Search

CN-122023244-A - Intelligent point cloud security detection method and device for multi-scene facilities and equipment

CN122023244ACN 122023244 ACN122023244 ACN 122023244ACN-122023244-A

Abstract

The intelligent point cloud safety detection method and device for the multi-scene facilities and equipment comprise the steps of roughly aligning two-period point clouds by adopting coarse granularity registration based on characteristic points, accurately aligning the two-period point clouds by adopting an ICP fine granularity algorithm based on principal component analysis, and calculating a global state change map of the two-period point clouds by adopting a local difference analysis algorithm based on point cloud triangularization. The method has good universality, can be widely applied to various industrial and infrastructure safety monitoring scenes such as thickness detection of special railway equipment, steel ladle corrosion evaluation, tunnel lining thickness analysis and the like, and remarkably improves the efficiency and the result reliability of detection work.

Inventors

  • LIU YUNJIE
  • ZHAO JUAN
  • ZHANG ZHIPENG
  • SHAO XUEJUN
  • PANG QING
  • GUO YOUWEI
  • WU HUIJIE
  • LI ZHIXUAN
  • WU HAOTIAN

Assignees

  • 中国铁道科学研究院集团有限公司标准计量研究所
  • 中国铁道科学研究院集团有限公司

Dates

Publication Date
20260512
Application Date
20251222

Claims (5)

  1. 1. An intelligent point cloud security detection method for multi-scene facilities and equipment is characterized by comprising the following steps: S1, roughly aligning two-period point clouds by adopting coarse granularity registration based on characteristic points; s2, accurately aligning two-period point clouds by adopting an ICP fine granularity algorithm based on principal component analysis; S3, calculating a global state change map of the two-period point cloud through a local difference analysis algorithm based on the point cloud triangulation.
  2. 2. The method according to claim 1, wherein the step S1 specifically comprises: (1) Dividing two-period point cloud into source point cloud And target point cloud For point clouds And Any one of (3) And Calculating covariance matrix in k neighbor And : Wherein, the Is a point cloud Any point in (3) K nearest neighbor (K) th The three-dimensional coordinates of the plurality of coordinates, Centroid coordinates for the adjacent point; Is a point cloud Any point in (3) K nearest neighbor (K) th The three-dimensional coordinates of the plurality of coordinates, K is the number of neighbors; (2) Calculation of And As local feature points: Wherein, the Is that Is used to determine the three characteristic values of (a), Is that Is defined by the three characteristic values of (a); (3) Calculation of And Is the mean square error of curvature: Wherein, the In k neighborhood Is defined as the local feature point mean value of (a), In k neighborhood Is a local feature point mean value; (4) For a pair of And Sorting from big to small, before selecting Each is taken as a typical characteristic point to construct a typical characteristic point set And ; (5) Calculation of Centroid of (2) And Centroid of (2) : (6) Constructing centroid-based transformation matrices : Based on a transformation matrix Will be And Preliminary alignment is carried out to realize source point cloud Cloud to target point Coarse grain registration of (2), and setting point clouds after coarse grain registration as respectively And 。
  3. 3. The method according to claim 2, wherein the step S2 specifically comprises: For the source point cloud And target point cloud Coordinate sets of typical feature points respectively constructed And Constructing a principal component matrix: Obtaining feature vectors And : Calculating source point cloud after coarse granularity registration Coordinate set of corresponding typical feature points Cloud to target point Coordinate set of corresponding typical feature points Is a transform matrix of (a) : Using the transformation matrix, for the source point cloud Is transformed by all points in (a): Wherein, the Representation of Any point in (3) Transformed coordinates; Solving a rotation matrix using a singular value decomposition algorithm Translation matrix Updating Coordinates: Defining registration errors : Wherein, the Is the target point cloud after coarse registration Any point in the above; through multiple iterations until Outputting the rotation matrix at the moment Translation vector Acting on source target point cloud after coarse registration Obtaining the fine-grained registered point cloud And 。
  4. 4. The method according to claim 2, wherein the step S3 specifically includes: target point cloud after fine granularity registration Constructing a triangulated model to form triangular patches Wherein For a coordinate representation of an arbitrary triangular patch, Then fine-grained post-registration source point cloud Any point in (3) To triangular dough piece The distance of (2) is: Wherein, the 、 And Is triangular dough piece Three coordinates of (a); Selecting Minimum value as Cloud relative to target point The current local state change value traverses the source point cloud in sequence All points in the map are obtained For characterizing states in a two-cycle point cloud.
  5. 5. A multi-scenario facility and equipment oriented intelligent point cloud security detection apparatus applying the method of any of claims 1-4, comprising: The coarse registration module is used for roughly aligning the two-period point clouds based on coarse granularity registration of the feature points; The fine registration module is used for accurately aligning the two-period point cloud based on an ICP fine granularity algorithm of principal component analysis; the state transformation calculation module is used for calculating a global state change map of the two-period point cloud through a local difference analysis algorithm based on the point cloud triangulation.

Description

Intelligent point cloud security detection method and device for multi-scene facilities and equipment Technical Field The invention relates to the technical field of facility and equipment safety detection, in particular to an intelligent point cloud safety detection method and device for multi-scene facilities and equipment. Background Currently, the detection of the state of industrial equipment and the safety monitoring of infrastructure construction still rely on manual inspection. For the detection objects with higher precision requirements, single-point measuring equipment such as a displacement sensor, a speed sensor and the like, or traditional measuring instruments such as a thickness gauge, a level gauge, a total station and the like are generally adopted to carry out local exploration at a selected position. The traditional manual detection method is highly dependent on experience of operators, is easily interfered by subjective factors, and is difficult to ensure detection precision and consistency, and the detection means adopting a sensor or a traditional measuring instrument can improve the precision of local point positions, but basically deduces the whole safety state through limited detection points, so that short plates with incomplete detection coverage exist, and the undetected areas still have hidden safety risks. With the development and popularization of the three-dimensional laser scanning technology, the non-contact three-dimensional visual detection method based on the point cloud data provides a new way for comprehensively evaluating the safety state of facilities. By acquiring high-density three-dimensional point cloud data of the surface of the measured object and combining an efficient and intelligent point cloud detection algorithm, the overall analysis of the overall state of the measured object can be realized. However, most of the current intelligent point cloud detection algorithms are only suitable for fixed detected objects in a single scene, and have insufficient generalization and universality, and particularly, the three-dimensional detection method based on deep learning needs to reconstruct and label a data set after replacing the detected objects, so that the application learning cost across scenes and targets is extremely high. Disclosure of Invention Aiming at the problems, the invention provides an intelligent point cloud safety detection method and device for multi-scene facilities and equipment, which aim to improve the adaptability of a point cloud algorithm to different detection objects and different environmental conditions, can be effectively applied to various industrial and infrastructure safety monitoring scenes such as thickness detection of special railway equipment, steel ladle corrosion state evaluation, tunnel lining thickness analysis and the like, and provide key technical support for improving the safety detection efficiency and reliability of the facilities and equipment. The intelligent point cloud security detection method for multi-scene facilities and equipment provided by the present disclosure mainly comprises the following steps: S1, roughly aligning two-period point clouds by adopting coarse granularity registration based on characteristic points; s2, accurately aligning two-period point clouds by adopting an ICP fine granularity algorithm based on principal component analysis; S3, calculating a global state change map of the two-period point cloud through a local difference analysis algorithm based on the point cloud triangulation. Further, the step S1 specifically includes: (1) Dividing two-period point cloud into source point cloud And target point cloudFor point cloudsAndAny one of (3)AndCalculating covariance matrix in k neighborAnd: Wherein, the Is a point cloudAny point in (3)K nearest neighbor (K) thThe three-dimensional coordinates of the plurality of coordinates,Centroid coordinates for the adjacent point; Is a point cloud Any point in (3)K nearest neighbor (K) thThe three-dimensional coordinates of the plurality of coordinates,K is the number of neighbors; (2) Calculation of AndAs local feature points: Wherein, the Is thatIs used to determine the three characteristic values of (a),Is thatIs defined by the three characteristic values of (a); (3) Calculation of AndIs the mean square error of curvature: Wherein, the In k neighborhoodIs defined as the local feature point mean value of (a),In k neighborhoodIs a local feature point mean value; (4) For a pair of AndSorting from big to small, before selectingEach is taken as a typical characteristic point to construct a typical characteristic point setAnd; (5) Calculation ofCentroid of (2)AndCentroid of (2): (6) Constructing centroid-based transformation matrices: Based on a transformation matrixWill beAndPreliminary alignment is carried out to realize source point cloudCloud to target pointCoarse grain registration of (2), and setting point clouds after coarse grain registration