EP2676240A1

EP2676240A1 - Method for detecting target objects in a surveillance region

Info

Publication number: EP2676240A1
Application number: EP12706022.6A
Authority: EP
Inventors: Julia ICK
Original assignee: Hella KGaA Huek and Co
Current assignee: Hella GmbH and Co KGaA
Priority date: 2011-02-18
Filing date: 2012-02-17
Publication date: 2013-12-25
Also published as: US20140049647A1; WO2012110653A1; DE102011011929A1; US20170161574A1; US9607225B2; US9904861B2

Abstract

The invention relates to a method for detecting target objects in an image sequence of a surveillance region (3), wherein interfering contributions are eliminated from pixel values of images (7) of the image sequence, wherein the interfering contributions are generated by at least one interfering object (5) contained in the surveillance region (3), and wherein the target objects are sought in the image sequence on the basis of the corrected pixel values, wherein a beginning of a predefined movement sequence of the interfering object (5) is detected, during the movement sequence an instantaneous position of the interfering object (5) within the images is determined and the instantaneous position of the interfering object (5) is taken into account when eliminating the interfering contributions of the interfering object (5) from the pixel values. The invention additionally relates to a system for carrying out a method of this type, and to a vehicle comprising such a system.

Description

Method for detecting target objects in a surveillance area

The invention relates to a method for detecting target objects in an image sequence of a surveillance area and to a system for carrying out such a method. The invention further relates to a vehicle with such a system.

Such methods are often used to count or track (track) people or other moving objects that are in a surveillance area or are moving through a surveillance area. Possible areas of application are, for example, the entrance area of a building or a door of a vehicle of passenger traffic, such as a train or bus. The image sequences are recorded with a camera and then evaluated with a suitably equipped evaluation, for example by means of an image segmentation, a pattern analysis, a motion analysis and / or another image analysis method based on image data of the image sequence.

Typically there is the problem that within the monitoring area other objects than the target objects, so-called disturbance objects that are not to be counted or "tracked", but that can be reliably distinguished or discriminated against by the target objects, and an attempt is usually made to solve this problem by assigning target and disturbing objects to different movement classes Detecting target objects having a typical speed are assigned to a first movement class, and interfering objects, which may be, for example, stationary objects, are assigned to a second movement class Subsequently, the images of the image sequence which have been cleaned up in this way are examined for the presence of the target objects.

In many applications, however, it is not possible to assign target and disturbing objects to different movement classes. For example, vehicle doors in buses and trains or other doors as well as persons are not (purely) stationary objects, so they can not be faded out in the conventional manner described above.

The present invention is therefore based on the object of proposing a method for detecting target objects in an image sequence of a monitoring area, which overcomes or at least alleviates the problems in the prior art. The method should therefore allow to detect target objects within a surveillance area. The surveillance area should also be able to contain disturbing objects which, like the target objects, may not be spatially stationary but mobile. In addition, a system is proposed, which is suitable for detecting moving target objects in a surveillance area, in addition to the moving target objects also also movable interfering objects can be located. Finally, it is intended to propose a vehicle for carrying persons suitable for detecting and counting persons entering or leaving the vehicle.

This object is achieved by a method having the features of the main claim and by a system having the features of the secondary claim. Advantageous embodiments and developments emerge with the features of the dependent claims.

The method according to the invention for detecting moving target objects, in particular persons, in an image sequence of a monitoring area therefore provides that pixel values of images of the image sequence are cleaned up from interference contributions, the interference contributions being generated by at least one interference object contained in the control area. Subsequently, the target objects in the image sequence are searched on the basis of the adjusted pixel values, for example by means of one of the above-mentioned image analysis methods (image segmentation, mouse analysis, motion analysis, etc.).

In this case, the term pixel value generally designates a value calculated in a predetermined manner from raw data of an image for a single pixel of the image, wherein the value can be, for example, numerical or even boolean. It can also be provided that a pixel value is calculated from the raw data of a contiguous region of adjacent pixels of the image and assigned to each of these pixels, for example by means of an averaging method.

By means of pixel values, objects can be recognized, for example, by grouping adjacent pixels (pixels) with similar pixel values into image segments and classifying image segments which correspond to predetermined selection criteria as target objects. Selection criteria include, for example, the size and / or the shape of a picture segment, as well as the criterion as to whether the pixel values of the pixels within a picture segment are within predetermined limits.

It is now decisive for the proposed method that a start, that is to say a start time, of a predetermined and thus known movement sequence of the at least one jamming object is detected and during the course of the movement an instantaneous, current position of the jamming object within the images is determined. During the subsequent clearing of the pixel values from the interference contributions of the disturbing object, the instantaneous position of the disturbing object is taken into account. In this case, the movement sequence of a faulty object is designated as predetermined or known, if it is known how the instantaneous position of the disturbing object can be calculated or simulated during the course of the movement.

With the proposed method, therefore, disturbing objects within the surveillance area whose motion sequences are known and the beginning of the motion sequence can be detected can be masked out of the pictures of the picture sequence in a particularly accurate and reliable manner or the contributions (interference contributions) of the disturbing objects to the pixel values of the pictures to be counted out.

It can thus be ensured that pixels or pixels belonging to target objects in the image sequence are not confused with pixels or pixels of the interfering object, or pixel regions (segments) belonging to interfering objects merge with pixel regions which belong to target objects, so that the distinction between target object and Disturbing object and the detection of the target object is easier and more reliable.

In one embodiment of the method, it is provided that the pixels of each image are transferred into a three-dimensional coordinate system, a so-called world coordinate system. Preferably, the origin of this coordinate system lies at the level of a base area of the surveillance area, for example, a floor of the surveillance area, and further extends an XY plane of this coordinate system within this area. The height of each pixel is specified perpendicular to this XY plane or determined by means of the evaluation unit. It is particularly useful if the

Coordinate origin is positioned directly below the camera, with which the image sequence is generated.

Preferably, the image sequence is detected with a stereo camera. Each of the images then contains two individual images captured from two different perspectives, with which a three-dimensional representation of the monitoring region is possible. In particular, each (stereo) pixel of a stereo image, which is given in each case by two pixels of the two individual images of the image, can be transmitted into the three-dimensional world coordinate system as described above. In addition, it may be provided that a height map and / or an occupancy map of the surveillance area is or will be calculated for each image of the image sequence. For this purpose, the base area of the monitoring area, ie the XY plane of the world coordinate system, is decomposed into a map grid, with the individual grid elements corresponding, for example, to 5 × 5 cm area areas of the base area of the monitored area. In order to calculate the height map and the occupancy map for an image, in a first step for each of these raster elements in the image, those pixels are determined in which any object, such as a target object or an interfering object, or a part of this object is imaged. For example, such pixels can be recognized by the fact that in them the base is covered by this object.

For each of these pixels, the coordinates in the world coordinate system and the associated raster element below the pixel are determined. In the elevation map, the height of the highest pixel in which an object is displayed is then entered for each raster element. This maximum height of a raster element is also referred to below as the height value of the raster element. The assignment card contains the number of pixels for each raster element, in each of which an object is mapped. This number is also referred to below as the occupancy value of a raster element.

In a preferred embodiment of the method, therefore, a stereo camera system is used for detecting the image sequence, and each pixel of the three-dimensional images of this image sequence is assigned a screen area within the base area of the monitoring area. After this assignment, as described above, a height map and an occupancy map are calculated. The pixel values stored in these maps, ie the height values and occupancy values associated with the grid elements, are subsequently cleaned up from interference contributions from interfering objects.

In one embodiment of the method, it is provided that, after the position of the disturbing object is determined in an image, for those raster elements of the image within which the disturbing object is located, the altitude te and / or the occupancy values are set to zero. In this way, the pixel values of the image are cleaned in a particularly reliable manner from interference contributions of the disturbing object.

In a preferred embodiment of the method, it is provided that the instantaneous position of the disturbing object is calculated on the basis of a movement model, that is to say a predetermined calculation method. That is, the motion of the jamming object is simulated by means of the motion model as soon as the beginning of the movement of the jamming object has been detected. The motion model may include one or more model parameters, either in one

Initialization phase or during a first

Training phase is determined. The at least one

Model parameters may be given for example by a numerical value, a Boolean value or even a multidimensional vector of such quantities, such as a set of world coordinates or

Row and column indexes. Alternatively, it would also be possible to determine the instantaneous position of the interfering object

each by observing the disturbing object after the

Detected detected start of the movement. In a typical case where the obstruction object is through a door and the movement is through a door opening, it could be e.g. Any door recognition can be used.

The model parameters mentioned in connection with the preferred embodiment may be, for example, a rest position of a specific subarea of the clutter object within an image. Such a subarea can be, for example, a lower edge of an interfering object, for example a door. Another model parameter can be given by an opening region belonging to one or more interfering objects within an image, that is, for example, a passage opening within a disturbing object given by a door. Another possible model parameter represents a period of time which lies between a start of the movement sequence and a signal. This signal may, for example, be a trigger signal with which the motion sequence of the interfering object is triggered, for example a triggering signal of an auto-triggering signal. matic door opener. However, the signal can also be a measuring signal of a sensor with which movements of the disturbing object are monitored. Thus, for example, a motion sensor of any type can be used to generate such a signal on a door, for example with an inductive, capacitive or optoelectronic sensor element. It can be provided in particular that the beginning of the movement sequence is determined on the basis of the pixel values of the image sequence acquired with the camera and / or by evaluating images recorded with another camera and / or on the basis of a measurement signal of a sensor different from the camera. The aforementioned ways of recognizing the beginning of the course of motion can be combined with one another. For example, it may be provided that it is necessary for detecting the beginning of the movement sequence that the beginning is detected both on the basis of images with the evaluation unit and also generates a corresponding measurement signal from a further motion sensor or a trigger signal of an operating unit and from the evaluation unit Will be received.

The motion model may include, for example, an equation of motion for the interfering object. This equation of motion may include one or more of the model parameters. Typical model parameters for such an equation of motion are the already mentioned (average) speed of the disturbing object, the time difference between the measuring or triggering signal and the beginning of the movement and the rest position of the disturbing object. The motion model can additionally include a method with which model parameters of the motion model, for example those that enter into the equation of motion of the disturbing object, can be determined and / or optimized.

For example, it can be provided that the above-mentioned model parameter, which indicates the time delay between a trigger signal of the movement sequence and the beginning of the movement sequence, is determined by one or more measurements of this time delay. In a corresponding method for determining this model parameter, provision may be made, for example, for the triggering signal, such as a door opener, to be received by the evaluation unit and for the time t _{0 of} this signal reception to be stored. Then the time is determined by the evaluation unit, which is determined by the recording time ("Timestamp"). is defined that image in which the beginning of the triggered by the trigger signal sequence of movements has been detected by the evaluation. In the example of an opening movement of a door, the beginning of this opening movement can be recognized, for example, by means of a forming door gap, which for example must have a predetermined minimum width Ex, by the evaluation unit. The required time delay is finally calculated by the difference between these two points in time, ie by tj-1 _{0 /} with the evaluation unit and assigned to the corresponding model parameter.

The method for determining and / or optimizing the model parameters may, for example, be used during the initialization phase mentioned or during the first or another training phase.

The speed of the disturbing object during the course of the movement can, for example, be given by the quotient of the distance traveled during the course of the movement and the duration of the course of the movement. The named distance can be specified or measured, for example, as a number of pixels passed through or raster elements. For example, positions of objects or parts of objects can be specified or measured in world coordinates or in row and column indices of pixels or pels.

In one possible embodiment of the method, in which pixels of the images of the surveillance area captured by the camera are transferred to a height map and / or occupancy values, it is provided that one of the model parameters is given by a line index of a start line or an end line of the interference object within one such height or occupancy map of an image, by a column index of a center column between two parts of the interfering object in a height or occupancy map of an image or by an average time difference between a trigger signal for the movement of the disturbing object or a measurement signal of a (motion) sensor and in a Occupancy card or height map of an image recognizable beginning of the movement. It can also be provided that one of the model parameters is given by an average speed of movement of the interfering object, whereby the speed of movement can be defined as the number of pixels per time unit covered in an occupancy card. Additional Mordell parameters used can define a direction - eg horizontal, vertical or oblique - of the movement sequence or a movement curve.

Within the first training phase, it can be provided, for example, that, starting from starting values or initial values for one or more model parameters, a plurality of measurements of these or other model parameters are carried out and, if appropriate, subsequently the model parameters thus generated are averaged after each measurement. Preferably, a distinction is made between successful and unsuccessful measurements. An unsuccessful measurement can occur, for example, when a relevant image area for the measurement of a model parameter has been obscured by a target object, for example by a person in the surveillance area.

As described above, it can be provided that the value of one or more model parameters is specified in an initialization phase. This so-called initial value can then be maintained as a constant value for the model parameter or further optimized in the following procedure within training phases, as described below. One way of specifying the initial value of a model parameter defined by an idle position of the clutter object is to input it via a graphical user interface. For example, it can be provided that an image of the monitoring unit is displayed with a screen during an initialization phase of the method and marks the position of the interfering object, for example a lower edge of a door, by means of a mouse, a touch screen or another suitable interface in the image on the screen becomes. Subsequently, for example based on the marking, the world coordinates of the rest position, or the course of the lower edge of the door (ie the height and the length of the lower edge) are calculated and stored as corresponding values of the model parameter in an evaluation unit. In addition, by means of the screen or in another suitable manner, as another model parameter, the resting position of a center between two adjoining door leaves of a door can be entered. Subsequently, the world coordinators th of this position, for example, based on the corresponding mark in the screen, calculated and stored as a corresponding value of the associated model parameter in an evaluation.

In a further development, it is provided that initial values of model parameters are calculated as functions of input initial values of other (model parameters.) It can be provided that an initial search range for the disturbance object is determined from a rest position of a disturbance object A bottom and top row of pixels may be determined as described above, and this bottom and top row of pixels may serve as a search area for the object of disturbance within the images and within the height maps, respectively and / or the occupancy maps between these rows of pixels or rasters.

In addition, it can be provided that one or more model parameters of the movement model are continuously optimized in a second training phase of the method, which can be connected directly to the first training phase or the initialization phase, for example. In this case, those measured values which deviate very greatly from the current or the preceding model parameter values are preferably neglected or less heavily weighted.

The faulty object may be, for example, a door and the movement sequence may be an opening or closing of the door. In this case, the door may have two leaves that can be pushed apart or unfolded, for example. In particular, it is possible that

these two door leaves are symmetrical, thus forming a central passage opening through the door when opening the two door leaves. The door may in particular be a passenger door of a bus or a passenger train.

Is the disturbing object characterized by the fact that the disturbing object or a part of the disturbing object frequently, predominantly or exclusively in predetermined and can be provided that the current position of the disturbing object or the said part of the disturbing object is compared with these positions. In addition, it is then possible to determine and indicate whether the disturbing object or the said part of the disturbing object is currently in one of these positions or is currently moving to one of these positions and, if so, which of the positions it is in or soft on the positions it moves to.

The named predetermined and known positions of the disturbing object or of the part of the disturbing object can be the abovementioned rest position (s) of the disturbing object or the part of the disturbing object. For example, if the obstruction object is a door, as described above, the rest positions may be a closed position and an open position where the door is closed or opened, and accordingly, an opening movement and an opening Closing movement of the door. It may therefore be provided in this case that the current position of the door or a part of the door is compared with these rest positions and is specified according to whether the door is currently in the open position or the closed position or is currently open or closes. In addition, a corresponding signal can be generated in which this information is coded.

In a preferred embodiment of the method, it is further provided that target objects in the surveillance area are tracked by the evaluation unit and / or counted In particular, it is possible for persons who move through the surveillance area in a predetermined direction to count become.

A system according to the invention for detecting target objects within an image sequence of a surveillance area comprises a camera for capturing the image sequence and an evaluation unit connected to the camera for data transmission, which is set up to clean up pixel values of images of the image sequence of interference contributions. In this case, the mentioned interference contributions originate from at least one faulty object contained in the control area. In this case, the evaluation unit is also set up on the basis of the pixel values to search the target objects in the image sequence. For the system according to the invention, it is now decisive that the evaluation unit is also set up to detect a start of a predetermined movement sequence of the interference object, to determine a momentary position of the interference object within the images during the movement sequence and to determine the instantaneous position of the interference object during the cleaning up of the interference object To take account of pixel values of the interference contributions of the disturbing object.

Thus, the system according to the invention has the capability of eliminating or eliminating disturbing influences of disturbing objects whose motion sequences are known from the pixel values of the images and thus improving or eliminating separation or discrimination of target objects and disturbing objects in the image sequence of the surveillance area to increase a hit rate and / or a counting accuracy of the target objects.

It may also be provided that the evaluation unit is set up to calculate or simulate the instantaneous position of the at least one interfering object on the basis of a movement model, as has already been described above for the method according to the invention. Therefore, corresponding model parameters can also be predefined or the system can be optimized in a training phase, wherein the evaluation unit is set up to generate measured values for the parameters, to mittein and to replace the model parameters by such averaged parameter values.

Finally, the evaluation unit of the system programmatically to perform one or a combination of several of the embodiments described above or developments of the proposed method. This results in the advantages described for the method.

In a further embodiment of the system, it is provided that the camera is a stereo camera and is thus suitable for the generation of images which each comprise two individual images. The images of a stereo camera, which are often referred to as stereo images, allow a three-dimensional representation of the surveillance area. In particular, steels Reimaging the transmission of pixels of the stereo images in a three-dimensional image coordinate system. As has already been explained in the context of the method described above, the evaluation unit can also be set up to convert the pixels transmitted into the world coordinate system into height maps or occupancy maps.

The system may include a motion sensor, for example, in addition to the camera used for the generation of the image sequence, which is adapted to fully or partially detect the motion sequence of one or more interfering objects (in particular the beginning) and to transmit appropriate Messsignaie to the evaluation unit. In this case, in particular the beginning of the movement sequence can be coded in these measurement signals. The evaluation unit is preferably set up to use these measurement signals to determine the beginning or even the entire course of the relevant interfering object. The sensor may, for example, comprise an electrical, electronic, inductive, capacitive and / or optoelectronic sensor element, such as a CMOS chip or CCD chip.

There is also proposed a passenger transport vehicle comprising at least one vehicle door and a control unit adapted to generate a trigger signal for automatically opening the vehicle door, the control unit being connected to the vehicle door for transmitting the trigger signal. The vehicle also comprises a system of the type proposed here, wherein the evaluation unit of the system is connected to the door or the control unit of the door for receiving the trigger signal, which signals the beginning of the opening of the door, that is, the faulty object.

In the following, the invention will be explained in more detail with reference to specific embodiments shown in Figures 1 to 10. It shows:

1 shows a vehicle with a system proposed here,

FIG. 2 shows an image of a monitoring area acquired with the system shown in FIG. 1,

FIG. 3 shows an occupancy or height map of the image shown in FIG. 2 during an initialization phase of the system, FIG. 4 shows another occupancy or altitude map during a training phase of the system,

FIG. 5 shows another occupancy or height map during a training phase of the system,

FIG. 6 shows another occupancy or altitude map after a first one

Training phase of the system,

7 shows an occupancy or height map with an interfering object and a target object before the pixel values of the map are cleaned up,

FIG. 8 shows the occupancy or altitude map shown in FIG

Cleanup of pixel values,

FIG. 9 shows an occupancy or height map with an interference object and a target object before the pixel values of the map are cleaned up,

FIG. 10 shows the occupancy or altitude map shown in FIG

Clean up the pixel values.

Identical or similar features are identified by the same reference numerals.

1 shows a cross section through a vehicle 1 for transporting people, for example a bus or a train, shown schematically. It contains a specific embodiment of the system 2 proposed here, which is set up to carry out a specific embodiment of the method proposed here.

The system 2 comprises a stereo camera 4 'for acquiring an image sequence of stereo images, also referred to below as images for short, of a monitoring area 3, which is identified by a dashed line in FIG. 1 and also in the following figures. Each of the images comprises two individual images in which the surveillance area 3 is depicted in two different perspectives. The stereo camera 2 is connected to an evaluation unit 4 of the system for transmitting raw data of the images to the evaluation unit 4.

Within the Ü monitor area 3, an automatic vehicle door 5 of the vehicle 1 is arranged. This is connected to a Bed ieneinheit 6, which is adapted to, with appropriate operation by a passenger P to generate a trigger signal and transmit to the door. The door 5 is arranged to open after the trigger signal has been received. There is also a further motion sensor 6 'is provided, with which the movement of the door is detected, for example with a mechanical, electronic, optoelectronic, inductive or other sensor element, which may be connected for example directly to the door. This motion sensor 6 'is also connected to the evaluation unit 4 for transmitting corresponding measurement signals, in which in particular the beginning of the opening movement of the door 5 is coded. The evaluation unit 4 is set up to process the measurement signals and, in particular, to determine the time of the beginning on the basis of the measurement signals and on the basis of the trigger signal of the operator control unit 6.

FIG. 2 diagrammatically shows one of the images 7 of the surveillance area 3 which has been acquired by the stereo camera 4 'of the system 2 shown in FIG. 1 in an initialization phase of the system and displayed by means of a screen (not shown here) of the system 2.

The door 5 has left and right doors 8, 9 which contact each other in a center line 10 'of the door 5 when the door 5 is closed.

In the illustrated image 7, the position of a lower edge 10 of the door 5 is marked by means of a marking line 11 extending along an entire width of the door 5. Further, in the image 7, the position of the center line 10 'by means of another marker 12. The two mark 11, 12 have been entered by means of an input interface (not shown) in the evaluation unit 4, for example by means of a touch screen function of the screen or a mouse or a other for setting markers within the image 7 suitable input interface.

The evaluation device 4 is programmed to transfer the pixels of the image 7 (as well as all other images captured by the stereo camera 4 ') into a three-dimensional coordinate system (world coordinate system) and to calculate the corresponding (world) coordinates of the pixels. As can be seen in FIG. 2, the origin of this coordinate system is at the level of a base area 13 of the monitoring system. 3, ie at the level of a floor of the surveillance area, which in this example extends beyond the surveillance area 3. The height of each pixel of the image 7 (as well as all other images) is thus determined perpendicular to this XY plane along the Z axis of the world coordinate system by means of the evaluation unit 4.

In addition, the evaluation unit 4 calculates a height map and an occupancy map of the surveillance area for each image of the image sequence.

In FIG. 3, the elevation map or occupancy map 14 of the image 7 shown in FIG. 2 is shown schematically by way of example.

Evident is a section of a map grid 15 with raster elements 16, in which the base area 13 is disassembled within the surveillance area 3. In this case, the individual grid elements 16 in this example correspond to surface elements of the base area 13 with dimensions of 5 × 5 cm in each case. In order to calculate the height map and the occupancy map, evaluation unit 4 determines in a first step for each of these raster elements 16 those pixels of image 7 in which an arbitrary object, such as a target object, in this example a passenger, or an interfering object, In this example, the door 5, or a part of these objects will be mapped. Such "occupied" pixels are recognized by the fact that the base 13 in them

this object is obscured.

For each of these occupied pixels, the associated coordinates in the world coordinate system and then the associated raster element 16 are determined directly below the pixel. In the height map of each image, the height of the highest "occupied" pixel in which an object is imaged is then entered for each raster element 16. This height is also referred to below as the height value of the associated raster element 16. The occupancy map of each image contains for each raster element 16 the number of all "occupied" pixels, in which so in each case an object is displayed. This number is also referred to below as the occupancy value of a raster element 16.

In the height map or occupancy map 14 of the image 7 shown in FIG. 3, image regions in which the height values or occupancy values have a value other than zero are shown hatched. In the illustrated example, only the door 5 is located within the surveillance area 3.

As already mentioned above, the system 4 is in an initialization phase at the time the image 7 is taken. In this phase, initialization values for model parameters of a movement model for simulating the opening movement of the door 5 are set or measured. Based on this movement model for the opening movement of the door 5, an instantaneous position of the door 5 during the opening movement in each image detected during the opening movement by the camera 4 'can be determined. Based on the determined in this way instantaneous position of the door 5 then all the grid elements 16 are determined within or above which the door 5 is currently located. Thereafter, the height values and the occupancy values of these raster elements are set to zero. In this way, the height values or occupancy values stored in these cards, that is to say in general the pixel values derived from the raw data of the images, are corrected by interference contributions from the door 5.

A first model parameter is a rest position of the lower edge 10 of the door 5, and a second model parameter is a rest position of the center line 10 'of the door 5. The position of this center line 10' simultaneously defines the position of a passage opening of the door during the opening movement of the door 5 between edge regions 17 and 18 of the two door leaves 8 and 9 is formed. A third model parameter is given by a period of time which lies between the beginning of the opening movement and the triggering signal emitted by the operating unit 6 or a measuring signal of the motion sensor 6 'which signals the beginning of the opening of the door. A fourth model parameter is given by a mean velocity of the edge regions 17 and 18 of the door 5 during the opening movement. This speed is determined by the quotient from during the movement through the edge regions 17, 18 distances traveled and the duration of the movement given. The mentioned distances are indicated or measured in each case as a number of raster elements 16 which have passed through the edge regions 17, 18 during the opening movement.

The above-mentioned positions of said parts 10, 10 ', 17, 18 of the door are indicated or measured both in world coordinates and in row and column indices of the associated raster elements 16 of the occupancy card and the height map 14.

The first model parameter, given by the world coordinates and the row and column indices of the rest position of the lower edge 10, is calculated by means of the evaluation unit 4 from the course of the marking line 11 in FIG. From the thus determined coordinates or row and column indices of the lower edge 10 in the rest position of the door 5, the initial values of two further model parameters are determined, namely the line index of a start line 19 (fifth model parameter) and a line index of an end line 20 (sixth model parameter) Door 5 within the height and occupancy map of the image 7. By means of these two initial values, an initial door area 21 between start and end line 19, 20 is defined, in which the course of the door 5 in the height or the occupancy card 14 in the following pictures is expected.

In addition, an initial value of the second model parameter, given by a column index of the rest position of the center line 10 'of the door 5 in the elevation map or the occupancy map 14, from the marker 12 in the image 7 with the evaluation unit 4 is calculated. A center column 21 defined by this column index thus runs in the occupancy map or elevation map 14 between the edge regions 17, 18 of the door 5.

In addition, in the initialization phase, initial values of the third and the fourth model parameters are also input by means of an input interface. The initialization phase is followed by a first training phase in which the initial values of the model parameters 2 to 5 are calibrated or more precisely determined by means of measurements. These measurements are carried out automatically with the evaluation unit 4 on the basis of images recorded during this first training phase, with generated measured values of the model parameters being averaged. Furthermore, a distinction is made between successful and unsuccessful measurements, with a measurement of, for example, the second parameter not being successful, for example, if the center line 10 'has been covered by a person standing in front of the door 5.

The beginning of the opening movement is defined in this example as the time in which between the two edge regions 17, 18 of the door 5 a gap with a width 0x, for example, at least 3 grid elements (ie 15 cm) with the evaluation unit 4 in the height or Occupancy card 14 can be seen. This time is shown in FIG. The third model parameter, that is, the elapsed time between reception of the opening signal of the operation unit (trigger timing) and the beginning of the opening movement (opening timing), is measured by counting the number of images detected between the triggering timing and the opening timing have been. Furthermore, at this time, the second model parameter, the position of the center line 21, is measured as the midpoint between the two edge regions 17, 18 of the door 5. It would also be possible to use this time span as a time difference between the above-defined beginning of the movement and a specific one Define measurement signal of the motion sensor 6 ', such as the measurement signal, the beginning of the opening movement of the

Door 5 signals. In this way, for example, the time difference between the beginning of the actual, physical opening movement of the door and the recognizable by means of the height maps or occupancy maps start of the movement process could be considered.

The end of the opening movement is defined as the time at which the two edge regions 17, 18 move out of the monitoring region 3 (end time). This time is shown in FIG. The fourth model parameter, ie the speed with which the door leaves 8, 9 move outward, is measured by counting the number of images which has been recorded between the triggering time and the end time.

The repeated measurements taking place after the initialization phase for optimizing the model parameters two to four, which enter into an equation of motion of the movement model for determining the instantaneous position of the door 5, are carried out until a sufficient constancy of these model parameters is achieved. In addition to the model parameters two to four, two further model parameters, model parameters seven and eight, are also determined. To determine these model parameters, all raster elements 16 are stored in the evaluation unit 4, which are covered by the door 5 in at least one image during the opening process. The seventh model parameter is determined to be the smallest row index of these stored raster elements, and the eighth model parameter is determined to be the largest row index of these stored raster elements. In FIG. 6, the pixel lines belonging to these two model parameters are shown as horizontal dashed-dotted lines 22 and 23 above and below the door leaves 8 and 9. These lines are also referred to below as start and end lines.

The phase of these measurements is also referred to as the first training phase. Experience has shown that about 5 to 10 successful measurements are required for each parameter. After this first training phase can be deducted with the system 2 interference contributions of the door 5 from the height and occupancy map. For this purpose, the instantaneous positions of the edge regions 17 and 18 with the evaluation unit are calculated at any time on the basis of the model parameters 2 to 4, and the height values and the occupancy values for all those raster elements 16 are set to zero over which the door leaves 8 or 9 currently extend , These are all raster elements which are located between the start and end lines 22, 23 and at the same time left of the current position of the left edge portion 17 of the left door leaf 8 or between the start and end lines 22, 23 and at the same time right of the current position of the right edge area 17 of the right door 9 are located. During the opening movement of the door, the instantaneous position of the left and right edge regions 17, 18 of FIG Door calculated by means of the already mentioned equation of motion of the movement model with the evaluation unit 4.

In this way, the height and occupancy values of the interference contributions of the door 5 are adjusted and the door 5 is hidden from the occupancy and height map 14. The passengers can then be detected particularly reliably on the basis of the thus corrected altitude and occupancy values, for example on the basis of altitude and occupancy maxima in the altitude or occupancy map. The height and occupancy maxima of the passengers can no longer merge with the height and occupancy maxima of the door.

This will be illustrated by the first situation shown in FIGS. 7 and 8 and the second situation shown in FIGS. 9 and 10.

In the first situation, a passenger is just in front of the door 5. In Figure 7, an occupancy or height map 14 of an image of this situation is shown before the above-described cleanup of the pixel values has been performed. The hatched area in Figure 7 is an image segment 24 which includes raster elements with similar (unadjusted) altitude and occupancy values. In Figure 8, the altitude or occupancy map associated with this situation is shown after the pixel values have been cleaned up. Now, a segment 25 can be seen, which can be clearly detected as a person or passenger due to its shape. The interference contributions of the door have been eliminated, so that the pixel values in the image area 26 belonging to the door are constantly zero. The door 5 is thus hidden from the elevation map or from the occupancy card 14.

In the second situation, a passenger has already moved into the forming door opening between the two door leaves 8, 9 during the opening of the door 5. FIG. 9 once again shows a coherent segment 25 in which the passenger appears merged with the door and therefore can not or only very difficultly be detected. After the described correction of the pixel values from the interference contributions, the segment 25 shown in FIG. 10 results, which is clearly detectable as a passenger. Finally, the evaluation unit is set up to track and count the passengers. In particular, passengers entering and leaving the door 5 are counted. It is also envisaged that a second training phase follows directly after the first training phase in which the model parameters two to four as well as seven and eight are continuously measured and further optimized as described above. In this way, these model parameters can be adapted to a changing over the time opening behavior of the door. Again, such readings become very strong of current ones

Values of the model parameters differ, neglected or less weighted.

Claims

claims

Method for detecting target objects in an image sequence of a surveillance area (3), in which pixel values of images (7) of the image sequence are cleared of interference contributions, wherein the interference contributions are generated by at least one interference object (5) contained in the surveillance area (3) and wherein looking for the target objects in the image sequence on the basis of the adjusted pixel values,

characterized,

in that a start of a predetermined movement sequence of the disturbing object (5) is detected, during the course of movement an instantaneous position of the disturbing object (5) within the images is determined and the instantaneous position of the disturbing object (5) in clearing the pixel values from the disturbance contributions of the disturbing object (5).

The method of claim 1, characterized in that the instantaneous position of the interfering object (5) is calculated on the basis of a motion model, wherein preferably the motion model includes at least one model parameter.

The method of claim 2, characterized in that at least one of the at least one model parameter is specified during an initialization phase of the method in the form of an initial value of the model parameter.

Method according to one of claims 2 or 3, characterized in that at least one of the at least one model parameter is determined during a first training phase of the method.

Method according to one of claims 2 to 4, characterized in that at least one of the at least one model parameter is continuously optimized in a second training phase of the method.

Method according to one of claims 2 to 5, characterized in that at least one of the at least one model parameter by a rest position of a portion (10) of the interfering object (5) within an image (7), by one belonging to one or more interfering objects (5) Opening area (10 ') within an image, by a period between the beginning of the movement sequence and a trigger signal that triggers the movement sequence, and / or given by a speed of the disturbing object (5) during the movement sequence of the disturbing object (5).

Method according to one of the preceding claims, characterized in that the pixel values include height values and / or occupancy values of pixels of the images (7).

Method according to claim 7, insofar as it relates to one of claims 2 to 6, characterized in that one of the model parameters is given by a row index of a start line (19) or an end line (20) of the disturbing object (5) in the height or An occupancy map (14) of an image, by a column index of a center column (21) between two parts of the interfering object (5) in the height or occupancy map of an image, by an average time difference between a triggering signal for the motion sequence of the interfering object (5) and a first one Detecting the movement sequence in an occupancy map of an image or by a movement speed of the disturbance object.

Method according to one of the preceding claims, characterized in that the disturbing object is a door (5) and the movement sequence is an opening or closing of the door (5).

System (2) for detecting target objects in an image sequence of a surveillance area (3), the system comprising a camera (4 ') for capturing the image sequence and an evaluation unit (4) connected to the camera (4') for data transmission is arranged to clean up pixel values of images (7) of the image sequence of interference contributions, the interference contributions of at least one in the control range clutter object (3) contained (5) originate, and wherein the off ^¬ evaluation unit (4) is further adapted to search based on the adjusted pixel values of the target objects in the image sequence,

characterized,

in that the evaluation unit (4) is set up to detect a beginning of a predetermined movement sequence of the interference object (5), to determine a momentary position of the interference object (5) within the images (7) during the course of the movement, and to determine the instantaneous position of the interference object (5 ) in correcting the pixel values from the jamming contributions of the jamming object (5).

11. System according to claim 10, characterized in that the Auswer ^¬ teeinheit (4) is arranged to evaluate a of a movement sensor (6 ') or an operating element (6) coming trigger signal to detect the start of the motion sequence, and / or is set up, the beginning of the movement through an image analysis of the

Picture sequence of the camera (4 ') and / or other camera images to recognize.

12. System according to any one of claims 10 or 11, characterized in that the evaluation unit (4) is adapted to calculate the current position of the movement sequence based on a movement model.

13. System according to any one of claims 10 to 12, characterized in that the camera (4 ') is a stereo camera for generating two individual images comprising images (7).

14. System according to any one of claims 10 to 13, characterized in that the evaluation unit (4) is adapted to carry out a

Method according to one of claims 1 to 9.

15. System according to any one of claims 10 to 14, characterized in that the interfering object is a door (5) and the movement is an opening movement of the door (5).

16. A vehicle (1) for transporting persons, comprising at least one vehicle door (5) and a Bedienein unit (6), which is adapted to a trigger signal for automatically opening the vehicle door (5) wherein the vehicle (1) further includes a system (2) according to any one of claims 10 to 15, wherein the interfering object (5) is provided by the vehicle door.