EP1672604A1 - Method and apparatus for detection of tampering with a surveillance camera - Google Patents

Abstract

Method for detection of sabotage attempts on a monitoring camera using images recorded by the camera, wherein each image recorded by the camera is compared with a reference image previously recorded by the camera in order to generate a correlation value (qcorr), which is then compared with a threshold value (thresAlarm). An independent claim is made for an arrangement for detecting surveillance camera sabotage attempts that has an processor for implementing the method steps.

Description

The present invention relates to a method for detecting sabotage attempts on a surveillance camera on the basis of the images taken by the camera.

With the proliferation of surveillance cameras in so-called CCTV systems, the need for an effective and reliable means of detecting sabotage on these cameras is increasing. In EP-A-1 079 350 a device for room surveillance is described with a camera, which has means for detecting sabotage of the camera. These means are designed for the storage and verification of stable features of the image taken by the camera, wherein the recorded image is examined for stable features and the stable features found are compared with the stored ones. This method allows you to detect partial coverage of the camera or fake false scenes with a photo held in front of the camera.

The invention is now a method for detecting sabotage attempts on a surveillance camera based on the images taken by the camera are specified, which allows the detection of manipulations of all kinds on the camera.

The method according to the invention is characterized in that each image is compared with a previously recorded reference image and from this a correlation value is obtained and this is compared with an alarm threshold.

• Extraktion von für bestimmte Eigenschaften des Bildes charakteristischen Koeffizienten;
• Bildung eines Mittelwerts für jeden Koeffizienten;
• Transformation der Koeffizienten in eine Informationsmatrix;
• Korrelation der aktuellen Informationsmatrix mit einer früher berechneten Informationsmatrix;
• Bestimmung des Korrelationswerts; und
• Vergleich des Korrelationswerts mit der Alarmschwelle.

A first preferred embodiment of the method according to the invention is characterized by the following method steps:

• Extraction of coefficients characteristic of certain properties of the image;
• Forming an average for each coefficient;
• Transformation of the coefficients into an information matrix;
• Correlating the current information matrix with an earlier computed information matrix;
• Determination of the correlation value; and
• Comparison of the correlation value with the alarm threshold.

• Bildung eines Mittelwerts für jedes Pixel des betrachteten Bildes oder eines Ausschnitts von diesem;
• Extraktion von für bestimmte Eigenschaften des Bildes wie Mittelwert-Bild oder Hintergrund-Bild charakteristischen Koeffizienten;
• Transformation der Koeffizienten in eine Informationsmatrix;
• Korrelation der aktuellen Informationsmatrix mit einer früher berechneten Informationsmatrix;
• Bestimmung des Korrelationswerts; und
• Vergleich des Korrelationswerts mit der Alarmschwelle.

A second preferred embodiment of the method according to the invention is characterized by the following method steps:

• Forming an average for each pixel of the viewed image or a portion thereof;
• Extraction of coefficients characteristic of certain properties of the image, such as averaged image or background image;
• Transformation of the coefficients into an information matrix;
• Correlating the current information matrix with an earlier computed information matrix;
• Determination of the correlation value; and
• Comparison of the correlation value with the alarm threshold.

Preferably, said particular characteristic is formed by the color or intensity of the individual pixels. Coefficients representative of certain properties of the image are Fourier coefficients, DCT coefficients, wavelet coefficients, edge or corner values.

Further advantageous and preferred embodiments of the inventive method are claimed in claims 6 to 14.

The invention further relates to a device for detecting sabotage attempts on a surveillance camera on the basis of the images taken by the camera, with a processor for evaluating the recorded images. The device according to the invention is characterized by a program implemented on this processor for the comparison of each image with a previously recorded reference image and the acquisition of a correlation value and for its comparison with an alarm threshold.

The processor can either be integrated in the surveillance camera or several surveillance cameras can be connected to a common processor, which is designed for the parallel processing of multiple data words.

• Extraktion von für bestimmte Eigenschaften des Bildes charakteristischen Koeffizienten;
• Bildung eines Mittelwerts für jeden Koeffizienten;
• Transformation der Koeffizienten in eine Informationsmatrix;
• Korrelation der aktuellen Informationsmatrix mit einer früher berechneten;
• Bestimmung des Korrelationswerts; und
• Vergleich des Korrelationswerts mit der Alarmschwelle.

A first preferred embodiment of the device according to the invention is characterized in that the program implemented on the processor is designed to carry out the following method steps:

• Extraction of coefficients characteristic of certain properties of the image;
• Forming an average for each coefficient;
• Transformation of the coefficients into an information matrix;
• Correlation of the current information matrix with an earlier calculated one;
• Determination of the correlation value; and
• Comparison of the correlation value with the alarm threshold.

• Bildung eines Mittelwerts für jedes Pixel des betrachteten Bildes oder eines Ausschnitts von diesem;
• Extraktion von für bestimmte Eigenschaften des Bildes wie Mittelwert-Bild oder Hintergrund-Bild charakteristischen Koeffizienten;
• Transformation der Koeffizienten in eine Informationsmatrix;
• Korrelation der aktuellen Informationsmatrix mit einer früher berechneten;
• Bestimmung des Korrelationswerts; und
• Vergleich des Korrelationswerts mit der Alarmschwelle.

A second preferred embodiment of the device according to the invention is characterized in that the program implemented on the processor is designed to carry out the following method steps:

• Forming an average for each pixel of the viewed image or a portion thereof;
• Extraction of coefficients characteristic of certain properties of the image, such as averaged image or background image;
• Transformation of the coefficients into an information matrix;
• Correlation of the current information matrix with an earlier calculated one;
• Determination of the correlation value; and
• Comparison of the correlation value with the alarm threshold.

Further preferred and advantageous embodiments of the device according to the invention are claimed in claims 20 to 24.

• Fig. 1 ein schematisches Flussdiagramm eines Ausführungsbeispiels des erfindungsgemässen Verfahrens; und
• Fig. 2 ein Diagramm zur Funktionserläuterung.

In the following the invention will be explained in more detail with reference to an embodiment and the drawings; it shows:

• 1 shows a schematic flow diagram of an embodiment of the method according to the invention; and
• Fig. 2 is a diagram for explaining the function.

In the method according to the invention, images recorded quite generally by a camera are examined for changes. One sets a picture as a reference picture and compares a later taken picture with this reference picture, whereby the whole picture contents or only a part of this or even selected pixels are compared with each other. In this comparison, a correlation value expressing the degree of coincidence between the current image and the reference image is obtained, which is compared with an alarm threshold. Based on the idea that sabotage usually occurs by swiveling the camera away from the surveillance field and by masking or spraying the camera lens, sabotage detection is based on the detection of changes (image content / image information) in the examined image.

You can either compare the images directly, or you can use as in the embodiment described below, an edge image or examine corners, intensity or color of the pixels, or you can make a DCT, Fourier or wavelet transformation and the DCT-, Use Fourier or wavelet coefficients (DCT: Discrete Cosine Transformation, see eg US Provosional Application No. 60/549 457). Generally one can say that characteristic coefficients are extracted from the image for certain properties of the image.

This means that a temporal average of the pixels is formed for each pixel examined or for each examined image region, resulting in a stable background image. Then characteristic coefficients are extracted and transformed into an information matrix, which is referred to in the English-language literature as "feature space". Alternatively, the coefficients can already be formed from the image and then transformed into an information matrix. The current information matrix is then compared with the information matrix of a reference image.

According to FIG. 1, the images recorded by a camera are supplied in a plurality of channels at a specific frequency of, for example, 2 images per second to a first stage 1 of a sabotage detection process in which a stable background image B is obtained without movements, of which FIG the sequence is assumed to be a reference. For this purpose, for example, a Gaussian mixture model is used, see for example C. Stauffer, W.E.L. Grimson's "Adaptive Background Mixture Model for Real-Time Tracking" in CVPR '99, Vol. 2, pages 246-252, June 1999.

$${\mathrm{\chi }}_{\mathrm{k}}=N_{\mathrm{k}}\mathrm{\hspace{1em}exp}\left\{-{\left({\mathrm{\mu }}_{\mathrm{k}}-\mathrm{Y}\right)}^2/2{\mathrm{\sigma }}_{\mathrm{k}}^{\mathrm{\hspace{1em}}2}\right\}$$

$${\mathrm{Y}}_{\mathrm{mod}}\left(\mathrm{t}\right)={\mathrm{\Sigma }}_{\mathrm{k}}{\mathrm{\hspace{1em}\omega }}_{\mathrm{k}}{\mathrm{\hspace{1em}\chi }}_{\mathrm{k}}$$

wobei µ_k den Mittelwert bezeichnet, σ_k ² die Abweichung (Kovarianz Matrix) und ω_k die Gewichtung der k-ten Verteilung.

• Zuerst wird die Verteilung der Gewichtungen (zur Zeit t-1) so geordnet, dass die Verteilung mit der höchsten Gewichtung an der Spitze steht: ω₁ ≥ ω₂ ≥ ω₃
• Anschliessend wird überprüft, ob die aktuelle Luminanz mit einer der Verteilungen |µ_k- Y| ≤ γσ_k überein stimmt, wobei γ = konstant ist.
• Wenn mindestens eine Übereinstimmung festgestellt wird, werden µ_k, σ_k und ω_k aktualisiert.
  • $${\mathrm{\mu }}_{\mathrm{k}}\left(\mathrm{t}+1\right)=\mathrm{\alpha }{\mathrm{\hspace{1em}\mu }}_{\mathrm{k}}\left(\mathrm{t}\right)+\left(1-\mathrm{\alpha }\right)\mathrm{Y}\left(\mathrm{t}\right)\mathrm{\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.17em}}\mathrm{\alpha }=\mathrm{konstant}\in \left[0,1\right]$$
    
  • $${\mathrm{\sigma }}_{\mathrm{k}}\left(\mathrm{t}+1\right)=\mathrm{\alpha }{\mathrm{\hspace{1em}\sigma }}_{\mathrm{k}}\left(\mathrm{t}\right)+\left(1-\mathrm{\alpha }\right)\left|{\mathrm{\mu }}_{\mathrm{k}}-\mathrm{Y}\right|$$
    
  • wenn Übereinstimmung: ω_k(t+1) = α ω_k(t) + (1-α)
  • wenn keine Übereinstimmung: ω_k(t+1) = α ω_k(t) + 0
• Wenn keine Übereinstimmung festgestellt wird, dann wird k=3 reinitialisiert:
  • $${\mathrm{\mu }}_3\left(\mathrm{t}+1\right)=\mathrm{Y}\left(\mathrm{t}+1\right)$$
    
  • $${\mathrm{\sigma }}_3\left(\mathrm{t}+1\right)=\mathrm{L}\mathrm{\hspace{0.17em}\hspace{1em}\hspace{1em}\hspace{1em}L}=\mathrm{konstant}\mathrm{\hspace{1em}}\left(\mathrm{grosser\; Wert}\right)$$
    
  • $${\mathrm{\omega }}_3\left(\mathrm{t}+1\right)=\mathrm{S}\mathrm{\hspace{0.17em}\hspace{1em}\hspace{1em}\hspace{1em}S}=\mathrm{konstant\hspace{1em}}\left(\mathrm{kleiner\; Wert}\right)$$
    

µ₁ repräsentiert den stabilen Hintergrund B.Only luminance components of the individual pixels are considered, and each pixel of the scene is modeled by a mixture of k Gaussian distributions: $${\mathrm{\chi }}_{\mathrm{k}}\left({\mathrm{\mu }}_{\mathrm{k}}.{\mathrm{\sigma }}_{\mathrm{k}}\right)\mathrm{}\mathrm{With}\mathrm{k}=1.2.3$$

$${\mathrm{\chi }}_{\mathrm{k}}=N_{\mathrm{k}}\exp \left\{-{\left({\mathrm{\mu }}_{\mathrm{k}}-\mathrm{Y}\right)}^2/2{\mathrm{\sigma }}_{\mathrm{k}}^{\mathrm{}2}\right\}$$

$${\mathrm{Y}}_{\mathrm{mod}}\left(\mathrm{t}\right)={\mathrm{\Sigma }}_{\mathrm{k}}{\mathrm{\omega }}_{\mathrm{k}}{\mathrm{\chi }}_{\mathrm{k}}$$

where μ _k denotes the mean value, σ _k ² the deviation (covariance matrix) and ω _k the weighting of the k-th distribution.

• First, the distribution of weights (at time t-1) is ordered so that the distribution with the highest weighting is at the top: ω ₁ ≥ ω ₂ ≥ ω ₃
• Subsequently, it is checked whether the current luminance with one of the distributions | μ _k - Y | ≤ γσ _k is true, where γ = constant.
• If at least one match is found, μ _k , σ _k, and ω _{k are} updated.
  • $${\mathrm{\mu }}_{\mathrm{k}}\left(\mathrm{t}+1\right)=\mathrm{\alpha }{\mathrm{\mu }}_{\mathrm{k}}\left(\mathrm{t}\right)+\left(1-\mathrm{\alpha }\right)\mathrm{Y}\left(\mathrm{t}\right)\mathrm{}\mathrm{\alpha }=\mathrm{constant}\in \left[0.1\right]$$
    
  • $${\mathrm{\sigma }}_{\mathrm{k}}\left(\mathrm{t}+1\right)=\mathrm{\alpha }{\mathrm{\sigma }}_{\mathrm{k}}\left(\mathrm{t}\right)+\left(1-\mathrm{\alpha }\right)\left|{\mathrm{\mu }}_{\mathrm{k}}-\mathrm{Y}\right|$$
    
  • if match: ω _k (t + 1) = α ω _k (t) + (1-α)
  • if no match: ω _k (t + 1) = α ω _k (t) + 0
• If no match is found then k = 3 is reinitialized:
  • $${\mathrm{\mu }}_3\left(\mathrm{t}+1\right)=\mathrm{Y}\left(\mathrm{t}+1\right)$$
    
  • $${\mathrm{\sigma }}_3\left(\mathrm{t}+1\right)=\mathrm{L}\mathrm{L}=\mathrm{constant}\mathrm{}\left(\mathrm{Great\; value}\right)$$
    
  • $${\mathrm{\omega }}_3\left(\mathrm{t}+1\right)=\mathrm{S}\mathrm{S}=\mathrm{constant}\left(\mathrm{small\; value}\right)$$
    

μ ₁ represents the stable background B.

$$\mathrm{m}={\mathrm{n}}_{3\times 3}\mathrm{B},$$

wobei $$h_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]\mathrm{\hspace{0.17em}\hspace{1em}\hspace{1em}\hspace{1em}}{h}_y=\left[\begin{array}{ccc}-1& -2& -1\\ 0& 0& 0\\ 1& 2& 1\end{array}\right]$$

In the next stages of the sabotage detection process, an edge extraction takes place in which the edge values of the individual pixels of the stable background B are determined. For this purpose, a so-called Sobel operator 2 and a mean value operator 3 are applied to B and a value d _i (Sobel) and m _i (mean value) is calculated for each pixel i: $$\mathrm{d}=\left|{\mathrm{H}}_{\mathrm{x}}\mathrm{B}\right|+\left|{\mathrm{H}}_{\mathrm{y}}\mathrm{B}\right|$$

$$\mathrm{m}={\mathrm{<figure-callout\; id="3"\; label="n"\; filenames="imgf0001.png"\; state="\{\{state\}\}">n</figure-callout>}}_{3\times 3}\mathrm{B}.$$

in which $$H_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]H_y=\left[\begin{array}{ccc}-1& -2& -1\\ 0& 0& 0\\ 1& 2& 1\end{array}\right]$$

For mean value operator 3, all elements are equal to 1.

Thus, at the output of stage 2, the individual pixels i of (d) and at the output of stage 3 the individual pixels i of (m) are obtained, from which an edge image is subsequently obtained. Subsequent to stages 2 and 3, in a stage 4, binarization of pixels i takes place according to the formula:
i (d, m) = 0 for dβ <mη, otherwise i (d, m) is equal to 1; β and η are constant.

The result of this binarization is a digital edge image Q, which is stored in a ring memory 5. The ring buffer 5 is substantially longer than the number of frames corresponding to the time constant. Preferably, the size of the ring memory is 1.5τ, where τ is linked to α by the following relationship: τ = 1 / (1-α).

$${\mathrm{S}}_{\mathrm{N}}={\displaystyle {\sum }_{\mathrm{i}}{\mathrm{Q}}_{\mathrm{i}}\left(\mathrm{N}\right)}$$

$${\mathrm{S}}_{\mathrm{corr}}={\displaystyle {\sum }_{\mathrm{i}}{\mathrm{Q}}_{\mathrm{i}}\left(0\right)\wedge {\mathrm{Q}}_{\mathrm{i}}\left(\mathrm{N}\right)}$$

Then, in a step designated by the reference numeral 6, the sum of all pixels is calculated on the basis of a correlation of the current edge image with an earlier one. The first binary image Q (0) and the last binary image Q (N) in the ring buffer are used for this purpose. Count the number of bits present in the first binary image Q (0) (sum S ₀ ), the number of bits present in the last binary image Q (N) (sum S _N ), and finally the number of bits which are present both in the first and in the last binary image Q (0) + Q (N) (sum S _corr ). $${\mathrm{<figure-callout\; id="0"\; label="S"\; filenames="imgf0001.png"\; state="\{\{state\}\}">S</figure-callout>}}_0={\displaystyle {\Sigma }_{\mathrm{i}}{\mathrm{Q}}_{\mathrm{i}}\left(0\right)}$$

$${\mathrm{S}}_{\mathrm{N}}={\displaystyle {\Sigma }_{\mathrm{i}}{\mathrm{Q}}_{\mathrm{i}}\left(\mathrm{N}\right)}$$

$${\mathrm{S}}_{\mathrm{corr}}={\displaystyle {\Sigma }_{\mathrm{i}}{\mathrm{Q}}_{\mathrm{i}}\left(0\right)\wedge {\mathrm{Q}}_{\mathrm{i}}\left(\mathrm{N}\right)}$$

The correlation value q _corr is defined as q _corr = S _corr / min (S ₀ , S _N ), with 0 <q _corr <1. The closer the two edge images Q (0) and Q (N) overlap, the closer the correlation value q _corr at 1. In the case of camera movement caused by sabotage, the camera captures another image section, which results in a change in the correlation value. The correlation value q _corr then passes into a designated by the reference numeral 7 stage in which the alarm decision is made.

The alarm decision is made on the basis of a threshold value for the change δ _q = 1 - q _corr . For the triggering of an alarm the equation δ _q = 1 - q _corr ≤ thres _alarm applies. The alarm threshold thres _Alarm depends on the alarm sensitivity and the edge component f _edge = min (S ₀ , S _N ) on the number of considered pixels (= size of the respective image). For the same sensitivity setting, a higher threshold value δ _{q is} required for a small edge component f _edge than for a high edge component. As a rule, the alarm threshold for δ _q = 1 - q _{corr will be set} at 0.5 and shifted upwards for edge portions below 0.2. For this reason, the alarm threshold δ _{q is} not a fixed quantity but a function of the edge component f _edge .

The setting of the alarm threshold in function of the edge portion is shown in FIG. 2, in which the alarm threshold thres _{alarm is} plotted on the ordinate axis and the edge component f _{edge is} plotted on the abscissa axis. It can be seen that the sensitivity range of the algorithm extending between the limits 0 and 1 is dynamically mapped to a variable band of thresholds for δq. The upper limit for the threshold value is shown with a solid line, the lower limit with a dashed line and the middle value with a dotted line. While for an edge component f _edge ≥ 0.6 the lower limit is 0.15, the upper limit is 0.85 and the mean is 0.50, these values for an edge component of f _{edge are} 0.34, 0.63 and 0.92.

Other sabotaging options besides moving the camera are masking or spraying of their lens. In these cases, the lack of overlap of the edge images results in the correlation value q _corr = 0, ie a maximum deviation of the image sections.

The method steps illustrated in the flow chart of FIG. 1 are implemented in the form of a program on a processor which is either integrated in a surveillance camera and forms a functional unit with it, or to which a plurality of, for example, 32 surveillance cameras can be connected and which for this purpose the parallel processing of several, for example 32, data words is formed.

The description of the method of detecting sabotage attempts on a surveillance camera by examining edge values is not intended to be limiting.

Rather, it is pointed out that the study of edge values is only a special case of examining the recorded images for changes. As mentioned earlier, such an examination for changes can also be made by directly comparing the images, or by examining the corners, intensity or color of the pixels, or by DCT, Fourier or wavelet transform and using the DCT , Fourier or wavelet coefficients. Generally one can say that characteristic coefficients are extracted from the image for certain properties of the image.

This means that a temporal average of the pixels is formed for each pixel examined or for each examined image region, resulting in a stable background image. Then characteristic coefficients are extracted and transformed into an information matrix. Alternatively, the coefficients can already be formed from the image and then transformed into an information matrix. The current information matrix is then compared with the information matrix of a reference image.

Claims (24)

A method for detecting sabotage attempts on a surveillance camera on the basis of the images taken by the camera, characterized in that each image compared with a previously recorded reference image and obtained therefrom a correlation value (q _corr ) and this is compared with an alarm threshold (thres _alarm ). Method according to claim 1, characterized by the following method steps: - extraction of coefficients characteristic of certain properties of the image; - averaging for each coefficient; - transformation of the coefficients into an information matrix; Correlation of the current information matrix with an earlier calculated information matrix; - determination of the correlation value (q _corr ); and - Comparison of the correlation value (q _corr ) with the alarm threshold (thres _Alarm ). Method according to claim 1, characterized by the following method steps: Averaging for each pixel of the viewed image or a section thereof; Extraction of coefficients characteristic of certain properties of the image, such as mean image or background image; - transformation of the coefficients into an information matrix; Correlation of the current information matrix with an earlier calculated information matrix; - determination of the correlation value (q _corr ); and - Comparison of the correlation value (q _corr ) with the alarm threshold (thres _Alarm ). A method according to claim 2 or 3, characterized in that said specific characteristic is formed by the color or intensity of the individual pixels. Method according to Claim 4, characterized in that Fourier coefficients, DCT coefficients, wavelet coefficients, edge or corner values are examined as coefficients representative of certain properties of the image. A method according to claim 5, characterized in that in the examination of edge values from the camera image, a stable background extracted without movements and for this an edge image (B) is calculated, and that the current Edge image correlated with a previously calculated and thereby the correlation value (q _corr ) is determined. A method according to claim 6, characterized in that the extraction of the stable background (B) is performed by a statistical method (1), and by applying a Sobel operator (2) and a mean operator (3) to the individual pixels of the stable background (B) the edge values of the individual pixels are determined. A method according to claim 7, characterized in that in a binarization stage (4) a digital edge image (Q) obtained and stored in a ring memory (5). A method according to claim 8, characterized in that by correlation (6) of the current edge image with an earlier calculation of the edge total over all pixels. A method according to claim 9, characterized in that the correlation by comparison of the first binary image (Q (0)) with the last binary image (Q (N)) in the ring memory (5). Method according to claim 10, characterized in that a first sum (S ₀ ) of the edges in the first binary image (Q (0)), a second sum (S _N ) of the edges in the last binary image (Q (N)) and a third sum (S _corr ) of the edges occurring in the first and in the last binary image (Q (0) and Q (N)) is determined. A method according to claim 11, characterized in that the correlation value (q _corr ) is defined as the quotient of the third sum (S _corr ) and the smaller of the two sums first sum (S ₀ ) and second sum (S _N ). A method according to claim 12, characterized in that the correlation value (q _corr ) is fed to a stage (7) for alarm decision and is compared in this with an alarm threshold (thres _alarm ). A method according to claim 13, characterized in that the alarm threshold (thres _alarm ) is set as a function of the edge portion of the number of pixels considered and for this purpose the alarm sensitivity (δq) is dynamically mapped to a variable band of alarm thresholds. Device for detecting sabotage attempts on a surveillance camera on the basis of the images taken by the camera, with a processor for evaluating the captured images, characterized by a program implemented on this processor for comparing each image with an earlier one recorded reference image and obtaining a correlation value (q _corr ) and for its comparison with an alarm threshold (thres _alarm ). Device according to claim 15, characterized in that the processor is integrated in the surveillance camera. Device according to claim 16, characterized in that a plurality of surveillance cameras are connected to a common processor, which is designed for the parallel processing of a plurality of data words. Device according to claim 16 or 17, characterized in that the program implemented on the processor is designed to carry out the following method steps: - extraction of coefficients characteristic of certain properties of the image; - averaging for each coefficient; - transformation of the coefficients into an information matrix; Correlation of the current information matrix with a previously calculated one; - determination of the correlation value (q _corr ); and - Comparison of the correlation value (q _corr ) with the alarm threshold (thres _Alarm ). Device according to claim 16 or 17, characterized in that the program implemented on the processor is designed to carry out the following method steps: Averaging for each pixel of the viewed image or a section thereof; Extraction of coefficients characteristic of certain properties of the image, such as mean image or background image; - transformation of the coefficients into an information matrix; Correlation of the current information matrix with a previously calculated one; - determination of the correlation value (q _corr ); and - Comparison of the correlation value (q _corr ) with the alarm threshold (thres _Alarm ) _. Device according to claim 18, characterized in that the coefficients representative of certain properties of the image are formed by Fourier coefficients, DCT coefficients, wavelet coefficients, edge or corner values. Device according to claim 20, characterized in that in the examination of edge values from the camera image a stable background is extracted without movements and for this an edge image (B) is calculated, and that the current edge image correlates with an earlier calculated and thereby the correlation value (q _corr ) is determined. Device according to Claim 21, characterized in that the edge image is calculated by applying a Sobel operator (2) and a mean operator (3) to the pixels of the stable background (B). Device according to Claim 22, characterized in that a binarization stage (4) for obtaining a digital edge image (Q) is provided from the edge values of the individual pixels, and that the binarization stage (4) has a ring memory (5) for storing the digital edge images (Q ) is connected downstream. Device according to Claim 23, characterized in that the correlation takes place by comparison of the first binary image (Q (0)) with the last binary image (Q (N)) in the ring memory (5).

Images