CN104751468A - Method for extracting human body refining center line in depth image - Google Patents

Abstract

The invention discloses a method for extracting a human body refining center line in a depth image. The method includes: acquiring a human body profile diagram, performing image smoothing processing on the human body profile diagram, adopting a depth image level division processing mode to acquire an image of a self-occlusion position, refining a body area image and a self-occlusion area image through an image refining algorithm respectively to acquire the human body refining center line. The method ingeniously integrates a refining line acquiring concept into occlusion area extraction, so that the problem of self-occlusion is solved while refining line containing depth information is acquired, and the method has the advantages of instantaneity, high efficiency and the like.

Description

A kind of method extracting human body refinement center line in depth image

Technical field

The present invention relates to technical field of image processing, particularly relate to a kind of method extracting human body refinement center line in depth image.

Background technology

Along with the fast development of computer technology, " body sense " operative technique enters into the life of people day by day as a kind of new man-machine interaction mode, the human skeleton of robust is identified in man-machine interaction, game, safety, remote monitoring, even medical aspect has to be applied very widely, and the appearance of depth camera makes human skeleton model identification more cheap and is easy to realize.

Skeleton maintains original figure topological structure, is the strong abstract means describing object basic topology in two dimension or three dimensions.Utilize depth image carry out human skeleton extract can well overcome illumination variation, shade, object block and the interference of the factor such as environmental change.In depth image, the extraction of human body refinement center line obtains people to complete the very important condition precedent of of skeleton.

And the thinning lines of blocking and blocking position in face of position and health cannot be obtained from traditional 2D human body refinement centerline algorithms.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, a kind of method extracting human body refinement center line in depth image is provided, solving in traditional 2D thinning algorithm and cannot be obtained from the problem of blocking the thinning lines of blocking position in face of position and health, is a kind of new method that can adapt to depth data.

The object of the invention is to be achieved through the following technical solutions: a kind of method extracting human body refinement center line in depth image, it comprises following multiple step:

S1: to the smoothing filtering process of human body contour outline figure.

S2: carry out the process of depth image Hierarchical Segmentation for human body contour outline figure, obtains the image of occlusion area in face of health.

S3: by image thinning algorithm respectively to inside and outside two-layer image, namely occlusion area image in face of body region image and health, carries out thinning processing, obtains human body refinement center line.

The detailed process of the disposal of gentle filter described in step S1 is: first the human body contour outline figure of acquisition is carried out convergent-divergent, use medium filtering to its smoothing process, again the human body contour outline figure after smoothing processing is amplified to life size, reuse medium filtering to its smoothing process, obtain level and smooth human body contour outline figure.

In face of acquisition health described in step S2, the detailed process of the image of occlusion area is:

S201: the deletion template T calculating each pixel, adopt iterative processing mode to calculate the depth value P of each pixel of body region ₀with the depth value P of its 3 × 3 neighborhood territory pixel _i(i=1,2 ..., 8).

S202: the depth value P judging current point ₀with the depth value P of neighborhood territory pixel _idegree of depth difference DELTA P, if this degree of depth difference DELTA P meets Δ P=P _i-P ₀> Threshold, then make the deletion template T of this field pixel _i(i=1,2 ..., 8) value be T _i=1, otherwise template T is deleted in order _i=0, wherein Threshold value is the threshold value preset.

S203: mated with the deletion template T ' in delete list by current deletion template T, if containing this current deletion template T in delete list, then by the depth value P of this current pixel point ₀change P ' into ₀, P ' ₀computing formula be:

$P_0^{\′}=\frac{1}{n}{\mathrm{\Σ}}_{i=1}^8{P}_i(T_i=1)$

In formula, P ' ₀for the amended depth value of current pixel point, n is T _ithe number of=1;

Again by the original depth value P of this current pixel point ₀be kept in an empty image, as extracting occlusion area image;

If current deletion template T does not delete template T with any one in delete list ^'coupling, then proceed the judgement of next pixel, do not modify to currency;

S204: after the process that iterates, occlusion area image in face of body region image and health can be obtained.

The detailed process obtaining human body refinement center line described in step S3 is:

S301: the deletion template calculating each pixel of body region image and occlusion area image respectively, adopts iterative processing mode to calculate the depth value P of each pixel ₀with the depth value P of its 3 × 3 neighborhood _i(i=1,2 ..., 8).

S302: the depth value P judging neighborhood territory pixel _iif meet P _i> 0, then make the deletion template T of this field pixel _i(i=1,2 ..., 8) value be T _i=1, otherwise the deletion template T making this field pixel _i=0.

S303: mated with the deletion template T ' in delete list by current deletion template T, if containing this current deletion template T in delete list, then delete this current pixel point.

S304: if do not comprise this current deletion template T in delete list, then carry out the calculating of next pixel, repeats step S301 ~ S303.

S305: finally remaining pixel forms human body refinement center line, obtains blocking position thinning lines in health thinning lines in body region image and occlusion area image simultaneously.

The invention has the beneficial effects as follows: the refinement center line that can obtain human depth's image, cleverly thinning lines being obtained thought is dissolved in occlusion area extraction, solve from occlusion issue, obtain the thinning lines comprising depth information simultaneously, there is the advantages such as real-time, efficient.

Accompanying drawing explanation

Fig. 1 is a kind of schematic flow sheet extracting the method for human body refinement center line in depth image of the present invention;

Fig. 2 is the schematic diagram of current pixel and neighborhood distribution thereof in the present invention;

Fig. 3 is the schematic diagram being used for recording the deletion template calculating gained in the present invention;

Fig. 4 is the schematic diagram of delete list in the present invention;

Fig. 5 is the effect schematic diagram of different step gained in the present invention.

Embodiment

Below in conjunction with accompanying drawing, technical scheme of the present invention is described in further detail, but protection scope of the present invention is not limited to the following stated.

As shown in Figure 1, a kind of method extracting human body refinement center line in depth image proposed by the invention, first human body contour outline figure is obtained, picture smooth treatment is carried out to this human body contour outline figure, then depth image Hierarchical Segmentation processing mode is adopted, obtaining the image from blocking position, carrying out thinning processing to body region image with from occlusion area image respectively finally by image thinning algorithm, obtaining human body refinement center line.

The concrete implementation step of the inventive method is as follows:

Step one, to the smoothing filtering process of human body contour outline figure.

The Kinect for Windows that available Microsoft provides obtains the human body contour outline figure of 640 × 480 pixel sizes.The human body contour outline figure of acquisition is carried out zooming to 160 × 120 pixel sizes, use medium filtering to its smoothing process, depth value absent region is compensated, be amplified to life size again, and then use the smoothing process of the profile of medium filtering to this image, the human body contour outline figure that final acquisition is level and smooth.

Step 2, depth image Hierarchical Segmentation, obtains the image of occlusion area in face of health, namely from the image blocking position, comprises following multiple sub-step:

S201: current pixel point is set up to the domain template of 3 × 3

Calculate the deletion template T of each pixel, adopt iterative processing mode to calculate the depth value P of each pixel of body region ₀with the depth value P of its 3 × 3 neighborhood territory pixel _i(i=1,2 ..., 8).Foundation current pixel as shown in Figure 2 and the schematic diagram of neighborhood distribution thereof, wherein, P0 point is the current pixel point extracted, and P1 ~ P8 point is the field point around P0 point.

S202: set up the deletion template of 3 × 3

Judge the depth value P of current point ₀with the depth value P of neighborhood territory pixel _idegree of depth difference DELTA P, if this degree of depth difference DELTA P meets Δ P=P _i-P ₀> Threshold, in formula, Threshold value is the threshold value preset, then make the deletion template T of this field pixel _i(i=1,2 ..., 8) value be T _i=1, otherwise the deletion template T making this field pixel _i=0.Set up the schematic diagram being used for recording the deletion template calculating gained as shown in Figure 3, wherein, T0 point is current pixel point, and T1 ~ T8 point is the field point around T0 point.

S203: current deletion template and delete list are carried out mating contrast

Current deletion template T is mated with the deletion template T ' in delete list, by the deletion template T of current pixel ₀and 3 × 3 deletion template T of neighborhood territory pixel _imate with the deletion template T ' in delete list, if containing this current deletion template T in delete list, then by the depth value P of this current pixel point ₀change P ' into ₀, P ' ₀computing formula be:

$P_0^{\′}=\frac{1}{n}{\mathrm{\Σ}}_{i=1}^8{P}_i(T_i=1)$

In formula, P ' ₀for the amended depth value of current pixel point, n is T _ithe number of=1.

Again by the original depth value P of this current pixel point ₀be kept in an empty image, as extracting occlusion area image, as hand region image.

Mate if current deletion template T does not delete template T ' with any one in delete list, then proceed the judgement of next pixel, currency is not modified.

As shown in Figure 4, Fig. 4 is the schematic diagram of delete list, comprises eight kinds and deletes template T ' ₁~ T ' ₈, wherein, " 1 " represents that the depth value of this pixel is 1, and " 0 " represents that the depth value of this pixel is 0, and " * " represents that the depth value of this pixel can be 1, also can be 0, and that is, actual in the delete list shown in Fig. 4 have 32 to delete template.Delete template T ' for these eight kinds ₁~ T ' ₈there is certain rule.

When some deletion template T ' match in deletion template T and delete list, represent that this pixel meets the treatment conditions of its depth value of amendment, just by the depth value P of this current pixel point ₀change into

S204: after the process that iterates, namely repeat step S201 ~ S203, until in matching process, till not having the change of pixel depth value, occlusion area image in face of body region image and health can be obtained.

Described body region image is smooth body region image, and in face of described health, occlusion area image comprises the hand region image blocked in face of health, the head zone image etc. blocking the foot areas image in face of health and block in face of health.

Step 3, carries out refinement by image thinning algorithm respectively to inside and outside two-layer image, obtains human body refinement center line, comprises following multiple sub-step:

S301: set up the domain template of 3 × 3

Calculate the deletion template of each pixel of body region image and occlusion area image respectively, adopt iterative processing mode to calculate the depth value P of each pixel ₀with the depth value P of its 3 × 3 neighborhood _i(i=1,2 ..., 8);

S302: set up the deletion template of 3 × 3

Judge the depth value P of neighborhood territory pixel _iif meet P _i> 0, then make the deletion template T of this field pixel _i(i=1,2 ..., 8) value be T _i=1, otherwise the deletion template T making this field pixel _i=0;

S303: mate with delete list

Current deletion template T is mated with the deletion template T ' in delete list, if containing this current deletion template T in delete list, then delete this current pixel point;

S304: if do not comprise this current deletion template T in delete list, then carry out the calculating of next pixel, repeats step S301 ~ S303;

S305: obtain human body refinement center line

Finally remaining pixel forms human body refinement center line, obtains blocking position thinning lines in health thinning lines in body region image and occlusion area image simultaneously.

Block the hand thinning lines that position thinning lines comprises the hand region image blocked in face of health.

As shown in Figure 5, Fig. 5 is the effect schematic diagram of different step gained in the present invention, the result of different disposal step is shown in Fig. 5, human body contour outline figure after first behavior smoothing processing, occlusion area image in face of the health extracted after second behavior depth image Hierarchical Segmentation process, the lines in the third line are the human body refinement center line by obtaining after image thinning algorithm process.

The present invention accurately can obtain the refinement center line of occlusion area in face of the refinement center line of the person and the person, accurately can obtain human body both hands when all not blocking health, the human body refinement center line of human body one hand when blocking health and when human body both hands all block health.First row in Fig. 5 is the processing procedure schematic diagram of human body both hands when blocking health, secondary series is the processing procedure schematic diagram of human body one hand when blocking health, 3rd is classified as processing procedure schematic diagram when human body both hands all do not block health, and the 4th is classified as human body both hands blocks health and blocks position and body centre's line processing when interleaving process schematic.

Claims (4)

1. extract a method for human body refinement center line in depth image, it is characterized in that: it comprises following multiple step:

S1: to the smoothing filtering process of human body contour outline figure;

S2: carry out the process of depth image Hierarchical Segmentation for human body contour outline figure, obtains the image of occlusion area in face of health;

S3: by image thinning algorithm respectively to inside and outside two-layer image, namely occlusion area image in face of body region image and health, carries out thinning processing, obtains human body refinement center line.

2. a kind of method extracting human body refinement center line in depth image according to claim 1, it is characterized in that: the detailed process of the disposal of gentle filter described in step S1 is: first the human body contour outline figure of acquisition is carried out convergent-divergent, use medium filtering to its smoothing process, again the human body contour outline figure after smoothing processing is amplified to life size, reuse medium filtering to its smoothing process, obtain level and smooth human body contour outline figure.

3. a kind of method extracting human body refinement center line in depth image according to claim 1, is characterized in that: in face of the acquisition health described in step S2, the detailed process of the image of occlusion area is:

S201: the deletion template T calculating each pixel, adopt iterative processing mode to calculate the depth value P of each pixel of body region ₀with the depth value P of its 3 × 3 neighborhood territory pixel _i(i=1,2 ..., 8);

S202: the depth value P judging current point ₀with the depth value P of neighborhood territory pixel _idegree of depth difference DELTA P, if this degree of depth difference DELTA P meets Δ P=P _i-P ₀> Threshold, then make the deletion template T of this field pixel _i(i=1,2 ..., 8) value be T _i=1, otherwise template T is deleted in order _i=0, wherein Threshold value is the threshold value preset;

S203: mated with the deletion template T ' in delete list by current deletion template T, if containing this current deletion template T in delete list, then by the depth value P of this current pixel point ₀change P ' into ₀, P ' ₀computing formula be:

$P_0^{\′}=\frac{1}{n}{\mathrm{\Σ}}_{i=1}^8{P}_i(T_i=1)$

In formula, P ' ₀for the amended depth value of current pixel point, n is T _ithe number of=1;

Again by the original depth value P of this current pixel point ₀be kept in an empty image, as extracting occlusion area image;

Mate if current deletion template T does not delete template T ' with any one in delete list, then proceed the judgement of next pixel, currency is not modified;

S204: after the process that iterates, occlusion area image in face of body region image and health can be obtained.

4. a kind of method extracting human body refinement center line in depth image according to claim 1, is characterized in that: the detailed process obtaining human body refinement center line described in step S3 is:

S301: the deletion template calculating each pixel of body region image and occlusion area image respectively, adopts iterative processing mode to calculate the depth value P of each pixel ₀with the depth value P of its 3 × 3 neighborhood _i(i=1,2 ..., 8);

S302: the depth value P judging neighborhood territory pixel _iif meet P _i> 0, then make the deletion template T of this field pixel _i(i=1,2 ..., 8) value be T _i=1, otherwise the deletion template T making this field pixel _i=0;

S303: mated with the deletion template T ' in delete list by current deletion template T, if containing this current deletion template T in delete list, then delete this current pixel point;

S304: if do not comprise this current deletion template T in delete list, then carry out the calculating of next pixel, repeats step S301 ~ S303;

S305: finally remaining pixel forms human body refinement center line, obtains blocking position thinning lines in health thinning lines in body region image and occlusion area image simultaneously.