CN108256517A - Laser radar-based insulator string identification method - Google Patents

Abstract

The invention relates to a laser radar-based insulator string recognition method, comprising the following steps: acquisition of a depth image: obtaining a depth image through a laser radar device installed on a robot; preprocessing of a depth image: preprocessing a depth image with an insulator string , to obtain the region of interest; feature extraction of the depth image: in the region of interest, establish a depth characteristic curve model, construct a depth periodic characteristic matrix and a width vector, and extract the characteristics of the number of insulator string segments and the change characteristics of the disk diameter; the type of insulator string Identification: Identify the type of insulator string based on the extracted features. According to the characteristics of the insulator string itself, the invention uses the gray-scale periodicity feature and the gray-scale threshold value feature to identify the insulator string. The accuracy of the recognition results is high, which improves the work efficiency and provides a basis for the subsequent substation inspection robot to complete the flushing task.

Description

Insulator String Recognition Method Based on LiDAR

technical field

The invention relates to a laser radar-based autonomous identification method applied to substation insulator strings, in particular to an automatic identification method for insulator strings based on machine vision.

Background technique

Insulators operating online are affected by atmospheric environments such as SO ₂ , nitrogen oxides, and granular dust in a natural environment, and a layer of dirt is gradually deposited on the surface. In the case of dry weather, these insulators can maintain a high insulation level, and their discharge voltage is close to that in a clean and dry state; Dirt absorbs water to dissolve the electrolyte in the polluted layer, resulting in a decrease in the insulation level of the insulator, an increase in the leakage current, and a flashover accident in severe cases. Therefore, insulator strings need to be cleaned regularly.

At this stage, the main means of cleaning insulator strings is to clean them by professionals in substations. When cleaning, there are problems such as high labor intensity and high risk.

As an advanced technical means, substation inspection robots can replace manual cleaning of insulator strings. To accomplish this task, the robot needs to automatically identify the insulator strings. Therefore, it is necessary to propose an automatic identification method for insulator strings with a high degree of automation.

Since the insulator string is in an outdoor environment, its illumination changes are complex, and the erection structure in the substation is complex. Through lidar detection, the influence of light on the sensor can be avoided, and at the same time, the influence of complex background can be overcome, and it can be well adapted to the outdoor environment. Moreover, the automatic identification method of insulator strings based on lidar has not been reported yet.

Contents of the invention

In view of the above problems, the technical problem to be solved by the present invention is to provide a laser radar-based automatic recognition method for insulator strings with high recognition accuracy.

The technical solution adopted by the present invention to solve the technical problems is: a laser radar-based insulator string identification method, comprising the following steps:

Depth image acquisition: Obtain depth images through the laser radar device installed on the robot;

Preprocessing of depth images: Preprocessing depth images with insulator strings to obtain regions of interest;

Feature extraction of depth image: In the area of interest, establish a depth characteristic curve model, construct a depth periodic characteristic matrix and a width vector, and extract the characteristics of the number of insulator strings and the variation of disk diameter;

Insulator string type identification: identify the type of insulator string based on the extracted features.

The preprocessing of the depth image comprises the following steps:

(2.1) binarize the depth image, separate the foreground and background in the depth image, and eliminate background interference;

(2.2) adopt expansion operation to the binary image after processing, then adopt corrosion operation, eliminate the interfering pixel point on the binary image;

(2.3) Then select the largest connected domain, extract the region of interest, and complete the preprocessing.

Described establishment depth characteristic curve model is as follows:

Extract the left edge of the target area:

li _k =k,

Extract the right edge of the target area:

ri _k =k,

In the formula, D(x, y) is a binary image, k represents the number of rows of the image, t represents the number of columns of the image, l _k and r _k are the pixel positions on the edge;

Position the center line of the target region by the left and right edges:

mi _k =k,

In the formula, m _k is the coordinates of the pixel points of the depth characteristic curve, mi _k is the row coordinate of the pixel point of the depth characteristic curve, mj _k is the column coordinate of the pixel point of the depth characteristic curve; n _I is the number of image rows;

Build the depth characteristic curve model:

f(x _k ,y _k )＝0

in:

x _k = mi _k

x _k is the abscissa of the characteristic curve, that is, the number of image lines, y _k is the ordinate of the characteristic curve, that is, the depth value; L(m _k ) represents the depth image.

The construction of the depth periodic feature matrix:

First find the maximum value and minimum value on the depth characteristic curve f(x _k ,y _k )=0; establish a matrix

In the formula:

v _2k is the abscissa of the extreme point on the depth characteristic curve f(x _k ,y _k )=0, v _3k is the ordinate of the extreme point on the depth characteristic curve f(x _k ,y _k )=0, namely the depth Value; k=1,2...n _I ;

Set thresholds λ _c1 , λ _c2 and λ _c1 < λ _c2 ; Perform column traversal, record V 1,k and V 2 _,k when V 3, _k+1 -V _3, k >λ _c1 and V _{3,k +1} -V 3 _,k <λ _c2 ; when At the end of the traversal, the values that meet the conditions are formed into a matrix in the order of records Record v _1,k in the matrix The first row of , v _2,k is recorded in the second row of the matrix, resulting in:

n ₃ is the number of matrix columns;

again To traverse by column, let

c _1,k ＝abs(v1 _1,k+1 -v1 _1,k )

c _2,k ＝v1 _2,k+1 -v1 _2,k

c _3,k = v1 _2,k

Get the depth periodic feature matrix

Build the width vector:

In the formula:

i=1,2...n _I

(x _i1 , y _i1 , z _i1 ) are the space coordinates of the left edge point, (x _i2 , y _i2 , z _i2 ) are the space coordinates of the right edge point; n _I represents the number of image lines.

The extraction of the feature of the number of insulator strings and the feature of disc diameter variation includes the following steps:

(1) by Extracting features of insulator strings

Firstly, the threshold α _c is set, the initialization flag δ _c is 0, and the segment information Num(C) is 0;

again Iterate by column:

When C _1,k =1, C _2,k <α _c and δ _c =0, set the flag δ _c to 1, and store C _3,k into the vector T as the upper edge of the insulator string;

When δ _c =1 and C _1,k =0 or C _2,k >α _c , set the flag δ _c to 0, store C _3,k in the vector T as the lower edge of the insulator string, and store the segment number information Num(C) plus 1;

The vector T is the upper and lower edges of the insulator string, where the odd-numbered element t _2n-1 is the upper edge of the insulator string, and the even-numbered element t _2n is the lower edge of the insulator string; n=1,2...k ₁ ; k ₁ is The length of the vector T.

(2) Extract the disc diameter change characteristics of the insulator string from W

Calculate the variance of the insulator string width according to the vector T:

When d ² <threshold α _w , the insulator has a constant disk diameter, otherwise, it has a variable disk diameter.

The identification of the insulator string type according to the extracted features includes the following steps:

If Num(C)=1, and d ² <threshold value α _w , the insulator string is a single-stage insulator string with equal disk diameter; if Num(C)=2, and d ² <threshold value α _w , then the insulator string is a double-stage insulator string For insulator strings with equal disk diameters, if Num(C)=1, and d ² >threshold value α _w , the insulator strings are single-stage variable disk insulator strings.

The present invention has the following beneficial effects and advantages:

1. The present invention utilizes laser depth periodic features, insulator string spacing features, and disk diameter features to identify insulators, which has high accuracy in identifying insulators, improves work efficiency, and provides a basis for subsequent substation inspection robots to complete tasks.

2. The present invention uses laser radar as a sensor, and can also accurately identify insulator strings under backlight conditions. In depth images, foreground and background can also be separated accurately, overcoming the influence of complex background on recognition accuracy.

Description of drawings

Fig. 1 is a schematic flow sheet of the present invention;

Figure 2 is a laser depth map;

Figure 3 is a preprocessed depth binary image;

Fig. 4 is a depth characteristic curve;

Figure 5 is a diagram of the recognition effect.

Detailed ways

The present invention will be further described in detail below in conjunction with the examples.

A method for automatic identification of insulator strings based on laser radar, comprising the following steps:

(1) Acquisition of depth image: Install laser radar equipment on the substation inspection robot, fix it on the platform, and rotate the laser radar at a fixed angle to obtain depth images;

(2) Preprocessing of the depth image: Preprocessing the depth image with insulator strings to obtain the region of interest.

(3) Feature extraction of the depth image: In the region of interest, the depth characteristic curve model is established, the depth periodic characteristic matrix and width vector are constructed, and the characteristics of the number of insulator strings and the variation characteristics of the disk diameter are extracted.

(4) Type identification of insulator strings: identify the type of insulator strings according to the extracted features.

The preprocessing of the depth image comprises the following steps:

(2.1) Through depth binarization processing, the foreground background in the depth image is separated, and the background interference in the depth image is eliminated.

(2.2) Using image morphology to process the depth binary image, first use the dilation operation, and then use the erosion operation to eliminate the interference pixels on the depth binary image.

(2.3) Select the largest connected domain in the depth binary image, extract the region of interest, and complete the preprocessing.

The feature extraction method of the depth image is as follows:

(3.1) Establish a depth characteristic curve model, the defined depth characteristic curve can reflect the characteristics of the insulator string, the method is as follows:

Extract the left edge of the target area:

li _k =k,

Extract the right edge of the target area:

ri _k =k,

In the formula, D(x, y) is the depth binary image, k represents the number of rows of the image, t represents the number of columns of the image, l _k , r _k are the pixel positions on the edge. n _I represents the maximum number of lines of the image. The position set on the center line of the target area can be determined through these two sets.

mi _k =k,

In the formula, m _k is the coordinate of the pixel point of the depth characteristic curve, mi _k is the row coordinate of the pixel point of the depth characteristic curve, and mj _k is the column coordinate of the pixel point of the depth characteristic curve.

Build the depth characteristic curve model:

f(x _k ,y _k )＝0

in:

x _k = mi _k

x _k is the abscissa of the characteristic curve, that is, the number of image lines, y _k is the ordinate of the characteristic curve, that is, the depth value; L(m _k ) represents the depth image.

(3.2) Construct the depth periodic characteristic matrix and width vector through certain rules, and extract the characteristics of the number of insulator strings and the variation characteristics of the disc diameter.

The insulator string segment number feature and width feature extracted from the depth image are analyzed to complete the identification of the insulator string.

The flow chart of the automatic identification method for insulator strings based on lidar is shown in Figure 1. The specific process is as follows:

1. LiDAR data collection

The laser radar scans the insulator string area to obtain the laser depth image L(x,y), as shown in Figure 2.

2. Image preprocessing

Binarize the depth image according to formula (1):

D(x,y) means binarization, and M is the threshold. Then carry out the dilation and corrosion operation, and mark the largest connected domain. After preprocessing, the foreground and background can be separated to obtain the insulator string and its erection. As shown in Figure 3.

3. Establish depth characteristic curve model

According to the periodic change of depth in the area where the insulator strings are located in the depth image, a depth characteristic curve model is established, the method is as follows:

Extract the left edge of the target area:

li _k =k,

Extract the right edge of the target area:

ri _k =k,

In the formula, D(x, y) is a binary image, k represents the number of rows of the image, t represents the number of columns of the image, l _k and r _k are the pixel positions on the edge. The left and right edges are respectively the left and right edges of the insulator string. Through these two sets, the position set of the center line of the target area can be determined.

mi _k =k,

In the formula, m _k is the coordinate of the pixel point of the depth characteristic curve, mi _k is the row coordinate of the pixel point of the depth characteristic curve, and mj _k is the column coordinate of the pixel point of the depth characteristic curve.

Build the depth characteristic curve model:

f(x _k ,y _k )＝0

in:

x _k = mi _k

As shown in Figure 4, the abscissa represents the number of image lines, and the ordinate represents the depth value.

4. Construct depth periodic features and width feature vectors

(1) Construct a deep periodic feature matrix:

First find the maximum value and minimum value on the depth characteristic curve f(x _k , y _k )=0. build matrix

In the formula:

v _2k is the abscissa of the extreme point on the depth characteristic curve f(x _k ,y _k )=0, v _3k is the ordinate of the extreme point on the depth characteristic curve f(x _k ,y _k )=0, namely the depth value. n ₁ means matrix The number of columns, n ₂ means matrix the number of columns.

Set the thresholds λ _c1 , λ _c2 , and λ _c1 <λ _c2 , which are 3 and 10 in this embodiment, respectively. again Perform column-wise traversal, and record v _{1,k and} v 2, _k when v 3,k+1 -v _3, k >λ _c1 and v ₃ ,k ₊₁ -v 3 _,k <λ _c2 . When the traversal ends, the values that meet the conditions are formed into a matrix in the order of records Record v _1,k in the matrix The first row of , v _2,k is recorded in the second row of the matrix, resulting in:

n ₃ is the number of matrix columns.

Then traverse V1 by column, so that

c _1,k ＝abs(v1 _1,k+1 -v1 _1,k )

c _2,k ＝v1 _2,k+1 -v1 _2,k

c _3,k = v1 _2,k

Get the depth periodic feature matrix.

(2) Build the width feature vector as follows:

In the formula:

i=1,2...n _I

(x _i1 , y _i1 , z _i1 ) are the space coordinates of the left edge point, and (x _i2 , y _i2 , z _i2 ) are the space coordinates of the right edge point. n _I represents the maximum number of lines of the image. here

5. Extract the characteristics of the number of insulator strings and the characteristics of the disc diameter

(1) by Extracting features of insulator strings

First, set the threshold α _c , which is 15 in this embodiment. The initialization flag δ _c is 0, and the segment number information Num(C) is 0.

again Iterates by column. When C _1,k =1, C _2,k <α _c and δ _c =0, set flag δ _c to 1, and store C _3,k into vector T as the upper edge of the insulator string. When δ _c =1 and C _1,k =0 or C _2,k >α _c , set the flag δ _c to 0, store C _3,k in the vector T as the lower edge of the insulator string, and store the segment number information Num(C) is incremented by 1. Traverse to the end of the matrix to complete the identification of insulator strings. The vector T is the upper and lower edges of the insulator string, where the odd-numbered element t _2n-1 is the upper edge of the insulator string, and the even-numbered element t _2n is the lower edge of the insulator string. n=1,2...k ₁ ; k ₁ is the length of the vector T.

(2) Extract the disc diameter change characteristics of the insulator string from W

Calculate the variance of the insulator string width according to the vector T:

When d ² <α _w , α _w is the threshold, which is 0.01 in this embodiment. ; Extract the disc diameter feature as equal disc diameter, otherwise, extract the disc diameter feature as variable disc diameter.

6. Identification of insulator string type

The three kinds of post insulator strings in the substation are identified according to the segment number information and disc diameter change information. Combining according to the obtained segment number features and disc diameter features, if Num(C)=1, and d ² <threshold α _w , then the insulator string is a single-segment equal-diameter insulator string, if Num(C)=2, and d ² <threshold α _w , then the insulator string is a double-segment insulator string with equal disk diameter, and if Num(C)=1, and d ² >threshold α _w , then the insulator string is a single-segment variable-disc insulator string. Complete the identification of three commonly used insulator strings in substations. As shown in Figure 5, it is a string of double-section insulators with equal disk diameters.

Claims (7)

1. the insulator chain recognition methods based on laser radar, which is characterized in that include the following steps：

The acquisition of depth image：Depth image is obtained by the laser radar apparatus installed in robot；

The pretreatment of depth image：The depth image for carrying insulator chain is pre-processed, obtains interested region；

The feature extraction of depth image：In interested region, depth characteristic curve model is established, structure depth is periodically special Matrix and width vector are levied, extracts insulator chain hop count feature and disk diameter variation characteristic；

Insulator chain type identification：Insulator chain type is identified according to the feature extracted.

2. the insulator chain recognition methods according to claim 1 based on laser radar, which is characterized in that the depth map The pretreatment of picture includes the following steps：

(2.1) to depth image binary conversion treatment, prospect, background in depth image are isolated, and eliminate background interference；

(2.2) to treated, bianry image uses dilation operation, then using erosion operation, eliminates the interference picture on bianry image Vegetarian refreshments；

(2.3) and then largest connected domain is chosen, extracts area-of-interest, complete pretreatment.

3. the insulator chain recognition methods according to claim 1 based on laser radar, which is characterized in that described to establish deeply It is as follows to spend indicatrix model：

Extract target area left hand edge：

li_k=k,

Extract target area right hand edge：

ri_k=k,

In formula D (x, y) be bianry image, the line number of k representative images, the columns of t representative images, l_k, r_kFor the pixel on edge Position；

Position on the center line of target area is determined by left hand edge and right hand edge：

mi_k=k,

M in formula_kFor the coordinate of depth characteristic curve pixel, mi_kFor depth characteristic curve pixel row coordinate, mj_kFor depth spy Levy curve pixel point range coordinate；n_IFor picturedeep；

Establish depth characteristic curve model:

f(x_k,y_k)=0

Wherein:

x_k=mi_k

x_kIt is characterized the abscissa of curve, i.e. picturedeep, y_kIt is characterized the ordinate of curve, i.e. depth value；L(m_k) represent deep Spend image.

4. the insulator chain recognition methods according to claim 1 based on laser radar, which is characterized in that the structure is deep Spend periodic feature matrix：

Depth characteristic curve f (x are first obtained_k,y_kMaximum and minimum on)=0；Establish matrix

In formula：

v_2kFor depth characteristic curve f (x_k,y_kThe abscissa of extreme point, v on)=0_3kFor depth characteristic curve f (x_k,y_kOn)=0 The ordinate of extreme point, i.e. depth value；K=1,2 ... n_I；

Given threshold λ_c1,λ_c2And λ_c1﹤ λ_c2；It is right againTraverse by row, by V_3,k+1-V_3,k＞ λ_c1And V_3,k+1-V_3,k＜ λ_c2When V_1,kWith V_2,kIt is recorded；At the end of traversal, the value for the condition that meets is formed into matrix by record sequence By v_1,kIt is recorded in matrixThe first row, v_2,kThe second row of matrix is recorded in, is obtained：

n₃For matrix columns；

It is right againIt is traversed, is enabled by row

c_1,k=abs (v1_1,k+1-v1_1,k)

c_2,k=v1_2,k+1-v1_2,k

c_3,k=v1_2,k

Obtain depth periodic feature matrix

5. the insulator chain recognition methods according to claim 1 based on laser radar, which is characterized in that structure width to Amount：

In formula：

(x_i1,y_i1,z_i1) be left hand edge point space coordinate, (x_i2,y_i2,z_i2) be right hand edge point space coordinate；n_IRepresent image Line number.

6. the insulator chain recognition methods according to claim 1 based on laser radar, which is characterized in that described to extract Insulator chain hop count feature and disk diameter variation characteristic include the following steps：

(1) byExtract insulator chain hop count feature

Given threshold α first_c, initialization flag position δ_cIt is 0, segment number information Num (C) is 0；

It is right againIt is traversed by row：

Work as C_1,k=1, C_2,k＜ α_cAnd δ_cWhen=0, flag bit δ is put_cIt is 1, and by C_3,kIt is stored in vector T as on insulator chain Edge；

Work as δ_c=1 and C_1,k=0 or C_2,k＞ α_cWhen, put flag bit δ_cIt is 0, by C_3,kIt is stored in vector T as under insulator chain Edge, and segment number information Num (C) is added 1；

Upper and lower edge of the vector T for insulator chain, wherein odd term element t_2n-1For the top edge of insulator chain, even item element t_2nLower edge for insulator chain；N=1,2 ... k₁；k₁Length for vector T.

(2) by the disk diameter variation characteristic of W extraction insulator chains

Insulator chain width variance is calculated according to vector T：

Work as d²＜ threshold alphas_wWhen, the insulator is waits disks diameter, otherwise, to become disk diameter.

7. the insulator chain recognition methods according to claim 1 based on laser radar, which is characterized in that the basis carries Insulator chain type, which is identified, in the feature of taking-up includes the following steps：

If Num (C)=1, and d²＜ threshold alphas_w, then insulator chain is the disks diameter insulator chains such as single hop；If Num (C)=2, and d²＜ Threshold alpha_w, then insulator chain is the disks diameter insulator chains such as double sections, if Num (C)=1, and d²＞ threshold alphas_w, then insulator chain is single Duan Bianpan is through insulator chain.