EP2786564A1

EP2786564A1 - Method and device for monitoring a monitoring region

Info

Publication number: EP2786564A1
Application number: EP12794848.7A
Authority: EP
Inventors: Markus HERRLI ANDEREGG; Jonas HAGEN; David Studer; Martin Wüthrich
Original assignee: Xovis AG
Current assignee: Xovis AG
Priority date: 2011-11-29
Filing date: 2012-11-23
Publication date: 2014-10-08
Also published as: WO2013078568A1; US9854210B2; US20140327780A1; EP2600611A1; EP2600611A8

Abstract

The invention relates to a method and a device (1) for monitoring a monitoring region (2) with at least two image sensors (3, 4). A sub-region (5, 6) of the monitoring region (2) is monitored by each of the image sensors (3, 4), wherein each image sensor (3, 4) detects objects (9.1, 9.2) to be monitored that are located within the sub-region (5, 6) monitored by said image sensor, and each image sensor (3, 4) outputs data relating to the detected objects (9.1, 9.2). The image sensors (3, 4) are disposed and oriented in such a way that the monitored sub-regions (5, 6) overlap and that each object (9.1, 9.2) to be monitored that is located in the monitoring region (2) is always detected by at least one image sensor (3, 4). A total of the objects to be monitored in the monitoring region (2) is determined from the data from the image sensors (3, 4), wherein by means of a computation unit (8) on the basis of the data from the image sensors (3, 4) the objects (9.1, 9.2) to be monitored in overlapping sub-regions (5, 6) which are detected by more than one image sensor (3, 4) can be co-ordinated with one another by means of evaluation of the match in order to determine the total of the objects to be monitored in the monitoring region (2). The method and the device are suitable for tracking people and for counting people and function with sensor overlap or with camera overlap.

Description

Method and device for monitoring a surveillance area

Technical area

The invention relates to a method and a device for monitoring a surveillance area with at least two image sensors. A partial area of the surveillance area is monitored by each of the image sensors in that localized objects to be monitored are detected by each image sensor within the subarea monitored by it and data about the detected objects is output by each image sensor. The image sensors are arranged and aligned such that the monitored subareas overlap and that each object to be monitored, which is located in the surveillance area, is always detected by at least one image sensor. From the data of the image sensors, a totality of the objects to be monitored in the monitored area is determined. State of the art

Various methods and devices are known, which belong to the above-mentioned prior art. An example of this is that in WO 2005/027533 of the Intellivid Corp. described computer-assisted video surveillance system, which can monitor different sections with different cameras. Each of these subareas is subdivided into subareas and in each subarea the detected objects are counted. Based on statistics on the presence of objects in the sub-areas, a probability is determined that sub-areas monitored by different cameras will match. As a result, objects which are located in such subareas can be assigned to one another.

A similar procedure is described in EP 1 676 239 B1 of Sensormatic Electronics. Here too, different subareas are monitored with different cameras, each subarea being subdivided into subareas and the detected objects being counted in each subarea. It is considered that objects in overlapping fields of view can be from multiple cameras. For this purpose, a central computer has information about the overlapping subregions, which are obtained on the basis of correlations between images of the cameras or which are explicitly specified. Objects that are in such subareas are assigned to each other. To improve this mapping, features of the objects, such as their color, can be used.

The disadvantage of the in WO 2005/027533 of the Intellivid Corp. and systems described in EP 1 676 239 B1 of the Sensormatic Electronics, the assignment of the objects takes place primarily via an allocation of the overlapping subregions. On the one hand, this leads to inaccuracies being too large for subregions because there may be several objects in such a subarea. On the other hand, this leads to misallocations occurring within a portion monitored by a camera during the transition of an object from one sub-area to another sub-area. A somewhat different approach is described in US 7,149,325 B2 to Honeywell. Here, the fields of view of several cameras are assembled into a common, superordinate picture. For this purpose, homography matrices are formed by matching excellent points in the overlapping areas. The pictures taken by the cameras are equalized for the combination to the higher picture. Foreground objects, such as As moving people are modeled as pixel collections and tracked in their movement. An important feature used to identify individuals is a color signature. The disadvantage of this approach is that in a first step, the images of the cameras must be put together. If the cameras are video cameras, the image sequences of the different cameras must be synchronized in order to be able to assemble the corresponding sequences of the superordinate images. This leads to an increase in the cost of installing the cameras. Instead of modeling the persons as pixel aggregates described in US Pat. No. 7,149,325 B2 to Honeywell, other types of modeling can also be used. Such an example is disclosed in US 2008/1 18106 A1 to Kilambi et al. described. Here, groups of people are modeled as elliptical cylinders. These cylinders are used for subsequent calculations, in particular for determining the number of persons in the group.

Another way to model people is in US 7,529,388 B2 to Brown et al. described. Here the individual persons are modeled as ellipses.

Presentation of the invention

The object of the invention is to provide a method of the technical field mentioned above and a corresponding device, which allow the data from the image sensors improved determination of the entirety of the monitored objects in the surveillance area. The solution of the problem is defined by the features of claim 1. According to the method according to the invention, the objects to be monitored in overlapping partial areas, which are detected by more than one image sensor, are assigned to one another by means of an evaluation of their agreement in order to determine the entirety of the objects to be monitored in the monitored area. According to the device according to the invention, the objects to be monitored in overlapping partial areas, which are detected by more than one image sensor, can be assigned to each other by means of a calculation unit based on the data of the image sensors in order to determine the entirety of the objects to be monitored in the monitored area. In this case, the calculation unit may be formed as a separate unit or may be integrated into one of the at least two image sensors. However, there is also the possibility that more than one or all of the at least two image sensors comprise such a calculation unit. The image sensors used for this can be any image sensor type. For example, they may be cameras that only occasionally capture an image. In this case, the detection of an image can each be triggered by a motion sensor. But there is also the possibility that the image sensors capture images at regular intervals. There can be temporal distances of any length between the individual pictures. These distances can also be arbitrarily short. In the latter case, the image sensors may be, for example, cameras which record film sequences. Regardless of whether the image sensors capture still images or film sequences, the image sensors can be cameras that record optical images in the visible range. It can also be any other camera type. For example, the image sensors may also be infrared cameras, ultraviolet cameras, or other cameras available per se that capture images of electromagnetic radiation of any other wavelength or range of wavelengths. Furthermore, the image sensors may, for example, also be cameras which record sound or ultrasound images. But there is also the possibility that the image sensors are laser sensors. Regardless of the type of image sensors, there is a possibility that some or all of the image sensors include two sensors. By such an image sensor, detection of the three-dimensional space may be possible. By way of example, such an image sensor can be an image sensor with two cameras arranged next to one another. For example, not only a position of an object in an image plane can be detected by the stereo image captured by the two cameras, but also a position of an object at a depth perpendicular to the image plane can be determined. However, the invention can be realized expressly without any detection of stereo images. However, there is also the possibility that the image sensors are sensors which enable detection of the three-dimensional space. For example, they can be 3D laser sensors which scan the room and can also detect a distance of the objects to be monitored from the respective SD laser sensor.

Just as different image sensors can be used, different objects can also be monitored. For example, the objects to be monitored may be persons. In this case, for example, a distinction can be made between adult persons and children. Furthermore, the objects to be monitored may, for example, also be animals. It is also possible to differentiate between different animal sizes. However, there is also the possibility that the objects to be monitored are vehicles or other moving objects. It can be distinguished, for example, between trucks, cars and motorcycles. Depending on the place of application and objects to be monitored, different image sensors can be used, which are optimized for the corresponding application. In this case, for example, the light conditions can be considered for a choice of image sensors.

The entirety of the objects to be monitored determined from the data of the image sensors relates to the objects to be monitored, which are located in the entire surveillance area. In this case, this entirety of the objects to be monitored may comprise only a number of the total detected objects or an identification for each object to be monitored. But there is also the possibility that the Entity of objects to be monitored for each of the objects includes more data. For example, the further data may include the positions of the objects. For example, you can also include the speeds of the objects. Furthermore, there is also the possibility that the further data capture a course of movement of the objects. If the totality of the objects to be monitored, together with the identification, comprises further data, then these further data can be listed, for example, as a list. The individual objects can be identified based on their position in the list, which eliminates the need for separate identification of the objects.

The evaluation of the match used for the assignment of the objects to be monitored in overlapping partial areas can take place in different ways. For example, it can be a positive rating. In doing so, the match of objects that is more likely to be scored is rated larger, while the score of objects that are less likely to score is rated smaller. There is also the possibility that the individual ratings for probabilities are normalized. Alternatively, the match score may be a negative score (cost score) where the match of items that is more likely to be scored is rated smaller, while the score of items that are less likely to score is to be valued with larger values. As a further variant, however, there is also the possibility that a certain evaluation value represents a highest probability of a match. In doing so, both overlying and underlying rating values may represent a smaller probability of a match. By way of example, underlying and underlying valuation values can thus represent a statement about a used valuation type or valuation criteria used.

The solution of the invention has the advantage that the reliability of the assignment of the objects is increased by the evaluation of the agreement. Accordingly, an improved monitoring of the surveillance area is ensured. In addition, the method and the device are suitable for person tracking and for counting people and function across sensors or across cameras. Preferably, each object to be monitored, which is located in the monitored area, is always completely detected by at least one image sensor. This has the advantage that the reliability of the assignment is increased because the assignment of the objects detected by different image sensors takes place only for completely detected objects. Alternatively, however, there is also the possibility that the objects to be monitored are not always completely covered by at least one image sensor. This can be advantageous in order to also detect edge areas of the monitoring area and to reduce a number of the required image sensors.

In each case, a matrix is preferably created for the assignment of the objects to be monitored in overlapping subareas whose elements pjj contain the evaluation of how well an object detected by an image sensor with identification i matches an object with identification j detected by another image sensor. There is the possibility that the indices i and j run over all objects to be monitored, which are located in the subarea that is monitored by the corresponding image sensor. However, there is also the possibility that the indices i and j run only over those objects which are located in an overlapping region of the two subregions and are detected by the corresponding image sensor.

If three or more partial areas which are monitored by a corresponding number of image sensors overlap one another, then there is the possibility that, in a first step, the objects to be monitored are assigned to two of the partial areas. In a further step, an assignment of the objects to be monitored in the already processed subareas and the objects to be monitored in a further subarea can then be made. This further step can be repeated until all overlapping subareas have been processed.

In the case of three or more overlapping partial regions, however, in one variant the created matrix can also be more than two-dimensional. For example, the matrix may comprise the same number of dimensions as the number of overlapping subregions. Accordingly, the number of indices of the matrix elements pjj must be adjusted. In doing so, the indices can, in turn, contain all the objects to be monitored in the corresponding subarea run or only run over those objects that are within an overlap area of the subregions.

It is possible to determine only matches for matches where there is at least a certain probability of a match. There are no items for the unvalued matches. Accordingly, storage space and calculation time can be saved thereby.

Preferably, for each combination of possibly matching objects, a sum of the scores of the matching objects in this combination is formed. In this case, the assignment of the objects to be monitored preferably takes place by a choice of the combination, which results in an extreme of this sum. There is a possibility that the sum of the reviews will be maximized. This is advantageous when a large score value represents a high probability of a match, while a small score value represents a small probability of a match. The sum can also be minimized. This is advantageous when a small score value represents a high probability of a match, while a large score value represents a small probability of a match. Both maximizing the sum and minimizing the sum have the advantage that an optimal allocation of the objects to be monitored can be achieved in a simple manner. For example, if a particular score represents a maximum likelihood of a match, and both score and score below represent a smaller probability of a match, then a variant of the difference between the score with the highest probability and score values of the items matching that combination be formed. In this case, the assignment of the objects to be monitored can be achieved by selecting the combination with one extreme of this sum of the differences.

As a variant of these embodiments, which provide the formation of a sum, for example, a sum of functional values of the evaluation values or the Differences are formed. As a function, for example, the evaluation values or the differences can be squared before the summation, or the square root of the evaluation values or differences can be drawn in each case. However, there is also the possibility that the summands are calculated from the evaluation values or differences by means of any other formula or function. Depending on the specific embodiment, the use of such a formula or function may be advantageous since a more stable and reliable algorithm for the assignment of the objects to be monitored can be achieved.

As an alternative to these variants, however, the assignment of the objects to be monitored can also be carried out differently than by the formation of a sum of the evaluation values.

Advantageously, in the evaluation of the conformity of objects to be monitored, it is taken into consideration that the objects to be monitored can also be detected by only one image sensor. The fact that an object to be monitored can only be detected by an image sensor can be due, for example, to the fact that an object to be monitored is located at a position monitored only by an image sensor. However, it may also be due to the fact that, while an object to be monitored is located in an overlapping region of two or more subregions, it is only detected by an image sensor. In both cases, taking into account that the objects to be monitored can also be detected by only one image sensor has the advantage that an improved assignment of the objects to be monitored is achieved.

If, in each case, a matrix is created for the assignment of the objects to be monitored in overlapping subareas, such a co-consideration can be implemented, for example, by adding in the matrix a column or a row whose elements contain the evaluation that no match with one monitoring object in the sub-area, which is represented by the columns or rows.

As a variant, there is also the possibility that it is assumed for the evaluation of the matches of the objects to be monitored that each object is detected by at least two image sensors. This variant can be advantageous if an assignment takes place only for objects to be monitored, which are located in an overlapping area of two or more subareas. This allows the assignment to be made faster.

Preferably, unassigned objects are included as separate objects in the entirety of the objects to be monitored. This has the advantage that even non-assignable or only with low probability attributable objects are included in the entirety of the objects to be monitored. This allows a more complete and better determination of the entirety of the objects to be monitored. Alternatively, however, there is also the possibility that unassigned objects are not included in the entirety of the objects to be monitored.

Preferably, the data output by the image sensors is output anonymously. As a result, the data can be further processed to the entirety of the objects to be monitored, whereby data protection regulations can be adhered to without special precautions being taken to specifically protect further data processing. The anonymization consists, in particular, in the fact that the image sensors do not output any data which could directly or indirectly permit a conclusion on the identity of monitored persons. For example, the captured image data are not output but also no information about recorded colors and / or body mass of the monitored persons.

Alternatively, however, there is also the possibility that the data output by the image sensors are not output anonymously.

Preferably, in the data output by the image sensors, each detected object is represented by a parameterized model. There is the possibility that such a parameterized model consists solely of a position of the detected object or of a beam which, starting from the corresponding image sensor, passes through the respective detected object. This has the advantage that anonymized data about the objects to be monitored can be output by parameterizing the models. Furthermore, there is also the possibility that a parameterized model also is more complex. For example, a size, a shape or an orientation of the objects to be monitored can be characterized by the parameters. This has the advantage that, despite the output of such parameters, the data on the detected objects can be anonymized. As a variant, however, there is also the possibility that in the data output by the image sensors, the detected objects are not represented by a parameterized model.

If, in the data output by the image sensors, each detected object is represented by a parameterized model, preferably for the evaluation of the match, one from a minimum distance between a ray from an image sensor through a centroid of the parameterized model and a ray from takes into account a value calculated for another image sensor by a centroid of the parametric model. In this case, the center of gravity of the parameterized model is a reference point of the respective object to be monitored, which in the present document is also referred to as the center of gravity of an object. The use of this term does not mean that it is the physical center of gravity of the object. Although the point may be at the physical center of gravity of an object, it may be located elsewhere in or around the object. The exact arrangement of such a center of gravity can be determinable, for example, on the basis of the parameters output by the corresponding image sensor and the parametric model used. The center of gravity can, however, also be determinable, for example, directly in the corresponding image sensor on the basis of the parametric model used, with the image sensors each outputting only data to the beam starting from the respective image sensor through the center of gravity of the corresponding object. Accordingly, the arrangement of the center of gravity in the parameterized model or in the object to be monitored depends on the model. In addition, the center of gravity can be determined differently based on the parameters. For example, it may be a geometric center of the parameterized model. It can also be another point in or around the parameterized model. If it is the parameterized model For example, if it is an ellipse, the center of gravity may be, for example, a center of the ellipse. If the object to be monitored is a person, it may, for example, also be the position of the feet or the head of the person to be monitored, which is determined on the basis of the parameters of the parameterized model.

If, in the data output by the image sensors, each detected object is represented by a parameterized model, it is preferable for each object to be monitored of the entirety of the objects to be monitored to store a height of the center of gravity of the corresponding object, starting with a meaningful height as the value of that height is assumed and the value of this height is dynamically adjusted by determining, for each assignment of objects to be monitored in which the corresponding object to be monitored is detected by at least two image sensors, a height of the center of gravity of that object, by a height of a center of a shortest Connection line between a beam from the one image sensor is determined by the center of gravity of the parameterized model detected by him and a beam from the other image sensor by the center of gravity of the parameterized model he has detected. For example, the value of the height can be adapted dynamically to the last specific value. However, there is also the possibility that the value of the height is in each case dynamically adapted to an average of the previously determined values of the height. Both have the advantage that the positions of the objects to be monitored of the entirety of the objects to be monitored can be determined more precisely, by using a respective beam from the corresponding image sensor through the center of gravity of the parameterized model detected by it, the position of the respective object corresponds to the point of intersection of the beam with a height of the center of gravity of the corresponding object above the bottom of the corresponding subregion.

However, there is also the possibility that no height of the center of gravity of the objects to be detected is determined and stored or adapted dynamically.

If, in the data output by the image sensors, each detected object is represented by a parameterized model, it is preferred for the evaluation of the Agreement takes into account whether a center point of a shortest connecting line between the beam from the one image sensor through the center of gravity of the parameterized model detected by it and the beam from the other image sensor through the center of gravity of the parameterized model detected by it, at a reasonable height for a height of Center of gravity of the objects to be monitored. There is the possibility that a fixed range for a reasonable height is given. However, it is also possible that an area around a stored and dynamically adjusted value of the height of the center of gravity of the corresponding object is used for the determination of a meaningful height. Both have the advantage that the evaluation takes into account how plausible the coincidence of two objects is due to the height of their center of gravity. The use of a fixed range for a meaningful height also has the advantage that the objects to be monitored can be selected according to the height of their center of gravity. As a variant of this, there is also the possibility that, in the evaluation of the match, no height of the center of the shortest connecting line between the beam from one image sensor through the center of gravity of the parameterized model detected by it and the beam from the other image sensor through the center of gravity of the latter captured parameterized model. This is advantageous if no selection of the objects to be monitored should take place according to the height of their center of gravity. For example, this may be the case when the objects to be monitored include both adults and children of different ages and, where appropriate, animals such as dogs or cats.

If, in the data output by the image sensors, each detected object is represented by a parameterized model, a similarity of the parameterized models of the objects to be monitored for the evaluation of the match is preferably taken into account. There is the possibility that the parameters that are output from the different image sensors to the modeled objects are directly compared. However, there is also the possibility that, for example, a position of the models output by the image sensors is compared in space, wherein an orientation and positioning of the image sensors are taken into account. For example, if a value for the height of its center of gravity is stored and possibly dynamically adjusted for the detected objects to be monitored, the position of the respective object may be determined by the intersection of the beam from the corresponding image sensor through the center of gravity of the object with a height of the center of gravity of the corresponding object Object above the bottom of the corresponding portion can be determined, wherein the thus determined based on the output from the various image sensors data positions of the objects are compared. Furthermore, there is also the possibility that, for example, in the image sensors, the detected objects are modeled with a three-dimensional model such as, for example, an ellipsoid, a cylinder or a rectangular block. For example, there is the possibility that the data output by the image sensors does not contain any parameters from this three-dimensional model, but that the data output from the image sensors include parameters of a two-dimensional model representing a projection of the three-dimensional model onto a two-dimensional surface, the two-dimensional Area corresponds to a modeling of the monitored subareas. In this case, the data output by the image sensors may be, for example, parameters for an ellipse, a rectangle or another geometric shape. If, in this example, the similarity of the parameterized models of the objects to be monitored is taken into account for the evaluation of the conformity, then for example an orientation and a positioning of the image sensors as well as the fact that the parameters output are parameters of a model can be taken into account, which corresponds to a projection. As an alternative, there is also the possibility that a similarity of the parameterized models of the objects to be monitored is not taken into account for the evaluation of the match.

Advantageously, speeds and directions of movement of the objects to be monitored are taken into account for the evaluation of the conformity of the image sensors. This has the advantage of not being due to different movement obviously mismatched objects due to their movement behavior can be evaluated accordingly and held by an assignment.

As a variant of this, there is also the possibility that for the evaluation of the conformity only speeds or directions of movement of the objects to be monitored are taken into account, either by the image sensors. This also has the advantage that mismatched objects can be appropriately evaluated due to their movement behavior and prevented from being assigned.

Alternatively, however, there is also the possibility that for the evaluation of the conformity neither speeds detected by the image sensors nor directions of movement of the objects to be monitored are taken into account.

Preferably, the assignments of the objects to be monitored which have been made earlier are taken into account for the evaluation of the match. This has the advantage that a consistency of the assignments of the objects to be monitored can be achieved over a period of time. This is particularly advantageous when the image sensors provide data on the detected objects in short time intervals. For example, this may be the case when the image sensors are cameras that record film sequences.

Alternatively, however, there is also the possibility that no earlier assignments of the objects to be monitored are taken into account for the evaluation of the match. Such an alternative is indicated, for example, when the image sensors provide data on the detected objects at time intervals, with the objects to be monitored in each case being able to have moved over considerable distances in the meantime. Preferably, several of the above-mentioned evaluation criteria are used for the evaluation of the match. The different evaluation criteria can be weighted differently. This has the advantage that for different types of objects to be monitored and different situations in which the objects be monitored, an optimal evaluation of the agreement can be achieved.

However, as a preferred variant, there is also the possibility that only one of the above-mentioned evaluation criteria is used for the evaluation of the match. Depending on the objects to be monitored and the situations in which the objects are to be monitored, an optimal evaluation of the match can also be achieved.

Alternatively, however, it is also possible for one or more other evaluation criteria, such as, for example, a color of the objects to be monitored or another identification of the objects detected by the image sensors, to be used as an evaluation criterion for the evaluation of the match. There is also the possibility that such other evaluation criteria may be used together with the above evaluation criteria to evaluate the match. It is also possible that different scoring criteria or different combinations of scoring criteria are used to match the objects in different regions within the surveillance area. This can be advantageous, for example, if the lighting conditions within the monitoring area are very different. In this case, it may be useful, for example, to weight the speed of the objects to be monitored more than a size of the objects in a rather dark region, since a size estimation of the objects is less accurate due to the lighting conditions. At the same time, in a brighter region, a size of the objects may be relatively more weighted for the evaluation of the match. Preferably, the surveillance area is modeled as a two-dimensional area with a two-dimensional coordinate system, wherein each of the objects to be monitored of the entirety of the objects to be monitored is characterized by data relating to this coordinate system. In this case, the characterizing data, for example, an identification of the corresponding object contain. The data may also include, for example, information about a position, a speed, a course of movement or the like of the corresponding object. This has the advantage that a spatial assignment of the individual objects is made possible. As a variant of this, there is also the possibility that the surveillance area is modeled as a three-dimensional space with a three-dimensional coordinate system, wherein each of the objects of the entirety of the objects to be monitored is characterized by data relating to this coordinate system. The characterizing data may include, for example, an identification of the corresponding object. The data may also include, for example, information about a position, a speed, a course of movement or the like of the corresponding object. This variant also has the advantage that a spatial assignment of the individual objects is made possible.

As a further variant, there is also the possibility that the surveillance area is modeled as a one-dimensional space or as a line with a one-dimensional coordinate system, wherein each of the objects of the totality of the objects to be monitored is characterized by data relating to this coordinate system. This is advantageous, for example, if the surveillance area has an elongated shape and objects can only move along this elongate shape. For example, such a surveillance area may be a street, a corridor, a roller conveyor, or any other laterally limited, elongated space. Of course, however, a street, a corridor or a conveyor belt can also be modeled as a two- or three-dimensional space. The latter can be advantageous if, for example, vehicles on the road or persons or animals in the aisle or on the roller conveyor can overtake or cross one another.

If a movement course of the objects is to be recorded, there is also the possibility that the surveillance area is modeled as a one-, two- or three-dimensional space with an additional time dimension. Alternatively, there is also the possibility that the surveillance area is not modeled as one-, two-, or three-dimensional space. Such an alternative may be advantageous if, for example, only a number of objects to be monitored in the monitoring area to be determined. In this case, no data is needed on the location of the objects, which also requires less computational capacity. In this case, for example, the monitoring area can also be modeled only by a time dimension in order to determine a temporal change of the number of objects to be monitored in the monitored area.

Preferably, each partial area monitored by an image sensor is modeled as a two-dimensional area having a two-dimensional coordinate system, wherein the data output by each image sensor relating to the detected objects relate to the coordinate system of the partial area monitored by this image sensor. In this case, the data output by the image sensors may contain, for example, an identification of the corresponding object. For example, the data may also include information about a position, a speed, a course of movement or the like of the corresponding object. For example, the output data may also include information about a size or other characteristics of the object. This has the advantage that a spatial assignment of the individual objects is made possible and that, if appropriate, certain features of the objects to be monitored can be detected.

As a variant of this, it is also possible for each subarea monitored by an image sensor to be modeled as a three-dimensional space with a three-dimensional coordinate system, wherein the data output by each image sensor relating to the detected objects relate to the coordinate system of the subarea monitored by this image sensor. In this case, the data output by the image sensors may contain, for example, an identification of the corresponding object. For example, the data may also include information about a position, a speed, a course of movement or the like of the corresponding object. For example, the output data may also include information about a size or other characteristics of the object. This variant also has the advantage that a spatial assignment of the individual objects is made possible and that, if appropriate, certain features of the objects to be monitored can be detected.

As a further variant of this, it is also possible for each subarea monitored by an image sensor to be modeled as a one-dimensional space or as a line with a one-dimensional coordinate system, with the data output by each image sensor relating to the detected objects being based on the coordinate system of the one monitored by this image sensor Refer to subarea. This is advantageous, for example, if the corresponding subregion has an elongated shape and objects can only move along this elongated shape. For example, such a subarea may be a roadway, a hallway, a conveyor belt, or any other laterally limited, elongated space. Of course, however, a street or a roller conveyor can also be modeled as a two- or three-dimensional space. This can be advantageous if, for example, vehicles on the road or persons or animals in the corridor or on the roller conveyor can overtake or cross each other.

If a movement course of the objects is to be recorded, it is also possible for each subarea monitored by an image sensor to be modeled as a one-, two- or three-dimensional space with an additional time dimension.

Alternatively, however, there is also the possibility that not every partial area monitored by an image sensor is modeled as a one-, two- or three-dimensional space. For example, each sub-area monitored by an image sensor can be modeled as a different-dimensional space. In addition, there is the possibility that one, several or all monitored subareas are modeled as more than three-dimensional spaces. Advantageously, the data output by the image sensors to each of the detected objects to be monitored includes an identification and X and Y coordinates. In a preferred variant, the data output by the image sensors also includes a velocity vector for each of the detected objects to be monitored. Alternatively, however, there is also the possibility that the data output by the image sensors comprise different data for each of the detected objects to be monitored.

If the subregions monitored by the image sensors are modeled as two-dimensional surfaces with a two-dimensional coordinate system, the objects detected by the image sensors are advantageously modeled as ellipsoids whose projection onto the two-dimensional surface of the corresponding subregion yields an ellipse, the data output by the image sensors for each of the detected objects to be monitored, an identification, X and Y coordinates of a center of the ellipse, sizes of the major axes of the ellipse, an orientation angle of the ellipse, and a velocity vector. This has the advantage that the data output by the image sensors are adequately anonymized. At the same time, however, the data output by the image sensors enables the objects to be distinguished, as well as positioning and monitoring the speed of the objects. Furthermore, these output data have the advantage that they allow an estimation of a size and a position of the objects to be monitored in the room.

As a variant, there is also the possibility that the objects detected by the image sensors are modeled differently. For example, the objects can be modeled directly as ellipses. However, there is also the possibility that the objects are not modeled as ellipsoids but, for example, as cylinders or rectangular blocks. In the same way, however, the objects can also be directly modeled as projections of such a cylinder or as a projection of such a rectangular block or directly as a rectangle. In this variant, the data output by the image sensors may include corresponding parameters of the modeling of the objects used instead of the major axes of the ellipses and the orientation angle of the ellipse. For example, these parameters may be a height and a radius and an orientation angle of a cylinder. However, it may also be, for example, the side lengths and the orientation angle of a rectangle.

If the subregions monitored by the image sensors are modeled as three-dimensional spaces with a three-dimensional coordinate system, then advantageously the objects detected by the image sensors are modeled as ellipsoids, wherein the data output by the image sensors to each of the detected objects to be monitored an identification, X, Y and Z coordinates of a center of the ellipsoid, sizes of the major axes of the ellipsoid, two Orientation angle of the ellipsoid and a velocity vector include. This also has the advantage that the data output by the image sensors are anonymized. At the same time, however, the data output by the image sensors enable identification of the objects as well as positioning and monitoring of the speed of the objects. Furthermore, these output data have the advantage that they allow an estimation of a size and a position of the objects to be monitored in the room.

As a variant, there is also the possibility that the objects detected by the image sensors are modeled differently. For example, the objects can not be modeled as ellipsoids but, for example, as cylinders or rectangular blocks. In this variant, the data output by the image sensors may include, instead of the major axes of the ellipsoid and the two angles of orientation of the ellipsoid, corresponding parameters of the modeling of the objects used. For example, these parameters may be a height and a radius, as well as two orientation angles of a cylinder. However, it may also be, for example, the side lengths and two orientation angles of a rectangular block. Alternatively, there is also the possibility that the objects detected by the image sensors are modeled differently and that the data output by the image sensors comprise different data to the detected objects. For example, the data may include only one position of the detected objects. However, there is also the possibility that the data include information such as colors or identification codes of the detected objects.

If the data output by the image sensors relate to a one-, two- or three-dimensional coordinate system, the coordinate systems to which the data output from the image sensors relate are advantageously equalized. In this case, the coordinate systems are preferably equalized by straightening objectively caused curvatures of the coordinate systems. This can For example, be the case when the image sensors used are cameras that use a wide-angle lens. However, it is also possible, for example, for the coordinate systems to which the data output by the image sensors relate to be equalized by being adapted to the coordinate system to which the data output from one of the image sensors is adapted. Both have the advantage that the coordinate systems to which the data output from the image sensors relate can be compared more easily with each other.

As a preferred variant of this, there is also the possibility that, if the data output by the image sensors relate to a one-, two- or three-dimensional coordinate system, the coordinate systems are already equalized in the corresponding image sensor, as a result of which the data output by the image sensors is already on refer to equalized coordinate systems. This also has the advantage that the coordinate systems to which the data output from the image sensors relate can be compared more easily. Alternatively, however, there is also the possibility that if the data output by the image sensors relate to a one-, two- or three-dimensional coordinate system, the coordinate systems to which the data output from the image sensors relate are not equalized.

If the data output from the image sensors relates to a two-dimensional coordinate system and the surveillance area is modeled by a two-dimensional coordinate system, it is preferable to transmit the data output from the image sensors to the two-dimensional coordinate system of the surveillance area.

If the data output from the image sensors relates to a three-dimensional coordinate system and the surveillance area is modeled by a three-dimensional coordinate system, the data output from the image sensors are preferably transmitted to the three-dimensional coordinate system of the surveillance area. If the data output by the image sensors relate to a one-dimensional coordinate system and the surveillance area is modeled by a one-dimensional coordinate system, the data output by the image sensors are preferably transmitted to the one-dimensional coordinate system of the surveillance area.

This transmission can be done for example by a conversion, in which an alignment and positioning of the different image sensors in the coordinate system of the monitoring area are taken into account. Accordingly, the coordinate systems to which the data output from the image sensors relate may be rotated, for example. In addition, for example, length units of the coordinate systems to which the data output by the image sensors relate can be converted to the coordinate system of the monitored area.

If the data output by the image sensors relate to a two- or three-dimensional coordinate system and the surveillance area is modeled by a one-dimensional coordinate system, the data output by the image sensors are preferably transmitted to the one-dimensional coordinate system of the surveillance area.

If the data output by the image sensors relate to a three-dimensional coordinate system and the surveillance area is modeled by a one- or two-dimensional coordinate system, the data output by the image sensors are preferably transmitted to the one-dimensional or two-dimensional coordinate system of the surveillance area.

In both cases, when transferring the data to the coordinate system of the surveillance area, the data may be projected onto the coordinate system of the surveillance area. If, for example, the data contain parameters of a modeling of the objects to be monitored, then for example the model can be projected onto the coordinate system. This means that if the data output by the image sensors contain parameters of an ellipsoid, for example this ellipsoid is projected onto a two-dimensional coordinate system of the surveillance area. If, on the other hand, the data output by the image sensors contains, for example, parameters of an ellipsoid or an ellipse and the surveillance region is modeled by a one-dimensional coordinate system, then, for example, a longitudinal extension of the ellipsoid or the ellipse on the one-dimensional coordinate system can be taken into account in the transmission of the data.

If the data output from the image sensors relate to a coordinate system with fewer dimensions than the coordinate system of the monitored area, it is preferable to transmit the data output from the image sensors to the coordinate system of the surveillance area by spatially positioning an alignment of the coordinate systems of the image sensors in the coordinate system of the surveillance area become.

If the data output by the image sensors relate to a coordinate system of the respective image sensor and if the surveillance area is modeled by a two-dimensional coordinate system, preferably a position and an orientation of each image sensor are stored and transferred to the data output from the image sensors considered two-dimensional coordinate system of the surveillance area.

If the data output by the image sensors relate to a coordinate system of the respective image sensor and if the surveillance area is modeled by a one- or three-dimensional coordinate system, a position and an orientation of each image sensor are preferably stored and transmitted for transmission by the image sensors Data taken into account on the one- or three-dimensional coordinate system of the surveillance area. In both cases mentioned above, this has the advantage that an optimal transfer of the data to the coordinate system of the monitoring area is achieved.

As a variant of this, however, there is also the possibility that, if the data output by the image sensors relate to a coordinate system of the respective image sensor and if the monitoring area is modeled by a coordinate system, a position and an orientation of the coordinate systems of the image sensors is determined in each case on the basis of features in the image data of the image sensors, this position and orientation for the transmission of the data output from the image sensors are taken into account on the coordinate system of the monitoring area. Preferably, the image sensors are mounted overhead. This means that the image sensors are aligned substantially vertically downwards and accordingly capture images of the events below the image sensors. The image sensors are thus expediently arranged above the objects to be monitored. The overhead installation of the image sensors has the advantage that the subregions monitored by the image sensors are monitored from above. If the subregions are essentially horizontal surfaces which, although they may have bulges and inclinations, the subregions are thereby monitored from a position substantially perpendicular to their surface. Accordingly, the objects to be monitored move on a surface which is substantially parallel to the image plane of the image sensors. This enables optimal detection of positions and speeds of the objects to be monitored. In addition, this has the advantage that the objects to be monitored only in very few cases from the perspective of the image sensors move each other and can hide each other. If the image sensors are mounted overhead and the objects to be monitored are persons, there is a danger, for example, of marginal areas of the monitored subareas or of an adult person bending over a child, that persons hide each other. Otherwise, the persons can be optimally monitored by this arrangement of the image sensors.

As a variant, there is also the possibility that the image sensors are not mounted overhead. For example, they can be aligned obliquely downwards or horizontally laterally. Depending on the objects to be monitored and the geometry of the monitored area, such an arrangement of the image sensors may also be advantageous. This may be the case, for example, when monitoring a conveyor belt on which there are persons or animals that are not overtaking each other can. However, this can also be the case, for example, when objects transported on a conveyor belt are to be monitored.

From the following detailed description and the totality of the claims, further advantageous embodiments and feature combinations of the invention result.

Brief description of the drawings

The drawings used to explain the embodiment show:

Fig. 1 is a schematic representation of an inventive device for

Monitoring a surveillance area,

2a, b, c each show a schematic illustration for illustrating the detection of an object to be monitored by an image sensor,

Fig. 3 is a further schematic representation of the inventive

Device for monitoring the surveillance area and

Fig. 4 is a further schematic representation of the inventive

Device for monitoring the surveillance area.

Basically, the same parts are provided with the same reference numerals in the figures.

Ways to carry out the invention

1 shows a schematic representation of a device 1 according to the invention for monitoring a monitoring area 2. This device 1 comprises a first image sensor 3 and a second image sensor 4. Both image sensors 3, 4 each comprise a camera which can record film sequences. The first image sensor 3 monitors a first subarea 5 of the surveillance area 2, while the second image sensor 4 monitors a second subarea 6 of the surveillance area 2. The two subregions 5, 6 overlap in an overlapping area 7 and together cover the entire surveillance area 2.

Furthermore, the device 1 comprises a calculation unit 8. This calculation unit 8 can be, for example, a server or else a computer. The two image sensors 3 and 4 are connected to the calculation unit 8 and output data to the calculation unit 8. In another embodiment, however, the calculation unit 8 can also be integrated into one of the image sensors 3, 4. There is also the possibility that several of the image sensors 3, 4 comprise such a calculation unit 8. The image sensors 3, 4 can detect objects to be monitored within the subarea 5, 6 monitored by them. For example, the objects to be monitored may be persons. It can also be animals, vehicles or objects.

Figures 2a, 2b and 2c each show a schematic representation for illustrating the detection of an object to be monitored by an image sensor. In the following, reference is made in each case to the first image sensor 3 and the first subregion 5 for the image sensor shown and the subregion shown. However, this is an exemplary description, which also stands for other image sensors and other sub-areas. In particular, this description also stands for the second image sensor 4 described above and the second subregion 6 described above.

FIG. 2 a shows the first image sensor 3 and the first subregion 5 monitored by it. In the first partial area 5 there is a person 9. In order to detect this person 9, the first image sensor 3 comprises a processing unit (not shown) which processes the image data recorded by the first image sensor 3. For this purpose, it identifies objects to be monitored, which are thereby detected by the image sensor 3, and outputs data on the detected objects to be monitored. In the present case, therefore, the data output is data relating to the person 9, which is located in the first subarea 5. For processing the image data captured by the first image sensor 3, the objects to be detected are modeled as ellipsoids in three-dimensional space. As shown in Figure 2b, therefore, the person 9 is modeled by an ellipsoid 10. In this modeling, an outgoing of the image sensor 3 projection of this ellipsoid 10 on the first portion 5 an ellipse 1 1 (Figure 2c). A beam 12 emanating from the image sensor 3, which passes through the center of gravity or center of the ellipsoid 10, also extends through the center of gravity or center of the ellipse 11.

In order to simplify this modeling of the detected objects to be monitored, the image data acquired by the first image sensor 3 are equalized in the processing unit in a first step and the rectified image data are provided with a two-dimensional coordinate system which extends over the monitored first subregion 5. In a second step, the objects to be monitored in the image data are identified by applying a known method for object identification. By way of example, this method can be a method in which a still image (without any object in the monitored first subarea 5) is subtracted from the image data. In a third step, an ellipse 1 1 is placed on the detected objects. For this purpose, lengths of the two main axes of the ellipse 1 1, a position of the ellipse 1 1 and an angle between one of the two main axes of the ellipse 1 1 and a coordinate axis of the first portion 5 are fitted to the corresponding detected object. These data on the detected objects are output by the processing unit and the first image sensor 3 to the computing unit 8 (see FIG. 1) where they are further processed.

FIG. 3 shows, as already shown in FIG. 1, a schematic representation of the device 1 according to the invention for monitoring the monitoring area 2. In contrast to FIG. 1, however, there are two persons 9.1, 9.2 in the second subarea 6. A first of these two persons 9.1 is also located in FIG It is detected by both the first and the second image sensor 3, 4. By contrast, the second person 9.2 is located only in the second subarea 6 and is only detected by the second image sensor 4. For the first person 9.1, which is detected by both image sensors 3, 4, the rays 12.1, 12.2 are shown, which starting from the respective image sensor 3, 4 by the center of gravity or center of the modeled ellipsoid and the corresponding modeled ellipse 1 1.1, 1 1.2 run. In addition, the two ellipses 1 1.1, 1 1.2 are shown, which are detected by the two image sensors 3, 4 for the first person 9.1. The parameters of the two ellipses 1 1.1, 1 1.2 are output from the image sensors 3, 4 to the calculation unit 8.

Further, for the second person 9.2, which is detected only by the second image sensor 4, the beam 12.3 is shown, which extends from the second image sensor 4 through the center of gravity or center of the modeled ellipsoid and the corresponding modeled ellipse 1 1.3. Next, the ellipse 1 1.3 is shown, which is detected by the second image sensor 4 for the second person 9.2. The parameters of this ellipse 1 1.3 are output from the second image sensor 4 to the calculation unit 8.

From these data output by the two image sensors 3, 4, a totality of the objects to be monitored is determined by the calculation unit 8. For this purpose, in a first step, the data output by the two image sensors 3, 4 is converted to a two-dimensional coordinate system which extends over the monitoring area 2. In a second step, a matrix is created on the basis of the converted data. The elements of this matrix contain evaluations for the probability that an object detected by the first image sensor 3 coincides with one of the second image sensor 4. In the present example, these scores are for probabilities, i. normalized to one. The ratings could also be normalized to another value. In a third step, on the basis of the evaluations contained in the matrix, the most probable allocation of the detected objects is determined and the entirety of the objects to be monitored is determined. In this case, the most probable assignment is determined by the calculation unit 8 using the Hungarian method, which is also referred to as the Kuhn-Munkres algorithm.

A size of the matrix created by the calculation unit depends on the number of objects detected by the two image sensors 3, 4. Because it can be that one of the one Image sensor 3, 4 detected object is not detected by the other image sensor 4, 3, the matrix contains a number of lines, which is the number of objects detected by the one image sensor 3, 4 plus one. Further, for this reason, the matrix contains a number of columns, which is the number of objects detected by the other image sensor 4, 3 plus one. In the present case, only one object to be monitored is detected by the first image sensor 3 with the first person 9.1. Therefore, the matrix determined by the calculation unit 8 has two lines. By contrast, the second image sensor 4 detects two objects to be monitored with the first and the second person 9.1, 9.2. Accordingly, the matrix determined by the calculation unit 8 has three columns. Thus, the matrix determined by the calculation unit 8 has the form:

Pu Pn Pu

The first line of this matrix relates to assessments that the first person 9.1 detected by the first image sensor 3 coincides with one person or no person detected by the second image sensor 4. The second row of the matrix, on the other hand, refers to assessments that no person detected by the first image sensor 3 agrees with one person or with no person detected by the second image sensor 4. The first column of the matrix relates to assessments that the first person 9.1 detected by the second image sensor 4 coincides with one person or no person detected by the first image sensor 3. The second column of the matrix relates to evaluations that the second person 9.2 detected by the second image sensor 4 agrees with one person or with no person detected by the first image sensor 3. The third column, on the other hand, refers to assessments that no person detected by the second image sensor 4 coincides with or with any person detected by the first image sensor 3. Thus, since the element p ₂₃ includes the judgment that no object detected by the first image sensor 3 does not coincide with any object detected by the second image sensor 4, this element can be set to a fixed value. In the following, this element is set to zero. It could also be set to any other value. In the calculation unit 8, both the positions and the orientations of the two image sensors 3, 4 are stored. Therefore, based on the data output by the first image sensor 3, the calculation unit 8 can determine whether an object detected by the first image sensor 3 is only in the first subregion 5, or whether it is also in the second subregion 6 and thus in the overlap region 7. In the same way, based on the data output by the second image sensor 4, the calculation unit 8 can therefore determine whether an object detected by the second image sensor 4 is only in the second subarea 6, or if it is also in the first subarea 5 and thus in the overlap region 7. In the example shown in FIG. 3, the first person 9.1 is in the overlapping area 7, while the second person 9.2 is located only in the second area 6. Accordingly, the calculation unit 8 sets the score p to zero and the score p _2Z to 1:

p _2X \ 0

The elements p, p ₂ and _{13 of} the matrix are determined by the calculation unit 8 by the application of boundary conditions as well as certain evaluation criteria. The boundary conditions consist in that each object detected by one image sensor 3, 4 either coincides with or does not coincide with an object detected by the other image sensor 4, 3. This results in the following boundary conditions for the present matrix: p + p ₂ = 1, p _n + p ₂₂ = 1 and p _u + p _n + p = 1. In the example shown in FIG. 3 with the first and the second person 9.1, 9.2 in the monitoring area 2, this means that the matrix has the shape

Pu 0 (1 - p _n )

receives, wherein the value of the element p is determined by the calculation unit 8 according to the evaluation criteria explained below. In a modified case, which is constructed similar to the case illustrated in FIG. 3, but the second person 9.2, which is only detected by the second image sensor 4, is likewise located in the overlap region 7, the matrix set up by the calculation unit 8 looks different. In this case, it is taken into account that one of the two persons 9.1, 9.2 detected by the second image sensor 4 could coincide with the person 9.1 detected by the first image sensor 3. Accordingly, the matrix created by the calculation unit 8 looks as follows:

Pu Pu Q - Pn - Pn

- Pu) - Pn) o

The values of the elements p and ₁₂ are determined by the calculation unit 8 according to the evaluation criteria explained below.

As already shown in FIGS. 1 and 3, FIG. 4 shows a schematic representation of the device 1 according to the invention for monitoring the monitoring area 2. In contrast to FIG. 3, no persons are shown. However, the two ellipses 1.1, 1.2 in the overlap area 7 together with the beams 12.1, 12.2 emanating from the image sensors 3, 4 are shown, whose data according to FIG. 3 for the detected first person 9.1 (not shown here) is provided by the image sensors 3 4 are output to the calculation unit 8. FIG. 4 serves to illustrate the determination of the elements of the matrix.

As already mentioned, the elements of the matrix contain the evaluations for the probability that an object detected by the first image sensor 3 coincides with an object detected by the second image sensor 4. For this purpose, different evaluation criteria can be used.

For example, a minimum distance 13 between the rays 12.1, 12.2 emanating from the image sensors 3, 4 can be used as an evaluation criterion. In this case, a minimum distance of zero (intersecting rays 12.1, 12.2) can correspond to a maximum possible rating. If the scores are expressed in terms of probabilities, this equates to a value of / ?, = 1 and all other minimum distances are rated by a smaller probability. If, on the other hand, the valuations are expressed by means of cost valuations, this corresponds to the value of p, = 0, while all other minimum distances are evaluated by a larger valuation value.

As a second possible evaluation criterion, a similarity of the ellipse sizes can be used. However, it should be considered that the ellipses are projections of an ellipsoid. Accordingly, a distance in the plane of the monitoring area 2 to the respective image sensor 3, 4 can be taken into account for the determination of the ellipse sizes. In this case, a correspondence of the ellipse sizes taking account of distance to the respective image sensor can correspond to the greatest possible evaluation. If the scores are expressed in terms of probabilities, this corresponds to a value of p ₂ = 1 and all other ellipse size similarities are rated by a smaller probability. On the other hand, if the scores are expressed by cost scores, this equals the value of p ₂ = 0, while all other ellipse size similarities are scored by a larger score.

As the third possible evaluation criterion, a similarity measure for the speeds of the detected objects can be used. This may, for example, be the absolute value of the difference between the two velocity vectors. If this absolute value is zero, then it is the greatest possible match. If the scores are expressed in terms of probabilities, this corresponds to a value of p ₃ = 1 and all other speed differences are rated by a smaller probability. If, on the other hand, the valuations are expressed by means of cost estimates, this corresponds to the value of p ₃ = 0, while all other speed differences are evaluated by a larger score.

As the fourth possible evaluation criterion, a measure can be used which rates a meaningful height of the midpoint 14 of the line of the minimum distance 13 between the beams 12.1, 12.2 emanating from the image sensors 3, 4. The sensible height can be adapted to the objects to be monitored. For adult persons the reasonable height can be 80 - 100cm depending on the assumption of the person size. For example, the measure may be taken by the absolute amount of the difference between the meaningful height and the height of the midpoint 14. In this case, a value zero of the absolute value means the greatest possible match. If the scores are expressed in terms of probabilities, this corresponds to a value of ₄ = 1 and all other heights of the midpoint 14 are evaluated by a smaller probability. On the other hand, if the scores are expressed by cost scores, this equals the value of p = 0, while all other heights of the midpoint 14 are scored by a larger score.

The fifth possible evaluation criterion is a measure that takes into account the previous assignments. If the image sensors 3, 4 are cameras which record film sequences, a sum for example can be used for this via the ratings of the corresponding objects to be monitored during the last ten recorded individual images. In this example, the highest possible score means the highest possible score continuously during the last ten reviews. If the scores are expressed in terms of probabilities, this equates to a value of p ₅ = 1 and all other previous mappings are rated by a smaller probability. If, on the other hand, the valuations are expressed by means of cost valuations, this corresponds to the value of p ₅ = 0, while all other previous assignments are evaluated by a larger valuation value.

In each case one of these five above-mentioned evaluation criteria can be used to determine the elements of the matrix. However, there is also the possibility that another evaluation criterion not mentioned here is used. There is also the possibility that several evaluation criteria will be shared. For example, the matrix elements can each be formed from a sum of the different evaluation values. For example, the different evaluation values of the different evaluation criteria can also be weighted differently. Based on the above-mentioned evaluation criteria, a matrix element p _1} may be formed, for example, as follows: fl _j = + a ₂ p ₂ + a ₃ p ₃ + -? ₄ + a ₅ p _5, where _3k each represents a weighting factor. If the individual scores are 1 normalized probabilities then the weighting factors should also be normalized to 1: a, + a ₂ + a ₃ + a ₄ + a ₅ = 1.

As already mentioned, the assignment of the objects to be monitored on the basis of the matrix by the calculation unit 8 takes place in a manner known per se, using the Hungarian method, which is also referred to as the Kuhn-Munkres algorithm. From this assignment, the totality of the objects to be monitored is determined by the calculation unit 8. The entirety of the objects to be monitored can be output by the calculation unit 8 again. For this purpose, for example, the monitored objects can be displayed as points in the monitoring area 2 on a screen (not shown). However, there is also the possibility that the entirety of the objects to be monitored is output only in the form of numbers. For example, the calculation unit 8 may be connected to a further computer, which evaluates the time profile of the number of objects to be monitored. However, this additional computer can also, for example, record the movements of the objects and output congestion warnings if two objects accumulate in one area of the surveillance area. These possible uses of the totality of the objects to be monitored output by the calculation unit 8 are not conclusive. The data of the totality output can be used as needed. Also, the device 1 described above can be adapted according to the concrete needs.

The device 1 described above is not the only embodiment according to the invention. Various modifications of the device 1 are possible. For example, there is the possibility that the objects to be detected are not modeled by the image sensors as ellipsoids but differently. It is also possible for the data output by the image sensors to contain only the positions of the detected objects or only parameters for beams starting from the respective image sensor for the respective detected object. There is the possibility that the data acquisition of the image sensors essentially takes place in three steps as described above in connection with FIG. 2c. Of the data thus acquired can For example, for a detected object only the position of the ellipse midpoint from the processing unit and the corresponding image sensor are output to the calculation unit. However, it is also possible for the image sensors in each case to output only data on a beam starting from the corresponding image sensor through the ellipse center. As a variant, there is also the possibility that another form is fitted to the detected object. For example, it can be a circle or a rectangle.

In order to enable optimal determination of the positions of the objects to be monitored, it is possible, for example, to assume a height for the center of gravity of the objects to be monitored. In this case, an intersection point between a beam emanating from the respective image sensor and pointing to the detected object and a plane arranged above the ground at the assumed height of the center of gravity can be calculated. The intersection point thus determined can be used as the position of the corresponding object. Such a position determination is also possible if an object to be monitored is detected only by an image sensor.

With this procedure for position determination, there is the possibility that a fixed value is assumed as the value of the height of the center of gravity. However, there is also the possibility that an individual value for the height of the center of gravity is stored in the totality of the objects to be monitored for each object. If an object is newly recorded in the entirety of the objects to be monitored, a predetermined value can be used which corresponds approximately to the height of the center of gravity of the objects to be monitored. If this object moves into an overlap area in which it is detected by more than one image sensor, the stored value of the height of the center of gravity can be adjusted by taking into account the height of the center of the line of the minimum distance between the rays emanating from the image sensors. In each case, for example, the value of the height of the center of the line of the minimum distance can be stored directly as the height of the center of gravity. However, there is also the possibility that with repeated movement of an object into an overlap area, a mean value of the previously determined values of the height of the center of gravity is stored. The latter allows the consideration of several determinations of the height of the center of gravity, allowing a more precise determination of the height of the center of gravity. This leads to a more accurate position determination of the objects to be monitored, since the rays emanating from an image sensor and extending to the detected object can run in a very inclined manner when an object is located in the edge region of a subregion. Accordingly, even small deviations in the height of the center of gravity can lead to considerable positional deviations.

In summary, it should be noted that a method belonging to the technical field mentioned at the outset and a corresponding device are provided which allow an improved determination of the entirety of the objects to be monitored in the monitored area from the data of the image sensors.

Claims

claims

Method for monitoring a monitoring area (2) with at least two image sensors (3, 4), wherein a) a partial area (5, 6) of the monitored area (2) is monitored by each of the image sensors (3, 4) by 3, 4), objects to be monitored (9.1, 9.2) are located within the subarea monitored by it, and data about the detected objects (9.1, 9.2) are output from each image sensor (3, 4), b) the image sensors (3 , 4) are arranged and aligned in such a way that the monitored subregions (5, 6) overlap and that each object (9.1, 9.2) to be monitored, which is located in the surveillance area (2), is always controlled by at least one image sensor (3, 4 ) and c) from the data of the image sensors (3, 4) a totality of the objects to be monitored in the surveillance area (2) is determined, characterized in that d) on the basis of the data of the image sensors (3, 4) the objects to be monitored (9.1, 9.2) in overlapping partial areas (5, 6), which are detected by more than one image sensor (3, 4), are assigned to one another by means of an evaluation of their agreement in order to determine the entirety of the objects to be monitored in the monitoring area (2).

2. The method according to claim 1, characterized in that in each case for the assignment of objects to be monitored (9.1, 9.2) in overlapping subregions (3, 4) a matrix is created whose elements pjj contain the rating, as well as an image sensor (3, 4) detected object (9.1, 9.2) with identification i with an object (9.1, 9.2) with identification] detected by another image sensor (3, 4).

A method according to claim 1 or 2, characterized in that for each combination of possibly matching objects (9.1, 9.2) a sum of the evaluations of the matching in this combination objects (9.1, 9.2) is formed and that the assignment of objects to be monitored (9.1 , 9.2) is done by a choice of the combination which gives an extreme of this sum.

Method according to one of Claims 1 to 3, characterized in that the data output by the image sensors (3, 4) is output anonymously.

Method according to one of Claims 1 to 4, characterized in that, in the data output by the image sensors (3, 4), each detected object (9.1, 9.2) is represented by a parameterized model.

Method according to Claim 5, characterized in that, for the evaluation of the agreement, in each case one of a minimum distance (13) between a beam (12.1, 12.2, 12.3) from an image sensor (3, 4) is determined by a centroid of the parameterized model and Beam (12.1, 12.2, 12.3) from another image sensor (4, 3) is taken into account by a centroid of the parametric model calculated value.

A method according to claim 6, characterized in that for each object to be monitored of the entirety of the objects to be monitored, a height of the center of gravity of the corresponding object is stored, initially a meaningful height is assumed as the value of this height and the value of this height is dynamically adjusted, by determining a height of the center of gravity of this object for each assignment of objects to be monitored, in which the corresponding object to be monitored is detected by at least two image sensors is determined by determining a height of a midpoint of a shortest connecting line between a beam from the one image sensor through the centroid of the parametric model captured by it and a ray from the other image sensor through the center of gravity of the parametric model it detects.

8. The method according to any one of claims 5 to 7, characterized in that for the

Evaluation of whether a center (14) of a shortest connecting line between the beam (12.1, 12.2, 12.3) originates from the one image sensor (3, 4) through the center of gravity of the parameterized model it detects and the beam (12.1, 12.2, 12.3) starting from the other

Image sensor (3, 4) by the center of gravity of the parameterized model detected by him at a reasonable height for a height of a center of gravity of the objects to be monitored (9.1, 9.2) is located.

9. Method according to one of claims 5 to 8, characterized in that a similarity of the parameterized models of the objects to be monitored for the

Evaluation of the agreement is taken into account.

10. The method according to any one of claims 1 to 9, characterized in that for the

Evaluation of the conformity of the image sensors (3, 4) detected speeds and directions of movement of the monitored objects (9.1, 9.2) are taken into account.

1 1. The method according to any one of claims 1 to 10, characterized in that for the assessment of the match earlier assignments of the objects to be monitored (9.1, 9.2) are taken into account.

12. The method according to any one of claims 1 to 1 1, characterized in that the monitoring area (2) as a two-dimensional surface with a two-dimensional Coordinate system is modeled and that each of the objects to be monitored (9.1, 9.2) of the totality of the monitored objects is characterized by related to this coordinate system data.

13. The method according to any one of claims 1 to 12, characterized in that each of an image sensor (3, 4) monitored subregion (5, 6) is modeled as a two-dimensional surface with a two-dimensional coordinate system and that of each image sensor (3, 4) relate output data to the detected objects (9.1, 9.2) to the coordinate system of the sub-area (5, 6) monitored by this image sensor (3, 4).

14. The method according to claim 7, characterized in that the of the

Image sensors (3, 4) detected objects (9.1, 9.2) are modeled as ellipsoids (10) whose projection on the two-dimensional surface of the corresponding portion (5, 6) an ellipse (1 1.1, 1 1.2, 1 1.3) result and that the data output by the image sensors (3, 4) for each of the detected objects to be monitored (9.1, 9.2) an identification, X and Y coordinates of a center of the ellipse (1 1.1, 1 1.2, 1 1.3), sizes of Main axes of the ellipse (1 1.1, 1 1.2, 1 1.3), an orientation angle of the ellipse (1 1.1, 1 1.2, 1 1.3) and a velocity vector include.

15. The method according to claim 13 or 14, characterized in that the coordinate systems to which the output from the image sensors (3, 4)

Data are to be rectified.

16. The method according to claim 12 and one of claims 14 to 15, characterized in that the output from the image sensors (3, 4) data are transmitted to the two-dimensional coordinate system of the monitoring area (2).

17. The method according to claim 16, characterized in that a position and an orientation of each image sensor (3, 4) are stored and for the transmission of the output from the image sensors (3, 4) data to the two-dimensional coordinate system of the monitoring area (2) be taken into account.

18. The method according to any one of claims 1 to 17, characterized in that the

Image sensors (3, 4) are mounted overhead.

19. Device (1) for monitoring a monitoring area (2) with at least two image sensors (3, 4), wherein a) a partial area (5, 6) of the monitored area (2) can be monitored by each of the image sensors (3, 4), in that localized objects (9.1, 9.2) to be monitored can be detected by each image sensor (3, 4) within the subarea (5, 6) which can be monitored by it, and data from each image sensor (3, 4) relating to the detected objects (9.1, 9.2 ) can be output, b) the image sensors (3, 4) are arranged and aligned so that the monitored subregions (5, 6) overlap and that each object to be monitored (9.1, 9.2), which is located in the surveillance area (2) , is always detectable by at least one image sensor (3, 4) and c) the device (1) comprises a calculation unit (8), by which from the data of the image sensors (3, 4) a totality of the objects to be monitored in the surveillance area (2 ) can be determined, characterized in that ss d) by the calculation unit (8) on the basis of the data of the image sensors (3, 4) the objects to be monitored (9.1, 9.2) in overlapping subregions (5,, 6) which detects more than one image sensor (3, 4) are, by rating their Correspondence can be assigned to each other to determine the entirety of monitoring objects in the surveillance area (2).