Methods and systems for surface fitting based change detection in 3D point-cloud

Patent metadata

Field	Value
Publication number	US-10475231-B2
Application number	US-201815898160-A
Country	US
Kind code	B2
Filing date	Feb 15, 2018
Priority date	Sep 13, 2017
Publication date	Nov 12, 2019
Grant date	Nov 12, 2019

Abstract

Methods and systems for change detection utilizing three dimensional (3D) point-cloud processing are provided. The method includes detecting changes in the surface based on a surface fitting approach with a locally weighted Moving Least Squares (MLS) approximation. The method includes acquiring and comparing surface geometry of a reference point-cloud defining a reference surface and a template point-cloud defining a template surface at local regions or local surfaces using the surface fitting approach. The method provides effective change detection for both, rigid as well as non-rigid changes, reducing false detections due to presence of noise and is independent of factors such as texture or illumination of an object or scene being tracked for changed detection.

First claim

What is claimed is:

1. A processor implemented method for change detection in a surface represented by 3D point-cloud, the method comprising:
acquiring a reference point-cloud defining a reference surface and a template point-cloud defining a template surface, wherein the reference point-cloud and the template point-cloud correspond to the 3D point-cloud of the surface acquired at different time instances;
aligning the template surface with the reference surface by performing registration of the template point-cloud and the reference point-cloud, wherein the registration of the reference point-cloud and the template point-cloud is based on shape similarity based constraints;
equalizing point density of the registered reference point-cloud and the registered template point-cloud by performing subsampling to provide a processed reference point-cloud and a processed template point-cloud;
reducing noise from the processed reference point-cloud and the processed template point-cloud by estimating a smooth reference surface and a smooth template surface using a locally weighted Moving Least Squares (MLS) approximation;
voxelizing the smooth reference surface to generate a plurality of reference voxels and the smooth template surface to generate a plurality of template voxels, wherein each reference voxel is defined by a plurality of reference vertices and each template voxel is defined by a plurality of template vertices;
determining a corresponding reference vertex for every template vertex of each template voxel using a k-dimensional (kd)-tree, wherein the kd-tree is set with the plurality of reference vertices of a reference voxel;
determining vertex distance between every template vertex and the determined corresponding reference vertex, wherein the determined vertex distance above a predefined vertex distance threshold is indicative of the change in the surface represented by the 3D-point-cloud; and
generating an output point-cloud indicating change detected in the template point-cloud with respect to the reference point-cloud.

2. The method of claim 1 , wherein performing the subsampling to provide the processed reference point-cloud and the processed template point-cloud comprises:
identifying a first point-cloud and a second point-cloud among the reference point-cloud and the template point-cloud for sub-sampling, wherein the first point-cloud comprises higher density of points and the second point-cloud comprises lower density of points;
assigning a parameter to each point of the first point-cloud, wherein the parameter is initialized to zero value indicating that the point is not required to be sampled;
setting each point of the first point-cloud in the kd-tree, wherein each point is defined by 3D-coordinates;
identifying a closest point from the first point-cloud for each point of the second point-cloud, wherein the closest point is identified from the kd-tree and is defined by 3D-coordinates;
determining a sub-sampling distance between the each point of the second point-cloud and the corresponding closest point from the first point-cloud; and
generating a sub-sampled first point-cloud by retaining the closest point from the first point-cloud in the sub-sampled first point-cloud if the determined sub-sampling distance is below a threshold t 1 or above a predefined subsampling threshold t 2 , wherein t 2 is greater than t 1 .

3. The method of claim 1 , wherein generating the smooth reference surface and the smooth template surface comprises:
estimating a local reference surface for each point of the processed reference point-cloud and a local template surface for each point of the processed template point-cloud, wherein estimating the local reference surface and the local template surface comprises:
identifying local neighborhood comprising a plurality of neighbor points for each point of the processed reference point-cloud and each point of the processed template point-cloud; and
identifying a local reference planar surface represented by the corresponding plurality of neighbor points and a local template planar surface represented by the corresponding plurality of neighbor points, wherein the local reference planar surface and the local template planar surface is estimated such that the summation of the weighted distances between the plurality of neighbor points and projections of the plurality of neighbor points on the local reference surface and the local template surface is minimized;
projecting each point of the processed reference point-cloud on the corresponding local reference planar surface and each point of the processed template point-cloud on the corresponding local template planar surface, wherein each projected point from reference point-cloud is considered as origin of a local reference coordinate system and each projected point from template point-cloud is considered as origin of a local template coordinate system; and
estimating, each of the local reference surface and the local template surface by a corresponding polynomial in two variables, wherein values of the variables represent coordinates of corresponding projected point in terms of the local reference system, wherein the corresponding polynomials approximate height of the plurality of neighbor points from the local reference planar surface and the local template planar surface by minimizing the differences between the actual height of the plurality of neighbor points from the local reference planar surface and the corresponding heights as approximated by the polynomial, wherein the projected points form a smoothened surface manifold and smoothening uses differential geometry that enables approximation of a local surface by a function.

4. A system ( 102 ) for change detection in a surface represented by 3D point-cloud, wherein the system comprises a memory ( 204 ) operatively coupled to one or more hardware processors ( 202 ) and configured to store instructions configured for execution by the one or more hardware processors ( 202 ), further the system comprises:
a registration module ( 212 ) is configured to:
acquire a reference point-cloud defining a reference surface and a template point-cloud defining a template surface, wherein the reference point-cloud and the template point-cloud correspond to the 3D point-cloud of the surface acquired at different time instances; and
align the template surface with the reference surface by performing registration of the template point-cloud and the reference point-cloud, wherein the registration of the reference point-cloud and the template point-cloud is based on shape similarity based constraints;

a subsampling module ( 214 ) is configured to:
equalize point density of the registered reference point-cloud and the registered template point-cloud by performing subsampling to provide a processed reference point-cloud and a processed template point-cloud;

a change detection module ( 216 ) configured to:
reduce noise from the processed reference point-cloud and the processed template point-cloud by estimating a smooth reference surface and a smooth template surface using a locally weighted Moving Least Squares (MLS) approximation;
voxelize the smooth reference surface to generate a plurality of reference voxels and the smooth template surface to generate a plurality of template voxels, wherein each reference voxel is defined by a plurality of reference vertices and each template voxel is defined by a plurality of template vertices;
determine a corresponding reference vertex for every template vertex of each template voxel using a kd-tree, wherein the kd-tree is set with the plurality of reference vertices of a reference voxel; and
determine vertex distance between every template vertex and the determined corresponding reference vertex, wherein the determined vertex distance above a predefined v

Assignees

• Tata Consultancy Services Ltd

Inventors

• Bhowmick Brojeshwar
• Agarwal Swapna
• Sinha Sanjana
• Purushothaman Balamuralidhar
• Mallik Apurbaa

Classifications

• G06T15/04 Primary

  Texture mapping · CPC title

• G06T15/08

  Volume rendering · CPC title

• G06T2200/04

  involving 3D image data · CPC title

• G06T17/005

  Tree description, e.g. octree, quadtree · CPC title

• G06T2207/10028

  Range image; Depth image; 3D point clouds · CPC title