Intelligent subway purging system and method

Abstract

According to the intelligent subway purging system and method, a vehicle bottom intelligent purging unit and a vehicle side intelligent purging unit are arranged to clean a to-be-swept vehicle bottom and a to-be-swept vehicle side respectively and are uniformly controlled by an intelligent dispatching unit to work at the same time, and a control unit controls a corresponding moving unit to act, so that the to-be-swept vehicle bottom and the to-be-swept vehicle side are cleaned. The visual inspection unit is used for carrying out real-time shooting and transmitting to the intelligent scheduling unit to obtain the state of a blown part, a blowing scheme is obtained through analysis, and then the intelligent scheduling unit sends an instruction to the control unit; and the purging unit is controlled to perform automatic purging according to the selected purging scheme, simultaneous intelligent purging of the vehicle bottom and the vehicle side is completed, manual operation is not needed, the labor intensity of operators is reduced, and the purging efficiency of the metro vehicle is improved while the purging effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of rail vehicle cleaning technology, and in particular to a subway intelligent purge system and method. Background Art

[0002] With the continuous improvement of the national economic development level, people's demand for urban transportation has increased, and the number of urban rail transit vehicles has also increased rapidly. After a long period of operation, a large amount of dust will be attached to key parts such as the resistor box, electrical cabinet and motor under the car and on the side of the subway car. If it is not cleaned in time, it will affect the performance and service life of key components of the vehicle. In serious cases, it will cause vehicle operation failures and have an adverse impact on the economic security of the society.

[0003] At present, most of the subway cleaning work of domestic subway operating companies is done by manual handheld cleaning tools. Although this method of cleaning has a certain effect, the cleaning workers have high labor intensity and low cleaning efficiency. At the same time, it is also difficult to ensure the overall cleaning quality; although the cleaning workers wear protective equipment, it is difficult to ensure that dust particles are sucked into the body through the gaps, which has certain harm to their health. The Chinese invention patent with the authorization announcement number CN 112590715 B discloses an intelligent cleaning system and process for the underbody of a subway car. The cleaning system includes a mobile cleaning unit, a dust collection unit, a control unit, a detection and positioning unit, and an intelligent unit; one mobile cleaning unit corresponds to two dust collection units, and the two dust collection units are symmetrically arranged on both sides of the car body. During the cleaning operation, the dust collection units on both sides of the car body move or stop synchronously and at the same speed as the corresponding mobile cleaning units to achieve intelligent cleaning of the underbody.

[0004] The technical solution disclosed in the above patent document can only realize automatic cleaning of the bottom of the vehicle, but cannot realize automatic cleaning of the sides of the vehicle, and manual blowing is still required. Summary of the invention

[0005] The purpose of the present invention is to propose a subway intelligent blowing system and method to address the deficiencies of the prior art, which can simultaneously perform intelligent cleaning of the bottom and sides of the vehicle without the need for manual operation, thereby reducing the labor intensity of operators, and improving the blowing efficiency of subway vehicles while improving the blowing effect.

[0006] The technical solution to achieve the purpose of the present invention is:

[0007] A subway intelligent purge system comprises an intelligent dispatching unit, an undercarriage intelligent purge unit and a car-side intelligent purge unit, wherein the undercarriage intelligent purge unit and the car-side intelligent purge unit each comprise a mobile unit connected to the intelligent dispatching unit signal, a control unit, a purge unit and a visual detection unit, wherein the mobile unit and the purge unit are both electrically connected to the corresponding control unit, and the control unit is suitable for controlling the mobile unit and the purge unit to work under the dispatch of the intelligent dispatching unit.

[0008] Furthermore, the mobile unit is an AGV cart, and the purge unit includes a purge robot detachably mounted on the AGV cart, an air compressor, and a high-pressure nozzle mounted on the free end of the purge robot, the high-pressure nozzle is connected to the air circuit of the air compressor, the purge robot of the underbody intelligent purge unit is arranged at the front end or rear end of the AGV cart, the purge robot of the vehicle side intelligent purge unit is arranged at the left end or right end of the AGV cart, and the visual inspection unit is detachably mounted on the high-pressure nozzle.

[0009] Furthermore, the AGV cart includes a body, a mounting platform and an electrical cabinet arranged on the top of the body, the mounting platform is provided with mounting positions compatible with the electrical cabinet, the purge robot, the air compressor and the control unit, and the electrical cabinet, the purge robot, the air compressor and the control unit are detachably fixed to the corresponding mounting positions by screws, which is convenient for disassembly and assembly.

[0010] Furthermore, the visual detection unit adopts a 3D structured light vision system, which can obtain two-dimensional and three-dimensional image information.

[0011] Furthermore, anti-collision strips with adjustable sensitivity are arranged around the vehicle body to protect the entire device.

[0012] Furthermore, laser obstacle avoidance is arranged around the vehicle body, which can automatically sense the surrounding environment and stop the vehicle immediately. Preferably, the laser sensor has a three-level gradient deceleration information feedback algorithm.

[0013] Furthermore, the mobile unit is an AGV cart with a high-precision encoder, and the AGV cart of the underbody intelligent cleaning unit is provided with a visual positioning unit. The intelligent scheduling unit receives the positioning data sent by the high-precision encoder and the visual positioning unit, and sends instructions to the control unit after processing, and the control unit controls the corresponding mobile unit action.

[0014] Furthermore, the AGV trolley of the vehicle-side intelligent purge unit is signal-connected with the AGV trolley of the vehicle-bottom intelligent purge unit.

[0015] Furthermore, the high-pressure nozzle of the intelligent purge unit on the vehicle side includes a base and a circle of nozzles evenly arranged along the circumference of the base; the high-pressure nozzle of the intelligent purge unit under the vehicle includes a base and a plurality of nozzles arranged on the base and arranged in a straight line direction; the base is fixed to the purge manipulator by bolts. The two high-pressure nozzles adopt a customized differential design, which can provide stable and efficient airflow, and are more suitable for purge under and on the side of the vehicle. The structure is simple, the installation is convenient, the maintenance work is relatively easy, and the cost is reduced.

[0016] Furthermore, a subway intelligent purge method, using the purge system as described above, comprises the following steps:

[0017] Step S1: After the subway vehicle stops steadily on the track of the purge depot, the staff removes the cover of the part that needs to be purged and evacuates, exposing the part that needs to be purged;

[0018] Step S2: The underbody intelligent purge unit and the vehicle side intelligent purge unit are located at the standby point, and the control unit controls the mobile unit to move forward to the purge point;

[0019] Step S3: The visual detection unit obtains the real-time status of the current purge point and sends it to the intelligent scheduling unit for processing, obtains a purge plan, and sends it to the control unit, which controls the purge unit to purge. The purge plan includes a purge trajectory and a purge speed.

[0020] Step S4: After the current purge point is purged, the underbody intelligent purge unit and the vehicle side intelligent purge unit move forward synchronously to the next purge point, and repeat step S3. After the purge on one side of the vehicle is completed, the vehicle side intelligent purge unit detours to the other side of the vehicle, and the underbody intelligent purge unit turns around and repeats the purge until all purge points are purged;

[0021] Step S5: The undercarriage intelligent purge unit and the car side intelligent purge unit return to the standby point, and the staff resets the cover of the purge area on the subway car.

[0022] Further, in step S2, the mobile unit of the underbody intelligent purge unit is provided with a visual positioning unit, and a positioning method combining visual detection and pulse offset is adopted to make the mobile units of the underbody intelligent purge unit and the vehicle side intelligent purge unit reach the purge point. The positioning method includes the following steps:

[0023] Step S21: define the standby point of the mobile unit of the vehicle bottom intelligent purge unit to be in the non-interference area, create a template matching model, shield the non-detection area, the subway vehicle moves slowly at a speed of 1mm / s relative to the vehicle bottom intelligent purge unit, and the visual positioning unit automatically detects the reference object of the subway vehicle, obtains the offset data and transmits it to the vehicle side intelligent purge unit wirelessly. The reference object is a rectangular convex block at the rear of the vehicle. The non-interference area is the preset detection area where the vehicle bottom intelligent purge unit is located. If it is not in this interference area, the vehicle bottom feature cannot be detected;

[0024] Step S22: A magnetic stripe recognition device and an IC card recognition device are provided at the bottom of the mobile unit of the vehicle side intelligent purge unit. The magnetic stripe and IC card are used to track and locate the vehicle bottom intelligent purge unit, and the offset data is combined to accurately locate the unit to the purge point.

[0025] Further, in step S3, obtaining the purge scheme includes the following steps:

[0026] Step S31: image processing, identifying the parts at the purge point, acquiring a data set of the parts at the purge point through a visual inspection unit, the data set including the parts at the purge point and the cleaning status, and performing grayscale, morphological transformation and probabilistic Hough transformation on the data set for identification, and performing accurate extraction by using dynamic factor setting and secondary pseudo-edge elimination;

[0027] Step S32: Segment the purge area according to the processed image, that is, segment the dust and the parts at the purge point, perform real-time accurate dust segmentation on different types of parts according to complex working conditions, illumination changes, and the surface state of the parts at the purge point, establish a segmentation method based on a fully convolutional neural network, and obtain a segmented image;

[0028] Step S33: A purge level assessment framework based on RLRNM (segmentation network level recognition model) is established according to the data set, the segmented image is evaluated according to the purge level assessment framework, and a purge scheme of the corresponding level is selected. The purge level assessment framework includes feature extraction, ROI (Region of Interest) extraction and purge level assessment.

[0029] Furthermore, the image processing specifically includes the following steps:

[0030] Step S311: The scanned original image is processed using a weighted grayscale method. For the pixel point (i, j) of the original image, the weighted grayscale value is calculated according to the following formula:

[0031] I(i,j)=0.30R(i,j)+0.59G(i,j)+0.11B(i,j),

[0032] Among them, R(i,j) is the red component, G(i,j) is the green component, and B(i,j) is the blue component. A rectangular box is used to mark the area of ​​interest, and the mean and variance of the grayscale of each pixel are calculated to obtain a grayscale image, which is then segmented to obtain a binary image;

[0033] Step S312: using a morphological processing method to perform dilation and erosion processing on the obtained binary image, preferably using a 3×3 pixel structure element to perform dilation and erosion processing, and expanding the structure element e to obtain (DΘe),

[0034] (DΘe)(x,y)=max{D(x-x',yy;)+e(x',y;)(x',y;)∈S e},

[0035] Corrosion calculation of the structural element e yields (DΘe),

[0036] (D⊙e)(x,y)=min{D(x+x',y+y;)-e(x',y;)(x',y;)∈S e},

[0037] Subtract the expansion result from the corrosion result to obtain the edge information of the component at the purge point and obtain the edge image;

[0038] Step S313: using the probability Hough transform calculation formula ρ=xcosθ+ysinθ, sampling some points on the edge image and performing cumulative calculation to obtain a sampled image;

[0039] Step S314: Perform Canny edge detection on the sampled image. First, use the Gaussian function to smooth the image preprocessing, then use the threshold algorithm to perform optimal edge detection processing, and perform secondary pseudo-edge processing by comparing the difference between the pixel value of the pixel point and the arithmetic mean. Finally, perform regional fitting with the parts at the purge point, save the image, and identify the parts at the purge point. The secondary pseudo-edge processing formula is:

[0040]

[0041] Among them, n and m are the total number of columns and rows of the image respectively, and c is the dynamic threshold.

[0042] Furthermore, the segmentation method of the purge area using the RPLSN dust pixel level segmentation network includes the following steps:

[0043] Step S321: Divide the data in the image processed in step S31 into each layer, and each layer of data is a three-dimensional array of w×h×d, where w and h are spatial dimensions and d is the channel dimension. In the subsequent layers, the data vector of each pixel coordinate is added, and a large number of feature channels are introduced, so that the background features of the target image are propagated to a higher resolution layer to obtain a feature map;

[0044] Step S322: Softmax function is calculated on the feature map to obtain the output result, and the weight value of the edge pixel is calculated to balance the frequency of occurrence of the feature data.

[0045] The output formula of the Softmax function is as follows:

[0046]

[0047] Among them, α β (x) represents the activation value of the βth layer of the pixel at position x in the feature map;

[0048] Step S323: discard the fully connected layer, use the effective part of each convolutional layer, and calculate the average of the segmented areas. The calculation formula of the segmented areas is as follows:

[0049]

[0050] Among them, mIoU is the average intersection over union ratio, that is, the intersection of the predicted segmentation area and the actual segmentation area divided by the union of the predicted segmentation area and the actual segmentation area, i represents the true value, j represents the predicted value, and p ij is the total number of pixels of true pixel category i that are predicted to be of category j;

[0051] Step S324: Divide the segmented area into dust areas and component areas, and mark different areas with irregular graphics.

[0052] Furthermore, after the purge is completed, the visual inspection unit collects an image of the purge and evaluates the purge effect on the image. If the evaluation is qualified, the next step is performed; otherwise, the purge is repeated after re-acquiring the purge plan.

[0053] Furthermore, the establishment of the purge level assessment framework includes the following steps:

[0054] Step S331: Use the ResNET network to extract features of the parts at the purge point, apply an asymmetric residual function, activate it, and then calculate it with the weight w according to the following formula to complete feature extraction, reduce overfitting, and provide better feature extraction accuracy.

[0055] Among them, F is the residual function and f is the activation function;

[0056] Step S332: directly select the feature map frame as the ROI area, send it to the subsequent fully connected layer, then classify the dust and parts, regress the bounding box, and extract the ROI;

[0057] Step S333: Establish the following dust level classification sample evaluation after purge.

[0058] Dust level Dust particles Dust area Dust removal effect Q1 Particle diameter ≤5um Less than 10 per square centimeter excellent Q2 Particle diameter ≤5um Between 10 and 100 per square centimeter good Q3 5um≤Particle diameter≤15um Between 100 and 1000 per square centimeter Difference Q4 Particle diameter ≥15um More than 1000 per square centimeter Very bad

[0059] By adopting the above technical solution, the present invention has the following beneficial effects:

[0060] (1) The present invention provides an intelligent undercarriage purge unit and an intelligent side purge unit to clean the undercarriage and the side of the car respectively, and the intelligent dispatching unit controls the unit uniformly to realize simultaneous operation. The control unit controls the corresponding mobile unit to move, thereby driving the purge unit and the visual detection unit thereon to move. The visual detection unit takes real-time photos and transmits the photos to the intelligent dispatching unit to obtain the status of the purge area. The intelligent dispatching unit sends instructions to the control unit to control the purge unit to automatically purge, thereby completing the intelligent purge of the undercarriage and the side of the car simultaneously. No manual operation is required, thereby reducing the labor intensity of the operators, and improving the purge effect while improving the purge efficiency of the subway vehicle.

[0061] (2) The components of the purge unit of the present invention can be detachably mounted on the mobile unit, and the visual inspection unit can be detachably mounted on the high-pressure nozzle, thereby realizing a modular design, facilitating assembly and subsequent maintenance and replacement, and facilitating upgrading and replacement. At the same time, the modules of the underbody intelligent purge unit and the vehicle side intelligent purge unit adopt a differentiated structural layout, which is more convenient for purge the corresponding area of ​​the vehicle body.

[0062] (3) The positioning of the mobile unit of the underbody intelligent purge unit of the present invention adopts a combination of pulse offset realized by a high-precision encoder and visual positioning by a visual positioning unit, thereby improving the precise positioning of the mobile trolley on the purge components, facilitating the control of the purge safety distance, and avoiding damage to subway vehicle components.

[0063] (4) The vehicle-side intelligent purge unit of the present invention is connected to the vehicle-bottom intelligent purge unit by signal, thereby maintaining synchronous and precise movement with the moving unit of the vehicle-bottom intelligent purge unit, simplifying control, and also achieving precise positioning.

[0064] (5) In the purge method of the present invention, the intelligent scheduling system sends instructions to the control unit, which controls the simultaneous cleaning of the bottom and sides of the vehicle. The visual detection unit takes real-time photos of the status of the purge points, formulates corresponding purge plans, and performs differential purges on the purge points in different states, thereby improving the purge effect and the purge efficiency.

[0065] (6) The present invention adopts visual positioning, takes a fixed reference object as a reference, obtains the offset data of the intelligent purge unit under the vehicle, combines it with pulses to achieve precise positioning, and realizes the synchronous movement of the intelligent purge unit on the side of the vehicle through tracking and positioning, which can ensure high-precision positioning while having low cost, reliable positioning, and convenient later maintenance.

[0066] (7) The image processing when the purge scheme is acquired in the present invention extracts and segments the parts at the purge point, reduces the influence of complex working conditions, environment and surface state of parts, improves image recognition accuracy, and adopts probabilistic Hough transform to reduce the amount of data processing calculation and improve the image processing rate.

[0067] (8) The present invention optimizes the loss function in the image processing method, taking into full consideration the fact that there are many parts on the bottom of a subway vehicle and that the parts are uneven. Different colors are used to clearly distinguish them, thereby better identifying the key parts of the bottom of the subway.

[0068] (9) The present invention independently designs the RPLSN dust pixel-level segmentation network, which can work with fewer training images and achieve better performance even with limited data.

[0069] (10) The present invention adds a post-purge effect evaluation process to ensure that the cleaning is in place and to accumulate data for later optimization of the purge plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments and in conjunction with the accompanying drawings, wherein:

[0071] Figure 1 This is a schematic diagram of the structure of the vehicle bottom intelligent purge unit of the present invention;

[0072] Figure 2 It is a structural schematic diagram of the vehicle side intelligent purge unit of the present invention;

[0073] Figure 3 It is a structural schematic diagram of the high-pressure nozzle of the vehicle bottom intelligent purge unit of the present invention;

[0074] Figure 4 It is a structural schematic diagram of the high-pressure nozzle of the vehicle-side intelligent purge unit of the present invention;

[0075] Figure 5 is a flow chart of the purging method of the present invention;

[0076] Figure 6 The figure is a flow chart of the image processing method of the visual detection unit of the present invention.

[0077] The reference numerals in the accompanying drawings are:

[0078] Mobile unit 1, vehicle body 1-1, installation platform 1-2, electrical cabinet 1-3, anti-collision strip 1-4, laser obstacle avoidance 1-5, control unit 2, purge unit 3, purge manipulator 3-1, nozzle 3-1-1, air compressor 3-2, visual inspection unit 4, visual positioning unit 5. DETAILED DESCRIPTION

[0079] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0080] (Example 1)

[0081] like Figures 1 to 6 The subway intelligent purge system shown includes an intelligent dispatching unit, an intelligent purge unit under the vehicle and an intelligent purge unit on the vehicle side. The intelligent purge unit under the vehicle and the intelligent purge unit on the vehicle side both include a mobile unit 1, a control unit 2, a purge unit 3 and a visual detection unit 4 connected to the intelligent dispatching unit signal. The mobile unit 1, the control unit 2, the purge unit 3 and the visual detection unit 4 transmit data to the intelligent dispatching unit for background processing, wherein the mobile unit 1 and the purge unit 3 are both electrically connected to the corresponding control unit 2, and the control unit 2 accepts the instructions of the intelligent dispatching unit to control the mobile unit 1 and the purge unit 3 to work.

[0082] Specifically, the mobile unit 1 is an AGV trolley, and the purge unit 3 includes a purge manipulator 3-1 detachably mounted on the AGV trolley, an air compressor 3-2, and a high-pressure nozzle 3-3 mounted on the free end of the purge manipulator 3-1, wherein the high-pressure nozzle 3-3 is connected to the air circuit of the air compressor 3-2 to achieve purge. The AGV trolley includes a body 1-1, a mounting platform 1-2 and an electrical cabinet 1-3 arranged on the top of the body 1-1, and mounting positions compatible with the electrical cabinet, the purge manipulator, the air compressor and the control unit are arranged on the mounting platform 1-2, and the electrical cabinet, the purge manipulator, the air compressor and the control unit are detachably fixed on the corresponding mounting positions by screws, and the disassembly and assembly are convenient. The body 1-1 is surrounded by anti-collision strips 1-4 with adjustable sensitivity to protect the entire device. The body 1-1 is also surrounded by laser obstacle avoidance 1-5, which can automatically sense the surrounding environment and stop immediately. A laser sensor with a three-level gradient deceleration information feedback algorithm is preferred. The visual inspection unit 4 is detachably mounted on the high-pressure nozzle 3-3 and uses a 3D structured light visual system to obtain two-dimensional and three-dimensional image information. The overall structure is modularly designed, which is convenient for assembly and later maintenance and replacement, and is easy to upgrade.

[0083] Taking into account the differences in the use scenarios of the bottom and side of the vehicle, this embodiment has made differentiated designs for the two intelligent purge units, wherein the purge manipulator of the bottom intelligent purge unit is arranged at the front end or rear end of the AGV trolley, while the purge manipulator of the side intelligent purge unit is arranged at the left end or right end of the AGV trolley, so as to facilitate the purge of the side of the vehicle body. The high-pressure nozzle of the side intelligent purge unit includes a base and a circle of nozzles 3-3-1 evenly arranged along the circumference of the base, and the high-pressure nozzle of the bottom intelligent purge unit includes a base and a plurality of nozzles 3-3-1 arranged on the base and arranged in a straight line direction. The base is fixed to the purge manipulator by bolts. The two high-pressure nozzles adopt customized differential designs, which can provide stable and efficient airflow, and are more suitable for the purge of the bottom and side of the vehicle. The structure is simple, the installation is convenient, the maintenance work is relatively easy, and the cost is reduced.

[0084] The AGV trolleys of the two intelligent purge units are both AGV trolleys with high-precision encoders. The AGV trolley of the intelligent purge unit under the vehicle is equipped with a visual positioning unit 5. The intelligent scheduling unit receives the positioning data sent by the high-precision encoder and the visual positioning unit, and sends instructions to the control unit after processing. The control unit controls the corresponding mobile unit action to achieve more accurate automatic positioning. The AGV trolley of the vehicle-side intelligent purge unit is connected to the AGV trolley of the intelligent purge unit under the vehicle by signal, so as to send the precise positioning data to the intelligent purge unit under the vehicle. The bottom of the AGV trolley of the vehicle-side intelligent purge unit is equipped with a magnetic stripe recognition device and an IC card recognition device. The magnetic stripe and IC card are used to track and locate the intelligent purge unit under the vehicle to complete precise positioning, which can not only ensure high-precision positioning, but also have low cost, reliable positioning, and convenient later maintenance.

[0085] The intelligent dispatching system is equipped with a button function module and a touch-type intelligent monitoring and display interface. The button function module includes pause the current execution task, continue execution, and stop functions. The intelligent monitoring and display interface can display the operating status and real-time information of the intelligent blowing units under and on the sides of the vehicle (real-time position, battery power, blowing points, operating speed, etc.). The speed, acceleration, jerk and other data can be modified in real time by touch, which is convenient for manual intervention.

[0086] The intelligent blowing system for subway is used to perform intelligent blowing, including the following steps:

[0087] Step S1: After the subway vehicle stops steadily on the track of the purge depot, the staff removes the cover plates of the parts that need to be purged and then evacuates and stores them in a designated place, exposing the parts that need to be purged;

[0088] Step S2: The undercarriage intelligent purge unit and the vehicle side intelligent purge unit are located at the standby point, and the standby point of the mobile unit of the undercarriage intelligent purge unit is defined to be in the non-interference area. A template matching model is created, and the non-detection area is shielded. The subway vehicle moves slowly at a speed of 1 mm / s relative to the undercarriage intelligent purge unit. The visual positioning unit automatically detects the rectangular convex block at the rear of the subway vehicle as a reference object, obtains the offset data and transmits it to the vehicle side intelligent purge unit wirelessly, wherein the non-interference area is the detection area where the undercarriage intelligent purge unit is located. If it is not in this interference area, the bottom features of the vehicle cannot be detected;

[0089] The mobile unit of the intelligent purge unit on the vehicle side uses the magnetic strip and IC card at the bottom to track and locate the intelligent purge unit under the vehicle, and combines the offset data for precise positioning. Each control unit controls the mobile unit to move forward synchronously to the purge point.

[0090] Step S3: The visual detection unit obtains the real-time status of the current purge point and sends it to the intelligent scheduling unit for processing, obtains a purge plan including different purge trajectories and purge speeds, and sends it to the control unit, which controls the purge unit to purge, specifically including the following steps:

[0091] Step S31: image processing, identifying the parts at the purge point, acquiring a data set of the parts at the purge point through a visual inspection unit, wherein the data set includes the parts at the purge point and the cleaning status, performing grayscale, morphological transformation and probabilistic Hough transformation on the data set for identification, and performing precise extraction by using dynamic factor setting and secondary pseudo-edge elimination;

[0092] The grayscale processing method is as follows:

[0093] In order to better identify, the scanned original image is processed using the weighted grayscale method. For the pixel point (i, j) of the original image, the weighted grayscale value is calculated according to the following formula:

[0094] I(i,j)=0.30R(i,j)+0.59G(i,j)+0.11B(i,j),

[0095] Among them, R(i,j) is the red component, G(i,j) is the green component, and B(i,j) is the blue component. A rectangular box is used to mark the area of ​​interest, and the mean and variance of the grayscale of each pixel are calculated to obtain a grayscale image, which is then segmented to obtain a binary image;

[0096] The morphological transformation method is as follows:

[0097] Considering the illumination intensity, inconsistent shooting angles, and more noise in the image, the morphological processing method is used to dilate and erode the binary image. Preferably, a 3×3 pixel structure element is used for dilation and erosion. The structure element e is expanded and calculated to obtain (DΘe).

[0098] (DΘe)(x,y)=max{D(x-x',yy ； )+e(x',y ； )|(x',y;)∈S e},

[0099] Corrosion calculation of the structural element e yields (DΘe),

[0100] (D⊙e)(x,y)=min{D(x+x',y+y ； )-e(x',y;)|(x',y ； )∈S e},

[0101] Subtract the expansion result from the corrosion result to obtain the edge information of the component at the purge point and obtain the edge image;

[0102] The probabilistic Hough transform method is as follows:

[0103] Considering the large number of points in the target image, the large amount of calculation and low efficiency, the probability Hough transform calculation formula ρ = xcosθ + ysinθ is used to sample some points on the edge image and perform cumulative calculation to obtain the sampled image;

[0104] The edge detection and processing methods are as follows:

[0105] Canny edge detection is performed on the sampled image. First, the image is smoothed and preprocessed using the Gaussian function. Then, the threshold algorithm is used for optimal edge detection. Secondary pseudo-edge processing is performed by comparing the difference between the pixel value of the pixel point and the arithmetic mean. Finally, the area is fitted with the parts at the purge point, and the image is saved to identify the parts at the purge point. The secondary pseudo-edge processing formula is:

[0106]

[0107] Among them, n and m are the total number of columns and rows of the image respectively, and c is the dynamic threshold.

[0108] Step S32: Segment the purge area according to the processed image, that is, segment the dust and the parts at the purge point. In view of complex working conditions, illumination changes, and the surface state of the parts at the purge point, use the RPLSN dust pixel-level segmentation network to perform real-time accurate dust segmentation on different types of parts. Design a segmentation method based on a full neural network to establish a segmentation method based on a full convolutional neural network to obtain a segmented image. RPLSN can work with fewer training images. The specific steps are as follows;

[0109] Step S321: Divide the data in the image processed in step S31 into each layer, and each layer of data is a three-dimensional array of w×h×d, where w and h are spatial dimensions and d is the channel dimension. In the subsequent layers, the data vector of each pixel coordinate is added, and a large number of feature channels are introduced, so that the background features of the target image are propagated to a higher resolution layer to obtain a feature map;

[0110] Step S322: In order to obtain high-quality segmented images, a Softmax function is calculated on the feature map to obtain the output result, and the weight values ​​of edge pixels are calculated to balance the frequency of occurrence of feature data.

[0111] The output formula of the Softmax function is as follows:

[0112]

[0113] Among them, α β (x) represents the activation value of the βth layer of the pixel at position x in the feature map;

[0114] Step S323: discard the fully connected layer, use the effective part of each convolutional layer, and calculate the average of the segmented areas. The calculation formula of the segmented areas is as follows:

[0115]

[0116] Among them, mIoU is the average intersection over union ratio, that is, the intersection of the predicted segmentation area and the actual segmentation area divided by the union of the predicted segmentation area and the actual segmentation area, i represents the true value, j represents the predicted value, and p ij is the total number of pixels of true pixel category i that are predicted to be of category j;

[0117] Step S324: Divide the segmented area into dust areas and component areas, and mark different areas with irregular graphics.

[0118] Step S33: A purge level assessment framework based on RLRNM (segmentation network level recognition model) is established according to the data set, the segmented image is evaluated according to the purge level assessment framework, and a purge scheme of the corresponding level is selected. The purge level assessment framework includes feature extraction, ROI (Region of Interest) extraction and purge level assessment. The specific steps are as follows:

[0119] Step S331: Use the ResNET network to extract features of the parts at the purge point, apply an asymmetric residual function, activate it, and then calculate it with the weight w according to the following formula to complete feature extraction, reduce overfitting, and provide better feature extraction accuracy.

[0120] Among them, F is the residual function and f is the activation function;

[0121] Step S332: directly select the feature map frame as the ROI area, send it to the subsequent fully connected layer, then classify the dust and parts, regress the bounding box, and extract the ROI;

[0122] Step S333: Establish the following dust level classification sample evaluation after purge.

[0123]

[0124]

[0125] Step S4: After the purge is completed, the visual inspection unit collects the image of the purge and evaluates the purge effect on the image. If the evaluation is qualified, proceed to the next step, otherwise re-acquire the purge plan and repeat the purge. After the current purge point is completed, the underbody intelligent purge unit and the vehicle side intelligent purge unit move forward synchronously to the next purge point and repeat step S3. After the purge of one side of the vehicle is completed, the vehicle side intelligent purge unit detours to the other side of the vehicle, and the underbody intelligent purge unit turns around and repeats the purge until all purge points are completed;

[0126] Step S5: The undercarriage intelligent purge unit and the car side intelligent purge unit return to the standby point, and the staff resets the cover of the purge area on the subway car.

[0127] The underbody intelligent blowing unit and the vehicle side intelligent blowing unit are also equipped with power monitoring. If the power is insufficient after the blowing is completed, they will return to the designated charging location for charging for the next use.

[0128] The present invention adopts two sets of intelligent purge units under the vehicle and on the side of the vehicle to share one set of intelligent scheduling system. The intelligent purge unit under the vehicle contains a visual positioning unit. The control unit receives instructions from the intelligent scheduling system to control the movement of the AGV trolley and the in-place purge of the purge robot; the visual positioning unit is detailed and integrated with the high-precision encoder of the AGV trolley, and the visual positioning and pulse positioning are integrated to realize the precise positioning of the purge point.

[0129] At the same time, the visual inspection unit contains a set of specific recognition methods. For the problems of sample extraction, segmentation, recognition and prediction of parts of the subway to be purged in multiple samples and multiple working conditions, the sample classification network, precise segmentation network and parts recognition are analyzed and designed respectively, so as to realize the accurate recognition of the purging area. Taking into full consideration the influence of factors such as the lighting at the bottom of the subway vehicle, it is able to identify the parts at the bottom of the subway, and finely divide the dust and parts, improve the detection accuracy and the efficiency of purging, and create an evaluation framework. After purging, the purging evaluation is carried out, and the images before and after purging are analyzed and compared, and the dust classification level after purging is divided, which not only ensures that the purging meets the standards, but also collects data for the later optimization of the purging plan, which is convenient for later upgrades and optimizations.

[0130] The intelligent dispatching unit collects and shares information between the equipment under and on the sides of the vehicle. Through the set purge process, it dispatches other units and generates path planning instructions, thereby efficiently and quickly cleaning dust and debris attached to the subway, realizing intelligent and all-round purge of the bottom and sides of subway vehicles.

[0131] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A subway intelligent purge system, characterized in that: It includes an intelligent scheduling unit, an intelligent purge unit under the vehicle and an intelligent purge unit on the side of the vehicle. The intelligent purge unit under the vehicle and the intelligent purge unit on the side of the vehicle both include a mobile unit, a control unit, a purge unit and a visual detection unit which are signal-connected to the intelligent scheduling unit. The mobile unit and the purge unit are both electrically connected to the corresponding control unit. The control unit is suitable for controlling the mobile unit and the purge unit to work under the scheduling of the intelligent scheduling unit.

2. The intelligent subway purge system according to claim 1, characterized in that: The mobile unit is an AGV cart, and the purge unit includes a purge robot detachably mounted on the AGV cart, an air compressor, and a high-pressure nozzle mounted on the free end of the purge robot, the high-pressure nozzle is connected to the air circuit of the air compressor, the purge robot of the underbody intelligent purge unit is arranged at the front end or rear end of the AGV cart, the purge robot of the vehicle side intelligent purge unit is arranged at the left end or right end of the AGV cart, and the visual inspection unit is detachably mounted on the high-pressure nozzle.

3. The intelligent subway purge system according to claim 1, characterized in that: The mobile unit is an AGV cart with a high-precision encoder, and the AGV cart of the underbody intelligent cleaning unit is provided with a visual positioning unit. The intelligent scheduling unit receives the positioning data sent by the high-precision encoder and the visual positioning unit, and sends instructions to the control unit after processing, and the control unit controls the corresponding mobile unit action.

4. The intelligent subway purge system according to claim 3, characterized in that: The AGV trolley of the vehicle side intelligent purge unit is signal-connected with the AGV trolley of the vehicle bottom intelligent purge unit.

5. A subway intelligent purge method, using the purge system according to any one of claims 1 to 4, characterized in that: The following steps are involved: Step S1: After the subway vehicle stops steadily on the track of the purge depot, the staff removes the cover of the part that needs to be purged and evacuates, exposing the part that needs to be purged; Step S2: The underbody intelligent purge unit and the vehicle side intelligent purge unit are located at the standby point, and the control unit controls the mobile unit to move forward to the purge point; Step S3: The visual detection unit obtains the real-time status of the current purge point and sends it to the intelligent scheduling unit for processing, obtains the purge plan, and sends it to the control unit, which controls the purge unit to perform the purge; Step S4: After the current purge point is purged, the underbody intelligent purge unit and the vehicle side intelligent purge unit move forward synchronously to the next purge point, and repeat step S3. After the purge on one side of the vehicle is completed, the vehicle side intelligent purge unit detours to the other side of the vehicle and repeats the purge until all purge points are purged; Step S5: The underbody intelligent purge unit and the vehicle side intelligent purge unit return to the standby point, and the staff resets the cover of the purge area.

6. A subway intelligent purge method according to claim 5, characterized in that: In step S2, the mobile unit of the underbody intelligent purge unit is provided with a visual positioning unit, and a positioning method combining visual detection and pulse offset is adopted to make the mobile units of the underbody intelligent purge unit and the vehicle side intelligent purge unit reach the purge point. The positioning method includes the following steps: Step S21: define the standby point of the mobile unit of the undercarriage intelligent purge unit to be in the non-interference area, create a template matching model, shield the non-detection area, the subway vehicle moves slowly relative to the undercarriage intelligent purge unit, and the visual positioning unit automatically detects the reference object of the subway vehicle, obtains the offset data and transmits it to the vehicle-side intelligent purge unit wirelessly; Step S22: A magnetic strip and an IC card are provided at the bottom of the mobile unit of the vehicle side intelligent purge unit. The magnetic strip and the IC card are used to track and locate the vehicle bottom intelligent purge unit, and the offset data is combined to accurately locate the unit to the purge point.

7. The intelligent subway purge method according to claim 5, characterized in that: In step S3, obtaining the purge scheme includes the following steps: Step S31: image processing, identifying the parts at the purge point, acquiring a data set of the parts at the purge point through a visual inspection unit, and performing grayscale, morphological transformation and probabilistic Hough transformation on the data set for identification, and performing precise extraction by using dynamic factor setting and secondary pseudo-edge elimination; Step S32: Segment the purge area according to the processed image, that is, segment the dust and the parts at the purge point, perform real-time accurate dust segmentation on different types of parts according to complex working conditions, illumination changes, and the surface state of the parts at the purge point, establish a segmentation method based on a fully convolutional neural network, and obtain a segmented image; Step S33: Based on the purge level assessment framework established according to the data set, the segmented image is evaluated according to the purge level assessment framework, and a purge scheme of a corresponding level is selected, wherein the purge level assessment framework includes feature extraction, ROI extraction and purge level assessment.

8. A subway intelligent purge method according to claim 7, characterized in that: The image processing specifically includes the following steps: Step S311: The scanned original image is processed using a weighted grayscale method. For the pixel point (i, j) of the original image, the weighted grayscale value is calculated according to the following formula: I(i,j)=0.30R(i,j)+0.59G(i,j)+0.11B(i,j), Among them, R(i,j) is the red component, G(i,j) is the green component, and B(i,j) is the blue component. Mark the area of ​​interest, calculate the mean and variance of the grayscale of each pixel, obtain the grayscale image, and then segment the grayscale image to obtain a binary image; Step S312: using a morphological processing method to perform expansion processing and corrosion processing on the obtained binary image, subtracting the expansion result from the corrosion result to obtain the edge information of the component at the purge point, and obtaining an edge image; Step S313: using the probability Hough transform calculation formula ρ=xcosθ+ysinθ, sampling some points on the edge image and performing cumulative calculation to obtain a sampled image; Step S314: Perform Canny edge detection on the sampled image. First, use the Gaussian function to smooth the image preprocessing, then use the threshold algorithm to perform optimal edge detection processing, and perform secondary pseudo-edge processing by comparing the difference between the pixel value of the pixel point and the arithmetic mean. Finally, perform regional fitting with the parts at the purge point, save the image, and identify the parts at the purge point. The secondary pseudo-edge processing formula is: Among them, n and m are the total number of columns and rows of the image respectively, and c is the dynamic threshold.

9. The intelligent subway purge method according to claim 7, characterized in that: The segmentation method of the purge area using the RPLSN dust pixel level segmentation network comprises the following steps: Step S321: Divide the data in the image processed in step S31 into each layer, and each layer of data is a three-dimensional array of w×h×d, where w and h are spatial dimensions and d is the channel dimension. In the subsequent layers, the data vector of each pixel coordinate is added, and a large number of feature channels are introduced, so that the background features of the target image are propagated to a higher resolution layer to obtain a feature map; Step S322: Softmax function is calculated on the feature map to obtain the output result, and the weight value of the edge pixel is calculated to balance the frequency of occurrence of the feature data. The output formula of the Softmax function is as follows: Among them, α β (x) represents the activation value of the βth layer of the pixel at position x in the feature map; Step S323: discard the fully connected layer, use the effective part of each convolutional layer, and calculate the average of the segmented areas. The calculation formula of the segmented areas is as follows: Among them, mIoU is the average intersection over union ratio, that is, the intersection of the predicted segmentation area and the actual segmentation area divided by the union of the predicted segmentation area and the actual segmentation area, i represents the true value, j represents the predicted value, and p ij is the total number of pixels of true pixel category i that are predicted to be of category j; Step S324: Divide the segmented area into a dust area and a component area, and mark different areas.

10. The intelligent subway purge method according to claim 5, characterized in that: After the purge is completed, the visual inspection unit collects the image of the purge and evaluates the purge effect on the image. If the evaluation is qualified, proceed to the next step. Otherwise, repeat the purge after re-obtaining the purge plan.

Images