WO2018050644A1 - Procédé, système informatique et produit programme d'ordinateur pour détecter une altération de caméra de surveillance vidéo - Google Patents

Procédé, système informatique et produit programme d'ordinateur pour détecter une altération de caméra de surveillance vidéo Download PDF

Info

Publication number
WO2018050644A1
WO2018050644A1 PCT/EP2017/072909 EP2017072909W WO2018050644A1 WO 2018050644 A1 WO2018050644 A1 WO 2018050644A1 EP 2017072909 W EP2017072909 W EP 2017072909W WO 2018050644 A1 WO2018050644 A1 WO 2018050644A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
images
indicators
similarity
points
Prior art date
Application number
PCT/EP2017/072909
Other languages
English (en)
Inventor
Martí BALCELLS CAPELLADES
Nicolás HERRERO MOLINA
Josep Oriol HERRERA MARÍ
Jordi LLUÍS BARBA
Original Assignee
Davantis Technologies, S.L.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Davantis Technologies, S.L. filed Critical Davantis Technologies, S.L.
Publication of WO2018050644A1 publication Critical patent/WO2018050644A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B13/00—Burglar, theft or intruder alarms
    • G08B13/18—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength
    • G08B13/189—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems
    • G08B13/194—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems using image scanning and comparing systems
    • G08B13/196—Actuation by interference with heat, light, or radiation of shorter wavelength; Actuation by intruding sources of heat, light, or radiation of shorter wavelength using passive radiation detection systems using image scanning and comparing systems using television cameras
    • G08B13/19602—Image analysis to detect motion of the intruder, e.g. by frame subtraction
    • G08B13/19613—Recognition of a predetermined image pattern or behaviour pattern indicating theft or intrusion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/0021—Image watermarking
    • G06T1/0028—Adaptive watermarking, e.g. Human Visual System [HVS]-based watermarking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • G06T7/41—Analysis of texture based on statistical description of texture
    • G06T7/44—Analysis of texture based on statistical description of texture using image operators, e.g. filters, edge density metrics or local histograms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G06T9/20—Contour coding, e.g. using detection of edges
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2201/00—General purpose image data processing
    • G06T2201/005—Image watermarking
    • G06T2201/0051—Embedding of the watermark in the spatial domain
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2201/00—General purpose image data processing
    • G06T2201/005—Image watermarking
    • G06T2201/0201—Image watermarking whereby only tamper or origin are detected and no embedding takes place

Definitions

  • This invention relates to a method for detecting video surveillance camera tampering and a system, computer system and computer program product suitable for carrying out such a method.
  • this reference image may be a background image obtained by means of a background subtraction algorithm.
  • a method for detecting video surveillance camera tampering the camera being configured to generate a video signal that includes a sequence of images, each image comprising a plurality of pixels.
  • the method comprises acquiring at least a part of the image sequence and processing the acquired images using an image processing method to determine, from all or some of the pixels of the image, a set of predefined image characteristics.
  • This set of predefined characteristics includes at least one characteristic of points of interest in the image.
  • the method further comprises selecting the set of predefined characteristics of one or more pairs of acquired images, each pair comprising a first and second image, and determining, for one or more of the predefined characteristics, one or more indicators of similarity of the predefined characteristic between the first and second image of all or some of the pairs of images.
  • Such indicators of similarity include at least one indicator of correspondence between the points of interest in the first and second images.
  • the method also comprises providing the indicators of similarity to one or more classification engines, each classification engine being trained to generate, from all or some of the similarity indicators, a tamper indicator. In the event that some or all of the tamper indicators represent the existence of tampering, a tamper alarm can be generated, for example.
  • One aspect of the proposed method is that it can be effective and efficient for both visible cameras that produce well-lit images and cameras that produce thermal or infrared (IR) spectrum images.
  • Thermal cameras have features that cause the image to undergo sudden changes every time the camera carries out what is known as non-uniformity correction, which is a re- calibration of a thermal sensor that occurs periodically (e.g. every 5 minutes).
  • the thermal sensor usually has a resolution (e.g. 14 bits) that is much larger than the resolution of the image provided (e.g. 8 bits), the camera has to exclude information in order to generate the image.
  • the proposed method can offer good performance in this regard (reduced number of false detections) because it is based on determining points of interest (predefined characteristics) and correspondences between points on different images (indicators of similarity). It has been found that points of interest and correspondences between points on different images generate data that is more robust to changes produced by re-calibrations of the thermal camera, compared to other prior art approaches. Therefore, the suggested method offers a great deal of versatility with respect to prior art methods in that the proposed method can function effectively and reliably whether it processes images generated by a visible or thermal camera.
  • classification engines allows the proposed method to use many other predefined characteristics and corresponding indicators of similarity, as well as the above- mentioned points of interest and correspondences between points of interest.
  • Prior art methods are usually based on very small sets of characteristics so it is relatively easy to establish a decision criterion based on an empirically determined threshold.
  • the proposed method is very powerful and versatile in that it can allow the use of a large number of predefined characteristics and indicators of similarity, compared to prior art methods. This can result in greater power and reliability in estimating whether or not tampering has taken place.
  • Classification engines can be trained to take into account all of the indicators of similarity considered necessary, as described in other parts of the disclosure.
  • Another aspect of the proposed method is that new indicators (dependent on new or preexisting predefined characteristics) can be incorporated to estimate the degree of tampering relatively easily, re-training the corresponding classification engine with data that is representative of the new indicators.
  • An additional aspect of the proposed method is that it can detect different types of tampering using different classification engines; for example, one for each type of tampering to be detected, as described in more detail in other parts of the disclosure.
  • images from the camera may be able to undergo a background subtraction procedure to preserve only the static part of the scene.
  • this background subtraction can significantly reduce the number of false tampering detections and, therefore, may improve the effectiveness and reliability of the method.
  • the (predefined) characteristic of points of interest in the image may comprise a number of points of interest in the image, and the indicators of similarity may comprise a ratio of the number of points of interest between (first and second) images.
  • determining the indicator of correspondence between points of interest (of the first and second image) may comprise the application of a homography constraint, as described in more detail in other parts of the description.
  • This homography constraint may enable a good number of incorrect correspondences to be excluded and, therefore, provide a significant improvement in the reliability of the method.
  • the proposed method may use other predefined characteristics and/or indicators of similarity, apart from those relating to points of interest and correspondences between points in different images.
  • One aspect of combining multiple characteristics and indicators may be to improve the power and reliability of the method.
  • the method may use a Jaccard distance calculated from the points of interest detected and/or a movement indicator of the points of interest between (the first and second) images.
  • the set of predefined characteristics may comprise an image histogram
  • the indicators of similarity may comprise an entropy ratio between histograms (of the first and second images) and/or a ratio of width between histograms (of the first and second images) and/or a distance based on an intersection between histograms (of the first and second images).
  • the set of predefined characteristics may comprise a histogram of oriented gradients of the image
  • the indicators of similarity may comprise an entropy ratio between histograms of oriented gradients (of the first and second images) and/or a distance based on an intersection between histograms (of the first and second images).
  • the set of predefined characteristics may comprise image contours, and the indicators of similarity may comprise a ratio of contours between (the first and second) images.
  • a video surveillance camera tamper detector system configured to generate a video signal that includes a sequence of images, each image comprising a plurality of pixels and the system comprising:
  • the means included in the tamper detector system may be electronic or computer means used interchangeably, that is to say, one part of the described means may be electronic means and the other part may be computer means, or the means described may be all electronic or all computer means. Examples of systems comprising only electronic means may be a CPLD (complex programmable logic device), a FPGA (field programmable gate array) or an ASIC (application-specific integrated circuits).
  • a computer system for detecting video surveillance camera tampering the camera being configured to generate a video signal that includes a sequence of images, each image comprising a plurality of pixels, and the computer system comprising a memory and a processor, in which the memory stores computer program instructions executable by the processor, these instructions comprising functionalities for executing any one of the methods for detecting video surveillance camera tampering described above.
  • the computer system can be part of the detector system, i.e. a computer system included in the detector system, or it can be the detector system itself.
  • the invention provides a computer program product comprising program instructions for causing a (computer) system to execute any one of the above methods to detect video surveillance camera tampering.
  • Such a computer program may be stored in physical storage media, such as a recording medium, computer memory or read-only memory, or may be carried by an electrical or optical carrier wave.
  • Figure 1 shows a block diagram depicting a method for detecting video surveillance camera tampering, according to a first example.
  • Figure 2 shows a partial block diagram of a method similar to that shown in Figure 1 .
  • Figure 3 shows a block diagram depicting a method for detecting video surveillance camera tampering, according to a second example.
  • Figure 4 shows a block diagram depicting a method for detecting video surveillance camera tampering, according to a third example.
  • Figure 1 shows a block diagram depicting a method for detecting tampering of a video surveillance camera 100, according to an example.
  • the camera may be configured to generate a video signal including a sequence of images. Each image may comprise a plurality of pixels.
  • the method may comprise the acquisition of at least a part of the image sequence. Acquisition of images can be carried out analogically or digitally (via IP, for example) using specific hardware, and for any type of transmitter medium: cable, fiber, wireless, etc.
  • IP digitally
  • modules can be used to de-multiplex and decode video information independently from the protocol and encoding used by the camera to transmit the image via IP.
  • acquired images can be processed using image processing software to determine, from all or part of the pixels in the image, a set of predefined image characteristics suitable for comparing and detecting differences between images.
  • a set of predefined characteristics may also be referred to as an image descriptor, and may consist of characteristics of points of interest in the image and other characteristics such as pixel histogram, histogram of oriented gradients, etc.
  • image processing software determines, from all or part of the pixels in the image, a set of predefined image characteristics suitable for comparing and detecting differences between images.
  • image descriptor Such a set of predefined characteristics may also be referred to as an image descriptor, and may consist of characteristics of points of interest in the image and other characteristics such as pixel histogram, histogram of oriented gradients, etc.
  • the various predefined characteristics obtainable in the context of different examples of tamper detection are indicated in more detail.
  • Points of interest in an image describe some characteristic points in the image that can be located in another image taken from another point of view.
  • a background subtraction method Prior to determining the predefined characteristics of an image, a background subtraction method can be applied to the image. Therefore, determination of the predefined characteristics can be performed from the images obtained by the camera (without applying the background subtraction method) or, alternatively, from background images (resulting from the application of the background subtraction method).
  • a background image is an image that contains the static part of the video sequence.
  • the scene may exhibit a high degree of activity (e.g. people and/or vehicles moving around the scene), which may induce a false detection because this activity changes the appearance of the scene.
  • obtaining predefined characteristics can be carried out on what is called the background image.
  • This background image contains only the static part of the scene and is obtained by techniques known as background subtraction.
  • An example of a background subtraction algorithm is Mixture of Gaussians (MoG).
  • Image descriptors can be introduced into buffer or accumulator 103 as they are obtained.
  • Buffer 103 can be of fixed size either in time or in number of elements (descriptors). Thus, when a new image descriptor (or set of predefined characteristics) is introduced into buffer 103, the oldest descriptor can be extracted from the buffer, thereby keeping the size (time or number of elements) of accumulator 103 constant.
  • Block 104 can access buffer 103 to select the set of predefined characteristics of at least one image pair comprising a first and second image.
  • some of the descriptors can be obtained directly from the output of block 102 either before or after its storage in buffer 103.
  • Selection of descriptors can be performed according to different selection criteria depending on the type of tampering to be detected. For example, to detect a slow-moving scene change, two descriptors of images with little time difference between them can be selected. In contrast, to detect camera occlusion, the descriptors of two images that are more distant in time (within the sequence of images) can be selected.
  • one or more indicators of similarity of the predefined characteristic between the first and second image of each pair of images can also be determined for all or part of the predefined characteristics.
  • Indicators of similarity can also be termed tampering characteristics and can be provided to the next block 105 in vector form.
  • An indicator of similarity can, for example, be a value between 0 and 1 , in which a value close to 0 can indicate that the images are very similar and an indicator close to 1 can indicate that they are very different.
  • Indicators of similarity obtained in block 104 can comprise an indicator of correspondence between the points of interest in the first and second images and other indicators, such as an entropy ratio, a histogram width ratio, etc.
  • the different indicators of similarity that can be used in the context of different examples of tamper detection are indicated in more detail.
  • Block 105 can comprise one or more classification engines which can each receive a vector of tampering characteristics (or indicators of similarity) from block 104.
  • Each classification engine can have been previously trained to generate a tamper indicator from the corresponding indicators of similarity (or tampering characteristics).
  • each classification engine can use a pre-trained model or predefined thresholds for voting on whether tampering has or has not occurred. The result of this vote can produce, for example, a tamper indicator equal to '1 ' (tampering) or ⁇ ' (no tampering).
  • each resultant tamper indicator can be accumulated with previously obtained tamper indicators.
  • Figure 2 shows a partial block diagram of a method similar to that shown in Figure 1 .
  • classifier block 201 can receive tampering characteristics (or indicators of similarity) 200 and generate a binary (0/1 ) vote 202. It is also shown that classifier 201 can be based on a previously trained model 205 to generate vote (or tamper indicator) 202 as a function of the tampering characteristics 200 received.
  • Accumulator block 203 can receive vote (or tamper indicator) 202 to accumulate it to votes generated in previous iterations, in order to estimate the existence or otherwise of tampering.
  • Figure 3 shows a block diagram depicting a method for detecting video surveillance camera tampering, according to a second example.
  • the method can comprise accessing picture descriptor buffer 300 to select the set of predefined characteristics (or descriptors) of a given pair of images / 301 and / 302, in order to generate indicators of similarity (or tampering characteristics) 304.
  • These indicators of similarity 304 can be provided to a number of classifiers 305 that have been pre-trained to each generate a tamper indicator (or vote) for a particular type of tampering. It is possible, for example, to use as many classification engines as different types of tampering to be detected. Each classification engine can receive all or some indicators of similarity 304 depending on the type of tampering to be detected.
  • a first classifier can generate a camera occlusion tamper indicator
  • a second classifier can generate a fast-moving scene change tamper indicator
  • a third classifier can generate a slow-motion scene change tamper indicator, etc.
  • each vote (or tamper indicator) 306 generated by classifiers 305 can be accumulated to previously obtained indicators (in previous iterations) in, for example, a buffer or accumulator of constant size (in time or number of samples). Therefore, the entry of a new vote in that buffer can cause the exit (or elimination) of the oldest one. It is also possible to generate a percentage of votes that indicate tampering with respect to a total of votes accumulated in the buffer.
  • a global decision 308 can be generated based on votes 306 received and/or the percentages of votes indicating tampering.
  • the overall decision can be generated based on the following formula:
  • D (t) is the global decision in a time is the vote in time t and associated with
  • is the binary OR operator. In this particular example, therefore, with only one of the accumulated votes 306 indicating that tampering has occurred, final decision D (t) will also indicate that tampering has occurred.
  • Figure 4 shows a block diagram depicting a method for detecting video surveillance camera tampering, according to a third example.
  • Two tampering characteristic calculation blocks 405 and 414 (or indicators of similarity) are observed.
  • Block 405 can access buffer 400 to select the set of characteristics of two images / 401 and / 402 according to a first selection criterion.
  • Block 414 can access buffer 400 to select the set of characteristics of two images K 403 and L 404 according to a second selection criterion.
  • the first and second selection criteria can be different in order to determine different types of tampering.
  • the first criterion can define the selection of two images with a first elapsed time between them and the second criterion can define the selection of two images with a second time between images lower than the first.
  • the images selected with the first criterion can be used, for example, to detect camera occlusion, while those selected with the second criterion can be used to detect slow movement.
  • Blocks 408 and 409 can be similar to block set 305 in Figure 3.
  • Block 408 can consist of classifiers trained to detect different types of tampering from indicators of similarity 406 generated by block 405 and associated with images / 401 and / 402.
  • Block 409 can consist of classifiers trained to detect different types of tampering from indicators of similarity 407 generated by block 414 and associated with images K 403 and L 404.
  • Decision block 412 can be similar to block 307 in Figure 3.
  • block 412 can generate overall decision 413 based on votes 410 generated by the classifier blocks 408 and 409, according to an approach similar to that described with respect to block 307 in Figure 3.
  • the set of predefined characteristics (or descriptors) of an image can consist of points of interest in the image, a histogram (of the pixels) of the image, a histogram of oriented gradients of the image, image contours, etc.
  • a histogram of an image describes the frequency of occurrence of the possible values of a pixel in the image. In other words, it describes the probability of observing values or ranges of pixel values in the image. Taking into account image / and that possible values i (which may have a pixel) are between 0 and probability p / (i) that image / has value i can be calculated
  • ⁇ ⁇ is the total number of pixels whose value is equal to i
  • n is the total number of pixels in the image.
  • a histogram based on this formula can be a reference to the probability of each level of gray (or color, as the case may be) in image /.
  • a histogram of oriented gradients describes the frequency of occurrence of the orientations of the pixel gradients in an image also taking into account the force with which the gradients of the pixels are oriented in one direction or another.
  • a histogram of N-level oriented gradients can be defined by the following formula:
  • the contours of an image can be determined using any known technique or method for determining contours.
  • the contours of an image can give very valuable information to determine if a camera is pointing to another position (the contours change) or if it has been covered (the contours disappear).
  • the Sobel operator in the horizontal and vertical direction can be defined as follows:
  • a threshold can be applied to the magnitude image of the gradient to obtain the image of contours:
  • points of interest in image / can be defined as points x t of image / corresponding to a certain observed scene that can be equally locatable when the same scene is observed from another point of view.
  • Each point of interest can be described by coordinates which indicate the position at which this point was detected, and by descriptor d , which in some way
  • ⁇ ⁇ , .,. , ⁇ refers to the i-th point of the points detected in the image considered.
  • indicators of similarity or tampering characteristics
  • one or more of the following indicators can be considered: entropy ratio, histogram width ratio, entropy ratio of histograms of oriented gradients, ratio of number of points of interest, indicator of correspondence between points of interest, indicator of displacement of points of interest, contour ratio, distance based on an intersection of image histograms, distance based on an intersection of histograms of oriented gradients, etc.
  • Entropy of an image / can be calculated taking into account the definition of histogram provided above and based on the following formula:
  • Entropy ratio between a first image / and a second image / can be calculated
  • the entropy ratio between two images is a value between zero and one and that this value is closer to zero the more similar
  • the images are, and closer to one the more different they are.
  • the histogram width ratio can be used, for example, to detect camera occlusion tampering since camera occlusion is considered to cause a sudden change in the width of the image histogram.
  • the image that it returns is usually of constant color and, therefore, of very narrow histogram and centered in that color (or level of gray).
  • the histogram's standard deviation can be used as an indicator of the histogram width.
  • Standard deviation ⁇ can be calculated according to the following formula: in which ⁇ 1 is the average of the histogram of image /, which can be calculated according to the following formula:
  • a second image / can be calculated according to the following formula:
  • the histogram width ratio between two images is a value between zero and one and that this value is closer to zero the more similar the images are, and closer to one the more different they are.
  • Calculation of the entropy ratio of histogram of oriented gradients H HoG can be similar to the calculation of the image entropy ratio, with the difference that, in this case, histograms of oriented gradients (HoGs), as explained in another part of the description, are used:
  • the entropy ratio of HoGs between two images is a value between zero and one and that this value is closer to zero the
  • the ratio of the number of points of interest can also be used, for example, to detect a covered (or occluded) camera by assessing whether or not there is a significant drop in points of interest detected between two images. The transition from an image of a normal scene to an occluded image can cause a very significant drop in points of interest.
  • the ratio of the number of points of interest between an image / and an image / can be used, for example, to detect a covered (or occluded) camera by assessing whether or not there is a significant drop in points of interest detected between two images. The transition from an image of a normal scene to an occluded image can cause a very significant drop in points of interest.
  • the ratio of the number of points of interest between an image / and an image / can be used, for example, to detect a covered (or occluded) camera by assessing whether or not there is a significant drop in points of interest detected between two images. The transition from an image of a normal scene to an occluded image can cause a
  • M ⁇ is the number of points of interest detected in image /
  • M j is the number of points of interest detected in image /.
  • the indicator of correspondence between the points of interest in a first and second image can be calculated as the Euclidean distance of two vectors as described below.
  • Security cameras usually observe at least one plane and while there is no tampering it can be assumed that there will always be a set of points that will fulfill this homography constraint between a pair of images of the same scene.
  • a robust model estimation method can therefore be used (for example: RANSAC) to exclude incorrect correspondences, i.e. those that show a spatial error greater than a certain threshold T:
  • the indicator of correspondence between the points of interest of a first and a second image can be a Jaccard distance calculated as described below.
  • a Jaccard index can be calculated according to the following formula:
  • / ⁇ / refers to the set of points that correspond to (or match) both images
  • / U / refers to the set of all points detected in both images
  • the Jaccard distance is a value between zero and one and that this value is closer to zero the more similar the images are,
  • the proposed Jaccard distance can be used, for example, to detect camera occlusion, since the difference in points of interest between a normal image and an occluded image can be (very) significant. In this case, the number of correspondences between points of interest in the images would tend to zero and, according to formula 21 , Jaccard distance
  • Jaccard distance f jIc ix OJ Jaccard distance f jIc ix OJ
  • the point of interest displacement indicator can provide an estimate of the displacement of the points of interest in an image p relative to the points of a reference image m.
  • a matching or correspondence
  • Displacements calculated according to formula 22 can be used to calculate an average
  • Displacements can be calculated with respect to the first position of observation of the point to avoid the accumulation of displacement in cases of vibration of the camera. To calculate average displacement all of the points of interest in the image cannot be used, but a
  • each point can have a different time of existence, in which case the actual displacement of the camera would not be estimated, but the calculated average displacement could be below the actual displacement of the camera.
  • histograms of oriented gradients can provide global information about the contour content of the image, it can be of interest to locate the contours to, for example, detect a small movement of the camera. It can be the case that, after a slight displacement of the camera, there is no change in the histograms of oriented gradients because the new image has the same contours but slightly displaced (only a few pixels).
  • the contour ratio can be defined as the ratio between the number of pixels detected as contour in the two images and the total sum of pixels detected between the two images, which can be calculated according to the following formula:
  • the distance based on the intersection of image histograms can be determined by calculating the distance between the histograms as the intersection between them.
  • the intersection of the histogram of image / with the histogram of image / can be defined by means of the following formula:
  • This characteristic takes values in the range [0,1 ], 0 being when the
  • histograms are equal and 1 when the images do not share any pixel value.
  • the distance based on the intersection of histograms of oriented gradients can be determined in a manner similar to that described above with respect to the distance between image histograms.
  • intersection of a histogram of oriented gradients h, and another histogram of oriented radients h can be defined by means of the following formula: Therefore, given a pair of images / and /, the distance based on the intersection of the histograms of oriented gradients of these images can be determined according to the following formula:
  • This characteristic takes values in the range [0,1 ], 0 being when the
  • histograms are equal and 1 when the images do not share any pixel value.
  • tampering due to camera occlusion could produce at least some of the following alterations between a first image and a second image of a pair of images considered:
  • Drastic gray level change can be modeled by the entropy ratio and/or the intersection of histograms.
  • Narrowing of the image histogram can be modeled by the histogram width ratio.
  • Contour disappearance can be modeled by the entropy ratio of the histogram of oriented gradients and/or its intersection (of the histograms of oriented gradients) and/or the contour image.
  • a decrease in the number of points of interest can be modeled by the ratio of the number of points of interest.
  • a decrease in the number of correspondences between points of interest can be modeled by the indicator of correspondences between points of interest (e.g. Jaccard distance).
  • each image can have a label associated with value T if there is tampering or value ⁇ ' if there is no tampering (with respect to the previous related image).
  • each image i would have an observation , in which is the label that indicates the existence or absence of tampering and is a vector of indicators of similarity (or
  • the detection of tampering due to occlusion can now be treated as a pure classification problem, which can consist of a training phase and an assessment phase.
  • the set of videos can be divided into two groups so that each of these phases (training and evaluation) has its own subset of videos.
  • classification operator c() This operator enables the prediction process to be defined as in which can only have two output values ⁇ ' and To estimate which is operator c() during the
  • n can be obtained according to the following formula:
  • classifier block 201 can perform prediction function y and
  • model block 205 can be the classifier obtained during the training phase described above.
  • tampering due to sudden movement could produce at least some of the following alterations from a first image and a second image of a pair of images considered:
  • Contour disappearance can be modeled by the entropy ratio of the histogram of oriented gradients and/or the contour image.
  • ⁇ A decrease in the number of correspondences between points of interest can be modeled by the indicator of correspondences between points of interest (e.g. Jaccard distance).
  • training of one or more search engines similar to the one described above can be performed in relation to tampering due to camera occlusion, but, in this case, without considering some of the indicators of similarity used for tampering due to camera occlusion.
  • the entropy ratio of the histogram of oriented gradients and/or the correspondence indicator between points of interest (which can be the Jaccard distance) can be considered.
  • the detection of tampering due to a slow change in the scene refers to the detection, for example, of a scene change produced by poor camera fixation.
  • it is possible to estimate the movement of the camera by estimating the displacement of the points of interest of an image relative to another image (of a considered pair of images), as previously described.
  • a classification function can return 1 , while it can return 0 in all
  • the threshold can be predefined as a percentage of a diagonal of the image in order for it to be used for different image sizes.
  • the classification function can be expressed as follo
  • a user-defined exclusion region can be considered in order for the tamper detection method to take into account all the pixels in the image except those in the exclusion region. This can be useful if, for example, it is determined that a part of the image is usually obscured by an object parked in the scene (for example, a truck). Since this object can significantly alter the image, the detection method can generate false tamper indicators. Therefore, consideration of such an exclusion region can improve the reliability of the method.
  • the exclusion region can be defined, by a user, through a graphical interface.
  • Suitable drawing tools can allow the definition of an exclusion region that can have a rectangular shape, polygon with an undetermined number of edges, or any other shape.
  • Suitable brush tools can allow the user to paint one or more exclusion regions with various shapes.
  • Defining an exclusion region can affect the calculation of the predefined characteristics of an image. For example, to calculate the histogram of the image, only pixels in the image that are not part of the exclusion region can be taken into account. For the histogram of oriented gradients, the same criterion or similar can be applied. In the case of points of interest and contour image, they can be calculated for the whole image and later those (points and/or contours) that are within the exclusion region can be excluded.
  • the computer programs can be in the form of source code, object code or an intermediate code between source code and object code, such as in a partially compiled form, or in any other form suitable for use in the implementation of the processes according to the invention.
  • the carrier medium can be any entity or device capable of carrying the program.
  • the carrier medium can comprise a storage medium, such as a ROM, for example a CD ROM or a semiconductor ROM, or a magnetic recording medium, for example a floppy disc or a hard disk.
  • the carrier medium can be a transmittable carrier medium such as an electrical or optical signal which can be transmitted via electrical or optical cable or by radio or other means.
  • the carrier medium can be constituted by such a cable or other device or means.
  • the carrier medium can be an integrated circuit in which the computer program is embedded, the integrated circuit being adapted to perform or to be used in carrying out the relevant methods.
  • the invention can also be implemented by means of computer systems, such as personal computers, servers, computer networks, laptops, tablets or any other programmable device or computer processor.
  • Programmable electronic devices such as programmable logic controllers (ASICs, FPGAs, programmable controllers, etc.) can also be used complementarily or alternatively.
  • the invention can be implemented in hardware, software, firmware, or any combination of them.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

L'invention concerne des procédés de détection d'altération de caméra de surveillance vidéo, la caméra générant un signal vidéo qui comprend une séquence d'images comprenant une pluralité de pixels. Les procédés consistent : à acquérir de telles images et à les traiter pour déterminer un ensemble de caractéristiques prédéfinies de l'image qui comprennent des points d'intérêt dans l'image ; à sélectionner l'ensemble de caractéristiques prédéfinies d'une paire d'images comprenant une première et une seconde image ; à déterminer, pour les caractéristiques prédéfinies, des indicateurs de similarité pour la caractéristique prédéfinie entre les première et seconde images, y compris un indicateur de correspondance entre des points d'intérêt dans les première et seconde images ; et à fournir des indicateurs de similarité à un ou plusieurs moteurs de classification formés pour générer un indicateur d'altération à partir de ces indicateurs. L'invention concerne également des systèmes, des systèmes informatiques et des programmes informatiques qui sont appropriés pour mettre en œuvre de tels procédés.
PCT/EP2017/072909 2016-09-13 2017-09-12 Procédé, système informatique et produit programme d'ordinateur pour détecter une altération de caméra de surveillance vidéo WO2018050644A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
ESP201631188 2016-09-13
ES201631188A ES2659049B1 (es) 2016-09-13 2016-09-13 Procedimiento, sistema, sistema informático y producto de programa informático para detectar un sabotaje de una cámara de video-vigilancia

Publications (1)

Publication Number Publication Date
WO2018050644A1 true WO2018050644A1 (fr) 2018-03-22

Family

ID=59923409

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2017/072909 WO2018050644A1 (fr) 2016-09-13 2017-09-12 Procédé, système informatique et produit programme d'ordinateur pour détecter une altération de caméra de surveillance vidéo

Country Status (2)

Country Link
ES (1) ES2659049B1 (fr)
WO (1) WO2018050644A1 (fr)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110942456A (zh) * 2019-11-25 2020-03-31 深圳前海微众银行股份有限公司 篡改图像检测方法、装置、设备及存储介质
CN111260645A (zh) * 2020-02-20 2020-06-09 中国科学院自动化研究所 基于分块分类深度学习的篡改图像检测方法及系统
CN112101155A (zh) * 2020-09-02 2020-12-18 北京博睿维讯科技有限公司 一种显示内容核实方法、装置、系统及存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007067722A2 (fr) * 2005-12-08 2007-06-14 Lenel Systems International, Inc. Systeme et methode de detection d'une camera non valide dans une surveillance video
US20110064315A1 (en) * 2009-09-15 2011-03-17 Texas Instruments Incorporated Method and apparatus for image capturing tampering detection
US20140078307A1 (en) * 2012-03-14 2014-03-20 Samsung Techwin Co., Ltd. Method of detecting camera tempering and system thereof
WO2014042514A2 (fr) * 2012-09-12 2014-03-20 Mimos Berhad Système de surveillance et procédé de détection et de correction de sabotage

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5561524B2 (ja) * 2010-03-19 2014-07-30 ソニー株式会社 画像処理装置および方法、並びにプログラム

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007067722A2 (fr) * 2005-12-08 2007-06-14 Lenel Systems International, Inc. Systeme et methode de detection d'une camera non valide dans une surveillance video
US20110064315A1 (en) * 2009-09-15 2011-03-17 Texas Instruments Incorporated Method and apparatus for image capturing tampering detection
US20140078307A1 (en) * 2012-03-14 2014-03-20 Samsung Techwin Co., Ltd. Method of detecting camera tempering and system thereof
WO2014042514A2 (fr) * 2012-09-12 2014-03-20 Mimos Berhad Système de surveillance et procédé de détection et de correction de sabotage

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
DAW-TUNG LIN ET AL: "Real-Time Active Tampering Detection of Surveillance Camera and Implementation on Digital Signal Processor", INTELLIGENT INFORMATION HIDING AND MULTIMEDIA SIGNAL PROCESSING (IIH-MSP), 2012 EIGHTH INTERNATIONAL CONFERENCE ON, IEEE, 18 July 2012 (2012-07-18), pages 383 - 386, XP032223712, ISBN: 978-1-4673-1741-2, DOI: 10.1109/IIH-MSP.2012.99 *
TSESMELIS THEODORE ET AL: "Tamper detection for active surveillance systems", 2013 10TH IEEE INTERNATIONAL CONFERENCE ON ADVANCED VIDEO AND SIGNAL BASED SURVEILLANCE, IEEE, 27 August 2013 (2013-08-27), pages 57 - 62, XP032509897, DOI: 10.1109/AVSS.2013.6636616 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110942456A (zh) * 2019-11-25 2020-03-31 深圳前海微众银行股份有限公司 篡改图像检测方法、装置、设备及存储介质
CN110942456B (zh) * 2019-11-25 2024-01-23 深圳前海微众银行股份有限公司 篡改图像检测方法、装置、设备及存储介质
CN111260645A (zh) * 2020-02-20 2020-06-09 中国科学院自动化研究所 基于分块分类深度学习的篡改图像检测方法及系统
CN111260645B (zh) * 2020-02-20 2023-10-13 中国科学院自动化研究所 基于分块分类深度学习的篡改图像检测方法及系统
CN112101155A (zh) * 2020-09-02 2020-12-18 北京博睿维讯科技有限公司 一种显示内容核实方法、装置、系统及存储介质
CN112101155B (zh) * 2020-09-02 2024-04-26 北京博睿维讯科技有限公司 一种显示内容核实方法、装置、系统及存储介质

Also Published As

Publication number Publication date
ES2659049A1 (es) 2018-03-13
ES2659049B1 (es) 2018-11-22

Similar Documents

Publication Publication Date Title
US10242282B2 (en) Video redaction method and system
D'Avino et al. Autoencoder with recurrent neural networks for video forgery detection
St-Charles et al. Universal background subtraction using word consensus models
KR101764845B1 (ko) 다중 이동 물체의 겹침 제거 및 추적을 위한 영상 감시 장치 및 방법
CN108229297B (zh) 人脸识别方法和装置、电子设备、计算机存储介质
US9405974B2 (en) System and method for using apparent size and orientation of an object to improve video-based tracking in regularized environments
US20090041297A1 (en) Human detection and tracking for security applications
Ryan et al. Scene invariant multi camera crowd counting
US9811755B2 (en) Object monitoring system, object monitoring method, and monitoring target extraction program
WO2009049314A2 (fr) Système de traitement vidéo utilisant une soustraction de comportement entre des séquences d'images vidéo de référence et observées
WO2007143642A2 (fr) procédé et système pour la détection d'objets enlevés sur des images vidéo
Wei et al. Motion detection based on optical flow and self-adaptive threshold segmentation
JP4486594B2 (ja) 確率的外観集合体を使用するビデオに基づく顔認識
JP2018124689A (ja) 移動物体検出装置、移動物体検出システム、及び移動物体検出方法
Lejmi et al. Challenges and methods of violence detection in surveillance video: A survey
WO2018050644A1 (fr) Procédé, système informatique et produit programme d'ordinateur pour détecter une altération de caméra de surveillance vidéo
Khan et al. Review on moving object detection in video surveillance
CN103377472B (zh) 用于去除附着噪声的方法和系统
Farazi et al. Real-time visual tracking and identification for a team of homogeneous humanoid robots
Miramontes-Jaramillo et al. Descriptor-based tracking algorithm using a depth camera
Abdellaoui et al. Template matching approach for automatic human body tracking in video
Ata et al. Proposed enhancement for vehicle tracking in traffic videos based computer vision techniques
Mohammed et al. An improved CAMShift algorithm for object detection and extraction
Rabhi et al. Face tracking: state of the art
Jalal et al. A multi-resolution framework for multi-object tracking in Daubechies complex wavelet domain

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17771379

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17771379

Country of ref document: EP

Kind code of ref document: A1