CN103714555B - A kind of four-dimensional motion point cloud segmentation and method for reconstructing based on movement locus - Google Patents

Abstract

The invention relates to the field of computer graphics, and provides a four-dimensional motion point cloud segmentation and reconstruction method based on motion trajectories. Realized the skeleton extraction of motion trajectory, including: acquiring motion data, performing adjacent data registration, motion trajectory data extraction, cluster analysis of motion trajectory data, and consistent skeleton extraction of motion data after cluster analysis. The present invention does not need to define a template in advance, and does not need to first extract the skeleton of the data of each frame, and has lower requirements on the quality of the point cloud.

Description

A 4D Motion Point Cloud Segmentation and Reconstruction Method Based on Motion Trajectories

technical field

The invention relates to the field of computer graphics, in particular to a four-dimensional motion point cloud segmentation and reconstruction method based on motion trajectories.

Background technique

Existing motion capture and segmentation require templates or require high quality point clouds, or extract skeletons from single-frame point clouds for consistent optimization and processing of multi-frame skeletons.

Most of the segmentation of objects in the existing technology is based on the geometric features of the point cloud, which lacks the continuity and consistency of time and space; the existing point cloud skeleton reconstruction method that pursues the consistency of time and space has relatively high requirements for the quality of the input point cloud. High, because they need to extract a rough skeleton from the input time-series point cloud, and then optimize the consistency of time and space.

The present invention does not need to extract the skeleton in advance, and the consistency of the skeleton has been considered in the process of extraction.

Contents of the invention

The present invention adopts a four-dimensional motion point cloud segmentation and reconstruction method based on the motion trajectory, and realizes the skeleton extraction of the motion trajectory. The quality requirement is low. The present invention adopts following scheme:

A four-dimensional motion point cloud segmentation and reconstruction method based on motion trajectory, including:

S1. Obtain motion data;

S2. Perform adjacent frame data registration on the motion data;

S3. Extracting motion track data from the data registered in step S2;

S4. Perform cluster analysis on the motion track data described in step S3;

S5. Extracting a consistent skeleton from the data obtained after the cluster analysis described in step S4.

Preferably, the acquisition of motion data adopts a laser scanner-based motion data capture method,

Preferably, the laser scanner continuously scans and acquires the point cloud data of the moving object at a certain frame rate; the acquired data is stored in the computer in the form of one file per frame.

Preferably, the method for registering the data of adjacent frames of the motion data is implemented by using a non-rigid matching method, and the non-rigid matching method finds a corresponding point in the next frame for each point in each frame.

Preferably, the method of extracting the motion trajectory data of the data registered in step S2 is to use the depth-first method to connect and grow the motion trajectory between adjacent two adjacent frames with similar directions, so as to obtain The trajectory of movement between points.

Preferably, the method for performing cluster analysis on the motion trajectory data described in step S3 is, based on the distance formula of the motion trajectory, using spectral clustering or kmean clustering method, and dividing the moving object into different groups according to the similarity of the motion trajectory data. the movement part;

Preferably, it is characterized in that, extracting the consistent skeleton of the data obtained after the cluster analysis in step S4 further includes the step of calculating a consistent skeleton for the segmented part according to the adjacency relationship.

Preferably, the method of calculating the consistent skeleton according to the adjacency relationship is to find the corresponding relationship from the adjacent nodes and the adjacent nodes to the corresponding points of the next frame for the nodes of the skeleton, so as to convert the key nodes of the skeleton to The next frame, iteratively, can be converted into N frame data.

A four-dimensional motion point cloud segmentation and reconstruction method based on motion trajectory disclosed by the present invention obtains motion data, performs adjacent data registration, motion trajectory data extraction, motion trajectory data cluster analysis, and motion data after cluster analysis Consistent skeleton extraction is carried out to realize the skeleton extraction of the motion track. The present invention does not need to define a template in advance, and does not need to first perform skeleton extraction on the data of each frame, and has lower requirements on point cloud quality.

Description of drawings

1 is a flowchart of a four-dimensional motion point cloud segmentation and reconstruction method based on motion trajectory in Embodiment 1 of the present invention;

Fig. 2 is the four-frame point cloud of Embodiment 1 of the present invention as an example;

FIG. 3 is an example after adjacent frame data registration in Embodiment 1 of the present invention;

Fig. 4 is the connection relationship between point 1 and the point after alignment in the embodiment of the present invention;

Fig. 5 is the motion locus between points among multiple frames in Embodiment 1 of the present invention;

Fig. 6 is a visual display diagram of the distance definitions of different trajectories in Embodiment 1 of the present invention;

Fig. 7 is some trajectories before the cluster analysis of Embodiment 1 of the present invention;

Fig. 8 is the motion trajectory after cluster analysis of Embodiment 1 of the present invention.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

An embodiment of the present invention provides a four-dimensional motion point cloud segmentation and reconstruction method based on motion trajectory, including:

S1. Obtain motion data;

S2. Perform adjacent frame data registration on the motion data;

S3. Extracting motion track data from the data registered in step S2;

S4. Perform cluster analysis on the motion trajectory data described in step S3.

S5. Extracting a consistent skeleton from the data obtained after the cluster analysis described in step S4.

The embodiment of the present invention realizes the skeleton extraction of the motion trajectory by acquiring the motion data, performing adjacent data registration, motion trajectory data extraction, and cluster analysis of the motion trajectory data, and performing consistent skeleton extraction on the motion data after the cluster analysis. The invention does not require a template to be defined in advance, nor does it need to extract the skeleton of each frame of data first, and has lower requirements on the quality of the point cloud. The present invention is described in detail below.

Example 1:

Please refer to FIG. 1 , which is a flowchart of a four-dimensional motion point cloud segmentation and reconstruction method based on motion trajectory in the present invention. The method comprises the steps of:

S1. Obtain motion data;

This embodiment provides a four-frame point cloud as an example, as shown in Figure 2, using a motion data capture method based on a laser scanner, the laser scanner continuously scans and acquires point cloud data of moving objects at a certain frame rate; after acquisition The data is stored in the computer as one file per frame.

S2. Perform adjacent frame data registration on the motion data;

The adjacent frame data registration of motion data is realized by non-rigid matching method. The matching method is to find a corresponding point in the next frame for each point in each frame, as shown in Figure 3. The right point in each frame is The point after registration, the connection relationship between the point and the point after alignment is shown in Figure 4.

S3. Extracting motion track data from the data registered in step S2;

Extract the motion trajectory data from the data registered in step S2, and use the depth-first method to connect and grow the motion trajectories between adjacent two adjacent frames with similar directions, so as to obtain the point-to-point distance between multiple frames. The trajectory of the movement is shown in Figure 5.

S4. Perform cluster analysis on the motion track data described in step S3;

Carrying out cluster analysis on the motion trajectory data described in step S3, based on the distance formula of the motion trajectory, using spectral clustering or kmean clustering method, dividing the moving object into different moving parts according to the similarity of the motion trajectory data;

The Euclidean distance algorithm is used to calculate the similarity between the trajectory data, considering the relationship between the Euclidean distance between the trajectories and the direction of motion. During cluster analysis, N trajectories are randomly selected as cluster centers, and then similar trajectory data are continuously added to the nearest cluster, and the cluster centers are updated at the same time. This step is iterated until all trajectories are clustered or satisfied. other conditions specified by the user. The definition calculations for different distances use the following methods:

d1 = Euclidean distance;

d2 = angle α;

d = normalized(d1), normalized(d2)

The distance definitions of different trajectories are intuitively shown in Figure 6.

Several motion trajectories before cluster analysis are shown in Figure 7, and motion trajectories after cluster analysis are shown in Figure 8

S5. Performing consistent skeleton extraction on the data obtained after the cluster analysis described in step S4;

Calculate the consistent skeleton according to the adjacency relationship, including the nodes of the skeleton, find the corresponding relationship from the adjacent nodes and the adjacent nodes to the corresponding points of the next frame, so as to convert the key nodes of the skeleton to the next frame, and iterate It can be converted to N frame data.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention should be included in the protection of the present invention. within range.

Claims (7)

1. a kind of four-dimensional motion point cloud segmentation and method for reconstructing based on movement locus, it is characterised in that including：

S1, acquisition exercise data；

S2, consecutive frame Registration of Measuring Data is carried out to the exercise data；

S3, to by step S2 registration after data carry out motion trace data extraction；

S4, cluster analysis is carried out to the motion trace data described in step S3；

S5, the data to being obtained after cluster analysis described in step S4 carry out uniformity skeletal extraction；

It is that the range formula based on movement locus is adopted to the method that the motion trace data described in step S3 carries out cluster analysis By moving meshes it is different motion parts according to motion trace data with spectral clustering or kmean clustering procedures；

Calculated according to motion trace data and use Euclidean distance algorithm, during cluster analysis, randomly select N bars track as in cluster , then be constantly added to track data in the middle of nearest cluster by the heart, while updating cluster centre, this step iteration proceeds to Untill all of track is clustered, the definition of different distance is calculated and made with the following method：

D1=Euclidean distances；

D2=angle αs；

D=normalization (d1)+normalization (d2).

2. method according to claim 1, it is characterised in that the acquisition exercise data, using based on laser scanner Exercise data catching method.

3. method according to claim 2, it is characterised in that laser scanner continuously carries out a cloud number to moving object According to scanning and acquisition；During data after acquisition are stored in computer in the form of one file of every frame.

4. method according to claim 1, it is characterised in that the side of consecutive frame Registration of Measuring Data is carried out to the exercise data Method is to be realized using the method for non-rigid matching, the non-rigid matching process be each point in each frame in the next frame Find a corresponding points.

5. method according to claim 1, it is characterised in that to carrying out movement locus by the data after step S2 registrations The method that data are extracted is that movement locus is connected and grows between adjacent two frame using the method for depth-first by adjacent direction, So as to obtain the movement locus between multiframe between points.

6. method according to claim 1, it is characterised in that the data to being obtained after cluster analysis described in step S4 are carried out Uniformity skeletal extraction also includes the step of part completed to segmentation calculates the skeleton of uniformity according to syntople.

7. method according to claim 6, it is characterised in that the method that the skeleton of uniformity is calculated according to syntople To the node of skeleton, corresponding relation to be found from adjacent node and adjacent node to the corresponding points of next frame, so that will The key node of skeleton is transformed into next frame, and iteration is carried out, in may switch to N frame data.