An image recognition-based intelligent sensing method for cable arrangement and retraction control

Abstract

The application provides an image recognition-based intelligent sensing method for cable arrangement and release control, which is based on a neural convolution network target detection algorithm to intelligently control the release and arrangement. When the traditional release and arrangement control system relying on sensors fails, the technology can take over the release and arrangement task. The application detects the cable on the cable inlet and the coordinate position information of the cable arrangement mechanism through the target detection algorithm, converts the information into the corresponding percentage, and transmits the information to the control system to control the movement of the reel and the cable arrangement mechanism. Meanwhile, an edge correction formula is proposed to solve the problem of inaccurate edge percentage, and the transmission information with the control system is used to solve the problem of setting value change after the layer change is completed. The application can realize intelligent control of release and arrangement through the target detection algorithm, and after correction, the full-scale accuracy can be less than 5%.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of cable array paying-in and paying-out control, and particularly relates to an intelligent sensing method for cable array paying-in and paying-out control based on image recognition. BACKGROUND

[0002] The sensors used by the traditional cable array paying-in and paying-out control system are generally proximity switches and rotary encoders. The control system receives the information transmitted by the sensors, such as the angle of the drum rotation and the distance of the cable array mechanism movement, and calculates the control output of the cable array mechanism according to the difference between the position of the cable outlet of the drum and the position of the cable array mechanism, or the difference between the position of the cable inlet of the drum and the position of the cable array mechanism. In the application scenarios of unmanned or high-reliability, the sensing device needs to have appropriate redundant design to ensure that the paying-in and paying-out task can continue to be executed when the sensing device fails. However, it is difficult to realize the integrated installation of multiple sensors in the actual design of the paying-in and paying-out device. SUMMARY

[0003] In the cable array paying-in and paying-out control system, a camera is generally used as a monitoring device. In order to solve the above technical problems, when the original sensor of the paying-in and paying-out system fails, the camera can be used to replace the original sensor to continue the cable array paying-in and paying-out control. The application provides an intelligent sensing method for cable array paying-in and paying-out control based on image recognition.

[0004] The application adopts the following technical solutions:

[0005] An intelligent sensing method for cable array paying-in and paying-out control based on image recognition comprises the following steps:

[0006] S1 image shooting and conversion: the camera shoots and uses the rtsp stream mode to convert each frame of image data through the network port;

[0007] S2 target detection and information classification: target detection is performed, the image is input into an image recognition algorithm model to obtain the prediction box and position coordinate information of the cable at the cable outlet of the drum and the cable array mechanism after detection is completed, and the coordinate information of the cable position at the cable outlet of the drum and the position of the cable array mechanism is classified, or the prediction box and position coordinate information of the cable at the cable inlet of the drum and the cable array mechanism after detection is completed are obtained, and the coordinate information of the cable position at the cable inlet of the drum and the position of the cable array mechanism is classified;

[0008] S3 percentage calibration: the coordinate information of the two types obtained in step S2 is converted into the cable movement position percentage and the cable array mechanism movement position percentage in the corresponding image;

[0009] S4 data transmission: a TCP connection is established with the control system, and the two percentage values obtained in step S3 are sent to the control system through TCP;

[0010] S5 cable arrangement control: according to the obtained percentage value, the control system calculates the distance difference between the positions of the drum cable outlet and the cable arrangement mechanism, or the distance difference between the positions of the drum cable inlet and the cable arrangement mechanism, and through a control algorithm, the cable arrangement mechanism performs a cable arrangement operation or a cable release operation on the cable on the drum.

[0011] The training method of the image recognition model in step S2 includes:

[0012] S21 data collection: image information of the cable and the cable arrangement mechanism at different positions is obtained through a camera, and the obtained data is divided into a training sample set and a test sample set;

[0013] S22 model training: based on the self-made training sample set and test sample set, the image recognition algorithm model is trained to obtain the best training model.

[0014] Preferably, in step S21, the camera is vertically installed above the drum and the cable arrangement mechanism.

[0015] Preferably, in step S22, the image recognition model is compared after parameter adjustment, including modifying the initial image size, initial learning rate, training times and other training parameters, and multiple iterations are trained to obtain the optimal image recognition algorithm model.

[0016] In step S3, the midpoint of the detection frame when the cable and the cable arrangement mechanism move to the leftmost position is set to 0%, and the midpoint of the detection frame when the cable and the cable arrangement mechanism move to the rightmost position is set to 100%.

[0017] Preferably, in step S3, after the cable layer change is completed, the 0% and 100% set points of the new layer are adjusted to solve the problem of changing set values and improve the accuracy of percentage detection.

[0018] Preferably, the percentage obtained in step S3 is optimized and revised: the percentage of the cable movement position and the cable arrangement mechanism movement position is adjusted from 0%-100% to -50%-50% in proportion, and the following edge correction formula is established:

[0019]

[0020] In the formula, k is a constant, y is the corrected percentage, and i is the percentage before correction. By calculating the value of y through the formula, the percentage of the cable movement position and the percentage of the cable arrangement mechanism movement position are obtained, and the edge of the image is corrected.

[0021] Preferably, the k value of the edge correction formula is 0.04-0.07, and the distance prediction error is controlled within 5% through the edge correction formula.

[0022] The intelligent sensing method for retraction and extension control of the present invention produces the following beneficial effects:

[0023] Effect 1: When the original sensor of the cable reeling system fails, this invention can use the original camera used for redundancy backup to replace the original sensor and continue to perform cable reeling control.

[0024] Effect 2: This invention uses a target detection algorithm based on neural convolutional networks to intelligently control cable deployment and retraction. Compared to traditional methods that rely on numerical calculations based on various sensors, this invention significantly improves the accuracy and real-time performance of cable laying. When cable overlap occurs during cable laying, it will continue laying cables normally according to their current positions, instead of continuing to overlap at the same location as in traditional methods. Attached Figure Description

[0025] Figure 1 This is a top view schematic diagram of the cable deployment and take-up device of the present invention;

[0026] Figure 2 This is a design composition diagram of the cable deployment and take-up control system of the present invention;

[0027] Figure 3 The flowchart shows the intelligent sensing method for cable winding and unwinding control based on image recognition according to the present invention.

[0028] In the diagram, 1-reel, 2-cable, 3-cable laying device, 4-camera, 5-cable detection frame, 6-cable laying mechanism detection frame. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0030] Reference Figure 1 As shown, the drum 1 and the cable laying mechanism 3 are arranged parallel to each other. A camera 4 is positioned above the drum 1 and the cable laying mechanism 3. The camera 4 tracks and captures images of the cable 2 and the cable laying mechanism 3. Through image recognition using the cable detection frame 5 and the cable laying mechanism detection frame 6, the real-time status of the cable movement position at the cable outlet or inlet of the drum 1 and the movement position of the cable laying mechanism 3 is determined. In this invention, the center point of the cable detection frame 5 is selected as the coordinate information of the movement position at the cable outlet or inlet of the drum 1, and the center point of the cable laying mechanism detection frame 6 is selected as the coordinate information of the movement position of the cable laying mechanism 3.

[0031] The working principle of this mechanism is that the cable winding mechanism 3 winds the cable 2 layer by layer on the drum 1. The drum 1 rotates continuously, and the cable winding mechanism 3 delivers the cable 2 to the cable inlet of the drum 1 or sends the cable 2 out from the cable outlet of the drum 1. During this process, the cable winding mechanism 3 can move back and forth in a direction parallel to the length of the drum 1, thereby realizing the layer-by-layer winding or releasing action of the cable 2 on the drum from left to right and from inside to outside.

[0032] Referring to Figure 2 As shown in the drawings, the image recognition-based cable arrangement and take-up control intelligent sensing system of the application is composed of a camera, a video processing unit using a target detection network, and a control system. The detection network is embedded in the video processing unit, so that the detection network has good mobility and applicability.

[0033] The cable arrangement and take-up control intelligent sensing method first detects the position coordinate information of the cable exit of the drum and the cable arrangement mechanism using a target detection algorithm, and then transmits the data to the control system, which controls the cable arrangement and take-up according to the distance difference between the two.

[0034] Referring to Figure 2 As shown in the drawings, an image recognition-based cable arrangement and take-up control intelligent sensing method comprises the following steps:

[0035] S1 collects data, installs a camera above the positions of the drum and the cable arrangement mechanism, and shoots downward to obtain image information of the cable and the cable arrangement mechanism at different positions. The obtained data is divided into a training sample set and a test sample set.

[0036] S2 trains the image recognition model based on the self-made training sample set and test sample set, and compares the results after adjusting the parameters, such as modifying the initial image size, initial learning rate, and training times, and iteratively training multiple times to finally obtain the optimal parameter configuration training model.

[0037] S3 the camera shoots and uses the rtsp stream mode to convert each frame of image data through the network port.

[0038] S4 performs target detection, inputs the image into the neural network model trained in step S2, extracts features, and performs convolution calculation to obtain the cable prediction box and the cable arrangement mechanism prediction box of the cable exit of the drum or the cable exit of the drum, as well as the position coordinate information, and classifies the prediction box according to the feature comparison, and divides the two types of coordinate information into the cable position of the cable exit of the drum or the cable exit of the drum, and the cable arrangement mechanism position.

[0039] S5 performs percentage calibration to obtain the two types of coordinate information of the cable position of the cable exit of the drum or the cable exit of the drum, and the cable arrangement mechanism position, and converts them into the cable movement position percentage and the cable arrangement mechanism movement position percentage in the corresponding image. The cable movement position percentage refers to the length of the cable position of the cable exit of the drum or the cable exit of the drum from the initial end of the drum (left end or right end) in the length direction of the drum, which accounts for the percentage of the total length of the drum. The cable arrangement mechanism movement position percentage refers to the length of the cable arrangement mechanism position from the initial end of the cable arrangement mechanism in the movement direction of the cable arrangement mechanism, which accounts for the percentage of the total length of the cable arrangement mechanism.

[0040] In the embodiment, the midpoint of the detection frame when the cable and the cable arrangement are moved to the leftmost position is set as 0%, and the midpoint of the detection frame when the cable and the cable arrangement are moved to the rightmost position is set as 100%.

[0041] S6 optimizes the percentage, since the currently used camera is a monocular camera and is a point shooting type, the image edge becomes a curved surface, which causes inaccurate position information. When the cable layer is changed, the 0% and 100% set points of the new layer change, and if the original set points are still used, the percentage will be inaccurate, thereby affecting the control of the cable arrangement. By adjusting the 0% and 100% set points of the new layer, the accuracy of the percentage detection is improved.

[0042] Further, in the correction test of the present application, in order to better intuitively observe the difference between the real distance and the predicted distance, the percentage of the cable movement position and the cable arrangement movement position is temporarily adjusted from 0% to 100% to -50% to 50% in proportion. The difference between 0% and 50% and -50% is greater, so the error needs to be corrected. The present application proposes a formula The image edge is corrected, where k is a constant, the value range is 0.04-0.07, y is the corrected percentage, and i is the percentage before correction. The formula uses a linear function to linearize the predicted distance. The y value calculated by the formula is added to 50% to obtain the initial percentage (0%-100%).

[0043] For a given scene, according to the actual situation of the cable arrangement, the cable and the reel (such as the movement speed and distance of the cable arrangement, the diameter and the number of layers of the cable, the length and the rotation speed of the reel, etc.), the k value 0.055 (which can also be other values) can be obtained by training. By applying the above edge correction formula to correct the percentage of the cable arrangement movement position, when i is -50%, -25%, 0%, 25% and 45%, y is -33%, -8%, 10%, 32% and 48%, and the real value m measured by the sensor is -32%, -8%, 10%, 33% and 50%, that is, the corrected percentage y is close to the real value m measured by the sensor.

[0044] At the same time, according to the above edge correction formula and the k value, the percentage of the cable movement position can be calculated to obtain the effect that the corrected percentage y is close to the real position.

[0045] Therefore, after using the edge correction formula, the corrected percentage is fitted to be close to the real percentage. Through measurement, the maximum value of the real distance and the image predicted distance is 4.87%, which meets the accuracy requirement of less than 5% error.

[0046] S7 establishes a TCP connection with the control system, and sends the two percentage values obtained in step S6 to the control system through TCP.

[0047] S8 calculates the distance difference between the positions of the cable outlet of the winding drum and the cable arranging mechanism, or the distance difference between the positions of the cable inlet of the winding drum and the cable arranging mechanism according to the obtained percentage values, and makes the cable arranging mechanism perform the cable arranging operation or the cable releasing operation on the cable on the winding drum through a control algorithm.

[0048] It should be noted that the above embodiments are merely the preferred embodiments of the present application, and are not intended to limit the protection scope of the present application, and equivalent transformations made on the basis of the above embodiments all belong to the protection scope of the present application.

Claims

1. An intelligent sensing method for cable deployment and take-up control based on image recognition, comprising the following steps: S1 Image Capture and Conversion: The camera captures images and uses RTSP streaming mode, then transcodes each frame of image data via the network port. S2 Target Detection and Information Classification: Target detection is performed by inputting the image into the image recognition algorithm model to obtain the predicted bounding boxes and position coordinates of the cable and cable arrangement mechanism at the cable outlet of the drum after detection. The coordinates of the cable position and cable arrangement mechanism at the cable outlet of the drum are then classified. Alternatively, the predicted bounding boxes and position coordinates of the cable and cable arrangement mechanism at the cable inlet of the drum after detection are obtained, and the coordinates of the cable position and cable arrangement mechanism at the cable inlet of the drum are then classified. In step S2, the training method for the image recognition model includes: S21 Data Acquisition: The camera is vertically mounted above the drum and cable laying mechanism. Image information of the cable and cable laying mechanism at different positions is acquired through the camera. The acquired data is divided into training sample set and test sample set. S22 Model Training: Based on self-made training and test sample sets, the image recognition model is adjusted and compared. Training parameters, including initial image size, initial learning rate, and number of training iterations, are modified and trained multiple times to obtain the optimal image recognition algorithm model. S3 Percentage Calibration: The coordinate information of the two types obtained in step S2 is converted into the percentage of cable movement position and the percentage of cable laying mechanism movement position in the corresponding image; In step S3, the midpoint of the detection frame when the cable and cable laying mechanism move to the leftmost position is set to 0%, and the midpoint of the detection frame when the cable and cable laying mechanism move to the rightmost position is set to 100%. After the cable layer is changed, the 0% and 100% setting points of the new layer are adjusted. S4 Data Transmission: Establish a TCP connection with the control system and send the two percentage values ​​obtained in step S3 to the control system via TCP; S5 Cable Laying Control: The control system calculates the distance difference between the positions of the cable outlet of the drum and the cable laying mechanism, or the distance difference between the positions of the cable inlet of the drum and the cable laying mechanism, based on the obtained percentage value, and uses the control algorithm to enable the cable laying mechanism to perform cable laying or cable laying operations on the drum.

Images