Computer-implemented method for aligning source model with target model

Abstract

Aligning the source and target models includes calculating shape descriptors for a plurality of edge points, grouping the edge points according to the shape descriptors, and selecting representative points for each group, calculating target and source point pair features (PPFs) between representative point pairs on the target and the source models, where the PPFs define relative positions and orientations of the point pairs, matching the target PPFs with each source PPF, and determining the target PPFs. The method includes determining a target PPF point and a source PPF point, and transforming pairs of points associated with the matching PPFs, aligning a position and edge direction of the target PPF point with a position and edge direction of the source PPF point, determining an angle to align a second target PPF point with a second source PPF point, and finding a modal angle among the determined rotation angles, calculating an output transformation using a transformation associated with the modal angle.

Description

Technical Field

[0001] The present invention relates to three-dimensional modeling, and in particular to the alignment of a source model with a target model without an external reference system. Background Art

[0002] In many cases, it is beneficial to align two three-dimensional (3D) models whose relative orientation is unknown. For example, complex mechanical parts are difficult to measure in full 3D and evaluate for surface defects, and comparative measurement techniques are often used. 3D scanners are used to generate an actual model of the part, which can be compared with a template model, such as a design model generated using computer-aided design (CAD) software. Using this technique, dimensional errors and surface defects can be found in the model differences. However, the models must be precisely aligned, and the rotation and translation between the template model and the actual model are usually unknown.

[0003] US8830229B2 describes a "point pair feature" (PPF) method for aligning a template model with an actual model. Both the template model and the actual model are sampled and represented by discrete points, with the point direction perpendicular to the surface of the point. The template model is described using PPFs, which define the relative position and direction of each pair of points in the model. The PPFs of the template model are stored in a hash table so that one or more PPFs of the actual model can be used to search in the hash table to find similar PPFs from the template model.

[0004] Each point pair from the actual model is transformed by translating the first point in the point pair to the origin of the coordinate system and then rotating it so that the direction of the first point is aligned with a reference axis, such as the x-axis. The point pair in the template model with a similar PPF is transformed by translating the first point in the point pair to the origin of the coordinate system and then rotating it so that the direction of the first point is aligned with a reference axis, such as the x-axis. The rotation angle α between the two transformed PPFs is calculated, thereby calculating a series of rotation angles. The angle α with the most votes can be found through a voting scheme, and a transformation matrix is ​​calculated from a point pair in the actual model to a corresponding point pair in the template model.

[0005] Since there are many transformation matrices, the transformation matrices are clustered and the average matrix of the most common cluster is taken as the transformation matrix from the actual model to the template model.

[0006] This method has limitations for planar or smooth surfaces, for example, if two points come from a plane, the PPF has three ambiguous angles. To address this problem, Choi (2012) described defining boundary point to boundary point (B2B) features using only boundary points.

[0007] However, even with B2B features, there may still be alignment errors. For example, for a model that is mostly symmetrical but a little asymmetrical, if all boundary points are considered to have the same role, the alignment result may be wrong because the points in most symmetrical parts are well aligned, but the asymmetric points in a few parts are not aligned correctly. Therefore, the asymmetric points in a few parts are more important, and the misalignment problem can be solved by assigning appropriate weights to the points or features. US9280827B2 describes a learning-based weighting method, however, the actual model requires a large amount of 3D data, and the posture must be marked as input to the training network, so training the network is very time-consuming.

[0008] The object of the present invention is to provide a model alignment method that overcomes these problems. Summary of the invention

[0009] The features and advantages of the present invention will be explained below, and in part will be obvious from the description, or can be learned by practicing the principles of the present invention. The features and advantages of the present invention can be realized and obtained by the methods and combinations particularly pointed out in the appended claims.

[0010] According to a first example of the present invention, a computer-implemented method for aligning a source model with a target model is provided, comprising receiving a source model and a target model; identifying a plurality of edge points on each model, wherein each edge point contains edge point direction information; calculating a shape descriptor associated with each identified edge point; grouping the edge points of each model according to the associated shape descriptor, and selecting one or more representative points for each group of edge points; calculating a target point pair feature PPF between each pair of representative points on the target model, wherein the PPF defines the relative position and direction of the point pair; selecting one or more designated representative points on the source model, and forming a source PPF with each other representative point; searching for one or more designated representative points on the source model; and target PPFs to match each calculated source PPF; for each set of matched source PPFs and target PPFs: transforming point pairs associated with the source PPF and / or target PPF to align the position and edge direction of a first point associated with the target PPF with the position and edge direction of a reference point associated with the source PPF; and aligning a second point associated with the target PPF with a second point associated with the source PPF to determine a rotation angle; identifying a modal angle among a plurality of determined rotation angles; and calculating an output transformation from a source model to a target model using the transformation applied to the points associated with each set of matched source PPFs and target PPFs, wherein the associated rotation angle is equal to the modal angle.

[0011] According to a second example of the present invention, a data processing device is provided, including a processor, wherein the processor is configured to align a source model with a target model by the following method: receiving a source model and a target model; identifying a plurality of edge points on each model, wherein each edge point contains edge point direction information; calculating a shape descriptor associated with each identified edge point; grouping the edge points of each model according to the associated shape descriptor, and selecting one or more representative points for each group of edge points; calculating a target point pair feature PPF between each pair of representative points on the target model, wherein the PPF defines the relative position and direction of the point pair; selecting one or more designated representative points on the source model, and forming a source PPF together with each other representative point : Searching for one or more target PPFs to match each calculated source PPF; for each set of matched source PPFs and target PPFs: transforming point pairs associated with the source PPF and / or target PPF to align the position and edge direction of a first point associated with the target PPF with the position and edge direction of a reference point associated with the source PPF; and aligning a second point associated with the target PPF with a second point associated with the source PPF to determine a rotation angle; identifying a modal angle among a plurality of determined rotation angles; and calculating an output transformation from a source model to a target model using the transformation applied to points associated with each set of matched source PPFs and target PPFs, wherein the associated rotation angle is equal to the modal angle.

[0012] According to a third example of the present invention, a computer-readable medium including instructions is provided. When executed by a processor, the processor aligns a source model with a target model by the following method: receiving a source model and a target model; identifying multiple edge points on each model, and each edge point contains edge point direction information; calculating a shape descriptor associated with each identified edge point; grouping the edge points of each model according to the associated shape descriptor, and selecting one or more representative points for each group of edge points; calculating a target point pair feature PPF between each pair of representative points on the target model, wherein the PPF defines the relative position and direction of the point pair; selecting one or more designated representative points on the source model, and forming a source point pair with each other representative point. PPF: searching for one or more target PPFs to match each calculated source PPF; for each set of matched source PPF and target PPF: transforming point pairs associated with the source PPF and / or target PPF to align the position and edge direction of a first point associated with the target PPF with the position and edge direction of a reference point associated with the source PPF; and aligning a second point associated with the target PPF with a second point associated with the source PPF to determine a rotation angle; identifying a modal angle among a plurality of determined rotation angles; and using the transformation applied to the points associated with each set of matched source PPF and target PPF, calculating an output transformation from a source model to a target model, wherein the associated rotation angle is equal to the modal angle.

[0013] Identifying the edge points includes detecting a plurality of edges and sampling each edge to generate the plurality of edge points.

[0014] The shape descriptor of each point is calculated using multiple neighboring points within a preset radius.

[0015] The shape descriptor is calculated using the ratio between the length of the edge direction along the point across the neighboring points and the width of the edge direction perpendicular to the point across the neighboring points.

[0016] The shape symbol is calculated using a histogram of the distance between each neighboring point and the straight line passing through the point, and the direction of the straight line passing through the point is the edge direction of the point.

[0017] Shape descriptors can be calculated using values ​​of linearity, flatness, surface variation, and anisotropy.

[0018] Each point pair feature is computed using the distance between the points, the angle between the edge directions of the points, and the angle between the edge direction of each point and the connecting line extending between the points.

[0019] Each point pair feature also includes the computed shape descriptor of the point.

[0020] The target PPFs are stored in a hash table that can be searched using the source PPF.

[0021] Transforming the point pair includes translating a first point associated with the target PPF and a reference point associated with the source PPF to a common origin and rotating the point pair to align an edge direction of the first point associated with the target PPF and an edge direction of the reference point associated with the source PPF with a common first axis.

[0022] Identifying the modal angle involves quantifying the range of possible angles and taking a cumulative vote on the rotation angle for each set of matched source and target PPFs.

[0023] Computing the output transformation includes computing a transformation from the source model to the target model for each set of matched source PPFs and target PPFs, and clustering the computed transformations, wherein the associated rotation angle is equal to the modal angle.

[0024] The source model is an actual model based on real data, and the target model is a template model based on a computer-generated template. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the scheme of the present invention, a brief introduction is given below to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Preferred embodiments of the present invention will be further explained in detail below by way of examples and with reference to the accompanying drawings, in which:

[0027] Figure 1 A data processing device according to an embodiment.

[0028] Figure 2 are some examples of edge points.

[0029] Figure 3A and 3B is an example calculation of a shape descriptor according to some embodiments.

[0030] Figure 4A and 4B is the computation of a point pair function according to some embodiments.

[0031] Figure 5 A lookup in a hash table according to some embodiments.

[0032] Figure 6 is a transformation of a point-to-function according to some embodiments.

[0033] Figure 7 A method for aligning a source model with a target model according to an embodiment. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] With reference to the accompanying drawings, Figure 1 A data processing apparatus 1 is shown for aligning a source model 10 with a target model 20. The data processing apparatus 1 comprises an input unit 100, a point selection unit 200, a point pair feature (PPF) unit 300, a matching unit 400, a voting unit 500 and a transformation unit 600.

[0036] The input unit 100 may receive a source model 10 and a target model 20. In some embodiments, the source model 10 may be an actual model based on real data, and the target model 20 may be a template model based on a computer-generated model. For example, these models may all be related to mechanical parts, such as complex mechanical parts, where a 3D scanner may be used to generate an actual model of the part, and the template model may be generated using computer-aided design (CAD) software.

[0037] The point selection unit 200 may identify a plurality of edge points on each model. Each identified edge point is associated with an edge direction at the edge point location.

[0038] In some embodiments, the point selection unit 200 may detect a plurality of edges and sample each edge to generate a plurality of edge points.

[0039] The point selection unit 200 may also compute a shape descriptor associated with each identified edge point.

[0040] Figure 2 Four groups of edge points classified by shape are shown; each point and its neighboring points can be represented by a shape descriptor. In some embodiments, the point selection unit 200 can use multiple neighboring points within a preset radius. This Figure 2 The larger edge point and its neighboring points form a circle in the . A preset radius can be set based on the object size and / or 3D point density. There are several ways to calculate shape descriptors.

[0041] In some embodiments, the point selection unit 200 may use a ratio between a length spanning adjacent points along an edge direction of a point and a width spanning adjacent points perpendicular to an edge direction of a point.

[0042] Figure 3A shows the computation of the aspect ratio shape descriptor. i ) and its neighboring points to calculate the eigenvalues ​​and eigenvectors. The largest eigenvalue can be designated as λ1, and the smallest eigenvalue can be designated as λ3. The span along λ1 can be defined as the length, and the width can be defined as the span along λ3. If P i In such Figure 2 (c) on the straight line shown or in the Figure 2 (a) shows a large circle, the value of the aspect ratio shape descriptor is small. i In Figure 2 (d) or in the corners shown in Figure 2 In the small circle shown in (b), the value of the aspect ratio shape descriptor is larger.

[0043] In some embodiments, the point selection unit 200 may calculate the shape symbol using a histogram of the distance between each neighboring point and a straight line passing through the point, where the direction of the straight line passing through the point is the edge direction of the point.

[0044] Figure 3B shows the computation of a histogram-based shape descriptor. i and its neighboring points to fit a virtual plane. The distance from each point to the virtual plane can be calculated and stored in a histogram. The distance histogram represents the geometry of each point. If P iOn a straight line or plane, a histogram tends to have a peak. i lies on a smooth surface, then the histogram is smooth.

[0045] In some embodiments, the point selection unit 200 may use values ​​for linearity, flatness, surface variation, and anisotropy. These geometric parameters may be based on eigenvalues, for example, where λ1, λ2, and λ3 are P i and the eigenvalues ​​of its adjacent points. λ1 is the maximum eigenvalue.

[0046] In some embodiments, linearity can be calculated as:

[0047]

[0048] In some embodiments, flatness may be calculated as:

[0049]

[0050] In some embodiments, the surface variation may be calculated as:

[0051]

[0052] In some embodiments, anisotropy can be calculated as:

[0053]

[0054] The point selection unit 200 may also group the edge points of each model according to the associated shape descriptor. The point selection unit 200 is used to select one or more representative points for each group of edge points. In some embodiments, a predetermined number of representative points may be randomly selected from each group of edge points. In some embodiments, a weight may be assigned to each edge point based on the shape descriptor. The weighting scheme may assign the same weight to each geometric group, and each geometric group may be represented by the same number of selected points. In this way, each group has the same contribution to represent the model. In addition, since fewer points are selected, the speed of the device may be increased. Accuracy may also be improved because only discriminative points are considered. In addition, the weighting scheme may be directly implemented in an online process without the need for training.

[0055] The PPF unit 300 is configured to calculate a target PPF between each pair of representative points on the target model 20. The PPF defines the relative position and direction of the pair of points.

[0056] In some embodiments, each PPF may be calculated using distances between points, angles between edge directions of the points, and angles between the edge direction of each point and a connecting line extending between the points.

[0057] Figure 4Ashows how to calculate point-to-point relationships. Point-to-point relationships can be defined by three angles and one distance. Each edge point can have a main direction based on its neighboring points. For a pair of edge points P i and P j , the first two angles can be calculated using the connecting line between the two points. The first angle θ i It can be P i The angle between the main direction and the connecting line. The second angle θ j It can be P j The angle between the main direction and the connecting line. The third angle θ ij It can be P i The main direction and P j The angle between the main directions of ij Can be the distance between two edge points along a connecting line.

[0058] In some embodiments, each PPF may also include a computed shape descriptor of the point. Figure 4A Also shown is a local shape descriptor which can be combined with the point-to-point relationship to form the PPF. The local shape descriptor is substantially as described above.

[0059] Including local shape descriptors can improve the performance of PPF, such as Figure 4B As shown. If only the point-to-point relationship is used for the example point pairs in the template model shown in (a), two corresponding point pairs can be found in the actual model shown in (b). In some embodiments, more than two corresponding point pairs can be found in the actual model. As shown in Figure (b), the upper point pair is correct and the lower point pair is incorrect. If the local shape descriptor is included in the PPF, only the correct point pairs will be found in the actual model, as shown in (c).

[0060] The PPF unit 300 is further configured to calculate a source PPF between one or more designated representative points on the source model 10 and other representative points on the source model 10 .

[0061] The matching unit 400 is used to match one or more target PPFs with each calculated source PPF. The matching unit 400 is configured to search for one or more target PPFs to match each calculated source PPF.

[0062] In some embodiments, the target PPFs may be stored in a hash table that can be searched using the source PPF, such as Figure 5 As shown. The hash table can be configured to return "similar" PPFs within a certain tolerance threshold. In some embodiments, it may not be possible to match one or more source PPFs with any target PPF, and these unmatched PPFs will not participate in the subsequent process.

[0063] The voting unit 500 is configured to transform a first pair of points associated with a source PPF and / or a second pair of points associated with a target PPF. The point pairs are transformed so that the position and edge direction of the first point associated with the target PPF are aligned with the position and edge direction of the reference point associated with the source PPF. The transformation is repeated for each set of matching source PPFs and target PPFs.

[0064] In some embodiments, the voting unit 500 can be configured to translate the first point associated with the target PPF and the reference point associated with the source PPF to a common origin. The two pairs of points are rotated so that the edge direction of the first point associated with the target PPF and the edge direction of the reference point associated with the source PPF are aligned with a common first axis.

[0065] Figure 6 FIG. 5 shows an exemplary transformation performed by the voting unit 500. In the template model, the transformation target point pair (P i , P j ), P i Translate to the origin of the coordinate system, and then i The direction of is aligned with a reference axis such as the x-axis. The transformation matrix including translation and rotation can be called M Pij . The transformed (P i ',P j ') and the angle α1 between the z-axis. Similarly, in the actual model, the source point pair (Q m , Q n ) can be obtained by Qmn The angle α2 can be calculated based on the z-axis.

[0066] The voting unit 500 may determine a rotation angle for each set of matched source PPF and target PPF to align the second point associated with the target PPF with the second point associated with the source PPF. That is, the translated target point pair (P i ',P j ') and the translated source point pair (Q m ', Q n Thus, each set of matched point pairs has an associated rotation angle α around the z-axis.

[0067] The voting unit 500 is further configured to identify a modal angle among the plurality of determined rotation angles.

[0068] In some embodiments, voting unit 500 may quantify the range of possible angles and perform cumulative voting for the rotation angle for each set of matched source PPF and target PPF.

[0069] The transformation unit 600 is configured to compute an output transformation 30 from the source model 10 to the target model 20. The transformation unit 600 is configured to use a transformation applied to points associated with each set of matched source and target PPFs, wherein the associated rotation angle is equal to the modal angle.

[0070] In some embodiments, the transformation unit 600 may calculate a transformation from the source model 10 to the target model 20 for each set of matched source PPFs and target PPFs, and cluster the calculated transformations, wherein the associated rotation angle is equal to the modal angle.

[0071] Figure 7 A method for aligning a source model with a target model according to an embodiment is shown. The method starts at step S1.

[0072] In step S2, a source model and a target model are received.

[0073] In some embodiments, the source model may be an actual model based on real data, and the target model may be a template model based on a computer-generated template.

[0074] In step S3, a plurality of edge points are identified on each model. Each identified edge point is associated with an edge direction at the edge point position.

[0075] In some embodiments, identifying edge points may include detecting a plurality of edges and sampling each edge to generate a plurality of edge points.

[0076] In step S4, a shape descriptor associated with each identified edge point is calculated.

[0077] In some embodiments, the shape descriptor of each point may be calculated using a plurality of neighboring points within a preset radius.

[0078] In some embodiments, the shape descriptor may be calculated using a ratio between a length across neighboring points in a direction along an edge of a point and a width across neighboring points in a direction perpendicular to the edge of the point.

[0079] In some embodiments, a histogram of the distance between each neighboring point and a straight line passing through the point may be used to calculate the shape symbol, and the direction of the straight line passing through the point is the edge direction of the point.

[0080] In some embodiments, shape descriptors may be calculated using values ​​for linearity, planarity, surface variation, and anisotropy.

[0081] In step S5, the edge points of each model are grouped according to the associated shape descriptors, and one or more representative points are selected for each group of edge points.

[0082] In step S6, the target point pair feature (PPF) between each pair of representative points on the target model is calculated. The PPF defines the relative position and direction of the point pair.

[0083] In some embodiments, each point pair feature may be calculated using the distance between points, the angle between the edge directions of the points, and the angle between the edge direction of each point and a connecting line extending between the points.

[0084] In some embodiments, each point pair feature may also include a computed shape descriptor of the point.

[0085] In step S7, a source PPF is calculated between one or more designated representative points on the source model and other representative points on the source model.

[0086] In step S8, one or more target PPFs are matched with each calculated source PPF.

[0087] In some embodiments, the target PPFs may be stored in a hash table that may be searched using the source PPF.

[0088] In step S9, the first pair of points associated with the source PPF and / or the second pair of points associated with the target PPF are transformed so that the position and edge direction of the first point associated with the target PPF are aligned with the position and edge direction of the reference point associated with the source PPF. The transformation is repeated for each set of matching source PPFs and target PPFs.

[0089] In some embodiments, transforming the point pair may include translating the first point associated with the target PPF and the reference point associated with the source PPF to a common origin. The two pairs of points may be rotated to align the edge direction of the first point associated with the target PPF and the edge direction of the reference point associated with the source PPF with a common first axis.

[0090] At step S10 , a rotation angle is determined for each set of matched source PPF and target PPF to align a second point associated with the target PPF with a second point associated with the source PPF.

[0091] In step S11 , a modal angle is identified from among the plurality of determined rotation angles.

[0092] In some embodiments, identifying the modality angle may include quantizing a range of possible angles and taking a cumulative vote for the rotation angle for each matching source PPF and target PPF.

[0093] At step S12, the output transformation from the source model to the target model is calculated. The calculation uses a transformation applied to the points associated with each matching source PPF and target PPF, where the associated rotation angle is equal to the modal angle.

[0094] In some embodiments, computing the output transformation includes computing a transformation from the source model to the target model for each set of matched source PPFs and target PPFs, and clustering the computed transformations wherein the associated rotation angles are equal to the modal angles.

[0095] The method ends at step S13.

[0096] The above embodiments are described by way of example only. Many variations are possible without departing from the scope of the invention as defined by the appended claims.

[0097] For clarity of explanation, in some cases, the present technology may be presented as including separate functional blocks including devices, device components, steps or procedures in a method embodied in software or a combination of hardware and software.

[0098] The method according to the above example can be implemented by computer executable instructions, which are stored in computer readable media or can be obtained from computer readable media in other ways. For example, such instructions may include instructions and data that cause or otherwise configure a general-purpose computer, a special-purpose computer, or a special-purpose processing device to perform a specific function or function group. Some computer resources can be used by accessing a network. Computer executable instructions can be, for example, binary files, intermediate format instructions, such as assembly language, firmware, or source code. Examples of computer readable media that can be used to store instructions, information used, and / or information created during the method according to the example include disks or optical disks, flash memory, universal serial bus (USB) devices equipped with non-volatile memory, network storage devices, etc.

[0099] Devices implementing methods according to these inventions may include hardware, firmware, and / or software, and may take any of a variety of form factors. Typical examples of such form factors include laptop computers, smart phones, small personal computers, personal digital assistants, etc. The functionality described herein may also be embodied in peripheral devices or add-on cards. As a further example, such functionality may also be implemented on a circuit board between different chips or between different processes executed in a single device.

[0100] Instructions, the media for transmitting such instructions, computing resources for executing them, and other structures for supporting such computing resources are the means for providing the functions described in these inventions.

[0101] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A computer-implemented method for aligning a source model with a target model, comprising the following steps: receiving a source model and a target model; identifying a plurality of edge points on each model, wherein each edge point is associated with an edge direction at the edge point location; computing a shape descriptor associated with each identified edge point; Grouping edge points of each model according to the associated shape descriptors and selecting one or more representative points for each group of edge points; Calculate the target point pair feature PPF between each pair of representative points on the target model, where PPF defines the relative position and direction of the point pair; Calculate the source PPF between one or more specified representative points on the source model and other representative points on the source model: searching for one or more target PPFs to match each calculated source PPF; For each set of matching source and target PPFs: transforming point pairs associated with the source PPF and / or the target PPF to align the position and edge direction of a first point associated with the target PPF with the position and edge direction of a reference point associated with the source PPF; and determining a rotation angle to align a second point associated with the target PPF with a second point associated with the source PPF; identifying a modal angle among a plurality of determined rotation angles; and An output transformation from the source model to the target model is computed using the transformation applied to the points associated with each set of matched source and target PPFs, where the associated rotation angle is equal to the modal angle.

2. The computer-implemented method of claim 1, wherein: Identifying the edge points includes detecting a plurality of edges and sampling each edge to generate the plurality of edge points.

3. The computer-implemented method of claim 1, wherein: The shape descriptor of each point is calculated using multiple neighboring points within a preset radius.

4. The computer-implemented method of claim 3, wherein: The shape descriptor is calculated using the ratio between the length of the edge direction along the point across the neighboring points and the width of the edge direction perpendicular to the point across the neighboring points.

5. The computer-implemented method of claim 3, wherein: The shape symbol is calculated using a histogram of the distance between each neighboring point and the straight line passing through the point, and the direction of the straight line passing through the point is the edge direction of the point.

6. The computer-implemented method of claim 3, wherein: Shape descriptors are calculated using values ​​of linearity, flatness, surface variation, and anisotropy.

7. A computer-implemented method as claimed in any one of the preceding claims, characterized in that Each point pair feature is computed using the distance between the points, the angle between the edge directions of the points, and the angle between the edge direction of each point and the connecting line extending between the points.

8. The computer-implemented method of claim 7, wherein: Each point pair feature also includes the computed shape descriptor of the point.

9. The computer-implemented method of claim 1, wherein: The target PPFs are stored in a hash table that can be searched using the source PPF.

10. The computer-implemented method of claim 1, wherein: Transforming the point pair includes translating a first point associated with the target PPF and a reference point associated with the source PPF to a common origin and rotating the point pair to align an edge direction of the first point associated with the target PPF and an edge direction of the reference point associated with the source PPF with a common first axis.

11. The computer-implemented method of claim 1, wherein: Identifying the modal angle involves quantifying the range of possible angles and taking a cumulative vote on the rotation angle for each set of matched source and target PPFs.

12. The computer-implemented method of claim 1, wherein: Computing the output transformation includes computing a transformation from the source model to the target model for each set of matched source PPFs and target PPFs, and clustering the computed transformations, wherein the associated rotation angle is equal to the modal angle.

13. The computer-implemented method of claim 1, wherein: The source model is an actual model based on real data, and the target model is a template model based on a computer-generated template.

14. A data processing device, characterized in that: comprising a processor for performing the computer implemented method of any preceding claim.

15. A computer readable medium, characterized in that The method comprises instructions which, when executed by a processor, cause the processor to perform the method of any one of claims 1 to 13.

Images