Living body surface damage depth estimation method based on point cloud and medium

Abstract

The invention relates to a point cloud-based living body surface damage depth estimation method and a medium, and the method comprises the steps: pre-defining a living body surface damage depth, collecting a multi-modal image of a living body, and constructing an RGB-D-T data set of the living body surface damage depth based on the multi-modal image; performing living body surface damage candidate region extraction processing on the RGB-D-T data set to obtain a living body surface damage candidate region; according to the RGB-D-T data set and the living body surface injury candidate area, generating and obtaining living body surface injury candidate point clouds, and then screening the living body surface injury candidate point clouds to obtain living body surface injury target point clouds; and processing the living body surface damage target point cloud to obtain the depth of the living body surface damage. Thus, the multi-modal image of the living body surface is collected in a non-contact mode, and the living body surface damage point cloud is accurately extracted by processing the point cloud of the living body surface candidate area, so that the living body surface damage depth is further accurately estimated, and the estimation precision of the damage depth is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of living body surface damage detection, and in particular to a living body surface damage depth estimation method based on point cloud and a medium. BACKGROUND

[0002] Accurate quantitative evaluation of living body surface damage is a core technical requirement for animal welfare and production efficiency. In the field of economic animal breeding, early diagnosis of microscopic structural damage such as poultry feather damage and livestock skin ulceration has important industrial value.

[0003] In order to evaluate the living body surface damage, the current living body surface damage evaluation method mainly relies on manual visual evaluation or two-dimensional image analysis evaluation. Manual visual evaluation has the defects of relative subjectivity and poor repeatability, such as the scores of the same chicken by two evaluators may be different in the evaluation of poultry feather damage. Two-dimensional image analysis evaluation has a large error in dynamic breeding scenarios and is difficult to obtain damage depth values. In recent years, some scholars have used infrared thermal imaging to evaluate living body surface damage and have made some progress. However, the existing living body surface damage evaluation methods based on infrared thermal imaging are mostly based on two-dimensional thermal infrared images, which cannot evaluate the depth of living body surface damage. The depth of damage (or depth) directly affects the health of animals, and when the damage depth is deep, it may cause serious skin damage or even death.

[0004] Therefore, how to make more accurate and non-contact living body surface damage depth estimation for living body surface damage depth to improve the objectivity and automation of damage evaluation has become a technical problem to be solved in the field of current living body nondestructive testing technology development. SUMMARY

[0005] The first technical problem to be solved by the present application is to provide a living body surface damage depth estimation method based on point cloud which can accurately and non-contactly estimate the depth of living body surface damage.

[0006] The second technical problem to be solved by the present application is to provide a readable storage medium. The readable storage medium stores a computer program which is executed by a processor to implement the living body surface damage depth estimation method based on point cloud.

[0007] The technical solution adopted by the present application to solve the first technical problem is: a living body surface damage depth estimation method based on point cloud, characterized in that it comprises the following steps: Step 1, defining the depth of living body surface damage in advance; Step 2, collecting multi-modal images of the living body and constructing an RGB-D-T data set of the living body surface damage depth based on the multi-modal images; wherein the multi-modal images of the living body are synchronous color images, depth images and thermal infrared images; the RGB-D-T data set contains synchronous color images, depth images and thermal infrared images; Step 3, performing living body surface damage candidate region extraction processing on the RGB-D-T data set to obtain a living body surface damage candidate region; Step 4, generating a living body surface damage candidate point cloud according to the obtained RGB-D-T data set and the living body surface damage candidate region; Step 5, performing screening processing on the obtained living body surface damage candidate point cloud to obtain a living body surface damage target point cloud; Step 6, performing processing on the obtained living body surface damage target point cloud to obtain a living body surface damage depth.

[0008] In the improved method, in step 1, the living body surface damage depth is defined as follows: Step a1, selecting the longest diameter in all damage regions of the living body as the target straight line; Step a2, selecting the longest concave point to the target straight line in all damage regions of the living body as the target concave point; Step a3, defining the distance from the target concave point to the target straight line as the living body surface damage depth.

[0009] In the improved method, in step 2, the process of collecting multi-modal images of the living body and constructing an RGB-D-T data set of the living body surface damage depth based on the multi-modal images includes the following steps: Step b1, using a binocular visible light camera to collect a color left image and a color right image of the living body; wherein the color left image is marked as I L , and the color right image is marked as I R ; Step b2, using a thermal infrared camera to collect a thermal infrared image of the living body; wherein the thermal infrared image of the living body is marked as I ir_raw ; Step b3, performing stereo matching and disparity estimation processing on the obtained color left image and color right image to obtain a depth image; wherein the depth image is marked as I d ; Step b4, performing image registration processing on the obtained thermal infrared image and color left image to obtain a registered thermal infrared image; wherein the registered thermal infrared image is marked as I ir '; Step b5, image segmentation processing is performed on the obtained color left image to obtain a mask image; wherein the mask image is marked as I mask ; Step b6, logical AND operation processing is performed on the obtained mask image and the registered thermal infrared image to obtain a segmented thermal infrared image; wherein the segmented thermal infrared image is marked as I ir .

[0010] Further, in the point cloud-based living body surface damage depth estimation method, in step 3, the process of performing living body surface damage candidate region extraction processing on the RGB-D-T data set includes the following steps: Step c1, the segmented thermal infrared image is sequentially subjected to RGB three-channel separation and threshold segmentation processing to obtain an R channel image, a G channel image and a B channel image; wherein the R channel image is marked as I r , the G channel image is marked as I g , and the B channel image is marked as I b ; Step c2, logical operation processing is performed on the obtained R channel image, G channel image and B channel image to obtain a feature candidate image containing multiple potential target feature regions, and the feature candidate image is taken as the living body surface damage candidate region; wherein the feature candidate image is marked as M C .

[0011] Improved, in the point cloud-based living body surface damage depth estimation method, in step 4, according to the obtained RGB-D-T data set and the living body surface damage candidate region, the process of generating a living body surface damage candidate point cloud includes the following steps: Step d1, logical AND operation processing is performed on the segmented thermal infrared image and the feature candidate image to obtain a feature candidate thermal infrared image; wherein the feature candidate thermal infrared image is marked as M C_ir ; Step d2, the obtained feature candidate thermal infrared image is projected into a three-dimensional coordinate system to obtain a feature candidate point cloud; wherein the feature candidate point cloud is marked as P1; Step d3, a Radius-Outiler-Removal filtering algorithm is used to perform point cloud denoising processing on the feature candidate point cloud to obtain a denoised feature candidate point cloud, and the denoised feature candidate point cloud is taken as the living body surface damage candidate point cloud; wherein the denoised feature candidate point cloud is marked as P2.

[0012] Further, in the point cloud-based living body surface damage depth estimation method, in step 5, the process of performing screening processing on the obtained living body surface damage candidate point cloud includes the following steps: Step e1, the denoised feature candidate point cloud is clustered by using a DBSCAN clustering method to obtain a clustering set;Wherein, the clustering set is marked as C, C={C i},i=1,2,…,N;N is the total number of multiple potential target feature regions contained in the feature candidate graph;C i The i-th cluster in the clustering set, each cluster in the clustering set corresponds to each potential target feature region in the feature candidate graph one by one; Step e2, each cluster in the obtained clustering set is respectively subjected to concave-convex measurement processing, and the corresponding concave index is obtained, and all the obtained concave indexes form a concave index set;Wherein, the concave index set is marked as DI, DI={DI i},i=1,2,…,N;DI i The i-th cluster C i Corresponding concave index; Step e3, the point cloud corresponding to the cluster corresponding to the concave index with the largest value in the concave index set is taken as the living body surface damage point cloud;Wherein, the living body surface damage point cloud is marked as P3.

[0013] Further improvement, in the living body surface damage depth estimation method based on point cloud, in step 6, the process of evaluating the obtained living body surface damage target point cloud includes the following steps: Step f1, the living body surface damage point cloud is detected by using the minimum bounding box OBB detection to obtain the damage region minimum bounding box; Step f2, taking the height of the damage region minimum bounding box as the living body surface damage depth.

[0014] The technical scheme adopted by the present application to solve the second technical problem is: a readable storage medium, the readable storage medium stores a computer program, the computer program is executed by a processor, and any one of the living body surface damage depth estimation methods based on point cloud is realized.

[0015] Compared with the prior art, the present application has the following advantages: Firstly, the point cloud-based living body surface damage depth estimation method of the application defines the living body surface damage depth in advance, then collects multi-modal images of the living body, constructs an RGB-D-T data set of the living body surface damage depth based on the multi-modal images, and extracts the living body surface damage candidate region from the RGB-D-T data set to obtain the living body surface damage candidate region; then, the living body surface damage candidate point cloud is generated according to the obtained RGB-D-T data set and the living body surface damage candidate region, and the living body surface damage target point cloud is obtained by screening the obtained living body surface damage candidate point cloud; and the living body surface damage depth is obtained by processing the obtained living body surface damage target point cloud. In this way, the multi-modal images of the living body surface are collected in a non-contact manner, and the living body surface damaged point cloud is accurately extracted by processing the living body surface candidate region point cloud, so as to further accurately estimate the living body surface damage depth, thereby significantly improving the estimation accuracy of the damage depth.

[0016] Secondly, the point cloud-based living body surface damage depth estimation method of the application further separates different feature candidate point clouds by clustering the obtained living body surface damage target point cloud, obtains a clustering set, and further performs concave-convex measurement processing on each cluster in the clustering set to obtain a corresponding concave index, and forms a concave index set from all the obtained concave indexes, and the point cloud corresponding to the cluster corresponding to the concave index with the largest value in the concave index set is taken as the living body surface damage point cloud. In this way, by introducing the concave-convex index for the living body surface damage target point cloud, the actual situation of the existence of damage depth in the living body surface damage area is more in line with the actual situation, and the confusion between the living body surface damage area and the head of the living body can be avoided, thereby improving the estimation accuracy of the living body surface damage depth. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 Figure 1 is a point cloud-based living body surface damage depth estimation method flowchart in an embodiment of the application; Figure 2 Figure 2 is a living body surface damage depth definition diagram in an embodiment of the application; Figure 3 Figure 3 is a segmented thermal infrared image in an embodiment of the application; Figure 4 Figure 4 is a feature candidate image in an embodiment of the application; Figure 5 Figure 5 is a living body surface damage depth estimation diagram in an embodiment of the application. DETAILED DESCRIPTION

[0018] The application will be further described in detail below with reference to the embodiments of the drawings.

[0019] As Figure 1As shown, the embodiment provides a point cloud-based living body surface damage depth estimation method. In this embodiment, the living body is a chicken, and the living body surface is the feather of the chicken. The living body surface damage depth estimation method is used to evaluate the feather damage depth of the chicken. Specifically, the point cloud-based living body surface damage depth estimation method includes the following steps 1-6: Step 1, defining the living body surface damage depth in advance; in this embodiment, referring to Figure 2 As shown, the definition of the living body surface damage depth in advance is as follows: Step a1, selecting the longest diameter AB in all damage regions of the living body as the target straight line, wherein the longest diameter AB represents the transverse dimension of the largest damage region; Step a2, selecting the longest concave point O in all damage regions of the living body as the target concave point, wherein the longest concave point O represents the transverse dimension of the largest damage region; Step a3, defining the distance from the target concave point O to the target straight line AB as the living body surface damage depth d; Step 2, collecting multi-modal images of the living body and constructing an RGB-D-T data set of the living body surface damage depth based on the multi-modal images; wherein the multi-modal images of the living body are synchronous color images (RGB), depth images (Depth) and thermal infrared images (Thermal); the RGB-D-T data set includes synchronous color images (RGB), depth images (Depth) and thermal infrared images (Thermal); Specifically, in this embodiment, the process of constructing the RGB-D-T data set of the living body surface damage depth includes the following steps b1-b6: Step b1, using a binocular visible light camera to collect color left and right images of the living body; wherein the color left image is marked as I L , and the color right image is marked as I R ; in this embodiment, the image resolution of the binocular visible light camera is 960 pixel*540 pixel; Step b2, using a thermal infrared camera to collect thermal infrared images of the living body; wherein the thermal infrared images of the living body are marked as I ir_raw ; in this embodiment, the image resolution of the thermal infrared camera is 384 pixel*288 pixel; Step b3, performing stereo matching and disparity estimation processing on the obtained color left and right images to obtain a depth image; wherein the depth image is marked as I d ; for example, a mature adaptive aggregation network AANet model can be used to perform stereo matching and disparity estimation processing on the color left and right images; Step b4: Perform image registration processing on the obtained thermal infrared image and the left color image to obtain the registered thermal infrared image; wherein, the registered thermal infrared image is labeled as I. ir '; Step b5: Perform image segmentation on the obtained color left image to obtain a mask image; wherein, the mask image is labeled as I. mask For example, the SegNet semantic segmentation network can be used here to segment the color left image to obtain a mask image, which includes the outline region of the complete living body. Step b6: Perform a logical AND operation on the obtained mask image and the registered thermal infrared image to obtain the segmented thermal infrared image; wherein, the segmented thermal infrared image is labeled as I. ir For example, see the segmented thermal infrared image obtained in this embodiment. Figure 3 As shown; Step 3: Extract candidate regions for live surface damage from the RGB-DT dataset to obtain candidate regions for live surface damage. For example, the process of extracting candidate regions for live surface damage on the RGB-DT dataset includes the following steps c1~c2: Step c1, process the segmented thermal infrared image I ir RGB three-channel separation was performed, and threshold segmentation was applied using the Otsu method to obtain the R-channel image, G-channel image, and B-channel image, respectively; the R-channel image is labeled as I. r The G channel diagram is labeled as I. g The B channel diagram is labeled as I. b ; Step c2 involves performing logical operations on the obtained R-channel, G-channel, and B-channel images. By fusing multi-channel temperature features, the robustness of damage candidate region extraction is improved, resulting in a feature candidate image containing multiple potential target feature regions. This feature candidate image is then used as a candidate region for damage on the living surface. The feature candidate image is labeled M. C ; For example, in this embodiment, the obtained G-channel map I g Perform a logical inversion operation to obtain the inverted G-channel diagram I. g_r ; Segment the R channel into image I r Inverted G-channel diagram I g_r Performing a logical AND operation yields the initial transformation image M1, which then divides the B channel into image I. b Performing a logical AND operation between the primary transformed graph M1 and the secondary transformed graph M2 yields the secondary transformed graph M2. Then, performing a logical XOR operation between the primary transformed graph M1 and the secondary transformed graph M2 yields the feature candidate graph M. C The feature candidate map M C It contains N potential target feature regions, where N > 0; the feature candidate map obtained in this embodiment can be found in [reference needed].Figure 4 as shown; Step 4, generating a living body surface damage candidate point cloud according to the obtained RGB-D-T data set and the living body surface damage candidate region; wherein in this embodiment, the process of generating a living body surface damage candidate point cloud according to the obtained RGB-D-T data set and the living body surface damage candidate region includes the following steps d1~d3: Step d1, projecting the segmented thermal infrared image I ir and the feature candidate map M C into the three-dimensional coordinate system to obtain a feature candidate thermal infrared map; wherein the feature candidate thermal infrared map is marked as M C_ir ; Step d2, projecting the obtained feature candidate thermal infrared map M C_ir into the three-dimensional coordinate system to obtain a feature candidate point cloud; wherein the feature candidate point cloud is marked as P1; Step d3, performing point cloud denoising processing on the feature candidate point cloud P1 using a Radius-Outiler-Removal filtering algorithm to eliminate outliers on the surface of the point cloud, to obtain a denoised feature candidate point cloud, and taking the denoised feature candidate point cloud as the living body surface damage candidate point cloud; wherein the denoised feature candidate point cloud is marked as P2; Step 5, performing screening processing on the obtained living body surface damage candidate point cloud to obtain a living body surface damage target point cloud; wherein in this embodiment, the living body surface damage candidate point cloud often contains the head or other easily confused parts of the living body, and the process of performing screening processing on the obtained living body surface damage candidate point cloud includes the following steps e1~e3: Step e1, performing clustering processing on the denoised feature candidate point cloud P2 using a DBSCAN clustering method to separate different feature candidate point clouds, to obtain a clustering set; wherein the clustering set is marked as C, C={C i}, i=1, 2, …, N; N is the total number of a plurality of potential target feature regions contained in the feature candidate map M C ; C i is the i-th cluster in the clustering set, and each cluster C i in the clustering set corresponds to each potential target feature region in the feature candidate map M C one-to-one; Step e2, performing concave-convex measurement processing on each cluster in the obtained clustering set respectively to obtain corresponding concave indices respectively, and forming a concave index set from all the obtained concave indices; wherein the concave index set is marked as DI, DI={DI i}, i=1, 2, …, N; DI i is the concave index corresponding to the i-th cluster C i ; Specific to this embodiment, the process of performing concave-convex metric processing on each cluster in the cluster set to obtain the corresponding concave index includes the following steps: Step e2.1, for each cluster C in the cluster set C i , an arbitrary point , a k-neighbor point set is constructed; wherein the neighbor point set is as follows: ; In the formula, k is the number of neighbor points, denotes the k-neighbor index set measured by the Euclidean distance; Step e2.2, the Gaussian curvature of the k-neighbor point set is calculated; wherein: ; ; In the formula, is a 3x3 covariance matrix, is a neighborhood centroid, is the number of neighborhood point sets, λ1, λ2, λ3 are eigenvalues of the covariance matrix , satisfying 0≤λ1≤λ2≤λ3; is the Gaussian curvature of the point , and δ is a smoothing constant; Step e2.3, the initial unit normal vector of the point is calculated, and the normal vector direction is corrected to obtain the corrected unit normal vector ; wherein: ; ; In the formula, u1 is the eigenvector corresponding to the minimum eigenvalue λ1 of the covariance matrix , d normal is a reference normal vector, pointing to the outside of the living body; Step e2.4, the Gaussian curvature and the corrected unit normal vector are fused to obtain the concave index DI i of the corresponding cluster C i ; wherein: ; ; ; In the formula, is the normal vector consistency weight, reflecting the point the degree of coincidence between the normal vector of the point cloud and the overall direction of the cluster; η is a curvature sensitivity coefficient dynamically adjusted according to the resolution of the point cloud; represents that a positive contribution is ensured only when the greater the degree of concave, the higher the contribution; is a normalization factor representing the elimination of the influence of the size difference of the cluster on the index; is the average normal vector of the cluster C i , representing the overall surface orientation of the region; is the normal vector direction variance threshold, which controls the decay rate of the normal vector consistency weight; Step e3, the point cloud corresponding to the cluster corresponding to the concave index with the largest value in the concave index set is taken as the living surface damage point cloud; Step 6, the obtained living surface damage target point cloud is processed to obtain the living surface damage depth. In this embodiment, the process of processing the obtained living surface damage target point cloud includes the following steps: Step f1, the living surface damage point cloud is detected by using the minimum bounding box OBB detection to obtain the damage region minimum bounding box; Step f2, taking the height of the damage region minimum bounding box as the living surface damage depth. Wherein, Figure 5 is the living surface damage depth estimation schematic diagram obtained by the embodiment of the present application.

[0020] Specifically, the integrated development environment used in this embodiment is a high-performance server (Ubuntu 18.04), the experimental environment is Python 3.7.4, the server is equipped with Intel Gold 6139X2 processor (2.5 GHz / 8 Core), 256 GB memory and four NVIDIA RTX TITAN graphics cards. Through the feather damage depth estimation test of 10 chicken with damaged back feathers, the estimation results are shown in Table 1. As can be seen from Table 1, the average error between the living surface damage depth estimation value obtained by the living surface damage depth estimation method based on point cloud in this embodiment and the actual measured value is 1.93 mm. The results show that the living surface damage depth estimation method proposed in this embodiment has high enough accuracy and can accurately assist in evaluating the degree of living surface damage.

[0021] Table 1 Serial number Actual depth value (mm) Estimated depth value (mm) Error (mm) 1 5.60 4.30 1.30 2 7.10 5.60 1.50 3 8.20 6.20 2.00 4 9.50 8.30 1.20 5 10.50 8.70 1.80 6 11.00 9.10 1.90 7 12.40 10.30 2.10 8 13.00 10.80 2.20 9 14.10 11.20 2.90 10 15.30 12.90 2.40 The embodiment also provides a readable storage medium, and the readable storage medium stores a computer program. When the computer program is executed by a processor, the above-mentioned living surface damage depth estimation method based on point cloud is realized.

[0022] While the preferred embodiments of the application have been described above in detail, it is to be understood that various modifications and alterations to the preferred embodiments will occur to persons skilled in the art. Any such modifications or alterations are intended to fall within the scope of the application.

Claims

1. A method for estimating the depth of a lesion on a living surface based on a point cloud, characterized in that, The method comprises the following steps: Step 1, defining the depth of the surface damage of the living body in advance; Step 2, collecting multi-modal images of the living body and constructing an RGB-D-T data set of the depth of the surface damage of the living body based on the multi-modal images; wherein the multi-modal images of the living body are synchronous color images, depth images and thermal infrared images; the RGB-D-T data set comprises synchronous color images, depth images and thermal infrared images; Step 3, performing surface damage candidate region extraction processing on the RGB-D-T data set to obtain a surface damage candidate region of the living body; Step 4, generating a surface damage candidate point cloud of the living body according to the obtained RGB-D-T data set and the surface damage candidate region of the living body; Step 5, performing screening processing on the obtained surface damage candidate point cloud to obtain a target point cloud of the surface damage of the living body; Step 6, performing evaluation processing on the obtained target point cloud of the surface damage of the living body to obtain the depth of the surface damage of the living body.

2. The point cloud based in vivo surface lesion depth estimation method of claim 1, wherein, In step 1, the depth of the surface damage of the living body is defined in the following manner: Step a1, selecting the longest diameter of the damage region as a target straight line in all damage regions of the living body; Step a2, selecting the longest concave point to the target straight line as a target concave point in all damage regions of the living body; Step a3, defining the distance from the target concave point to the target straight line as the depth of the surface damage of the living body.

3. The point cloud based in vivo surface lesion depth estimation method of claim 2, wherein, In step 2, the process of collecting multi-modal images of the living body and constructing an RGB-D-T data set of the depth of the surface damage of the living body based on the multi-modal images comprises the following steps: Step b1, color left image and color right image of the living body are acquired by using binocular visible light camera; wherein, the color left image is marked as I L , and the color right image is marked as I R ; Step b2, acquiring a thermal infrared image of the living body using a thermal infrared camera; wherein the thermal infrared image of the living body is marked as I ir_raw ; Step b3, performing stereo matching and disparity estimation on the obtained color left image and color right image to obtain a depth image; wherein the depth image is marked as I d ; Step b4, image registration processing is performed on the obtained thermal infrared image and the color left image to obtain a registered thermal infrared image; wherein the registered thermal infrared image is marked as I ir ’; Step b5, performing image segmentation processing on the obtained color left image to obtain a mask image; wherein the mask image is marked as I mask ; Step b6, performing logical AND operation processing on the obtained mask image and the registered thermal infrared image to obtain a segmented thermal infrared image; wherein the segmented thermal infrared image is marked as I ir .

4. The method of claim 3, wherein the method further comprises: In step 3, the process of performing surface damage candidate region extraction processing on the RGB-D-T data set comprises the following steps: Step c1, the segmented thermal infrared image is sequentially subjected to RGB three-channel separation and threshold segmentation processing, and R channel map, G channel map and B channel map are obtained respectively; wherein, the R channel map is marked as I r , the G channel map is marked as I g , and the B channel map is marked as I b ; Step c2, performing logical operation processing on the obtained R channel image, G channel image and B channel image to obtain a feature candidate image containing a plurality of potential target feature regions, and taking the feature candidate image as the living body surface damage candidate region; wherein the feature candidate image is marked as M C .

5. The point cloud based in vivo surface lesion depth estimation method of claim 4, wherein, In step 4, the process of generating a surface damage candidate point cloud of the living body according to the obtained RGB-D-T data set and the surface damage candidate region of the living body comprises the following steps: Step d1, performing logical and operation processing on the segmented thermal infrared image and the feature candidate image to obtain a feature candidate thermal infrared image; wherein the feature candidate thermal infrared image is marked as M C_ir ; Step d2, projecting the obtained feature candidate thermal infrared image into a three-dimensional coordinate system to obtain a feature candidate point cloud; wherein the feature candidate point cloud is marked as P1; Step d3, performing point cloud denoising processing on the feature candidate point cloud by using a Radius-Outiler-Removal filtering algorithm to obtain a denoised feature candidate point cloud, and taking the denoised feature candidate point cloud as the surface damage candidate point cloud of the living body; wherein the denoised feature candidate point cloud is marked as P2.

6. A readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the point cloud-based depth estimation method of the surface damage of the living body according to any one of claims 1-5.

Images