Occluded target localization method based on salient primitives and topological relation invariance

Abstract

The application provides a kind of based on saliency element and topological relation invariability occlusion target positioning method, it is related to machine vision and target positioning technical field.The method comprises the following steps: step S1, the original three-dimensional point cloud data is preprocessed, and the three-dimensional point cloud to be measured is obtained;Step S2, a plurality of basic geometric primitives of the same type in the three-dimensional point cloud data to be measured are detected based on multiple RANSAC algorithms;Step S3, for the plurality of basic geometric primitives of the same type, the saliency calculation method is used to determine the saliency maximum geometric primitive;Step S4, according to the spatial topological relation invariability between the saliency maximum geometric primitive and the occlusion target, the occlusion target is positioned.The application is simple and efficient, and misidentification is not easy;The accurate positioning of the occlusion target can be realized, and the positioning speed is faster.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of machine vision and target positioning, and particularly relates to a method for positioning a blocked target based on saliency primitives and topological relation invariance. BACKGROUND

[0002] With the development of computer hardware and machine vision algorithms, computer vision technology is increasingly used for target recognition and defect detection. For some complex structure detection objects in industrial scenes, such as turbine engine blade wear detection, rail fastener wear detection, and automobile engine assembly defect detection, the measured objects are blocked, which cannot accurately position the detected target, resulting in defect mis-detection, and thus accurate positioning of the blocked target to be detected is required.

[0003] In order to solve the problem of accurate positioning of a blocked target, Sumiyoshi S proposes a method for estimating the three-dimensional point cloud of a blocked part of a target according to the obtained three-dimensional point cloud, using stable matching likelihood calculation to realize auxiliary point cloud filling of a large area of blocked parts, restore the information of the blocked part of the target, and determine the position of the target to be detected; Dewei Zou proposes a template matching target positioning method based on LINEMOD, which further subdivides a single template used for matching into several equal small templates, thereby reducing the influence of blocking on target positioning.

[0004] However, the accuracy of the filled auxiliary point cloud affects target positioning, and if the estimated approximate point cloud is not accurate, it will directly lead to target positioning error; if the number of template subdivisions is small, it cannot match the blocked target, and if the number of template subdivisions is too large, it will increase and waste the calculation amount. SUMMARY

[0005] In view of the above problems, the present application provides a method for positioning a blocked target based on saliency primitives and topological relation invariance.

[0006] The method for positioning a blocked target based on saliency primitives and topological relation invariance provided by the present application comprises the following steps:

[0007] Step S1, preprocessing the original three-dimensional point cloud data to obtain three-dimensional point cloud to be measured;

[0008] Step S2, detecting a plurality of basic geometric primitives of the same type in the three-dimensional point cloud data to be measured based on a plurality of RANSAC algorithms;

[0009] Step S3, determining a saliency maximum geometric primitive by using a saliency calculation method for the plurality of basic geometric primitives of the same type;

[0010] Step S4, positioning the blocked target according to the spatial topological relation invariance between the saliency maximum geometric primitive and the blocked target.

[0011] According to an embodiment of the present application, in step S1, the original three-dimensional point cloud data is preprocessed, including:

[0012] The three-dimensional point cloud data is uniformly down-sampled, and the down-sampled three-dimensional point cloud data is de-noised to obtain the to-be-tested three-dimensional point cloud data.

[0013] According to an embodiment of the present application, in step S2, the types of the basic geometric primitives include a plane, a cylinder and a cuboid.

[0014] According to an embodiment of the present application, when the type of the basic geometric primitive is a plane, the step S2 specifically includes:

[0015] Step S21, three non-collinear points are randomly selected from the to-be-tested three-dimensional point cloud data, and a plane P1 is fitted;

[0016] Step S22, for any selected point in the to-be-tested three-dimensional point cloud data, the distance between the selected point and the plane P1 is calculated, and the distance is compared with a pre-set threshold value, if the distance is less than the threshold value, the selected point is determined as an in-point of the plane P1, otherwise, the selected point is determined as an out-point of the plane P1;

[0017] Step S23, for the out-point, three non-collinear points are re-selected to fit a plane P2, the above step S22 is repeated, if the in-points of the plane P2 are more than the in-points of the plane P1, the in-points of the plane P1 are released, the plane P2 is determined as the plane geometric primitive of the iteration result, and the plane P2 is renamed as the plane P1;

[0018] Step S24, the above step S23 is repeated until a pre-set iteration termination number is reached, and the fitted plane is determined as the plane geometric primitive detected by the RANSAC algorithm;

[0019] Step S25, the above steps S21-S24 are repeated to obtain a plurality of plane geometric primitives detected by the RANSAC algorithm.

[0020] According to an embodiment of the present application, the step S3 specifically includes:

[0021] Step S31, the volume of each geometric primitive is calculated, and the volume saliency component of each geometric primitive is calculated according to the volume;

[0022] Step S32, the length, width and height of each geometric primitive are determined, the numerical values of the length, width and height are sorted in descending order, and the spatial size saliency component of each geometric primitive is calculated;

[0023] Step S33, each geometric primitive is divided into a plurality of volume units, and the voxel ratio significance component of each geometric primitive is calculated;

[0024] Step S34, the volume significance component, the space size significance component and the voxel ratio significance component of each geometric primitive are combined, and the significance of each geometric primitive is calculated by weighting;

[0025] Step S35, the significance of each geometric primitive is compared, and the geometric primitive with the largest value is determined as the most significant geometric primitive.

[0026] According to the embodiment of the present application, in step S31, the volume significance component of each geometric primitive is calculated according to the following formula:

[0027]

[0028] In the formula, v i is the volume of the i-th geometric primitive; n is the total number of geometric primitives; S v i is the volume significance component of the i-th geometric primitive.

[0029] According to the embodiment of the present application, in step S32, the space size significance component of each geometric primitive is calculated according to the following formula:

[0030]

[0031] In the formula, d1 i , d2 i are the maximum value and the second largest value of the length, width and height of the i-th geometric primitive after sorting; n is the total number of geometric primitives; S d i is the space size significance component of the i-th geometric primitive.

[0032] According to the embodiment of the present application, in step S33, the voxel ratio significance component of each geometric primitive is calculated according to the following formula:

[0033]

[0034] In the formula, n e i is the number of volume units without point cloud; n t i is the total number of volume units; n is the total number of geometric primitives; S o i is the voxel ratio significance component of the i-th geometric primitive.

[0035] According to an embodiment of the present application, in step S34, the saliency of each geometric primitive is calculated by weighting according to the following formula:

[0036]

[0037] wherein S v i is the volume saliency component of the i-th geometric primitive; S d i is the spatial size saliency component of the i-th geometric primitive; S o i is the voxel ratio saliency component of the i-th geometric primitive; w v is the weight of the volume saliency component; w d is the weight of the spatial size saliency component; w o is the weight of the voxel ratio saliency component; n is the total number of geometric primitives; S i is the saliency of the i-th geometric primitive.

[0038] According to an embodiment of the present application, when the type of the basic geometric primitive is a plane, the saliency maximum geometric primitive is a saliency maximum plane geometric primitive, and step S4 specifically comprises:

[0039] Step S41, the centroid of the saliency maximum plane geometric primitive is calculated according to the following formula:

[0040]

[0041] wherein n is the total number of point clouds of the saliency maximum plane geometric primitive; (x i , y i , z i ) is the three-dimensional coordinates of point cloud i; (x, y, z) is the three-dimensional coordinates of the centroid of the saliency maximum plane geometric primitive;

[0042] Step S42, according to the rigidity topological relation invariability between the centroid of the saliency maximum plane geometric primitive and the centroid of the occluded target, the occluded target is located according to the following formula:

[0043]

[0044] wherein d is the distance from the centroid of the saliency maximum plane geometric primitive to the centroid of the occluded target; respectively are the pitch angle and the azimuth angle of the centroid of the occluded target relative to the centroid of the saliency maximum plane geometric primitive in the spherical coordinate system; (O X , O Y , O Z ) is the three-dimensional coordinates of the occluded target.

[0045] Compared with the prior art, the occluded target positioning method based on the saliency primitive and the topological relation invariability has at least the following beneficial effects:

[0046] (1) The method uses the saliency primitive and the direct spatial invariable topological relation between the occluded target and the saliency primitive to realize target positioning, uses the most salient geometric primitive of the measured object, is simple and efficient, and is not easy to misidentify; and the spatial topological relation invariability is used to identify the occluded target, so that accurate positioning of the occluded target can be realized, and the positioning speed is faster.

[0047] (2) The method is accurate in target positioning under the condition that the target is seriously occluded, has strong robustness, and can accurately and quickly position the occluded target. Compared with the existing occluded target positioning method, the method avoids the problems of occluded target mispositioning caused by incomplete information based on feature recognition and large amount of calculation based on the subdivision template matching method.

[0048] (3) The method meets the requirements of rapidly growing machine vision applications and intelligent manufacturing in industrial scenes, and effectively solves the problem of target mispositioning caused by incomplete information of the occluded target. BRIEF DESCRIPTION OF DRAWINGS

[0049] The above and other objects, features and advantages of the present application will become more apparent from the following description of embodiments of the present application with reference to the accompanying drawings, in which:

[0050] Figure 1 A flowchart of the occluded target positioning method based on the saliency primitive and the topological relation invariability according to the embodiments of the present application is schematically shown;

[0051] Figure 2 A process diagram of detecting a planar geometric primitive based on the multiple RANSAC algorithm is schematically shown;

[0052] Figures 3(a) to 3(c) A result diagram of multiple planar geometric primitives of the automobile engine cylinder head lock clip detected based on the multiple RANSAC algorithm is schematically shown;

[0053] Figure 4 A result diagram of the saliency maximum planar geometric primitive of the automobile engine cylinder head lock clip is schematically shown;

[0054] Figure 5 A spatial topological relation diagram between the saliency maximum geometric primitive and the occluded target is schematically shown;

[0055] Figure 6 A topological relation diagram of the automobile engine cylinder head lock clip and the saliency maximum planar geometric primitive is schematically shown;

[0056] Figure 7An effect diagram of an occluded target of a cylinder head lock clip of an automobile engine is schematically shown. DETAILED DESCRIPTION

[0057] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to specific embodiments and drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0058] The terms used herein are only used to describe specific embodiments, and are not intended to limit the present application. The terms "comprise", "contain" and the like used herein indicate the existence of the described features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0059] All terms used herein (including technical and scientific terms) have meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having meanings consistent with the context of the present specification, and should not be interpreted in an idealized or overly formal manner.

[0060] Figure 1 An effect diagram of an occluded target of a cylinder head lock clip of an automobile engine is schematically shown.

[0061] As Figure 1 shown, the occluded target positioning method based on salient primitives and topological relation invariance according to the embodiment can include the following steps S1-S4.

[0062] Step S1, the original three-dimensional point cloud data is preprocessed to obtain the three-dimensional point cloud to be measured.

[0063] Specifically, the preprocessing of the original three-dimensional point cloud data includes uniformly downsampling the three-dimensional point cloud data, and then removing noise from the downsampled three-dimensional point cloud data to obtain the three-dimensional point cloud to be measured.

[0064] For example, the original three-dimensional point cloud data can be the original three-dimensional point cloud data of a cylinder head lock clip of an automobile engine, the original three-dimensional point cloud data is uniformly downsampled by uniform sampling, and the downsampled point cloud is denoised by statistical outlier removal, so as to obtain the three-dimensional point cloud to be measured of the lock clip.

[0065] Step S2, a plurality of basic geometric primitives of the same type in the three-dimensional point cloud to be measured are detected based on a plurality of RANSAC algorithms.

[0066] The type of the basic geometric primitive may include, for example, a plane, a cylinder, and a cuboid. That is, the basic geometric primitive includes a plane geometric primitive, a cylinder geometric primitive, and a cuboid geometric primitive.

[0067] Figure 2 A process diagram for detecting a plane geometric primitive based on a multiple RANSAC algorithm is schematically shown.

[0068] As shown in Figure 2 particular, when the type of the basic geometric primitive is a plane, step S2 specifically includes the following sub-steps S21-S25.

[0069] Step S21, three non-collinear points are randomly selected from the to-be-detected three-dimensional point cloud data, and a plane P1 is fitted.

[0070] Step S22, for any selected point in the to-be-detected three-dimensional point cloud data, the distance from the selected point to the plane P1 is calculated, and the distance is compared with a pre-set threshold value d th , if the distance is less than the threshold value d th , it is determined that the selected point is an inlier of the plane P1, otherwise it is determined that the selected point is an outlier of the plane P1.

[0071] Exemplarily, the threshold value d th may be 0.5 mm, and if the distance from each selected point to the plane P1 is less than 0.5 mm, the selected point is an inlier of the plane P1, otherwise it is an outlier of the plane P1.

[0072] Step S23, for the outliers, three non-collinear points are reselected to fit a plane P2, and the above step S22 is repeated, if the number of inliers of the plane P2 exceeds that of the plane P1, the inliers of the plane P1 are released, the plane P2 is determined as the plane geometric primitive of the iteration result, and the plane P2 is renamed as the plane P1.

[0073] Step S24, the above step S23 is repeated until a pre-set iteration termination number is reached, and the obtained fitted plane is determined as the plane geometric primitive detected by the RANSAC algorithm.

[0074] Exemplarily, the iteration termination number may be 10000, and the above step S23 is repeated until 10000 iterations are completed, and the obtained fitted plane is the plane geometric primitive detected by the RANSAC algorithm.

[0075] Step S25, the above steps S21-S24 are repeated to obtain multiple plane geometric primitives detected by the multiple RANSAC algorithm.

[0076] Figures 3(a) to 3(c)A result map of multiple planar geometric primitives detected based on multiple RANSAC algorithms for a cylinder head lock clip of an automobile engine is schematically shown.

[0077] Referring to Figures 3(a) to 3(c) Three planar geometric primitives are obtained by detecting the cylinder head lock clip of the automobile engine by three RANSAC algorithms.

[0078] In step S3, the salient geometric primitive with the largest saliency is determined by using a saliency calculation method for multiple basic geometric primitives of the same type.

[0079] In the embodiment of the present application, step S3 specifically comprises the following sub-steps S31-S35.

[0080] In step S31, the volume of each geometric primitive is calculated, and a volume saliency component of each geometric primitive is calculated according to the volume.

[0081] Specifically, the volume saliency component of each geometric primitive is calculated according to the following formula:

[0082]

[0083] In the formula, vi is the volume of the ith geometric primitive; n is the total number of geometric primitives; and Svi is the volume saliency component of the ith geometric primitive. i v i In the formula, vi is the volume of the ith geometric primitive; n is the total number of geometric primitives; and Svi is the volume saliency component of the ith geometric primitive.

[0084] In step S32, the length, width and height of each geometric primitive are determined, the values of the length, width and height are sorted in descending order, and a spatial size saliency component of each geometric primitive is calculated.

[0085] Specifically, the spatial size saliency component of each geometric primitive is calculated according to the following formula:

[0086]

[0087] In the formula, d1, d2 and d3 are the maximum value and the second largest value of the values of the length, width and height of the ith geometric primitive after sorting; n is the total number of geometric primitives; and Sdi is the spatial size saliency component of the ith geometric primitive. i i In the formula, d1, d2 and d3 are the maximum value and the second largest value of the values of the length, width and height of the ith geometric primitive after sorting; n is the total number of geometric primitives; and Sdi is the spatial size saliency component of the ith geometric primitive. d i In the formula, d1, d2 and d3 are the maximum value and the second largest value of the values of the length, width and height of the ith geometric primitive after sorting; n is the total number of geometric primitives; and Sdi is the spatial size saliency component of the ith geometric primitive.

[0088] In step S33, each geometric primitive is divided into multiple volume units, and a voxel ratio saliency component of each geometric primitive is calculated.

[0089] Specifically, the voxel ratio saliency component of each geometric primitive is calculated according to the following formula:

[0090] ​​

[0091] wherein n e i is the number of volume units without point cloud; n t i is the total number of volume units; n is the total number of geometric primitives; S o i is the voxel ratio saliency component of the i-th geometric primitive.

[0092] In step S34, the saliency of each geometric primitive is calculated by weighting the volume saliency component, the space size saliency component and the voxel ratio saliency component of each geometric primitive.

[0093] Specifically, the saliency of each geometric primitive is calculated by weighting according to the following formula:

[0094]

[0095] wherein w v is the weight of the volume saliency component; w d is the weight of the space size saliency component; w o is the weight of the voxel ratio saliency component; n is the total number of geometric primitives; S i is the saliency of the i-th geometric primitive.

[0096] In step S35, the saliencies of each geometric primitive are compared, and the geometric primitive with the largest value is determined as the saliency maximum geometric primitive.

[0097] Figure 4 A result graph of the saliency maximum planar geometric primitive of the automobile engine cylinder cover lock clip is schematically shown.

[0098] Referring to Figure 4 , specifically, when the type of the basic geometric primitive is a planar, the saliency maximum geometric primitive is the saliency maximum planar geometric primitive. In combination with Figures 3(a) to 3(c) and Figure 4 , the saliency maximum planar geometric primitive can be determined through the above-mentioned sub-steps S31-S35.

[0099] wherein d1 1 = 95 mm, d2 1 = 80 mm for the first planar geometric primitive shown in Fig. 3(a); d1 2 = 26 mm, d2 2 = 26 mm for the second planar geometric primitive shown in Fig. 3(b); and d1 3 = 18 mm, d2 3= 18mm. Since it is a planar geometric primitive, the weight coefficients take values of w v = 0, w d = 2, w o = 1.

[0100] On this basis, taking the first detected planar geometric primitive as an example, its saliency is calculated according to the following formula:

[0101]

[0102] In the formula, S v 1 , S d 1 , S o 1 are the volume saliency component, the spatial size saliency component and the voxel ratio saliency component of the first detected planar geometric primitive respectively; S 1 is the saliency of the first detected planar geometric primitive.

[0103] By analogy, the saliencies of the three detected planar geometric primitives are calculated as 2.4, 1.178 and 0.836 respectively, and the first planar geometric primitive is determined as the most salient planar geometric primitive according to the calculation result.

[0104] Step S4: locating the occluded target according to the spatial topological relation invariability between the most salient geometric primitive and the occluded target.

[0105] In the embodiment of the present application, the occluded target is located by calculating the distance and angle values between the most salient geometric primitive and the occluded target according to the spatial topological relation invariability between the most salient geometric primitive and the occluded target.

[0106] Figure 5 The spatial topological relation between the most salient geometric primitive and the occluded target is schematically shown in the figure.

[0107] Referring to Figure 5 , specifically, when the type of the basic geometric primitive is a plane, the most salient geometric primitive is the most salient planar geometric primitive, and step S4 specifically includes the following sub-steps S41-S42.

[0108] Step S41: calculating the centroid P c of the most salient planar geometric primitive according to the following formula:

[0109]

[0110] In the formula, n is the total number of point clouds of the most salient planar geometric primitive; (x i , y i , zi ) is the three-dimensional coordinate of the point cloud i; (x, y, z) is the center of mass P of the saliency maximum plane geometric primitive c .

[0111] For the automobile engine cylinder cover lock clamp, the center of mass P of the saliency maximum plane geometric primitive can be calculated c The numerical value of the three-dimensional coordinate of the center of mass P is P c (x, y, z) = (24.62, 2.11, 0.00).

[0112] Step S42, according to the rigidity topological relation invariability between the center of mass P of the saliency maximum plane geometric primitive c and the center of mass O of the occluded target (X，Y，Z) , the occluded target is positioned according to the following formula:

[0113]

[0114] In the formula, d is the distance from the center of mass P of the saliency maximum plane geometric primitive c to the center of mass O of the occluded target (X，Y，Z) ; and respectively, the pitch angle and the azimuth angle of the center of mass O of the occluded target (X，Y，Z) relative to the center of mass P of the saliency maximum plane geometric primitive c in the spherical coordinate system; (O X , O Y , O Z ) is the three-dimensional coordinate of the occluded target.

[0115] In an optional embodiment of the present application, Figure 6 a topological relation diagram of the automobile engine cylinder cover lock clamp and the saliency maximum plane geometric primitive is schematically shown.

[0116] Referring to Figure 6 , in the space rectangular coordinate system, for the automobile engine cylinder cover lock clamp, the spatial topological relation of the present embodiment is the relation between the spatial positions, the center of mass O of the occluded target (X，Y，Z) relative to the center of mass P of the saliency maximum plane geometric primitive c is known, and the distance d x , d y , d z between the three directions in the space rectangular coordinate system are respectively:

[0117]

[0118] Then the center of mass of the occluded target can be calculated according to the formula:

[0119]

[0120] Thus, the target lock clip centroid coordinates are (9.85mm, 26.74mm, -33.39mm). In combination with the lock clip centroid coordinates, the target point cloud is repositioned according to the target features, as shown in Figure 6

[0121] Figure 7 An effect diagram of the occluded target of the automobile engine cylinder cover lock clip is schematically shown.

[0122] When the target to be detected is occluded, the target to be detected cannot be positioned through the salient feature of the target to be detected, and the occluded target can be positioned through the spatial topological relationship invariability of the salient element and the target to be detected, as shown in Figure 7

[0123] From the above description, it can be seen that the above-mentioned embodiments of the present application at least achieve the following technical effects:

[0124] (1) The present method uses the spatial invariable topological relationship between the salient element and the occluded target and the salient element to realize target positioning, uses the most salient geometric element of the measured object, is simple and efficient, and is not easy to misidentify; and then uses the spatial topological relationship invariability to identify the occluded target, so that the occluded target can be accurately positioned, and the positioning speed is faster.

[0125] (2) The present method is accurate in target positioning under the condition that the target is severely occluded, has strong robustness, and can accurately and quickly position the occluded target. Compared with the existing occluded target positioning method, the present method avoids the problems of occluded target mispositioning caused by incomplete information based on feature recognition and large amount of calculation based on the subdivision template matching method.

[0126] (3) The present method meets the needs of rapidly growing machine vision applications and intelligent manufacturing in industrial scenes, and effectively solves the problem of target mispositioning caused by incomplete information of the occluded target.

[0127] Some of the blocks in the drawings can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, so that these instructions can create a means for implementing the functions / operations specified in the block diagram and / or flowchart when executed by the processor.

[0128] ​​In addition, the terms "first", "second", etc. are used only for the purpose of description, and should not be understood as indicating or implying relative importance or implying the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is at least two, such as two, three, etc., unless otherwise explicitly and specifically limited. In addition, the word "one" or "an" before an element does not exclude the presence of multiple such elements.

[0129] The above specific embodiments describe the purpose, technical solutions and beneficial effects of the present application in further detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. An occluded object localization method based on salient primitives and topological relation invariance, characterized in that, The method comprises the following steps: Step S1, preprocessing the original three-dimensional point cloud data to obtain a three-dimensional point cloud to be measured; Step S2, detecting a plurality of basic geometric primitives of the same type in the three-dimensional point cloud data to be measured based on a plurality of RANSAC algorithms; Step S3, determining a geometric primitive with the maximum significance from the plurality of basic geometric primitives of the same type by using a significance calculation method; Step S4, locating an occluded target according to the spatial topological relationship invariability between the geometric primitive with the maximum significance and the occluded target; The step S3 specifically comprises the following steps: Step S31, calculating the volume of each geometric primitive, and calculating the volume significance component of each geometric primitive according to the volume; Step S32, determining the length, width and height of each geometric primitive, sorting the numerical values of the length, width and height in descending order, and calculating the spatial size significance component of each geometric primitive; Step S33, dividing each geometric primitive into a plurality of volume units respectively, and calculating the voxel ratio significance component of each geometric primitive; Step S34, combining the volume significance component, the spatial size significance component and the voxel ratio significance component of each geometric primitive, and calculating the significance of each geometric primitive by weighting; Step S35, comparing the significance of each geometric primitive, and determining the geometric primitive with the maximum numerical value as the geometric primitive with the maximum significance; In step S31, the volume significance component of each geometric primitive is calculated according to the following formula: where v i is the volume of the ith geometric primitive; n is the total number of geometric primitives; is the volume significance component of the ith geometric primitive; In step S32, the spatial size significance component of each geometric primitive is calculated according to the following formula: wherein are the maximum and second maximum values of the sorted values of the length, width and height of the i-th geometric primitive, respectively; n is the total number of geometric primitives; is the spatial dimension significance component of the i-th geometric primitive. In step S33, the voxel ratio significance component of each geometric primitive is calculated according to the following formula: wherein is the number of volume elements without point clouds; is the total number of volume elements; n is the total number of geometric primitives; is the voxel-wise saliency component of the i-th geometric primitive; In step S34, the significance of each geometric primitive is calculated by weighting according to the following formula: wherein is the volume saliency component of the i-th geometric primitive; is the spatial size saliency component of the i-th geometric primitive; is the voxel ratio saliency component of the i-th geometric primitive; w v is the weight of the volume saliency component; w d is the weight of the spatial size saliency component; w o is the weight of the voxel ratio saliency component; n is the total number of geometric primitives; is the saliency of the i-th geometric primitive.

2. The salient primitive and topological relationship invariance based occluded object localization method according to claim 1, characterized in that, In step S1, the preprocessing of the original three-dimensional point cloud data comprises the following steps: Uniformly down-sampling the three-dimensional point cloud data, and then removing noise from the down-sampled three-dimensional point cloud data to obtain the three-dimensional point cloud to be measured.

3. The salient primitive and topological relationship invariance based occluded object localization method of claim 1, wherein, In step S2, the types of the basic geometric primitives include a plane, a cylinder and a cuboid.

4. The salient primitive and topological relationship invariance based occluded object localization method according to claim 3, characterized in that, When the type of the basic geometric primitive is a plane, the step S2 specifically comprises the following steps: Step S21, randomly selecting three non-collinear points from the three-dimensional point cloud data to be measured to fit a plane P1; Step S22, calculating the distance from any selected point in the three-dimensional point cloud data to be measured to the plane P1, comparing the distance with a pre-set threshold value, and determining the selected point as an in-point of the plane P1 if the distance is less than the threshold value, otherwise determining the selected point as an out-point of the plane P1; Step S23, re-selecting three non-collinear points to fit a plane P2 for the out-point, repeating the step S22, and releasing the in-point of the plane P1 if the number of in-points of the plane P2 is greater than that of the plane P1, determining the plane P2 as the plane geometric primitive of the iteration result, and renaming the plane P2 as the plane P1. Step S24, repeating the above step S23 until a preset iteration termination number is reached, and determining the obtained fitting plane as a plane geometric primitive detected by the current RANSAC algorithm; Step S25, repeating the above steps S21-S24 to obtain a plurality of plane geometric primitives detected by the plurality of RANSAC algorithms.

5. The salient primitive and topological relationship invariance based occluded object localization method according to claim 3, characterized in that, When the type of the basic geometric primitive is a plane, the saliency maximum geometric primitive is a saliency maximum plane geometric primitive, and the step S4 specifically includes: Step S41, calculating the centroid of the saliency maximum plane geometric primitive according to the following formula: wherein n is the total number of point clouds of the saliency maximum planar geometric primitive; (x i , y i , z i ) are the three-dimensional coordinates of point cloud i; are the three-dimensional coordinates of the center of mass of the saliency maximum planar geometric primitive; Step S42, positioning the occluded target according to the rigidity topological relation invariability between the centroid of the saliency maximum plane geometric primitive and the centroid of the occluded target according to the following formula: In the formula, d is the distance from the center of the saliency maximum plane geometric primitive to the center of the occluded target; θ, φ are respectively the pitch angle and the azimuth angle of the center of the occluded target relative to the center of the saliency maximum plane geometric primitive in the spherical coordinate system; is the three-dimensional coordinate of the occluded target.

Images