θ1 and a part of a sinusoidal function which obeys the following equation: ρ = xicosθ + yisinθ with θε [θi - 15 °, θ + 15 °] Generating a table of values (ρj, θj), referred to as a Hough raster, wherein each field is incremented each time a portion of the sine function corresponding to a point of a contour line of the image passes through; and - selecting the maxima in the Hough grid associated with the accumulation points and whose coordinates (ρ, θ) define the parameters of the polar equations of the principal directions. Device for detecting a web in an image, characterized in that it comprises a processing processor contains, according to the implementation of the method one of the preceding claims is formed.