Search

CN-122023028-A - Intelligent detection and evaluation method and system for damage of accident vehicle

CN122023028ACN 122023028 ACN122023028 ACN 122023028ACN-122023028-A

Abstract

The invention belongs to the technical field of image recognition, and particularly relates to an intelligent detection and evaluation method and system for damage of an accident vehicle, wherein the method comprises the steps of constructing a normalized deformation energy diagram of a local damage image of the accident vehicle, and initializing a particle swarm to form a ring topology structure; and (3) carrying out iterative search by adopting a particle swarm optimization algorithm, calculating inertia weight according to a distortion energy value at the current position of the particle in each iteration, updating a speed vector and a position of the particle according to the inertia weight and the distortion energy value at the position of the particle in a topological neighbor set of the particle, connecting the particles according to a topological sequence when the iteration is stopped to obtain a damage closed polygonal contour, further calculating a physical area and a depth index of a damage region, and generating a damage assessment report. The invention realizes automatic repair and evaluation of damaged edges of accident vehicles, eliminates subjective errors of manual damage assessment, and improves objectivity and accuracy of vehicle risk damage assessment claims.

Inventors

  • HUANG ZHEN
  • LIU CONG

Assignees

  • 深圳市全保通网络科技有限公司

Dates

Publication Date
20260512
Application Date
20260203

Claims (10)

  1. 1. The intelligent detection and evaluation method for the damage of the accident vehicle is characterized by comprising the following steps of: The method comprises the steps of constructing a Heisen matrix of each pixel point in a local damage gray level image of an accident vehicle, determining distortion energy values of the pixel points according to two characteristic values of the Heisen matrix to form a normalized distortion energy map, identifying the pixel point with the largest distortion energy value in the normalized distortion energy map as a seed point, and uniformly sampling on the circumference with a preset radius to generate an initial population containing a plurality of particles; the method comprises the steps of carrying out iterative search by adopting a particle swarm optimization algorithm based on an initial population, dynamically adjusting inertia weights according to distortion energy values at positions of particles and local neighborhood average values thereof in each iteration of the particle swarm optimization algorithm, updating speed vectors of each particle according to the inertia weights of each particle and the distortion energy values at the positions of the particles in a topological neighbor set, and updating the positions of each particle according to the speed vectors; According to the number of pixels enclosed in the outline of the damage closed polygon, the physical area of the damage is obtained by combining with camera calibration parameters, the arithmetic average value of distortion energy values of all pixel points in the outline of the damage closed polygon is counted and used as a depth index, and an assessment report is generated according to the physical area and the depth index.
  2. 2. The method for intelligently detecting and evaluating damage to an accident vehicle according to claim 1, wherein the method for acquiring two eigenvalues of the hessian matrix is as follows: solving the characteristic equation Obtaining two characteristic values And In which, in the process, Representing a hessian matrix; representing the characteristic value; Represents a second order identity matrix and det (·) represents a determinant operation of the matrix.
  3. 3. The method for intelligently detecting and evaluating damage to an accident vehicle according to claim 1, wherein the determining the distortion energy value of the pixel point comprises: Based on two eigenvalues of the hessian matrix, determining the original curvature intensity of the pixel point: In which, in the process, Representing pixel points Original curvature intensity at the point; 、 and carrying out maximum and minimum value normalization processing on the original curvature intensity of the pixel point to obtain a distortion energy value of the pixel point.
  4. 4. The method for intelligently detecting and evaluating damage to an accident vehicle according to claim 1, wherein the step of uniformly sampling the predetermined radius on the circumference to generate an initial population including a plurality of particles comprises the steps of: evenly sampling on circumference with preset radius to generate the product containing An initial population of particles, an initial velocity of each particle set to a zero vector, an initial position of each particle satisfying the expression: In which, in the process, Is the first Initial positions of individual particles; as the abscissa of the seed point, Is the ordinate of the seed point; is the first The polar angle of the individual particles is such that, , Indicating the number of particles; representing a preset radius; Establishing topological neighborhood relation among particles, and enabling the first step to be performed according to the order of polar angles from small to large The topological neighbor set of individual particles contains the first And (d) The particles form a closed ring topology.
  5. 5. The method for intelligently detecting and evaluating damage to an accident vehicle according to claim 1, wherein the inertial weight satisfies the expression: ; In the formula, Represents the number of iterations sequence number, Represent the first The particles are at the first Inertial weights in the secondary iterations; Representing a preset maximum inertia weight; representing a hyperbolic tangent function; representing a rheology-sensitive factor; Represent the first The particles are at the first The position of the iteration; Represent the first The particles are at the first A distortion energy value at a location at which the iteration is performed; Expressed in terms of Mean distortion energy values within a central local neighborhood window; a preset numerical stability constant is indicated for preventing the denominator from being zero.
  6. 6. The method for intelligently detecting and evaluating damage to an accident vehicle according to claim 1, wherein the updating the velocity vector of each particle comprises: ; In the formula, Represents the number of iterations sequence number, Represent the first The particles are at the first The updated velocity vector in the iteration; Represent the first The particles are at the first Inertial weights in the secondary iterations; Represent the first The particles are at the first A velocity vector at the time of the iteration; Represent the first The particles are at the first The position of the iteration; Represent the first A topological neighbor set of individual particles; Represent the first Sequence numbers of particles in the topological neighbor set of the individual particles; Represent the first The particles are at the first The position of the iteration; Representing a base stiffness coefficient; Represent the first The particles are at the first Distortion energy values at the locations where the iterations were taken.
  7. 7. The method for intelligently detecting and evaluating damage to an accident vehicle according to claim 1 or 6, wherein updating the position of each particle according to the velocity vector comprises: ; In the formula, Represents the number of iterations sequence number, Represent the first The particles are at the first The updated position coordinates in the iteration; Represent the first The particles are at the first The updated velocity vector in the iteration; Represent the first The particles are at the first The position at which the iteration is performed.
  8. 8. The method for intelligently detecting and evaluating damage to an accident vehicle according to claim 1, further comprising: and stopping iteration in response to reaching a preset maximum iteration number or the average displacement of particles in continuous preset iteration times is smaller than a preset convergence threshold.
  9. 9. The method for intelligently detecting and evaluating damage to an accident vehicle according to claim 1, wherein the generating the damage-assessment report according to the physical area and the depth index comprises: Generating an assessment report by integrating the physical area and the depth index, recommending a fitting replacement scheme in response to the depth index being larger than a preset stretching threshold, recommending a sheet metal paint spraying scheme in response to the physical area being larger than the preset sheet metal threshold and the depth index not being larger than the preset stretching threshold, and recommending a traceless repair scheme otherwise.
  10. 10. An intelligent detection and assessment system for damage to an accident vehicle, comprising a processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement an intelligent detection and assessment method for damage to an accident vehicle according to any one of claims 1 to 9.

Description

Intelligent detection and evaluation method and system for damage of accident vehicle Technical Field The present invention relates to the field of image recognition. More particularly, the invention relates to an intelligent detection and evaluation method and system for damage of an accident vehicle. Background In the current motor vehicle insurance claim settlement business, facing to the mass accident vehicle damage assessment demands, the vehicle appearance damage assessment is carried out by utilizing the computer vision technology to assist or replace manual work, so that the intelligent damage assessment system aims at automatically identifying the damage types such as scraping, sinking and the like and quantifying the damage degree by analyzing the vehicle images acquired at the accident scene, thereby realizing rapid claim settlement, reducing claim settlement leakage and improving customer satisfaction. The current damage identification technology mainly relies on traditional digital image processing methods, which are generally based on the gray gradient or texture characteristics of images, and a general edge detection operator or threshold segmentation algorithm is adopted to extract damaged areas of the vehicle body surface, wherein the basic logic is that damaged parts and normal vehicle paint surfaces are considered to have obvious differences in visual characteristics, the boundary range of damage is determined by capturing the differences, and the damaged areas are calculated according to the differences so as to match corresponding maintenance schemes. However, in practical applications, the quality of damaged images of vehicles is extremely easily disturbed by external environmental factors, and as the surface of the vehicle body is generally covered with a varnish layer with high reflectivity, the varnish layer has optical characteristics similar to a convex mirror, and has a strong reflection effect on ambient light and sceneries, under the outdoor complex investigation environment, the specular reflection effect can cause a large number of highlight spots or shadows to appear in the images, so that the real damaged edges are submerged or broken, and in this case, physically continuous concave or deformed areas often show intermittent and discrete characteristics in the images. The existing traditional image processing algorithm lacks robustness to the complex optical interference, and when the traditional image processing algorithm faces discontinuous damaged edges, effective automatic repair and connection cannot be carried out, and the identified damaged outline is incomplete, position offset and even complete omission are easy to cause, so that huge errors occur in physical area calculation of an assessment result directly, and objective and accurate assessment conclusion cannot be output. Disclosure of Invention In order to solve the technical problems that the damage edge is broken due to illumination and paint reflection interference in the prior art, and the damage assessment is inaccurate due to the fact that a traditional algorithm cannot be effectively repaired, the invention provides the scheme in the following aspects. In a first aspect, the present invention provides a method for intelligently detecting and evaluating damage to an accident vehicle, including: The method comprises the steps of constructing a Heisen matrix of each pixel point in a local damage gray level image of an accident vehicle, determining distortion energy values of the pixel points according to two characteristic values of the Heisen matrix to form a normalized distortion energy map, identifying the pixel point with the largest distortion energy value in the normalized distortion energy map as a seed point, and uniformly sampling on the circumference with a preset radius to generate an initial population containing a plurality of particles; the method comprises the steps of carrying out iterative search by adopting a particle swarm optimization algorithm based on an initial population, dynamically adjusting inertia weights according to distortion energy values at positions of particles and local neighborhood average values thereof in each iteration of the particle swarm optimization algorithm, updating speed vectors of each particle according to the inertia weights of each particle and the distortion energy values at the positions of the particles in a topological neighbor set, and updating the positions of each particle according to the speed vectors; According to the number of pixels enclosed in the outline of the damage closed polygon, the physical area of the damage is obtained by combining with camera calibration parameters, the arithmetic average value of distortion energy values of all pixel points in the outline of the damage closed polygon is counted and used as a depth index, and damage evaluation is carried out according to the physical area and the depth index. Preferably, the acquiring