DD289837A5 - METHOD AND ARRANGEMENT FOR THE AUTOMATIC DETECTION OF THE NUMBER OF PERSONS - Google Patents

Abstract

The invention relates to a method and an arrangement for automatic detection of the number of persons in a security gate for the purpose of person separation in automatic access control systems. The aim of the invention is the Abloesung weight-measuring or other methods for determining the number of persons whose subsequent installation in existing buildings is complicated or impossible or not wear and maintenance-free or their detection security does not meet high security requirements. The object of the invention is to provide an arrangement which can be installed with little constructional effort and a method that detects contactless and with high reliability, the number of people in the security gate. According to the invention, a ceiling mounted electronic camera connected to an image processing computer observes the lock scene from which image content is extracted by analyzing the intensity, texture, shape, and texture features to distinguish the classes "empty." "," 1 person "and" 1 person ". Fig. 5 {automatic access control; Number of persons; Traffic separation; Image processing; Image content; Image signal characteristics; Feature; Feature extraction; Classification}

Description

For this 4 pages drawings

Application blot of the invention

The invention relates to a method and an arrangement for the automatic detection of the number of persons in a security gate for the purpose of person separation in automatic access control systems.

Characteristic of the known state of the art

For complete verification of the stay of authorized persons In Sporrbelelchen and to prevent the entrainment of unauthorized persons or objects automatically acting Personenunterzelungseinrichtungen be used. From DD-PS 169273 a two-door personal lock with Weighing unit is known in which it is determined by weight measurement and comparison with the individual target weight of the person passing whether one or more than one person bofinden in the person lock. The installation of weighing equipment in existing buildings or the project-based integration are often not possible from bautochnischon Gründon. Andoro forced-release devices such as revolving doors or pendulum locks enjoy little acceptance because of their narrowness. This should also apply to the personal sluice described in DE-OS 2647061, in which dlo singulation with a narrow, adapted to the human coarse contour shape Idontifizlerungsnlsche oroller takes place in which the body shape is scanned in a forced attitude by contacts and optoelectronic sensors and the thus obtained Data and the measured weight are compared with target data.

From DE-PS 3623792 a device for determining the number of persons and their direction of movement Innorhalb a space to be monitored or a lock for the purpose of singling is known in the particular Welso arranged IR-Sonsoron determine the change in ambient temperature due to the heat of corporeal transit people , A disadvantage of this method is the dependency dos Sensorausgangsslgnales of dor speed passing persons, whereby dio Anwondung dos procedure for the Personenverolnzolung mitlols Schlouso, In wolchor dlo person at least until Schlioßon the first door odor to the completion of identification by example blomotrischo data must persist , not suitable.

Furthermore, DE-OS 3509 241 discloses a method for optically determining and / or locating a obiocto in transport and handling devices which incurs the following significant morkmalo:

- Creation of a Difference Blld from the environment image with object and environment image without object (Referonzblld),

- Using a fixed installed in relation to the environment Kamora with CCD Matrlx as an image receiver. Nachtelle of the mentioned solution are

The need for an optically structured ambient ambient orbit,

- the need to pre-polish different Objoktblldor

The necessity of forming the roforon image from a number of violets of umbilical pick-up at different light-scattering sounds,

- Necessity of documentation olnos to the Anglolch dor Holllgkolten of Umgobungsblld with object and Roforoniblld oriordof llchon, not from the objokt predockbaron Roferoniborolches.

The above-mentioned image processing systems are known, eg for character and graphic design, for aerial image analysis and for medical diagnostics. Such systems, however, are very expensive and too expensive and oversized for klolnoro applications.

Object of the invention

Object of the invention Is the replacement of weight-measuring or other methods for determining the number of persons In a security lock for persons separated by a method with high detection reliability and the creation of an arrangement that is wear and maintenance-free and their subsequent installation in existing building without complicated and expensive work possible is.

Explanation of the essence of the invention

The invention has the object of providing a method and an arrangement for determining the number of persons located in a person lock for separating persons. The method should have at least the same detection reliability as known weighing methods. To carry out the method, a non-contact arrangement with a CCD matrix camera, at the optical resolution of which no high demands should be made, and be used with an 8-bit image processing computer low power and with low construction costs in the installation. Furthermore, the object is based acceptance problems of the user, as they occur in known Personenvereinzelungseinrichtungen with revolving doors, pendulum locks or Verminnzelungsnischen, to exclude by generous dimensioning of the lock chamber. Furthermore, no special requirements should be placed on the user behavior, as in a known, working with IR detectors means for determining the number of persons and their direction of movement, in which a certain throughput speed must be maintained.

According to the invention, the lock scene observed by a CCD matrix camera installed on the ceiling of the lock is evaluated in the following method steps:

First, the recording of a reference image (image of the empty lock) takes place until its average brightness has reached an optimum value for further processing by successively changing the amplification value of a video amplifier. Thereafter, the reference image thus obtained is stored in a reference image memory, and features describing intensity and texture characteristics of the image content are extracted therefrom and stored as a reference feature in the reference feature memory.

For this purpose, in a first method step, the intensity properties are utilized by dividing the image area into segments of any shape and number and calculating the mean segment brightnesses NHR _n normalized to the average image brightness HELL according to Eq. (1):

NH ο ^H

ⁿ HELL H _n = average segment brightness (1)

N η = 1 N = number of segmonte

In a further method step, the texture properties of selected image segments are evaluated. They are reflected in the spectrum of their image signals, but they are also included in the zero crossing histogram or in further simplification in the number of zero crossings of the image signals (for unipolar image signals, the average brightness value of the image segment is the reference line for counting the passes). The Toxtureigonschaften is also used by dividing the image area into Sogmente beliobigor form and number and subsequent count of the segment zero crossing NDR _n .

After the extraction and printing of the Roforonzbildos and the Roferenzmorkmalo, the cycle begins, in which periodically intervals test images (Bildor dor real Schlousenszono) are taken and are exhausted.

First, at the Roforon feature calibration, the normlorton segment couplet NHT _n and the zero-continuity NDT _n do'i test pattern are determined, and then in the classifier I, the classification "loor" / "not loor" is prephromed. Dioser orkonnt "loor", if all difforonzboträgo between the characteristics NHT _n dos testimage and the ontsprochenden Referon.imorkmolon NHR _n klelnor as oin given Holligkoitsschwollwert Sn and all Difforonzboträgo zwlschon (lon characteristics NDT _n the test image and don entsprochondon Roforonzmorkmalon NDR _n kloinor as 0 in the zero crossing threshold word S _N are (GI. (3l):

(V | NHTn-NHRn | <8 _H) * (V | _n ~ NDT NDTn | <8 _N) ->"empty" (3)

Bonet -> "not empty" (4)

Then the decision is made "1 Poreon" / _H > 1 Poreon ". For this purpose, the knowledge base obtained from oillography is modeled on the image of a "Poreon In the orifice" to orarbolton, the statements about which are taken, and about dio Orloso, form and structure of the figure. , Form and Structure Oligonchofton dos Blldlnhnltos kloselflilorbnro Morkmolo oxtrahlort.

The following process steps take place:

- Alignment of the average brightness of the test image by arithmetic operations to the average brightness of the reference image to eliminate influences of lighting changes within certain limits. A complete brightness equalization takes place only if the classifier I "empty" recognizes, otherwise the stored brightness difference value of the last brightness adjustment is used with empty lock as a correction value.

- Obtaining a difference image by pixel-wise subtraction of the image signal values of the reference image of those of the test image.

- Generation of a black-and-white contrast image by zeroing all pixels whose values and gradation tone are smaller than predefined threshold values.

Smoothing the contours of the contrast image and eliminating such pixels as linear and less than three contiguous pixels that do not conform to the model by analyzing the neighborhood relationships.

- Analysis of the structure of the smoothed contrast image with the aim of soparation of independent subsets, the individual individuals map, also by analysis of the neighborhood relationships.

As a result of these method steps, the features "number of white pixels" (AP) and "number of subsets" (AM) are created. If classifier I has detected "not empty", classifier Il is activated and recognizes "1 person" if AP is less than a personal, body size dependent threshold S _f and AM = 1:

(AP <Sp) & (AM = 1) => JPorson "(5)

otherwise => "> 1 Perjon". (6)

AusfOhrungsbelsplel The invention will be explained below using an exemplary embodiment. In the accompanying drawings show 1 is a flowchart illustrating the most essential ancestor steps, 2: standardized quadrant brightnesses, Fig.3: nine-neighbor relationships, Fig.4: Special neighborhood relationships as examples, 5 shows the block diagram of the arrangement according to the invention, 6 shows the block diagram of a preferred embodiment according to FIG. 5.

1. procedural lessons

According to the invention, the method is described in the introduction by din in the following, program-controlled steps

realized. FIG. 1 is used for explanation. Reference is made to equations (1M6) mentioned in the essence of the invention.

1.1. Obtaining a reference daton

1.1.1. Recording and saving the reference image

Upon power up of the device or during an external reference refresh, it is expected that the airlock will be empty to receive a reference image. For optimum brightness adjustment, the average brightness of the recorded image and a setpoint value are used to determine a digital control variable that changes the amplification factor of the video amplifier. Image acquisition and gain change are carried out until the average image brightness has reached the target value with sufficient accuracy. The image signal values of this last image with optimum brightness have been stored as the reference image, and the manipulated variable is recorded until the next reference event.

1.1.2. Extraction of reference features

For feature extraction, intensity and texture gantry of the reference image counterpart are used for the purpose of exploration

"Leor" / "not empty" are suitable.

From the intensity gradients, the normalized Sogmenthelli (ikoiton NHR) NHR ₄ according to Eq. (1) determines dio by Division of the square image area has been goblldet in quadrants (Flg. 2). When using the Toxturolgenschaften is on dio decomposition of the image area renounced in segments. As Elmzelmorkmal will

the number of passes NDR of the envelope of the pixel values of the total postage picture is counted by the center line Holllgkolt.

The Roferonkorkmalo NHRi-NHR ₄ and NDR have been gospoichort for the dio lateoro bono usage.

1.2. Acquisition of the aktuollon Schlousonszeno

After the extraction of the period data begins an endless, periodic cycle of Tostblildaufnahmon (pictures dor nktuollon Schlousontzone) and their evaluation

 1.2.1. Detection "emptyV.nlchtloor"

Accordingly, under point 1.1.2. These methods are also extracted from dom test pattern diox features normalized suction dendrograms NHTi-NHT ₄ and the number of passes NDT. A Abstandsklasslfikatorerkonnt _H leor ^v Wonn all "Dlfferenzbeträgo zwischon don Morkmalon NHTrNHT ₄ dos Tostblldes and don appropriate Roforonzmorkmolen NHRrNHR ₄ small al · a vorgegoboner Holllgkollsschwflllwort Sm and dor Dlfforonzbetrag between dom feature NDT of Testbildos and dom Roforonzmorkmal NDR klolnor as a Nulldurchganrjsschwollwort Sn Is (Equation | 7 |):

1.2.2. brightness correction

A change in the illuminance, in particular by changing Tagoslichteinfall, affects the values of the individual pixels approximately linearly. Before carrying out further process steps, a brightness correction of the test image is required. To avoid actually required multiplications, the correction is done simply by subtracting the difference value of the average brightnesses from the test and reference image from each pixel value. The correction will only be done to this extent if "empty" has been detected, otherwise the stored difference value of the last correction made at "loer" will be used for correction.

1.2.3. Detection, 1 person "

1.2.3.1. model

For the distinction "1 person /> 1 person", the following model has been derived from an experimented knowledge base:

- The body of a standing person is seen from above as a bright area with decreasing brightness from the center.

- The surface is compact, d. that is, the ratio of the largest to the smallest axis is small (u 2, 1).

- The brightness and size of the surface depend on the person's body measurements, the position in the lock and the reflectivity of the clothing.

- The edges of the surface are not continuous.

- In addition to the image of the person's body, depending on the position, shading and lightening in the environment are formed.

- Several standing in the lock persons with a sufficiently large distance are shown as separate) surfaces.

- Densely standing people are depicted in an area with possibly only one peak, but

• the form differs from that of a person,

• The size exceeds that for one person.

For the recognition "1 person", features are made of the intonality, shape and structure properties of the difference image, including the knowledge base in the steps

- formation of a contrast image,

- smoothing the contours of the contrast image and

- Structure analysis extracted and classified.

The following items deduce the functions required for these steps.

1.2.3.2. Contrast imaging

The task of contrast imaging is to detect in the difference image obtained from test and reference image by pixolweise subtraction peak lights, even those with small amplitudes, and assign them the value 1 and large scatterers with small gradients assign the value 0. The method requires three functionally related thresholds:

Contrast threshold KSW

Upper contrast threshold OKSW - h χ KSW h> 1

Gradient threshold GSW = gx KSW · g a 1 The formation rules apply to the generation of the contrast image:

- zeroing all pixel values y (m) <KSW:

y (m) <KSW => y (m): = O (9)

- zeroing of all pixel values in the area KSW s y (m) <OKSW, the gradient <GSW is:

(KSW s y (m) <OKSW) & grad {y (m)) <GSW -> y (m): = 0 (10)

- All non-zeroed pixel words are assigned the word 1.

The gradient is calculated from the neighborhood buzlohungon of the given sampling word y (m) to its vlor (horizontal and vertical) neighbors:

grad (y, j) - | y ,,, - y i-1,) | + | y ,,, - y ,,,. |

+ lYi.i-Yi.i-il + IVi.i - V..I * ti W)

mlty ,, | - y (m).

The formation of a fact is mathematically unoxactic. In the solicitation is slo abor gorochtfortlgt, in order not to eliminate poakos with klolnon words. The sampling word at the picture-forming level becomes zero by degrees.

1.2.3.3. Gluttony of the Contrattbildos

In order to smooth out the contour of the contrast picture and the ellmineration not in the dome of Modollvorstollung suitable Dlldolomente, wlo single or double points and to Union anolnandorgoroihto points, ollt Flltorvorfnhron olngosotzt. Boi zwoldlmenslonalon Objectignaion in dor Bildornung must dor Flltoinlgorlthmus allo Nochbarn dos botrnchtoton Pixels oinbozlohon. Dio Minimalform analyzes dlo Noun-Nochbnrn-ooilohung olnos Bildausschnlttos, bol dom dos botrnchtot Pixel χ in the center of noun adjacent Pixoln + llogt (Flg.3).

Since the contrast image to be smoothed contains only 1-Plxel and O-pixels, the filter problem simplifies the following formation rules:

- A 1-pixel is preserved if it has at least two neighbors that together with it form a right-angled triple of 1-pixel (Fig.4a), otherwise it becomes an O-pixel. A triple is thus the smallest permissible picture element.

An O pixel becomes a 1 pixel if it has at least three 1 pixels as direct horizontal and vertical neighbors (Fig. 4b). Because of the finiteness of the image plane, it must be noted that the pixels lying at the edges have no neighbors outside the image plane.

In practice, it has proven expedient for the subsequent structural analysis to dispense with the filling of O pixels in order to enhance the Strukturoigenschaften.

1.2.3.4. .Strukturanalyse

The filtered contrast image is the result of the analysis of the intensity and shape properties of the image content in conjunction with the knowledge base about the lock scene. The structural analysis still to be performed has to determine the number of isolatable subsets of the entire 1-pixel set.

In order to isolate the subsets, the nine-neighbor relationship of the considered 1 pixel to the following neighboring pixels is again used (see FIG.

upper left neighbor => OLN, upper neighbor = ON,

Upper right neighbor »ORN, right neighbor« RN,

lower right neighbor = URN, lower neighbor »UN,

lower left neighbor «ULN, left neighbor = LN.

To solve the problem, two runs must be made, one with increasing pixel addresses and the second with backward falling addresses.

The following regulations apply to the first run:

If a 1-pixel does not have neighborhood relationships to be checked in the noted order (see Fig. 3)

ORN & RN OF & ORN BY & OLN ν LN,

than it will be

Incremented a pointer set to zero at the beginning of the analysis,

• The 1-pixel is assigned to the subset displayed by the pointer.

- If a 1-pixel has at least one of the neighborhood relationships, then

• assigned the 1-pixel dor Toilmenge that the participating neighbors belong to,

• The pointer is not Incremented

 For the second run:

- The first found 1-pixel sets the pointer to its quantity number.

- If a 1-pixel is assigned to a andar subset, as indicated by the pointer, then the neighborhood relations (see Fig.3) are in the noted order

LN & ULN ν UN & ULN ν UN & URN v RN.

considered.

If the 1-pixol bites at least one of these relations, then it is assigned to the subset of these afterwords,

• If none of these neighborhood relations exists, then the 1 pixel retains its mongonnum and the pointer is set to this.

1.2.3.5. classification

The features available are the "Number of Weak Pixels" (AP) of the smoothed contrast image and the "Number of Toilmongone" (AM) from the structural analysis. The feature AP is needed to increase the security of the system. If AM «1, but AP exceeds the individual threshold Sp, there are at least two persons in the airlock who are close together to deceive the systom. The classification is carried out according to QIn, (5) and (Θ).

2. Hardware solution

In Fig. 5 the block diagram of the inventive arrangement is given. The oloktronischo Kamora 11st over the controllable video amplifier 2 with dom Bildverroboitungsrochnor 3 vorbundon who over the feedback line ST don video amplifier 2 stouert sowlo receives the porsononabhänglge Schwollwortsignal Sp and the signals .loer, "1 person" and "> 1 person" to außon surrenders The output of the video amplifier 2 is preamplified with an analyzer β as well as the Roforone frame 4. The analyzer 6 passes through the video amplifier 2 via the feedback line ST and receives a referential image from the reference mirror 4. Dor Analyzer β is also associated with a Reforenzmerkmalspolchor 6 sowlo via the Olgnalloltungon NHT _n and NDT _n with olnom Klaiaifikator 17, which stands with Dom Referenzmorkmalspolchor B in Vorbindung Dor Klaoslfikator 17 »tolls a signal" leor "iur extornen evaluation ready. Furthermore, it internally emits a signal "not empty" to a classifier HB, which receives from the analyzer 6 the morkmoles AP and AM and from outside the porsonon-dependent decay word signal Sp. Dor classlifter Il β represents dlo Slgnalo .1 Porton "and"> 1 person "to oxtornon evaluation borolt.

In Flg. β is a preferred embodiment of the orgungsungsgomäßon arrangement dargostollt. Woitoro embodiment with don dom expert bokannton «qulvolonton means are possible.

Dor image receiver of Malrlxkemora 11st olno CCD-Matrlx 11 of type L216C with 04 inches and 16 columns, of which 10 have been used in Zontrum llogondo customs. DIO gonutzto photosensitive surface dor CCD Motrlx 11 botrAgt 0.8 x O ₁ mm B '. On her is the SchleuseurundllKchovonca. 1.5 χ 1.6m ¹ Approximately at height. To the L0dorMatrlx camera 1 wordon unusual conditions due to the oxtromon Q'Oson priority of CCD Matrlx 11 and Schlousongoomotrlo

posed. It must have a very short focal length and a wide shooting angle. Because aberrations, such as

Image distortion and anastigmatism, due to the difference image method used in the image analysis none

play an essential role in determining quality, an objective 10 empirically constructed from the lenses of a commercial door splitter and further three lenses is used. For the basic adjustment of the optimum amount of light for the CCD matrix 11, gray filters of different densities are used. The fine adjustment is done with two mutually rotatable

Polarization filters. The clock generator 12 generates in addition to the six required to control the CCD matrix 11 clocks more clocks for Control of the video amplifier 13, the ADU 14 and the input register 15. At the video output of the CCD matrix 11 appear rectangular synchronous to the horizontal clock with an amplitude of 3 V ",

on its positive roof, the actual pixel information (3 bars dark reference, 1 empty bar and 16 bars

Image content) with a modulation of max. 10OmV ₁₁ are located. Tasks of the video amplifier 13 are

- holding the dark reference if the dark reference clock TDR is active,

Separating the image information from the composite signal with the recorded dark reference as zero reference value,

- Amplifying and holding the pixel signals for the duration of a horizontal clock as the input voltage VID for the ADU14.

To match the video signal to the input voltage range of the ADU14 (10 V max.), The gain factor * is Video amplifier 13 in 16 steps digitally adjustable with the 4-bit-signal ST in the ratio 1: 4,4. This will be the regulation of

medium brightness of the video signal by the image processing computer 3 allows.

The image processing computer 3 is preferably a single-board computer consisting of the Grundulemonton shown CPU 16 (U880), 8K byte EPROM 18.2K RAM 17, SIO 20 (U556), input register 15, output register 19, ADU14 (C571) and

further, not shown circuits for address decoding, system clock generation and caching of

ADC output signals and the bars BAN (beginning of image) and PIX (pixel value valid). The ADU14 begins after the arrival of the start signal TADU with the analog-to-digital conversion of during the Umsetzdaucr

captured video signal into an 8-bit word, the pixel value. After completion of the conversion (about 30ms) the signal PIX is generated. The control program, which constantly polls this signal via the input register 15, then takes over the

Pixel value in the RAM 17 and clears the signal PIX. This process is performed by the also requested by the control program Start of picture BAN started and executed 2B6 times per picture. Because of the projected in relation to the maximum Readout frequency of CCD matrix 11 of 5MHz (horizontal clock) Very low actual readout frequency of 17.8kHz That is Dark current noise big. In order to reduce its influence, four successive images become a mid-level image

formed, whose higher-order five bits are usable.

For communication with the computer of the personal lock, the input register 15 or output register 19 are used. The personal body measurements required for the classifier 118 are coded in the 2-bit value Sp. Output signals of Classifiers are "loor", "1 person" and "> 1 person". The SLO circuit 20 allows the communication of the Bildverarboitungsrechners 3 with the computer dor person lock

via a serial (IFSS) interface with the signals Sending and Fixed programs for commissioning and testing of the arrangement.

The inventive method guarantees a reliable detection of the presence of koiner, one or more Porsonon in

a security gate. It does not require high-performance computing technology or high-resolution image acquisition technology. With the arrangement according to the invention, the implementation of the method is possible in a straightforward manner. The arrangement does not require complicated construction measures and can also be built into existing Porsononschiousen.

The hardware cost is relatively low and requires no Spozialschaltkreise. Thus wurdo an economically favorable Realization for persononverzozoung created.

Claims (10)

1. A method for automatically detecting the number of persons in a security gate by recording the pictorial scene with a fixedly installed on the ceiling of the lock electronic camera whose digitized image signals from an image processing computer a) once as a reference picture of the empty lock and b) are stored periodically as a test image of the current lock scene, characterized in that features are extracted from the test image and the difference image calculated from the test and reference images and classified into the classes "empty", "1 person" and "> 1 person", firstly, for recognition "empty" / "not empty", the features "normalized segment brightnesses" (NH _n ) and "segment zero passages" (ND _n ) are extracted by analysis of the intensity and texture properties from reference and test image from an analyzer (6) and from a Classifier I (7) and then to recognize "1 person" / "> 1 person" from the difference image by analyzing the intensity, shape and texture properties of the difference image content, including a knowledge-based model about one or more than one person in the lock, the features "number of white pixels" (AP) and "number of subsets" (AM) in the analyzer (6) are extracted and from a classifier Il (8) are processed. 2. The method according to claim 1, characterized in that for the extraction of the features "normalized segment brightnesses" (NH _n ), the image field is divided into several arbitrarily shaped, approximately equal segments, but preferably in four quadrants whose average magnitudes with the average brightness of the entire picture are normalized. 3. The method according to claim 1, characterized in that for the extraction of the features "segment zero crossings" (ND _n ), the image field is divided into several arbitrarily shaped, approximately equal segments, but preferably in four quadrants and the number of polarity changes in a given Ordered pixel values of the segments with respect to the average sanctities of the respective segments is counted. 4. The method according to claims 1 to 3, characterized in that for the recognition "leor" / "not empty" the classifier I (7) the amounts of the differences of the characteristics "normalized segment brightnesses" (NH _n ) and "segment zero crossings" (ND _n ) of the current test image and the corresponding features of the reference image stored in the reference feature scope (5) and, when predetermined threshold values are exceeded, makes the decision "not empty". 5. The method according to claim 1, characterized in that in the process steps Generation of a black and white contrast latitude 3 using the values and gradients of the difference image signals, - Smoothing the contours and elimination of non-model picture elements of the contrast image from the intensity and shape characteristics of the image signals, the feature "number of white pixels" (AP) describing the size and size of one or more objects in the lock is extracted. 6. The method according to claim 1, characterized in that from the structural properties of the image signals in the process step - Analysis of the structure of the smoothed contrast image and separation of subsets the feature "number of subsets" (AM), which describes the number of isolated objects in the lock, is extracted. Method according to claims 1, 5 and 6, characterized in that the classifier Il (8) is activated only when the classifier I (7) classifies the result "not empty" and then makes the decision "1 Pereon", if AM is equal to 1 and AP is less than a personal, externally supplied threshold Sp. 8. The method according to claim 1, characterized in that boi the recording of the Referenzblldos the gain of the video amplifier (2) by evaluating dor average brightness of the image signals in the analyzer (6) by a manipulated variable (ST) is set and held until the next reference image recording in that the image signals assume optimum values for further processing, 9. The method according to claim 1, characterized in that the average brightness of the image signals of the periodically recorded test images is adjusted by arithmetic operations in the analyzer (6) to the average brightness of the reference image to eliminate influences of lighting changes within certain limits, the full Brightness equalization takes place only with empty lock and otherwise the stored difference value of the last correction with empty lock is used for brightness adjustment. 10. Arrangement for the automatic detection of the number of persons, characterized in that for scanning the lock scene, an electronic camera (1) with low resolution requirements, for. B. a CCD matrix camera with 16 x 1 δ pixels, is used, which is connected via the controllable video amplifier (2) with the reference picture memory (4) and with the analyzer (6) through a feedback line (ST) for Gain control is connected to the video amplifier (2) that the reference image memory (4) is connected to the analyzer (6) for the purpose of differential imaging, the outputs of the reference feature memory (5), with the classifier I (7) for classification "empty "/" Not empty "is connected, as well as with the classifier I (7) and the classifier Il (8) is connected, which each has a connection to the output" not empty "of the classifier I (7) and the threshold input (Sp) and the decision "1 person'7"> 1 person "hits.