RU2460056C1 - Method of detecting fraud by staging road accident and apparatus for realising said method - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: invention relates to expert evaluation of road accidents in order to detect fraud by staging a road accident. The essence of the fraud lies in that damage to a car in different circumstances are passed for damages done in another road accident. A fraud will be eliminated if it will be proven that characteristics of all damages belong to a relatively temporary factor of one general set. The invention proposes to use such a characteristic in form of intensity of colour shades of corroded surfaces where the body of the car has been damaged in red, blue and green, as well as the sum of intensities of these colours. To detect fraud, the places where the damages are located are photographed and the obtained images are processed using a RGB colour model. The image processing results are analysed for identity of characteristics of the obtained images of one general set relative the temporary factor, for which known mathematical statistics algorithms are used.

EFFECT: simple procedure for assessing identity of the same general set of all damages to the body of a car.

2 cl, 3 dwg

Description

The invention relates to the examination of a traffic accident in order to detect fraud in the stage of the accident.

According to estimates of Russian insurance companies in the field of compulsory motor third-party liability insurance, the volume of insurance fraud is 15–20% of the total amount of payments made. One of the most common types of fraud in car insurance is the staging of an accident, mainly the stage of damage to a vehicle as a result of an accident [1].

One of the ways to identify insurance fraud is the express information analysis program - “Traffic Accident Indicator” [2]. The program “Traffic Accident Indicator” contains 81 indicators of fraud in road accidents with material damage, 56 indicators of fraud in road accidents related to harm to life and health, as well as 250 typical measures for the investigation of insurance cases under compulsory motor third-party liability insurance, motor third-party liability insurance, and comprehensive insurance. The “Indicator” analyzes the information on any insurance event related to an accident, gives a final conclusion indicating the degree of falsehood of the information and forms a list of events that should be carried out on a particular insurance event.

The main disadvantage of the program is that the computer program cannot evaluate the technical condition of the car. In particular, in the case of a staging of an accident, a staging of body damage, the program does not allow to determine the culprit of the accident, does not allow to determine the time of receipt of individual defects in the body of the vehicle. Thus, the Indicator program does not actually solve the technical side of the problem. This requires technical solutions.

An analogue of such devices can be a device for identifying (detecting) the culprits of a collision or collision with vehicles as USSR No. 1516415. However, this device allows you to find a participant in an accident who has left the scene of an accident, but does not allow you to identify the culprit of an accident, to establish or exclude the fact of a staging of an accident, to assess the size of defects in motor vehicles in road traffic accidents.

There is a method of determining the size of defects in vehicles during traffic accidents according to the patent for the invention of the Russian Federation No. 2276404 [3]. The method consists in the fact that the image of the zones of visible defects of the elements or components of the car is transmitted to electronic media. This is done using an objective monitoring tool to record the traces of an incident in an electronic form of storing information, for example, with a digital camera, and hidden ones using non-destructive means of objective monitoring, for example, a flaw detector. In this case, the recorded electronic signal of the defective zone is sent to the computer software, which compares the drawn areas of the control zone and the undeformed surface in the part or assembly under study, and determines the corresponding percentage of the defect of the studied object by the ratio of the areas obtained.

The main disadvantage of the object is the inability to determine the time at which the damage was received. Damage and defects can be received in various accidents, all of them are declared as received in one accident, which is the essence of fraud. However, the time of damage to the car body can be determined by the thickness of the oxide film formed at the site of damage. This is due to the fact that with damage to the body, as a rule, the paint and varnish protective layer is damaged, and the metal without a protective coating begins to rust. By measuring the thickness of the oxide film, it is possible to determine the time of damage. The thickness of the oxide film is determined by the corrosion rate. So, there is a known method for determining the corrosion rate of metals according to the patent of the Russian Federation for invention No. 1245071 [4]. In accordance with [4], a method for determining the corrosion rate of metals consists in placing samples in contact with each other in a channel of a nuclear reactor that are exposed to a corrosive environment and radiation. In this case, the parameters of the oxide film formed on the samples during the tests are periodically measured, and the corrosion rate is judged by these parameters.

Due to the technical complexity of this method according to the patent [4] cannot be used to determine the time to obtain deformation by the car body.

A known method of detecting fraud in the staging of an accident described in [5], which is taken as a prototype. In accordance with [5], it is proposed to evaluate the corrosion rate by color shades of the surface of the corroded layer, for which the so-called RGB color model is used [6]. The method is based on the features of the corrosion process, consisting in the fact that the corrosion process leads to a change in color shades of corrosion products over time. Changes in color shades for the three main color components of the image of corrosion products allows us to find differences in color shades for two corrosion processes that began on the same surface, but at different times. In corrosion products, shades of red, green, and blue change over time. The intensity of the above shades decreases over time. Typical dependences of the color change of corrosion for some seasons, taken from [5], are shown in figure 1 and figure 2.

Figure 1 shows the dependence of the color change of corrosion on time for autumn conditions, and figure 2, respectively, for spring conditions.

As follows from figure 1 and figure 2, the dependence of the color change of corrosion have a monotonous decreasing character. The curve corresponding to the change in the intensity of the sum of the colors of corrosion also has the same character. The monotonicity of the curves makes it possible to construct analytical regression models of the process of corrosion color change from time to time, each of which has its own dispersion. Having these models and determining the intensity of the colors of corrosion by photographing the corroded layer, we can determine the time elapsed since the beginning of the corrosion process. If this time parameter is different for different damages to the car body, this will mean that the damage to the body was not received on the same day and cannot correspond to one accident. However, to implement this prototype method, it is necessary to know empirical models of corrosion color changes as a function of time. At the same time, such models must take into account specific climatic, weather, seasonal and other conditions, for example, characteristics (corrosion properties) for this particular steel grade used for the car body, etc. All this makes the practical use of the method described in [5] for determining the time of receipt of damage by the car body impossible, because it does not allow to solve the problem with the necessary accuracy and reliability, which is the main disadvantage of the prototype method.

However, when detecting fraud, it is enough to establish that not all damage to the body reported for consideration belongs to one general population. Therefore, the aim of the invention is to simplify the procedure for assessing the ownership of the same general population of all damage to a car body. The goal is achieved by using the regularities of the corrosion processes noted above: the corrosion process leads to a monotonic change in color shades of corrosion products over time, which is characterized by the color intensity (red, blue, green) and the total intensity of light reflected by the corrosion products. Since for all damages to the car body received at the same time, the corrosion processes will proceed the same way, the color shades of the corrosion products for all damage sites within the statistical error will be the same, i.e. the intensity of the color shades of the corrosion deposits at the damage site will be the same. Thus, the characteristics of the surfaces of the perforated metal layers at the sites of damage should belong to one general population: the nominal characteristic of one or another color shade for all the damage sites and the corrected sample variances should characterize the same general population.

Thus, in order to answer the question whether all the damages of the car body belong to the same general population, that is, whether they were all inflicted at the same time, it is necessary to determine the characteristics of the surfaces of the corroded metal in all places of damage. For this, it is possible to use the methodology corresponding to the prototype method and described in [5]. Using a digital camera, vehicle body damage is photographed. The resulting photos are processed in the MathCAD system using the RGB color model [6].

During the processing of each image, it is represented in the form of an n × 3m matrix, where n is the image height in pixels, m is the image width in pixels. The width of the original image matrix is divided into three equal parts of width m, each of which is responsible for its own base color, namely for red, green and blue [5]. After that, the matrix of the original image is divided into three separate images: each image corresponds to its base color. Each cell is 8-bit with a range of numbers from 0 - black - up to 255 - white. By averaging the color values in each image, the color characteristic of the original image is obtained. Color characteristics (color saturation) for the base colors and for the entire image (the sum of the saturations of the base colors) for all surface images should belong to the same general populations, if all the damages were received at the same time.

To identify the fact of fraud, it is enough to show that the damages presented as an insured event relate to at least two different populations. As is known, the main characteristics of the sample are the sample mean and sample variance [7], which are determined by known formulas

- выборочное среднее;Where

- sample mean;

X _i - the value of the studied parameter in each i-th experiment (color saturation);

N - sample size (number of measurements);

S ² - sample variance.

Statistical evaluation of the characteristics of the sample is made with a certain reliability and accuracy. The sample size must be at least 7 [8], therefore, for the formation of two samples according to the image processing results, it is necessary to have at least 15 images of corroded damage sites. Moreover, not only images of individual damage can be processed, but also individual fragments of damage images if the damage area is relatively large.

Then suppose that N images or fragments of images of deformations of a car body were subjected to processing. As a result, for each image or image fragment, the color saturation values of the corroded lesions were obtained (measured), and for each of the basic colors, as well as for the overall saturation characteristic, the corresponding values for each experiment were obtained

(3) According to the formula (1) and (2), the characteristics of the sample are calculated

Further, the initial sample is divided into two samples for each of the parameters: those parameter values that are less than the sample average determined by formulas (4) form one sample of parameters with a volume of N ₁ . Those parameter values that are larger than the sample means determined by formula (4) form the first sample. Those parameter values that are less than the selected average form the second sample. Moreover, the first sample has a volume of N ₁ , and the second sample is equal to N ₂ . Moreover, the sum of the volumes of the first and second samples is equal to the size of the general population.

For each of the generated samples, statistical characteristics are determined - sample average and sample variances, the first and second samples being formed for each of the basic colors and for the overall saturation characteristic.

From the obtained samples of volumes N ₁ and N _2, it is necessary to conclude that they belong to the same general population, or to prove that they belong to different general populations, that is, to conclude that damage to the car was received at one or at different times. . To do this, we use the theory of testing statistical hypotheses [8].

If both samples belong to the same general population, it is necessary to compare the average values of the populations. We assume that both samples are uncorrelated. There are theoretically four possible variants of possible assumptions: a) both variances are known and equal to each other; b) both dispersions are known, but not equal; c) both variances are unknown, but it is assumed that they are equal to each other; d) both variances are unknown, and the assumption of their equality has not been made. It should be noted that for each of the color shades and for the general characteristic of the color of the lesions, we will have our own two sample sets, and the volumes of the first and second sets for each of the basic colors and for the general characteristic of the colors may differ, but in total must be equal to the volume of the general population - N.

и

являются оценками одной и той же дисперсии совокупности σ². Индексом δ обозначен базовый цвет или общая цветовая характеристика. В соответствии со сказанным введем в рассмотрение объединенную оценку дисперсии с числом степеней свободы, равным сумме степеней свободы исходных оценок, то есть равным N. Объединенная оценка имеет вид [9]We assume that for each base color for both samples the variances are unknown, but it is assumed that they are equal. As estimates of the variances, we take the sample variances of the first and second corresponding populations. Both sample variances

and

are estimates of the same variance of the population σ ² . The subscript δ indicates the base color or general color characteristic. In accordance with the foregoing, we introduce the combined estimation of variance with the number of degrees of freedom equal to the sum of the degrees of freedom of the initial estimates, that is, equal to N. The combined estimate has the form [9]

Statistics of the test criterion for the hypothesis of equality of sample means has the form

moreover, the criterion itself has a Student t-distribution with ν = N-2 degrees of freedom. The criteria T _δ calculated by formula (12) for each color shade and the overall color characteristic of the damage, based on the necessary accuracy and reliability of the assessment, are compared with the corresponding values of the Student distribution for the corresponding value of the number of degrees of freedom. It is advisable to take critical distribution points of Student for a reliability of 0.975. If all the values of the criterion T _δ calculated by formula (12) are less than the critical values of this criterion taken from the table corresponding to the Student distribution, then there is no reason to reject the hypothesis that all injuries belong to the same population. That is, all damage to the car body with a reliability of 0.975 will correspond to the condition that they were received as a result of the same accident.

Thus, the implementation of the method for detecting fraud in the staging of an accident consists of the following.

1. Using a digital camera, damage to the body of a car is captured.

2. The received images of body damage are processed using the RGB color model to determine the color intensity of the received images in such basic colors as red, green, blue, and in the sum of their intensities in three colors.

3. Four samples of color intensities obtained in accordance with Clause 2, which are further considered as four populations, are processed to calculate sample averages and sample variances for each population.

4. Of the four general samples for each of them, two samples are formed, one of which includes values less than the values of the corresponding general sample average, and the other includes values greater than the values of the corresponding general sample average.

5. For each pair of samples formed in accordance with clause 4 from the corresponding general population, sample mean and sample variances are determined.

6. Next, the membership of the measured values of the color intensity for each pair of samples corresponding to one population is determined. To this end, the equality of sample averages in each pair of four populations according to claim 2 and the equality of the corresponding sample variances by using the Student criterion and the Fisher criterion are checked.

7. If for all pairs of samples in accordance with claim 4, the calculated values of the statistical criteria are less than their corresponding critical values corresponding to the Student and Fisher distributions determined by the size (number of degrees of freedom) of each sample and the reliability of the estimate, then there is no reason to suspect fraud in actions road accident participant.

The method for detecting fraud described above can be implemented in a device consisting of a digital camera with a processor connected to it for processing photographic images. The processor includes a damage image processing unit to determine the color saturation of the images both in the base colors - red, green and blue, and in the sum of the color intensities. From the image processing unit, the signals corresponding to the color saturation value are sent to the memory unit, forming four general sets of image color saturation characteristics for each of the three basic colors and for the sum of the colors red, green and blue. Algorithms for determining sample average and sample variances are “sewn up” in the logic block, according to which sample average and sample variances are calculated for each of the four populations (expressions (4) and (5)). In the division block, each of the four populations recorded in the memory block is divided relative to the corresponding sample means into two samples, one of which includes those color saturation values that are less than the sample average, and the other sample contains those color saturation values that are greater corresponding sample mean.

Using re-wired software in the logic block, sample mean and sample variances are calculated for each of the eight samples. In the comparison block, the hypothesis about the belonging of each pair of samples for the same color of the general population is checked. For this, sample variances are compared by the Fisher test, and sample means are used by the Student test. In the comparison block, the percentage points of the t-student distribution and the percentage points of the Fisher distribution are “sewn up”.

The relations between the calculated and critical values of the corresponding criteria (more or less) obtained from the tabular data “wired” in the comparison unit, the tabular data of the Fisher and Student distribution are displayed on the display, which is part of the device.

The block diagram of the device shown in figure 3. Figure 3 shows the main blocks of the device and the relationship between them.

The device consists of a camera 1; image processing unit 2; memory block 3; block logic and division 4; logic blocks of the second stage 5 and 6, block comparison 7; display 8 and keyboard 9. The device blocks 2 through 7 form the device’s processor, and the display and keyboard form the device’s monitor.

The device operates as follows. The camera 1 takes pictures of places of damage to the car body, and the number of images or image fragments for large damage areas should be at least 15. Next, the images in digital form go to image processing unit 2, where for each image or image fragment using A “built-in” RGB model evaluates (measures) the saturation of the color characteristics of the image in the basic colors — red, green, and blue — and the sum of the saturations of these colors. The measurement results of the characteristics of the images are received in the memory unit 3, where the database is formed. After the database is formed, the reliability and reliability of determining the statistical characteristics of the general sample populations formed in the database are set using the keyboard 9. The reliability and reliability of the ratings is set from the keyboard of the device, and the values of these parameters are displayed on the display 8, which is included in the device. For a given reliability and reliability of statistical estimates, the database formed in the memory block 3 is processed in the logic block 4 in accordance with the algorithms for calculating sample average and sample variances. The 4 sample averages of the populations calculated for each of the base colors, calculated in the logic block, are the centers of dividing the populations into two samples, one of which contains values smaller than the sample averages, and the other contains values larger than the sample averages . The formation of eight sampling sets, two for each general population from the memory block 3, is carried out in the logic and division block 4, and the values of each samples are stored and processed in the logic blocks of the second stage 5 and 6. Where in the logic block of the second stage 5 samples are formed with the values of the arguments smaller than the corresponding sample general averages, and in block 6, samples are formed with argument values that are larger than the corresponding sample general averages. In the same logic blocks of the second stage 5 and 6, each of the eight samples is processed to determine the corresponding sample average and sample variances. The obtained statistical characteristics for each pair of samples obtained from the corresponding populations are compared in the comparison unit 7 with each other to reveal the validity of the hypothesis that each pair of samples obtained from one population, one population is valid. By the Fisher criterion, the sample variances are compared, and by the Student criterion, the sample means are compared, and the tabular values of the Fisher and Student distributions are “sewn” into the comparison unit. The calculated values of the Fisher and Student criteria are compared with the critical values that are selected from the Fisher and Student tables depending on the specified reliability and the number of degrees of freedom for each of the criteria. The values of the calculated Fisher and Student criteria and the corresponding tabular values are displayed on display 8. If all four Fisher criteria and four student criteria are less than the critical values of these criteria, there is no reason to reject the hypothesis that all damages sustained by the car body belong to one set.

Thus, the proposed method for detecting fraud, based on testing the hypothesis that all damage to the car body belongs to one general population, and the device for implementing the method allow you to quickly or reliably prove or disprove the hypothesis immediately and reliably at the location of the tested car. The method and device for its implementation can be effectively used by insurance companies to identify fraud or its absence in the analysis of the circumstances of the declared insurance cases.

Information sources

1. Algazin A.I., Larichev V.D. Insurance fraud. Identification methodology and methods of counteraction. - Book 1. Auto Insurance: Method. allowance. - M.: Publishing. the house "Regulation", 2008. - 185 p.

2. www.antiobman.ru

3. RF patent for the invention No. 2276404 "Method for determining the size of defects in vehicles in road traffic accidents." / Melnikov K.A. Publ. 2006.05.10 IPC G06F 19/00; B62O 41/00.

4. RF patent No. 1245071 “Method for determining the corrosion rate of a metal by the thickness of an oxide film”. / Seredkin S.V., Samsonov B.V., Gremyachkin V.A.

5. Kuznetsov N. P., Tarasova A. A., Yurtikov R. A. About one way to detect fraud in the staging of an accident. / Bulletin of IzhSTU, 2010, No. 3 (47). - S. 39-42.

6. RWG Hunt. The Reproduction of Color (Gthed) - UK, Chichester: Wiley, 2004. IS&T Series in Imaging Science and Technology. - ISBN 0-470-02425-9.

7. Hild A. Mathematical statistics with technical applications. - M .: IL, 1956.

8. N. Johnson, F. Lyon. Statistics and experimental design in engineering and science. Data processing methods. / Translation from English. / M.: Mir Publishing House, 1980 - 610 p.

Claims (2)

1. A method for detecting fraud in the staging of an accident by determining that all claimed car damage belongs to the same insured event, which consists in using a digital apparatus to photograph the damage to the car body, and the resulting photographic images or fragments thereof are processed using a color model RGB for determining the intensity of color shades of red, green and blue, as well as the sum of these intensities characterizing corrosion excellence location at the places of damage to the body, after which the methods of mathematical statistics determine the sample average and sample variances for the samples of each color and the total intensity, after which the belonging of all the damage to the car is determined by the same population, and upon confirmation of which there is no reason to reject the hypothesis of the absence of fraud the lack of a staging of an accident. 2. A device for detecting fraud in the stage of an accident by determining whether all the claimed damages belong to the same insured event, consisting of a camera and a processor, characterized in that the camera is digital and the camera is connected to the processor through a USB connector, and the processor itself consists of a unit processing photographic images using the RGB color model to determine the intensity of color shades in the three primary colors of photographic images of damage to the body, memory unit, where samples are formed for each of the color shades, logic blocks, where, in accordance with the algorithms of mathematical statistics, the hypothesis is checked on the belonging of the characteristics of the images of the damaged areas of the car body to one general population, as well as a display with a keyboard to enter the necessary numerical values and display the results of the hypothesis test.

Images