Claim data anti-counterfeiting detection method, device, equipment and medium

Abstract

The application relates to the technical field of image processing, and discloses a claim settlement data anti-fake detection method, device, equipment and medium, belongs to the technical field of image processing, is applied to a financial technology scene, and comprises the following steps: performing shooting posture recognition on a shooting device to obtain a real shooting posture; performing edge line segment detection and vanishing point recognition on an original image obtained by the shooting device to perform shooting posture inference according to the recognized object vanishing point position, and obtain an inferred shooting posture; and performing anomaly detection on the original image according to the inferred shooting posture and the real shooting posture to obtain an image category obtained by whether the shooting device truly shoots a damaged object. In the application, the real shooting posture at the shooting moment of the shooting device is obtained, and the inferred shooting posture is inversely deduced according to the object vanishing point position detected from the original image, so that the authenticity of the claim settlement image is detected according to the inferred shooting posture and the real shooting posture, and the accuracy of the claim settlement anti-fraud detection is improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, applicable to financial technology scenarios, and particularly to a method, apparatus, equipment, and medium for anti-counterfeiting detection of claims data. Background Technology

[0002] Traditional anti-counterfeiting methods for claims data typically use object detection models (such as the YOLO model) to identify the insured object category and damage extent in claims images, combining this with policy information to verify the claims data. For example, after an accident involving an insured vehicle, the user uploads images of the accident scene through a claims app. The object detection model identifies the location and damaged areas of the insured vehicle in the accident scene image, and optical character recognition technology extracts insured vehicle data such as license plate number and vehicle registration information. This data is then compared with the insured vehicle information in the policy database. If the information matches and the accident scene image detects damage features that meet expectations, the claims data passes verification. However, this method only focuses on the specific content captured in the claims image, lacking verification of how the image was captured. This allows fraudsters to display photos of past accidents on a high-definition monitor and then take a picture of the same image directly in front of the monitor with their mobile phone. Because the copied image is highly similar to the original image at the pixel level, current technology struggles to effectively distinguish whether the image is copied, leading to instances of malicious insurance fraud using copied old images during the claims process. Therefore, improving the accuracy of claims image authenticity detection to enhance the accuracy of claims fraud detection has become an urgent problem to be solved. Summary of the Invention

[0003] This invention provides a method, apparatus, computer equipment, and medium for anti-counterfeiting detection of claims data, in order to solve the technical problem of how to improve the accuracy of claims image authenticity detection and thus improve the accuracy of claims fraud detection.

[0004] Firstly, a method for detecting counterfeit claims data is provided, including: In response to a claim evidence collection request, the system acquires the original image captured by the target shooting device and performs shooting posture recognition on the target shooting device to obtain real shooting posture data; wherein, the original image includes an image of the damaged object, and the damaged object is the insured object; Edge line segment detection is performed on the original image to obtain object edge line segment data; Based on the object edge line segment data, the vanishing point of the damaged object in the original image is identified to obtain the object vanishing point location; Based on the location of the object's vanishing point, the shooting posture of the target shooting device is inferred to obtain inferred shooting posture data; Based on the inferred shooting posture data and the actual shooting posture data, anomaly detection is performed on the original image to obtain an image anomaly category; wherein, the image anomaly category includes a first image category and a second image category, the first image category indicating that the original image is a real image obtained by the target shooting device from the real damaged object, and the second image category indicating that the original image is a fake image obtained by the target shooting device from the damaged object.

[0005] Secondly, a claims data anti-counterfeiting detection device is provided, including: The shooting device posture recognition module is used to respond to a claim evidence collection request, acquire the original image captured by the target shooting device, and perform shooting posture recognition on the target shooting device to obtain real shooting posture data; wherein, the original image includes an image of a damaged object, and the damaged object is the insured object; The image edge line segment detection module is used to perform edge line segment detection on the original image to obtain object edge line segment data; The image vanishing point recognition module is used to identify the vanishing point of the damaged object in the original image based on the object edge line segment data, and obtain the vanishing point position of the object. The device attitude inference module is used to infer the shooting attitude of the target shooting device based on the vanishing point position of the object, and obtain inferred shooting attitude data. The image anti-spoofing detection module is used to perform anomaly detection on the original image based on the inferred shooting posture data and the actual shooting posture data to obtain an image anomaly category; wherein, the image anomaly category includes a first image category and a second image category, the first image category indicating that the original image is a real image obtained by the target shooting device from the real damaged object, and the second image category indicating that the original image is a fake image obtained by the target shooting device from the damaged object.

[0006] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned anti-counterfeiting detection method for claims data.

[0007] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned anti-counterfeiting detection method for claims data.

[0008] In the aforementioned scheme implemented by the anti-counterfeiting detection method, device, computer equipment, and storage medium for claims data, the original image captured by the target shooting device can be obtained in response to a claims evidence collection request. The shooting posture of the target shooting device is then recognized to obtain true shooting posture data. The original image includes an image of a damaged object, which is the insured object. Edge line segment detection is performed on the original image to obtain object edge line segment data. Based on the object edge line segment data, vanishing point identification is performed on the damaged object in the original image to obtain the object vanishing point position. Based on the object vanishing point position, the shooting posture of the target shooting device is inferred to obtain inferred shooting posture data. Anomaly detection is performed on the original image based on the inferred shooting posture data and the true shooting posture data to obtain either a first image category indicating that the original image is a genuine image captured by the target shooting device of a real damaged object, or a second image category indicating that the original image is a forged image captured by the target shooting device of a damaged object. In this invention, by simultaneously collecting the actual shooting posture data of the shooting device at the moment of taking a picture and the edge line segment data and vanishing point position of the damaged object in the original image, the inferred shooting posture of the shooting device is deduced based on the vanishing point position of the object. Then, the inferred shooting posture is cross-validated with the actual shooting posture, thereby realizing the authenticity detection of the original image. This effectively avoids the insurance fraud loophole caused by the inability of traditional methods to distinguish between real shooting and screen copying, significantly improving the accuracy of the authenticity detection of claims images, and thus improving the accuracy of anti-fraud detection in claims. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a method for detecting counterfeit claims data in one embodiment of the present invention; Figure 2 yes Figure 1 A schematic diagram of a specific implementation of step S101; Figure 3 yes Figure 1 A schematic diagram of a specific implementation method for step S102; Figure 4 yes Figure 1 A schematic diagram of a specific implementation method for step S103; Figure 5 yes Figure 1 A schematic diagram of a specific implementation of step S104; Figure 6 yes Figure 1 A schematic diagram of a specific implementation of step S105; Figure 7 yes Figure 6 A schematic diagram of a specific implementation method for step S602; Figure 8 This is a schematic diagram of a claims data anti-counterfeiting detection device in one embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 10 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0011] This application provides a method, apparatus, device, and medium for anti-counterfeiting detection of claims data, aiming to solve the technical problem of how to improve the accuracy of claims image authenticity detection, thereby improving the accuracy of claims fraud detection.

[0012] The claims data anti-counterfeiting detection method, apparatus, equipment and medium provided in the embodiments of this application are specifically described through the following embodiments. First, the claims data anti-counterfeiting detection method in the embodiments of this application is described.

[0013] The anti-counterfeiting detection method for claims data provided in this application relates to the field of image processing technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the policy claims data filtering method, but is not limited to the above forms.

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] like Figure 1 As shown, Figure 1A flowchart illustrating the anti-counterfeiting detection method for claims data provided in this embodiment of the invention may include, but is not limited to, steps S101 to S105: Step S101: In response to the claim evidence collection request, the original image captured by the target shooting device is obtained, and the shooting posture of the target shooting device is recognized to obtain the real shooting posture data; wherein, the original image includes the image of the damaged object, and the damaged object is the insured object.

[0016] Step S102: Perform edge line segment detection on the original image to obtain object edge line segment data.

[0017] Step S103: Based on the object edge line segment data, identify the vanishing point of the damaged object in the original image to obtain the object vanishing point location.

[0018] Step S104: Based on the vanishing point of the object, infer the shooting posture of the target shooting device to obtain inferred shooting posture data.

[0019] Step S105: Perform anomaly detection on the original image based on the inferred shooting posture data and the actual shooting posture data to obtain the image anomaly category; wherein, the image anomaly category includes a first image category and a second image category, the first image category indicates that the original image is a real image obtained by the target shooting device from a real damaged object, and the second image category indicates that the original image is a fake image obtained by the target shooting device from a damaged object.

[0020] Steps S101 to S105 as shown in the embodiments of this application involve, in response to a claim evidence collection request, acquiring the original image captured by the target shooting device, and performing shooting posture recognition on the target shooting device to obtain real shooting posture data. The original image includes an image of a damaged object, which is the insured object. Edge segment detection is performed on the original image to obtain object edge segment data. Vanishing point identification is performed on the damaged object in the original image based on the object edge segment data to obtain the object vanishing point position. Shooting posture inference is performed on the target shooting device based on the object vanishing point position to obtain inferred shooting posture data. Anomaly detection is performed on the original image based on the inferred shooting posture data and the real shooting posture data to obtain a first image category representing that the original image is a real image captured by the target shooting device of a real damaged object, or a second image category representing a forged image captured by the target shooting device of a damaged object. In this invention, by simultaneously collecting the actual shooting posture data of the shooting device at the moment of taking a picture and the edge line segment data and vanishing point position of the damaged object in the original image, the inferred shooting posture of the shooting device is deduced based on the vanishing point position of the object. Then, the inferred shooting posture is cross-validated with the actual shooting posture, thereby realizing the authenticity detection of the original image. This effectively avoids the insurance fraud loophole caused by the inability of traditional methods to distinguish between real shooting and screen copying, significantly improving the accuracy of the authenticity detection of claims images, and thus improving the accuracy of anti-fraud detection in claims.

[0021] In step S101 of some embodiments, specifically, in response to a claim evidence collection request, an original image captured by the target shooting device is obtained, wherein the original image includes an image of the damaged object, and the damaged object is the insured object.

[0022] Specifically, requests for evidence collection for claims are usually initiated by users (i.e., policyholders or claim applicants) in insurance applications (such as car insurance apps).

[0023] Specifically, the target shooting device refers to a portable smart terminal device used by users to collect images as evidence for claims. It is usually a smartphone or tablet. The device has a built-in inertial measurement unit, which includes microelectromechanical system sensors such as a three-axis accelerometer and a three-axis gyroscope.

[0024] Specifically, the original image refers to a digital image taken by the target shooting device that has not undergone post-editing. The original image includes images of damaged objects, which are the insured objects, i.e., the insured objects that have suffered physical damage in the insured event.

[0025] For example, in a car insurance claim scenario, after a user's vehicle is involved in an accident, they need to take and upload photos of the damaged parts of the vehicle through a car insurance app. This photo is the original image, and the damaged bumper, door, and the entire vehicle in the photo are all damaged objects.

[0026] like Figure 2 As shown, in some embodiments, step S101 may include, but is not limited to, steps S201 to S203: Step S201: Obtain the three-dimensional components of the device acceleration of the target imaging device.

[0027] Step S202: Normalize the three-dimensional components of the device acceleration to obtain the three-dimensional components of unit acceleration; wherein, the three-dimensional components of unit acceleration include a first acceleration component along the left-right direction of the target imaging device, a second acceleration component along the up-down direction of the target imaging device, and a third acceleration component along the vertical direction of the target imaging device.

[0028] Step S203: Calculate the attitude angle of the target shooting device based on the first acceleration component, the second acceleration component, the third acceleration component, and the preset gravitational acceleration to obtain the actual shooting attitude data.

[0029] In step S201 of some embodiments, specifically, the three-dimensional component of device acceleration refers to the three-axis raw acceleration of the triaxial accelerometer sensor built into the target imaging device in the three-dimensional spatial coordinate system. Specifically, it includes a first raw acceleration component along the left-right direction of the device (usually defined as the X-axis), a second raw acceleration component along the up-down direction of the device (defined as the Y-axis), and a third raw acceleration component along the direction perpendicular to the device (defined as the Z-axis). The three-axis raw acceleration is usually quantified in units of multiples of gravitational acceleration g.

[0030] Furthermore, during the acquisition of the three-dimensional components of the device's acceleration, this application also simultaneously acquires the three-axis raw angular velocities collected by the gyroscope built into the target shooting device. This is used to assist in judging the dynamic state of the device. If all three-axis raw angular velocities are lower than a preset threshold (e.g., 5 degrees per second), the target shooting device is determined to be in a quasi-stationary state, and the accelerometer data can be directly used for attitude calculation. If all three-axis raw angular velocities are higher than or equal to the preset threshold (e.g., 5 degrees per second), a dynamic interference warning for the target shooting device is marked, and the filtering needs to be strengthened in subsequent processing or the user needs to be prompted to reshoot.

[0031] In step S202 of some embodiments, specifically, the three-dimensional component of unit acceleration refers to the normalized three-dimensional component of device acceleration, which includes a first acceleration component along the left-right direction of the target shooting device, a second acceleration component along the up-down direction of the target shooting device, and a third acceleration component along the vertical direction of the target shooting device. The sum of the squares of the three components is equal to 1, which together constitutes a unit vector describing the direction of gravitational acceleration in the local coordinate system of the target shooting device.

[0032] Specifically, firstly, the square root of the sum of the squares of the first, second, and third original acceleration components is calculated, and then the square root value is multiplied by the gravitational acceleration g to obtain the original vector magnitude. Then, the first, second, and third original acceleration components are divided by this original vector magnitude to obtain the normalized first, second, and third acceleration components.

[0033] For example, if the three-dimensional components of the device's acceleration are [0, 0.5g, 0.866g], then the square root of the sum of the squares of 0, 0.5g, and 0.866g is calculated, and the square root value is multiplied by the gravitational acceleration g to obtain an original vector magnitude of 1.0g. The first, second, and third original acceleration components are then divided by this original vector magnitude to obtain the unit acceleration three-dimensional components [0, 0.5, 0.866]. The first acceleration component is 0, indicating that there is no gravitational projection in the left and right directions of the device; the second acceleration component is 0.5, indicating that the device bears 50% of the gravitational projection in the up and down directions; and the third acceleration component is 0.866, indicating that the device bears approximately 86.6% of the gravitational projection in the direction perpendicular to the screen. Moreover, the unit acceleration three-dimensional components only describe the direction of gravity and are no longer affected by the specific acceleration value or hardware differences.

[0034] In step S203 of some embodiments, specifically, the real shooting posture data refers to the set of angle parameters describing the spatial posture of the target shooting device. The real shooting posture data includes the real pitch angle and the real roll angle of the target shooting device relative to the horizontal plane. The real pitch angle is used to characterize the angle of rotation of the target shooting device around the X-axis, reflecting the degree of head tilting or head tilting when the device is shooting. The real roll angle is used to characterize the angle of rotation of the target shooting device around the Y-axis, reflecting the degree of left and right tilting when the device is shooting.

[0035] Specifically, the preset gravitational acceleration refers to the standard gravitational acceleration constant g, which serves as a reference during attitude calculation.

[0036] First, establish a sensor coordinate system with the center of gravity of the target imaging device as the origin. The X-axis is along the left-right direction, the Y-axis is along the front-back direction, and the Z-axis is perpendicular to the screen direction. Calculate the square root of the sum of the squares of the first and third acceleration components. Then, calculate the ratio of the second acceleration component to the square root value. The roll angle is obtained by calculating the arctangent function (arctan). Next, calculate the square root of the sum of the squares of the second and third acceleration components, and the ratio of the first acceleration component to the square root value. The pitch angle is then calculated by calculating the arctangent function (arctan).

[0037] For example, in a car insurance claim scenario, if the three-dimensional components of the unit acceleration are [0, 0.5, 0.866], substituting them into the arctangent function arctan, we get a roll angle of 0°, indicating that the phone is not tilted when shooting. We also get a pitch angle of 30°, indicating that the phone is tilted 30° relative to the horizontal plane when shooting.

[0038] Steps S201 to S203 reflect the actual spatial orientation of the shooting device relative to the horizontal plane at the moment the shutter is triggered. This orientation is independent of the image content itself and constitutes the physical benchmark for subsequent consistency verification. Compared with traditional anti-counterfeiting methods that rely solely on image content analysis, this method can effectively distinguish the essential differences in device posture between real shooting scenes and screen copying scenes, significantly improving the reliability and accuracy of claim image authenticity detection and providing solid data support for insurance anti-fraud.

[0039] like Figure 3 As shown, in some embodiments, step S102 includes, but is not limited to, steps S301 to S304: Step S301: Perform Gaussian blur processing on the original image to obtain a blurred original image.

[0040] Step S302: Perform line segment detection on the damaged object in the blurred original image to obtain object line segment data; wherein, the object line segment data includes line segment direction data and line segment length data.

[0041] Step S303: Calculate the slope of the object's straight line segments based on the line segment direction data and line segment length data to obtain the slope of the straight line segments.

[0042] Step S304: Filter the object's line segment data according to the slope of the line segment and the preset line segment filtering conditions to obtain the object's edge line segment data.

[0043] In step S301 of some embodiments, specifically, the blurred original image refers to the image obtained after Gaussian blurring, which has a reduced noise level but retains the edge structure.

[0044] Specifically, by setting the size of the Gaussian kernel (e.g., 5x5 pixels) and the standard deviation parameter (e.g., 1.5), the Gaussian kernel is slid across the original image pixel by pixel. A weighted average is calculated for each pixel and its neighborhood pixel values ​​to generate new pixel values ​​that are filled into the corresponding positions in the output image, resulting in a blurred original image. The standard deviation parameter is used to determine the degree of blurring; a larger standard deviation parameter will produce a smoother effect.

[0045] For example, in a car insurance claim scenario, if the original image contains a side scratch of a silver sedan, Gaussian blurring will soften the fine scratch texture and metallic reflections on the paint surface, while preserving the clarity of large-scale geometric edges such as door handle lines and waistlines.

[0046] In step S302 of some embodiments, specifically, the object line segment data refers to a set of line segments that describe the geometric contour of the damaged object in the blurred original image. The object line segment data includes line segment direction data (i.e., the orientation angle of the line segment on the image plane) and line segment length data (i.e., the pixel length of the line segment).

[0047] Specifically, the gradient magnitude and direction of each pixel in the blurred original image can be calculated using the LSD (Line Segment Detector) algorithm. Starting from the local maximum pixel of the gradient magnitude of the blurred original image pixel, adjacent pixels with similar gradient directions are connected along the gradient direction to form a line segment support region. Each line segment support region is approximated by a rectangle to determine the main direction of the rectangle as the line segment direction data, and the length of the rectangle as the object line segment data of the line segment length data.

[0048] For example, in a car insurance claim scenario, when a blurry original image shows a dent in the left front fender of a black SUV, the LSD algorithm can successfully extract the straight line data of objects such as the seam line between the fender and the hood (approximately horizontal, about 300 pixels long), the seam line between the fender and the A-pillar (approximately vertical, about 200 pixels long), and the straight line segment of the lower edge of the front bumper (approximately 45 degrees, 250 pixels long).

[0049] In step S303 of some embodiments, specifically, the slope of a straight line segment refers to the slope value of each straight line segment in the object's straight line segment data, that is, the ratio of the difference between the ordinate and the abscissa of the end point and the starting point of the straight line segment.

[0050] For example, in a car insurance claim scenario, if the straight line segment data of an object includes the front edge line segment of the vehicle's hood (starting point (400, 300), ending point (600, 400)), with a vertical coordinate difference of 100, a horizontal coordinate difference of 200, and a slope of 0.5; the upper edge line segment of the left front headlight (starting point (350, 450), ending point (430, 440)), with a slope of -0.125; and the spoke line segment of the left front wheel hub (starting point (300, 600), ending point (340, 520)), with a slope of -0.2.

[0051] In step S304 of some embodiments, specifically, the line segment filtering condition refers to the set of judgment rules used to determine whether a line segment belongs to the valid edge of a damaged object, mainly based on the slope and length of the line segment for filtering.

[0052] For example, line segment filtering criteria could include retaining line segments that are less than a line segment length threshold (e.g., 50 pixels, used to filter out overly short, noise-related lines) and less than a slope absolute value threshold (e.g., 5).

[0053] Specifically, object edge line segment data refers to the set of straight line segments that conform to the geometric characteristics of damaged objects after being filtered by screening conditions. That is, the object edge line segment data excludes excessively short noise interference lines and background line segments with abnormal slopes, and only retains straight line segments with clear geometric meaning and perspective convergence potential.

[0054] For example, in a car insurance claim scenario, the length of the front edge line segment of the vehicle's hood is 200 pixels and the slope is 0.5, which meets the line segment filtering criteria and is retained; the length of the upper edge line segment of the left front headlight is 80 pixels and the slope is -0.125, which meets the criteria and is retained; the length of the spoke line segment of the left front wheel hub is 60 pixels and the slope is -2.0, which meets the criteria and is retained.

[0055] Through steps S301 to S304, the main edge of the damaged object, the road surface, and the background can be effectively distinguished, so as to filter out excessively short stray line segments and interfering line segments with abnormal slopes, and retain high-quality edge line segments that conform to the geometric characteristics of the damaged object. This provides reliable data support for subsequent vanishing point identification and pose inference, significantly improves the accuracy of geometric feature extraction of damaged objects in claims images, and thus improves the reliability of claims anti-fraud detection.

[0056] like Figure 4 As shown, in some embodiments, step S103 includes, but is not limited to, steps S401 to S404: Step S401: Determine the edge segments of the damaged object in the original image based on the object edge segment data.

[0057] Step S402: Perform Gaussian spherical projection on the edge line segment to obtain the spherical projection coordinates of the line segment.

[0058] Step S403: Detect the spherical intersection of the edge line segments based on the spherical projection coordinates of the line segments to obtain the spherical intersection of the line segments.

[0059] Step S404: Perform image coordinate transformation on the intersection point of the line segment and the sphere to obtain the vanishing point position of the object.

[0060] In step S401 of some embodiments, specifically, the edge segment refers to the set of damaged object body structure segments represented by the object edge segment data.

[0061] For example, in a car insurance claim scenario, if the insured vehicle is involved in a side collision, the edge line segment dataset of the vehicle may include the waistline in the horizontal direction of the door, the door seam line in the vertical direction of the door, the horizontal frame of the upper and lower edges of the window, and the radial short line segments formed by the collision dent area, etc.

[0062] In step S402 of some embodiments, specifically, the spherical projection coordinates of a line segment refer to the set of line segment coordinates in the spherical coordinate system after the edge line segment is projected onto a Gaussian sphere.

[0063] Specifically, since parallel lines appear to never intersect in a planar image, but when mapped onto a sphere, these parallel lines converge into poles on the sphere, the location of the vanishing point can be transformed into the problem of finding the intersection points on the sphere.

[0064] Specifically, a virtual Gaussian spherical coordinate system can be established with the optical center of the target imaging device as the origin and the unit length as the radius. The difference between the horizontal coordinate of each pixel corresponding to the edge line segment and the horizontal coordinate of the original image center is calculated. This difference is then divided by the focal length of the imaging device's lens to obtain the horizontal component of the Gaussian spherical coordinate system. The difference between the vertical coordinate of each pixel corresponding to the edge line segment and the vertical coordinate of the original image center is also calculated. This difference is then divided by the focal length of the lens to obtain the vertical component of the Gaussian spherical coordinate system. The third directional component of the Gaussian spherical coordinate system is fixed at the unit length. The horizontal, vertical, and third directional components of the pixel are then used as the spherical projection coordinates of the line segment. The entire edge line segment appears as a great circle on the Gaussian sphere.

[0065] In step S403 of some embodiments, specifically, the intersection point of the line segment on the sphere refers to the coordinates of the intersection point on the Gaussian sphere.

[0066] Specifically, the RANSAC algorithm can be used to randomly select two edge segments based on the spherical projection coordinates of the line segments. The intersection of the corresponding great circle arcs on the sphere is determined as the direction of the candidate vanishing point. The remaining edge segments are then traversed, and the spherical distance between the great circle arc of each edge segment and the candidate intersection point is calculated. If the distance is less than a preset threshold (e.g., 3 degrees angular distance), the edge segment is marked as an interior point that supports the candidate intersection point. The above steps are repeated, such as for 500 iterations or until a sufficiently stable solution is found. The candidate intersection point with the most interior points is then determined as the optimal spherical intersection point of the line segment.

[0067] For example, in a car insurance claims scenario, if the line segment spherical projection coordinate dataset includes the hood edge group (great arcs corresponding to 2 parallel lines), the window frame group (great arcs corresponding to 2 parallel lines), the wheel hub spoke group (great arcs corresponding to 2 spoke lines), and the great arc corresponding to 1 background interference line. The RANSAC algorithm performs an iterative process. In the first iteration, it randomly selects the large arcs of the hood edge and the window frame, calculates the intersection point P1, and obtains two line segments with consensus (significant directional differences, low consensus). In the second iteration, it randomly selects the large arcs of the hood edge and the hood edge (parallel lines in the same group), calculates the intersection point P2, and obtains four line segments with consensus (two from the hood group and two from the window group). In the third iteration, it randomly selects the large arcs of the wheel spokes and the background interference lines, calculates the intersection point P3, and obtains only one line segment with consensus, which is judged as an incorrect intersection point and discarded. After a sufficient number of iterations (e.g., 500), it identifies P2 as having the highest consensus support (four line segments), and takes intersection point P2 as the spherical intersection point of the line segments. This point is the strongest intersection point locked on the sphere, corresponding to the main vanishing direction of the vehicle's main structure.

[0068] In step S404 of some embodiments, specifically, the vanishing point position of the object refers to the coordinate position of remapping the three-dimensional intersection direction vector on the Gaussian sphere back to the two-dimensional image plane, which is used to reflect the perspective convergence point position of the damaged object on the image plane.

[0069] Specifically, the horizontal coordinates of the intersection point of the line segments on the sphere can be divided by the vertical coordinates to obtain a horizontal ratio. This ratio is then multiplied by the lens focal length and added to the horizontal coordinates of the original image center to obtain the horizontal pixel position of the intersection point on the image plane. Similarly, the vertical coordinates of the intersection point can be divided by the vertical coordinates to obtain a vertical ratio. This ratio is then multiplied by the lens focal length and added to the vertical coordinates of the original image center to obtain the vertical pixel position of the intersection point on the image plane. The horizontal and vertical pixel positions of the intersection point are then used to determine the vanishing point of the object.

[0070] For example, in a car insurance claim scenario, the direction vector of the intersection point of a line segment on a Gaussian sphere has a horizontal component of 0, a vertical component of -0.5, and a vertical component of approximately 0.866. This indicates that the vanishing point is tilted downwards by about 30 degrees relative to the optical axis of the shooting device. If the focal length of the shooting device is 800 pixels, and the original image center coordinates are located at 640 pixels horizontally and 360 pixels vertically, then the horizontal coordinate of the vanishing point can be calculated to be 640 pixels, and the vertical coordinate to be approximately -102 pixels. This means that the vanishing point is located approximately 102 pixels above the image plane, which matches the perspective rule of the parallel lines of the vehicle's front bumper extending upwards to infinity.

[0071] Steps S401 to S404 effectively overcome the difficulty that parallel lines in a planar image never intersect. By using Gaussian spherical projection and spherical intersection detection, the perspective structure of damaged objects is identified, significantly improving the accuracy of vanishing point detection. This helps to improve the accuracy and reliability of subsequent anti-fraud detection in claims processing.

[0072] like Figure 5 As shown, in some embodiments, step S104 includes, but is not limited to, steps S501 to S503: Step S501: Obtain the device intrinsic parameter data of the target shooting device; wherein, the device intrinsic parameter data includes focal length and principal point coordinates.

[0073] Step S502: Vanishing point offset detection is performed based on the principal point coordinates and the vanishing point position of the object to obtain vanishing point offset data; wherein, the vanishing point offset data is used to characterize the degree of offset of the vanishing point position of the object relative to the center of the damaged object in the original image.

[0074] Step S503: Based on the vanishing point offset data and focal length, the attitude angle of the target shooting device is inferred to obtain the inferred shooting attitude data.

[0075] In step S501 of some embodiments, specifically, the device intrinsic data refers to the calibration parameters that describe the internal geometry of the optical imaging of the target imaging device. The device intrinsic data includes focal length and principal point coordinates. The focal length is usually expressed in pixels and is used to reflect the optical magnification characteristics of the imaging device lens. The principal point coordinates are used to characterize the position of the intersection of the optical axis of the imaging device and the image plane in the pixel coordinate system, and are usually close to the coordinates of the image center.

[0076] For example, in a car insurance claim scenario, if the claimant uses a mobile phone to photograph the accident vehicle, different models of mobile phones have different camera focal lengths and principal point coordinates. For example, the main camera of the mobile phone may have a focal length of about 800 pixels and principal point coordinates of (640, 360), which corresponds to the center of an image with a resolution of 1280×720.

[0077] In step S502 of some embodiments, specifically, the vanishing point offset data is used to characterize the degree of offset of the vanishing point position of the object relative to the center of the damaged object in the original image.

[0078] Specifically, the vanishing point offset data includes horizontal offset components and vertical offset components.

[0079] For example, if the vanishing point of the object is (640, 200) and the principal point coordinates are (640, 360), the horizontal offset component is calculated as 640-640=0 and the vertical offset component is 200-360=-160 pixels. This vanishing point offset data indicates that the vanishing point of the object is located 160 pixels above the center of the image. It is used to reflect the degree of downward view of the shooting equipment relative to the accident vehicle. The larger the offset (absolute value), the larger the pitch angle, and the more the shooting equipment is tilted downwards.

[0080] In step S503 of some embodiments, specifically, the inferred shooting attitude data refers to the spatial attitude estimation data of the target shooting device obtained solely based on the geometric features of the image, which includes the inferred pitch angle.

[0081] Specifically, by calculating the ratio of the vertical offset component to the focal length, the arctangent function is applied to the ratio to obtain the inferred pitch angle. The direction of the pitch angle is determined based on the sign of the vertical offset component. If the vertical offset component is negative, it indicates that the vanishing point of the object is above the principal point, and the shooting device is in a downward-looking posture. If the vertical offset component is positive, it indicates that the vanishing point of the object is below the principal point, and the shooting device is in a downward-looking posture.

[0082] Through steps S501 to S503, a mathematical inverse deduction from image geometric features (i.e., vanishing point position) to spatial pose parameters is realized, establishing an image verification dimension independent of device sensor measurements. This facilitates subsequent cross-comparison with actual shooting pose data, effectively identifying contradictions between the physical pose of the shooting device and image geometric perspective, significantly improving the reliability and accuracy of claims image authenticity detection, and providing solid data support based on perspective geometry principles for claims anti-fraud.

[0083] like Figure 6 As shown, in some embodiments, step S105 includes, but is not limited to, steps S601 to S604: Step S601: Perform posture difference detection based on the inferred shooting posture data and the actual shooting posture data to obtain device posture difference data.

[0084] Step S602: Based on the device posture difference data and preset image anomaly conditions, perform image anomaly recognition on the original image to obtain the image anomaly category.

[0085] Step S603: If the device posture difference data does not meet the image anomaly conditions, then the image anomaly category is determined to be the first image category.

[0086] Step S604: If the device posture difference data meets the image anomaly conditions, then the image anomaly category is determined to be the second image category.

[0087] In step S601 of some embodiments, specifically, the device attitude difference data refers to a value used to quantify the degree of inconsistency between the inferred shooting attitude and the actual shooting attitude. It is usually expressed as the absolute value of the difference between the actual pitch angle and the inferred pitch angle, and is used to reflect the degree of logical contradiction between the sensor measurement value of the shooting device and the image geometric inverse value.

[0088] like Figure 7 As shown, in some embodiments, the image anomaly condition includes a pose difference threshold, and step S602 includes, but is not limited to, steps S701 to S704: Step S701: Obtain the accelerometer drift data of the target shooting device, and determine the device sensing noise error based on the accelerometer drift data.

[0089] Step S702: Obtain the three-dimensional parametric model of the damaged object, and calculate the perspective deviation of the damaged object in the original image based on the three-dimensional parametric model to obtain the image perspective error. Step S703: Obtain the lens distortion parameters of the target shooting device, and determine the lens distortion error of the device based on the lens distortion parameters.

[0090] Step S704: Based on the integration of device sensing noise error, image perspective error and device lens distortion error, the image abnormality conditions are obtained.

[0091] In step S701 of some embodiments, specifically, accelerometer drift data refers to the fluctuation data of the accelerometer output value measured under static conditions as a function of time, which reflects the static deviation and random walk error of the accelerometer caused by factors such as temperature changes, mechanical stress, and electrical noise.

[0092] Specifically, device sensing noise error refers to the measurement uncertainty introduced by the inherent characteristics of accelerometer hardware.

[0093] Specifically, accelerometer output data is continuously collected while the shooting device is stationary. The sampling frequency is set to the same frequency as when the image is captured. The collection time is usually 10 to 30 seconds to obtain accelerometer drift data. The mean of the accelerometer drift data is calculated as the static deviation, and the standard deviation is calculated as random noise. The device sensing noise error is synthesized based on the static deviation and random noise.

[0094] For example, in a car insurance claim scenario, if the standard deviation of the accelerometer Z-axis output of the target mobile phone is 0.015g during a static test, the arctangent function can be used to calculate the standard deviation, resulting in a device sensing noise error of 0.86 degrees. Considering the safety factor, this can be rounded to 2 to 3 degrees.

[0095] In step S702 of some embodiments, specifically, the three-dimensional parametric model refers to a mathematical model that describes the geometric structure of a damaged object, usually represented in the form of a rigid cuboid.

[0096] Specifically, image perspective error refers to the uncertainty of the vanishing point position of an object caused by the deviation between the actual geometry of the damaged object and the three-dimensional parametric model.

[0097] For example, in car insurance claims, since the actual damaged vehicle is not a perfect rigid cuboid, the edge extraction may have an image perspective error of 3 to 5 degrees.

[0098] In step S703 of some embodiments, specifically, the lens distortion parameter refers to the calibration parameter that describes the geometric distortion of the optical imaging of the photograph, including the radial distortion coefficient.

[0099] Specifically, lens distortion error refers to the uncertainty in the position of the vanishing point of an object caused by lens distortion of the shooting equipment.

[0100] Specifically, the pixel displacement at different image positions can be calculated based on the lens distortion parameters. Based on the lens distortion parameters and the vanishing point position of the object, the maximum offset of the vanishing point position caused by lens distortion can be calculated and converted into the device lens distortion error.

[0101] For example, in a car insurance claim scenario, if the radial distortion coefficient of the wide-angle lens of the target mobile phone is -0.3, the pixel offset caused by radial distortion at the image edge is calculated to be about 50 pixels. Combined with a focal length of 800 pixels, the angle error corresponding to this offset is about arctan(50 / 800), or 3.6 degrees. Considering the overall lens distortion error of the device, it can be taken as 5 degrees.

[0102] In step S704 of some embodiments, specifically, the attitude difference threshold refers to the attitude difference value of the combined device sensing noise error, image perspective error and device lens distortion error, used to determine whether the original image is a real image or a fake image obtained by the target shooting device from the damaged object.

[0103] Specifically, the square root of the sum of the squares of the device sensing noise error, image perspective error, and device lens distortion error can be calculated, and the square root of the result can be multiplied by a preset error correction value to obtain the attitude difference threshold.

[0104] For example, in a car insurance claims scenario, if the device's sensing noise error is 3 degrees, the image perspective error is 8 degrees, and the device's lens distortion error is degrees, the sum of the squares and the square root of all errors can be calculated to be 9.9 degrees. To ensure that the posture difference threshold can cover 95% of the real captured image samples (such as a posture difference threshold of 10 to 15 degrees), the preset error correction value can be 1.2, that is, the posture difference threshold is rounded up to 12 degrees.

[0105] Through steps S701 to S704, the attitude difference threshold is a confidence interval composed of the superposition of sensor noise, perspective approximation error, and lens distortion error. This avoids both the false rejection of claims due to misjudgment of real images caused by an excessively low threshold and the false fraud caused by missed judgment of forged images due to an excessively high threshold. This significantly improves the accuracy of authenticity detection of claims images and the accuracy of anti-fraud detection in claims.

[0106] In step S603 of some embodiments, specifically, the image anomaly category includes a first image category, which represents that the original image is a real image obtained by the target imaging device from a real damaged object.

[0107] For example, in a car insurance claim scenario, if the claimant takes a picture of the front collision damage of the vehicle at the accident scene, and the difference between the calculated actual pitch angle and the inferred pitch angle is 3 degrees, which is less than the attitude difference threshold of 12 degrees, does this mean that the device attitude difference data does not meet the image anomaly conditions? If the image anomaly category of the original image is determined to be a real image, the degree of vehicle damage can be judged and the compensation amount can be calculated based on the real image.

[0108] In step S604 of some embodiments, specifically, the image anomaly category includes a second image category, which indicates that the original image is a forged image obtained by the target imaging device from the damaged object.

[0109] For example, in a car insurance claim scenario, if a claim applicant takes a picture of the front collision damage of a vehicle at the accident scene, and the difference between the calculated actual pitch angle and the inferred pitch angle is 18 degrees, which is greater than the attitude difference threshold of 15 degrees, then the image anomaly category of the original image is determined to be a forged image, immediately triggering a claim fraud alarm. The claim application is automatically blocked and transferred to the manual investigation process, effectively preventing malicious insurance fraud using historical images.

[0110] Through steps S601 to S604, reliable detection of screen copy fraud is achieved by comparing the consistency of the physical and geometric aspects of the shooting device. This effectively avoids the insurance fraud loopholes caused by the inability of traditional methods to distinguish between real shooting and screen copy, significantly improving the accuracy of the authenticity detection of claims images, thereby enhancing the accuracy of anti-fraud detection in claims.

[0111] In one optional embodiment of this application, the inferred device attitude data may further include inferring the roll angle. If the original image is taken at an angle (i.e. the target shooting device is tilted left or right), the horizon in the original image will no longer be horizontal. Two vanishing points in two directions are required, and the line connecting the two vanishing points is the horizon in the original image.

[0112] Specifically, the inferred roll angle of the target imaging device can be obtained by obtaining the coordinates of the first and second vanishing points in the vanishing point location of the object, connecting the first and second vanishing point coordinates to form a line, calculating the angle between the line and the horizontal direction of the original image, and combining the inferred pitch angle and the inferred roll angle into the inferred attitude data of the device.

[0113] Furthermore, the anomaly detection of the original image based on the inferred shooting posture data and the actual shooting posture data also includes the following steps: calculating the roll residual value between the actual roll angle in the actual shooting posture data and the inferred roll angle in the inferred shooting posture data; comparing the roll residual value with the roll angle error threshold; if the roll residual value is less than the roll angle error threshold, the image anomaly category is determined as the first image category; if the roll residual value is greater than or equal to the roll angle error threshold, the image anomaly category is determined as the second image category; and combining the image anomaly categories corresponding to the roll angle and the image anomaly categories corresponding to the pitch angle, the final image anomaly category is obtained.

[0114] As can be seen, in the above scheme, in response to the claim evidence collection request, the original image captured by the target shooting device is obtained, and the shooting posture of the target shooting device is identified to obtain the real shooting posture data. The original image includes the image of the damaged object, which is the insured object. Edge line segment detection is performed on the original image to obtain object edge line segment data. Based on the object edge line segment data, the vanishing point of the damaged object in the original image is identified to obtain the object vanishing point position. Based on the object vanishing point position, the shooting posture of the target shooting device is inferred to obtain inferred shooting posture data. Based on the inferred shooting posture data and the real shooting posture data, anomaly detection is performed on the original image to obtain a first image category indicating that the original image is a real image captured by the target shooting device of a real damaged object, or a second image category indicating that the original image is a forged image captured by the target shooting device of a damaged object. In this invention, by simultaneously collecting the actual shooting posture data of the shooting device at the moment of taking a picture and the edge line segment data and vanishing point position of the damaged object in the original image, the inferred shooting posture of the shooting device is deduced based on the vanishing point position of the object. Then, the inferred shooting posture is cross-validated with the actual shooting posture, thereby realizing the authenticity detection of the original image. This effectively avoids the insurance fraud loophole caused by the inability of traditional methods to distinguish between real shooting and screen copying, significantly improving the accuracy of the authenticity detection of claims images, and thus improving the accuracy of anti-fraud detection in claims.

[0115] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0116] Please see Figure 8 This application also provides a claims data anti-counterfeiting detection device, which can implement the above-mentioned claims data anti-counterfeiting detection method. The device includes: The shooting device posture recognition module is used to respond to the claim evidence collection request, acquire the original image captured by the target shooting device, and perform shooting posture recognition on the target shooting device to obtain the real shooting posture data; wherein, the original image includes the image of the damaged object, and the damaged object is the insured object; The image edge line segment detection module is used to detect edge lines in the original image and obtain object edge line segment data; The image vanishing point recognition module is used to identify the vanishing point of damaged objects in the original image based on the object's edge line segment data, and to obtain the object's vanishing point location. The device attitude inference module is used to infer the shooting attitude of the target shooting device based on the vanishing point of the object, and obtain the inferred shooting attitude data. The image anti-spoofing detection module is used to perform anomaly detection on the original image based on the inferred shooting posture data and the actual shooting posture data to obtain the image anomaly category; wherein, the image anomaly category includes a first image category and a second image category, the first image category indicates that the original image is a real image obtained by the target shooting device from a real damaged object, and the second image category indicates that the original image is a forged image obtained by the target shooting device from a damaged object.

[0117] In one embodiment, the posture recognition module of the shooting device is specifically used for: Acquire the three-dimensional components of the device acceleration of the target imaging device; The three-dimensional components of the device acceleration are normalized to obtain the three-dimensional components of unit acceleration; wherein, the three-dimensional components of unit acceleration include a first acceleration component along the left-right direction of the target imaging device, a second acceleration component along the up-down direction of the target imaging device, and a third acceleration component along the vertical direction of the target imaging device. The attitude angle of the target shooting device is calculated based on the first acceleration component, the second acceleration component, the third acceleration component, and the preset gravitational acceleration to obtain the actual shooting attitude data.

[0118] In one embodiment, the image edge line segment detection module is specifically used for: Apply Gaussian blur to the original image to obtain a blurred original image; Line segment detection is performed on damaged objects in the blurred original image to obtain object line segment data; the object line segment data includes line segment direction data and line segment length data; The slope of the straight line segment is calculated based on the line segment direction data and line segment length data. The line segment data of the object is filtered based on the slope of the line segment and the preset line segment filtering conditions to obtain the line segment data of the object's edge.

[0119] In one embodiment, the image vanishing point recognition module is specifically used for: Determine the edge segments of the damaged object in the original image based on the object edge segment data; Gaussian spherical projection is performed on the edge line segments to obtain the spherical projection coordinates of the line segments; Based on the spherical projection coordinates of the line segments, the spherical intersection points of the edge line segments are detected to obtain the spherical intersection points of the line segments. By performing image coordinate transformation on the intersection point of the line segment and the sphere, the vanishing point of the object can be obtained.

[0120] In one embodiment, the device attitude inference module is specifically used for: Obtain the intrinsic parameters of the target imaging device; these parameters include focal length and principal point coordinates. Vanishing point offset detection is performed based on the principal point coordinates and the vanishing point position of the object to obtain vanishing point offset data; the vanishing point offset data is used to characterize the degree of offset of the vanishing point position of the object relative to the center of the damaged object in the original image; Based on the vanishing point offset data and focal length, the attitude angle of the target shooting device is inferred to obtain the inferred shooting attitude data.

[0121] In one embodiment, the image anti-counterfeiting detection module is specifically used for: The device posture difference data is obtained by detecting the posture difference between the inferred shooting posture data and the actual shooting posture data. Based on the device posture difference data and preset image anomaly conditions, image anomaly identification is performed on the original image to obtain the image anomaly category; If the device posture difference data does not meet the image anomaly conditions, then the image anomaly category is determined to be the first image category; If the device posture difference data meets the image anomaly conditions, then the image anomaly category is determined to be the second image category.

[0122] In one embodiment, the image anomaly condition in the image anti-counterfeiting detection module includes a pose difference threshold, and is further used for: Acquire accelerometer drift data of the target imaging device, and determine the device sensing noise error based on the accelerometer drift data; A three-dimensional parametric model of the damaged object is obtained, and the perspective deviation of the damaged object in the original image is calculated based on the three-dimensional parametric model to obtain the image perspective error. Obtain the lens distortion parameters of the target shooting device, and determine the lens distortion error of the device based on the lens distortion parameters; The attitude difference threshold is obtained by integrating the device sensing noise error, image perspective error and device lens distortion error.

[0123] Specific limitations regarding the claims data anti-counterfeiting detection device can be found in the limitations of the claims data anti-counterfeiting detection method described above, and will not be repeated here. Each module in the aforementioned claims data anti-counterfeiting detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0124] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a claims data anti-counterfeiting detection method on the server side.

[0125] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a claims data anti-counterfeiting detection method on the client side.

[0126] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: In response to a claim evidence collection request, the system obtains the original images captured by the target camera and performs shooting posture recognition on the target camera to obtain the actual shooting posture data; the original images include images of the damaged object, which is the insured object. Edge line segment detection is performed on the original image to obtain object edge line segment data; Based on the object's edge line segment data, the vanishing point of the damaged object in the original image is identified to obtain the object's vanishing point location. Based on the vanishing point of the object, the shooting posture of the target shooting device is inferred to obtain inferred shooting posture data; Anomaly detection is performed on the original image based on the inferred shooting posture data and the actual shooting posture data to obtain the image anomaly category. The image anomaly category includes a first image category and a second image category. The first image category indicates that the original image is a real image obtained by the target shooting device from a real damaged object, and the second image category indicates that the original image is a fake image obtained by the target shooting device from a damaged object.

[0127] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: In response to a claim evidence collection request, the system obtains the original images captured by the target camera and performs shooting posture recognition on the target camera to obtain the actual shooting posture data; the original images include images of the damaged object, which is the insured object. Edge line segment detection is performed on the original image to obtain object edge line segment data; Based on the object's edge line segment data, the vanishing point of the damaged object in the original image is identified to obtain the object's vanishing point location. Based on the vanishing point of the object, the shooting posture of the target shooting device is inferred to obtain inferred shooting posture data; Anomaly detection is performed on the original image based on the inferred shooting posture data and the actual shooting posture data to obtain the image anomaly category. The image anomaly category includes a first image category and a second image category. The first image category indicates that the original image is a real image obtained by the target shooting device from a real damaged object, and the second image category indicates that the original image is a fake image obtained by the target shooting device from a damaged object.

[0128] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0129] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0130] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0131] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0132] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for detecting counterfeit claims data, characterized in that, include: In response to a claim evidence collection request, the system acquires the original image captured by the target shooting device and performs shooting posture recognition on the target shooting device to obtain real shooting posture data; wherein, the original image includes an image of the damaged object, and the damaged object is the insured object; Edge line segment detection is performed on the original image to obtain object edge line segment data; Based on the object edge line segment data, the vanishing point of the damaged object in the original image is identified to obtain the object vanishing point location; Based on the location of the object's vanishing point, the shooting posture of the target shooting device is inferred to obtain inferred shooting posture data; Based on the inferred shooting posture data and the actual shooting posture data, anomaly detection is performed on the original image to obtain an image anomaly category; wherein, the image anomaly category includes a first image category and a second image category, the first image category indicating that the original image is a real image obtained by the target shooting device from the real damaged object, and the second image category indicating that the original image is a fake image obtained by the target shooting device from the damaged object.

2. The method as described in claim 1, characterized in that, The step of performing anomaly detection on the original image based on the inferred shooting posture data and the actual shooting posture data to obtain image anomaly categories includes: Based on the inferred shooting posture data and the actual shooting posture data, posture difference detection is performed to obtain device posture difference data; Based on the device posture difference data and preset image anomaly conditions, the original image is subjected to image anomaly identification to obtain the image anomaly category; If the device posture difference data does not meet the image anomaly conditions, then the image anomaly category is determined to be the first image category; If the device posture difference data meets the image anomaly conditions, then the image anomaly category is determined to be the second image category.

3. The method as described in claim 2, characterized in that, The image anomaly condition includes a pose difference threshold, which is determined through the following steps: Acquire accelerometer drift data of the target imaging device, and determine the device sensing noise error based on the accelerometer drift data; A three-dimensional parametric model of the damaged object is obtained, and the perspective deviation of the damaged object in the original image is calculated based on the three-dimensional parametric model to obtain the image perspective error. Obtain the lens distortion parameters of the target imaging device, and determine the lens distortion error of the device based on the lens distortion parameters; The attitude difference threshold is obtained by integrating the device sensing noise error, the image perspective error, and the device lens distortion error.

4. The method as described in claim 1, characterized in that, The step of performing edge line segment detection on the original image to obtain object edge line segment data includes: The original image is subjected to Gaussian blurring to obtain a blurred original image; The damaged object in the blurred original image is subjected to line segment detection to obtain object line segment data; wherein, the object line segment data includes line segment direction data and line segment length data; The slope of the straight line segment is calculated based on the line segment direction data and the line segment length data. The object's line segment data is filtered based on the slope of the line segment and preset line segment filtering conditions to obtain the object's edge line segment data.

5. The method as described in claim 1, characterized in that, The step of identifying the vanishing point of the damaged object in the original image based on the object's edge segment data to obtain the object's vanishing point location includes: The edge segments of the damaged object in the original image are determined based on the object edge segment data. Gaussian spherical projection is performed on the edge line segment to obtain the spherical projection coordinates of the line segment; Based on the spherical projection coordinates of the line segment, the spherical intersection point of the edge line segment is detected to obtain the spherical intersection point of the line segment; The vanishing point of the object is obtained by transforming the image coordinates of the intersection point of the line segments on the sphere.

6. The method as described in claim 1, characterized in that, The step of inferring the shooting posture of the target shooting device based on the vanishing point position of the object to obtain inferred shooting posture data includes: Obtain the device intrinsic parameter data of the target imaging device; wherein, the device intrinsic parameter data includes focal length and principal point coordinates; Vanishing point offset detection is performed based on the principal point coordinates and the vanishing point position of the object to obtain vanishing point offset data; wherein, the vanishing point offset data is used to characterize the degree of offset of the vanishing point position of the object relative to the center of the damaged object in the original image; Based on the vanishing point offset data and the focal length, the attitude angle of the target shooting device is inferred to obtain the inferred shooting attitude data.

7. The method according to any one of claims 1 to 6, characterized in that, The step of recognizing the shooting posture of the target shooting device to obtain real shooting posture data includes: Obtain the three-dimensional components of the device acceleration of the target imaging device; The three-dimensional acceleration components of the device are normalized to obtain a unit acceleration three-dimensional component; wherein, the unit acceleration three-dimensional component includes a first acceleration component along the left-right direction of the target imaging device, a second acceleration component along the up-down direction of the target imaging device, and a third acceleration component along the vertical direction of the target imaging device. The target shooting device's attitude angle is calculated based on the first acceleration component, the second acceleration component, the third acceleration component, and the preset gravitational acceleration to obtain the actual shooting attitude data.

8. A claims data anti-counterfeiting detection device, characterized in that, include: The shooting device posture recognition module is used to respond to a claim evidence collection request, acquire the original image captured by the target shooting device, and perform shooting posture recognition on the target shooting device to obtain real shooting posture data; wherein, the original image includes an image of a damaged object, and the damaged object is the insured object; The image edge line segment detection module is used to perform edge line segment detection on the original image to obtain object edge line segment data; The image vanishing point recognition module is used to identify the vanishing point of the damaged object in the original image based on the object edge line segment data, and obtain the vanishing point position of the object. The device attitude inference module is used to infer the shooting attitude of the target shooting device based on the vanishing point position of the object, and obtain inferred shooting attitude data. The image anti-spoofing detection module is used to perform anomaly detection on the original image based on the inferred shooting posture data and the actual shooting posture data to obtain an image anomaly category; wherein, the image anomaly category includes a first image category and a second image category, the first image category indicating that the original image is a real image obtained by the target shooting device from the real damaged object, and the second image category indicating that the original image is a fake image obtained by the target shooting device from the damaged object.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the claims data anti-counterfeiting detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the claims data anti-counterfeiting detection method as described in any one of claims 1 to 7.

Images