Railway bulk cargo discharge hopper identification and positioning method based on machine vision

Abstract

The invention provides a railway bulk cargo discharge hopper identification and positioning method based on machine vision, and relates to the technical field of railway bulk cargo discharge hopper tracking and positioning. The method comprises the following steps: collecting a depth map sample of a target discharge hopper; preprocessing the depth map sample to generate a sample data set; extracting geometric features of the discharge hopper from the sample data set; taking the geometric features of the discharge hopper as input, and performing target identification and positioning training of the discharge hopper by using a CNN (Convolutional Neural Network) to obtain a discharge hopper identification model; and the real-time depth map captured by the camera is sent to the discharge hopper recognition model for real-time recognition and positioning of the discharge hopper, the discharge hopper recognition model outputs three-dimensional space information and the existence probability of the discharge hopper, the result is transmitted to the control system, and the mechanical arm is guided to execute discharge operation. The method is suitable for real-time monitoring in the unloading operation, the current position of the bucket is accurately positioned in real time, impact damage between the bucket and a cargo hold is avoided in time, and the unloading efficiency and safety are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of railway bulk cargo unloading hopper tracking and positioning, and particularly relates to a method for identifying and positioning a railway bulk cargo unloading hopper based on machine vision. Background Art

[0002] With the development of modern science and technology, machine vision technology has become an important part of automation and intelligence in multiple industries. Especially in the fields of industrial manufacturing, logistics, transportation, and healthcare, the characteristic of flexible conversion between the two-dimensional world and the three-dimensional world makes the application prospect of machine vision broad. Machine vision technology can perform tasks such as high-efficiency quality inspection, target recognition, and positioning navigation by image acquisition, processing, and analysis, replacing the human eye. Compared with traditional manual operations, machine vision systems have the advantages of high precision, high efficiency, low error, and being unaffected by human factors, becoming one of the key technologies in modern automation systems.

[0003] Through actual research, the unloading operation in the railway bulk cargo transportation industry is considered an important link to ensure transportation efficiency and safety. The unloading operation of bulk cargo not only requires high efficiency and precision but also needs to ensure the positioning accuracy of the unloading hopper. For a long time, the precise positioning of the unloading hopper has been the focus of industry attention. However, the existing unloading methods rely on the cooperation of video monitoring, manual operation, and mechanical equipment. Although this method can meet the basic operation requirements, it exposes a series of problems when facing complex unloading environments.

[0004] During the unloading process of bulk cargo, the cargo usually has irregular shapes and height variations, making visual monitoring extremely difficult. Especially when the surface of the cargo has similar colors, especially when the color identity of coal or other bulk materials is high, it is very difficult for traditional video monitoring systems to effectively distinguish and locate the position of the unloading hopper. The dependence of the vision system on color and shape means that when the color of the cargo is similar to the color of the background or equipment, the monitoring results will not be accurate enough. This problem of "color identity" poses challenges to visual monitoring, often resulting in the monitoring system being unable to judge the state of the unloading hopper in real time and accurately, thereby affecting the efficiency and quality of the unloading operation.

[0005] In addition, the application of traditional video monitoring systems is usually affected by environmental lighting conditions, and it is even more difficult to ensure image quality in complex weather or harsh environments (such as haze, smoke, etc.). When the lighting at the operation site is uneven or there is a large amount of smoke and dust, the video monitoring cannot clearly capture the position and state of the unloading hopper. This environmental dependence and hardware limitation lead to difficulties in accurately obtaining and real-time feedback of the position of the unloading hopper, seriously restricting the improvement of operation efficiency.

[0006] At the same time, the latency of the video surveillance system is also an issue that cannot be ignored. Due to the time delay in image acquisition, transmission, and processing in traditional video surveillance systems, the real-time performance is insufficient. For the rapidly changing unloading operation scenario, timely feedback is often not provided, resulting in the mechanical equipment being unable to make precise adjustments based on real-time data, thus affecting the smoothness and safety of the entire unloading process and making it difficult to meet the requirements of modern industrial development. Summary of the Invention

[0007] Object of the Invention: Aiming at the limitations of traditional visual monitoring methods, the present invention proposes a method for identifying and positioning a railway bulk cargo unloading bucket based on machine vision. The purpose is to use real-time image processing and analysis algorithms of machine vision to be able to accurately locate the current position of the bucket in real time, including the pitch angle, roll angle, yaw angle, and three-axis translation vector of the bucket. The positioning accuracy far exceeds that of traditional visual monitoring. At the same time, the method for identifying and positioning a railway bulk cargo unloading bucket based on machine vision is applicable to real-time monitoring during the unloading operation, thus being able to timely avoid impact damage between the bucket and the cargo hold and improve the efficiency and safety of unloading.

[0008] To solve the above technical problems, the present invention provides a method for identifying and positioning a railway bulk cargo unloading bucket based on machine vision, and the technical solution is as follows:

[0009] Collect depth map samples of the target unloading bucket;

[0010] Perform preprocessing on the depth map samples to generate a sample data set;

[0011] Extract geometric features of the unloading bucket from the sample data set;

[0012] Use the geometric features of the unloading bucket as input, and use a convolutional neural network (CNN) for target recognition and positioning training of the unloading bucket to obtain an unloading bucket recognition model;

[0013] Send the real-time depth map of the unloading bucket captured by the camera into the unloading bucket recognition model for real-time recognition and positioning of the unloading bucket. The unloading bucket recognition model outputs the three-dimensional space information and existence probability of the unloading bucket, and transmits the results to the control system to guide the robotic arm to perform the unloading operation.

[0014] In a further embodiment, performing preprocessing on the depth map samples includes:

[0015] For each pixel in the depth map sample Perform mean filtering processing, and the calculation formula is:

[0016]

[0017] In the formula, k is the size of the mean filter, Indicates a pixel The abscissa and ordinate are shifted by m and n pixel values respectively.

[0018] Smoothing is performed using a Gaussian kernel, and the calculation formula is:

[0019]

[0020] In the formula, is the value of the Gaussian kernel at the point and is the standard deviation of the Gaussian kernel.

[0021] In a further embodiment, geometric features of the discharge hopper are extracted from the sample dataset, specifically including:

[0022] Determine the edge region from the sample dataset, and the calculation formula is as follows:

[0023]

[0024]

[0025]

[0026] In the formula, and are the gradients of the image in the x and y directions respectively, G represents the edge intensity, and then the edge region in the image is determined.

[0027] Use K-means clustering for region segmentation, extract regions with similar depth values, and extract the region where the discharge hopper is located. The calculation formula is as follows:

[0028]

[0029]

[0030] In the formula, is the distance from the depth image pixel to each cluster center, is the cluster center, is the set of pixels belonging to cluster k, is the number of pixels in cluster k.

[0031] In a further embodiment, during the training of the convolutional neural network CNN, semi-supervised learning is used to optimize the convolutional neural network CNN, and the loss function of semi-supervised learning is:

[0032]

[0033] In the formula, is the classification loss, is the regression loss, is the depth map reconstruction loss, is the self-supervised loss, both of which are hyperparameters set through cross-validation and engineering experience.

[0034] In a further embodiment, the three-dimensional space information includes the three-dimensional coordinates of the center point of the discharge hopper, and the three-dimensional coordinates of the center point The calculation formula is:

[0035]

[0036]

[0037]

[0038] In the formula, is the pixel coordinate, is the depth value of the corresponding pixel, and are the focal lengths of the camera, and are the optical center positions of the camera.

[0039] In a further embodiment, a projection equation is constructed to project the three-dimensional space point of the center point three-dimensional coordinates into a two-dimensional image point :

[0040]

[0041] In the formula, is the two-dimensional image point, is the three-dimensional space point, and the superscript T represents the transpose.

[0042] Beneficial effects:

[0043] (1) Compared with the traditional manual visual monitoring method, the present invention has higher accuracy and more comprehensive operation capabilities. Traditional methods usually rely on manual judgment or simple mechanical control, are easily interfered by human factors, and it is difficult to achieve high-precision and high-efficiency operations. The recognition and positioning method based on machine vision can, through the stereo vision system and depth map technology, accurately capture the position of the discharge hopper in real time in a dynamic and complex environment, avoiding the errors caused by perspective problems or sensor limitations in traditional methods.

[0044] (2) By precisely analyzing the characteristics and geometric shape of the discharge hopper surface, and combining the depth map and convolutional neural network (CNN), this method can effectively identify the three-dimensional position and posture of the discharge hopper and provide real-time feedback to the control system for precise control. Compared with traditional single-sensor technologies, machine vision can not only provide richer visual information but also efficiently process data changes under different lighting conditions and angles, further improving the robustness and accuracy of the system.

[0045] (3) This method is not only applicable to the positioning of the discharge hopper in a static environment but also capable of providing real-time target recognition and positioning during dynamic operations, making the discharging process more automated and intelligent. Through real-time data processing and target tracking, the system can ensure the precise operation of the discharge hopper in a complex environment, effectively improving work efficiency and safety. In addition, the detection method based on machine vision does not require contact with the discharge hopper, avoiding potential damage or performance degradation problems in traditional sensors and significantly enhancing the reliability and durability of the overall system. Description of the Drawings

[0046] Figure 1 It is the schematic diagram of the convolutional neural network architecture in this method.

[0047] Figure 2 It is the specification diagram of the checkerboard for camera calibration.

[0048] Figure 3 It is the schematic diagram of three-dimensional ranging by stereo vision in this method.

[0049] Figure 4 It is the schematic diagram of the K-means clustering algorithm in this method.

[0050] Figure 5 It is the overall flow chart of the system in this method. Detailed Embodiment

[0051] In the following description, numerous specific details are given to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details. In other instances, some well-known technical features are not described to avoid confusion with the present invention.

[0052] Combined with Figure 1 the schematic diagram of the dual-channel convolutional neural network architecture, a method for identifying and positioning a railway bulk cargo discharge hopper based on machine vision is proposed. The specific steps are as follows:

[0053] Step 1: Perform camera calibration through the checkerboard calibration method. The size of the checkerboard is 11×7, and the size of each square is 25 mm.

[0054] Step 2: The calibrated camera collects the depth map samples of the discharge hopper by an optical method. The resolution of the camera is 2560×1440, and the depth FPS is 100Hz;

[0055] Step 3: Preprocess the depth map samples, apply a mean filter and a Gaussian kernel to smooth the images, and generate a high-quality sample dataset;

[0056] Step 4: Extract features from the training sample dataset, and extract the geometric features of the discharge hopper through an edge detection and depth map segmentation algorithm;

[0057] Step 5: Use the features as inputs, and use a convolutional neural network (CNN) to perform object recognition and localization training on the discharge hopper. During the training process, combine the position information and geometric features in the depth map, and optimize the model through semi-supervised learning so that it can efficiently recognize the discharge hopper in different working environments;

[0058] Step 6: Through a real-time data acquisition and transmission system, send the real-time depth map captured by the camera into the trained model for real-time recognition and localization of the discharge hopper. The model output includes the three-dimensional spatial information and the existence probability of the discharge hopper, and send the results to the control system to guide the robotic arm to perform accurate discharging operations.

[0059] As a preferred embodiment, the output of the three-dimensional spatial coordinate information includes the three-dimensional coordinates of the center point of the discharge hopper.

[0060] As a preferred embodiment, the resolution of the camera is 2560×1440, and the depth FPS is 100Hz, so as to better improve the spatial resolution and detail capture ability of the depth map. The intersecting binocular stereo vision measurement model is as Figure 3 shown. In the intersecting binocular stereo vision system, the optical axes of the two cameras intersect inward at a point. This model is more in line with the visual principle of the human eye. When the human eye observes a nearby object, the lines of sight of the two eyes intersect inward. Through the principle of triangulation, according to the imaging positions of the points on the imaging planes of the two cameras, use trigonometric function relationships to calculate the three-dimensional coordinates of the object.

[0061] As a preferred embodiment, the calibration method in Step 1 is the checkerboard calibration method, which is applicable to dynamic scenes. Compared with other static calibration methods, this method does not require particularly accurate control point positions. By taking checkerboard images at different angles and positions, it is especially suitable for application scenarios that require dynamic calibration.

[0062] The calibration checkerboard is Figure 2The specifications are 11×7 in size, with a grid size of 25mm. Place the checkerboard within the camera's field of view, ensuring that the checkerboard plane is as parallel as possible to the camera's field of view, and adjust the position of the checkerboard to obtain sufficient calibration angles and multiple effective perspectives. The purpose of this operation is to ensure that the camera can capture images of the checkerboard from different angles, thereby improving the calibration accuracy.

[0063] During the calibration process, any actions that may cause camera vibration should be avoided, such as ground vibration caused by monitoring poles or machinery. At the same time, avoid strong reflective surfaces in the calibration environment, especially glass or smooth surfaces within the camera's field of view, as these may cause light reflection or interference and affect the normal operation of the camera. In this embodiment, the calibration accuracy is ensured by measures such as avoiding vibration sources, controlling reflection interference, and environmental light control.

[0064] As a preferred embodiment, the acquisition quality of the depth map in step two is closely related to the surface characteristics and structural features of the discharge hopper. The surface smoothness, reflectivity, and material uniformity of the discharge hopper will directly affect the measurement accuracy of the depth sensor. Generally speaking, when the surface of the discharge hopper is relatively smooth and there are no obvious reflection changes, the depth sensor can more accurately capture the depth information of the target area, thereby generating a clear and accurate depth map. When there are irregular textures, stains, corrosion, or wear on the surface of the discharge hopper, the depth sensor may be interfered by lighting conditions and surface reflections, resulting in errors or omissions in the depth map.

[0065] As a preferred embodiment, step three preprocesses the depth map samples. Combining a mean filter with a Gaussian kernel to smooth the image is easy to improve the quality and detail retention of the depth map. For each pixel Perform mean filtering, and its calculation formula is:

[0066]

[0067] where k is the filter size, set to 3×3.

[0068] Further use a Gaussian kernel for smoothing, and its calculation formula is:

[0069]

[0070] where is the standard deviation of the Gaussian kernel, set to 0.6.

[0071] As a preferred embodiment, step four performs geometric feature extraction of the discharge hopper through the Sobel operator and K-means clustering depth map segmentation algorithm, effectively improving the accuracy of feature extraction. The calculation formula is:

[0072]

[0073]

[0074]

[0075] Among them, and are the gradients of the image in the x - direction and y - direction respectively, G represents the edge intensity, and then the edge region in the image is determined.

[0076] Figure 4 Shows the visualization effect of the principle of the K - means clustering algorithm. Three different perspectives respectively show the clustering process of data points in three - dimensional space in different stages of the K - means algorithm clustering. The points in each region represent different clustering clusters. The points in the figure are all black dots, representing data points. The convergence process of multiple clustering centers is shown through different perspectives. As the iteration progresses, the data points will gradually gather near their respective clustering centers until the clustering centers are stable.

[0077] The goal of the K - means clustering algorithm is to minimize the sum of the squared distances from each data point to its clustering center, that is, to optimize the objective function. The Euclidean distance is used to represent the distance between the data point and the clustering center, and its calculation formula is:

[0078]

[0079] Among them, is the feature vector of the data point , is the coordinate of the clustering center k, and m is the dimension of the data.

[0080] At this time, once the data points are assigned to the corresponding clustering centers, calculate the new center of each cluster. The new clustering center is the mean of all data points in the cluster, and the calculation formula is:

[0081]

[0082] Among them, is the number of data points belonging to cluster k, represents all data points in cluster k.

[0083] The goal of the K - means algorithm is to minimize the sum of the squared distances of each data point from its belonging clustering center. The calculation formula of the objective function J is:

[0084]

[0085] Among them, is the data point and the clustering center the Euclidean squared distance is the set of data points in cluster k

[0086] Finally, the K-means algorithm updates the cluster centers and assigns data points through continuous iterative loops until the objective function converges or reaches a predetermined number of iterations

[0087] As a preferred embodiment, in step five, a CNN architecture with multi-level and deep feature extraction is adopted (see the architecture schematic diagram in Figure 1 ). It integrates cross-layer feature fusion and semi-supervised learning, including convolutional layers, pooling layers, fully connected layers and classification layers. By integrating feature maps from different layers, especially low-level and high-level features, it enhances the multi-scale perception ability of the network and finally outputs a scalar value representing the probability of the discharge hopper existing in the input image. Among them, cross-layer feature fusion is one of the key steps in the network. At this stage, the convolution results of the depth map and the geometric feature map are fused into a unified feature map, better integrating the useful information from the two different inputs

[0088] Step a: Convolutional layer and pooling layer

[0089] Convolution operations are performed on the depth map and geometric features respectively. The input depth map has 3 channels, the input geometric feature map has 2 channels, and the output number of channels is 32, that is, the number of convolutional kernels is 32. Each channel first extracts local features through the convolutional layer, and then reduces the spatial dimension of the feature map through the pooling layer and extracts high-level features. The calculation formula of the convolutional layer is:

[0090]

[0091]

[0092] Among them, ReLU (Rectified Linear Unit) is a commonly used activation function, which can increase the non-linear representation ability of the network and are convolutional kernels has a dimension of (3, 3, 3, 32) has a dimension of (3, 3, 2, 32) and are biases has a dimension of (32) has a dimension of (32). For and The specific values are obtained by using He initialization, and its calculation formula is:

[0093]

[0094] Among them, is the number of input channels.

[0095] Step b: Cross-layer feature fusion:

[0096] Effectively combine features at different levels to enhance the model's expressive ability and improve the diversity of features. By adjusting the weighting coefficients, the influence of different feature maps on the final fused features can be flexibly controlled. The calculation formula is:

[0097]

[0098] Among them, and are the weighting coefficients used to control the weight of each feature map, set through domain knowledge and feature importance analysis is 0.6, is 0.4.

[0099] Step c: Fully connected layer processing and classification output:

[0100] The fused feature map is further processed through multiple fully connected layers to learn high-dimensional features and output classification results. Finally, a value is output, and binary classification is performed through the sigmoid activation function to output the probability of the existence of the discharge hopper. The calculation formula of the fully connected layer is:

[0101]

[0102] Among them, is the weight matrix of the fully connected layer, is the bias.

[0103] After the output of the fully connected layer, the final classification is performed. This classification task is a binary classification problem, and the goal is to output the probability of the existence of the discharge hopper. It is calculated through a classification layer:

[0104]

[0105] Among them, is the weight matrix of the classification layer, is the bias term, represents the Sigmoid activation function, which is used to map the output to between 0 and 1, representing the probability of the existence of the discharge hopper.

[0106] represents the probability of the existence of the discharge hopper. If the probability value is greater than the threshold set to 0.8, it can be determined that the discharge hopper exists; otherwise, it is determined that the discharge hopper does not exist.

[0107] As a preferred embodiment, the semi-supervised learning method adopted in step five combines the data advantages of a limited number of labeled samples and a large number of unlabeled samples, thereby effectively improving the accuracy of the target recognition and positioning of the discharge hopper. Traditional supervised learning methods rely on a large amount of labeled data, while semi-supervised learning methods can better capture the potential laws in the data by training with a small number of labeled samples and using unlabeled samples. The semi-supervised loss function is:

[0108]

[0109] Where is the classification loss, is the regression loss, is the depth map reconstruction loss, is the self-supervised loss, is a hyperparameter, which is set through cross-validation and engineering experience = 5, =1, =3, =0.5.

[0110] As a preferred embodiment, the central position of the discharge hopper in three-dimensional space is the three-dimensional coordinate values displayed on the display screen. These coordinate values are obtained through three-dimensional measurement by the guided stereo vision system and reflect the precise position of the discharge hopper in space. When the stereo vision system captures the depth information of the discharge hopper, after processing and calculation, the system can determine the specific position (X, Y, Z values) of the center point of the discharge hopper in the three-dimensional coordinate system.

[0111] Specifically, in step six, according to the input depth map and geometric features, through the PnP algorithm combined with the camera internal parameters, reasoning and calculation are carried out to output the coordinates of the center point of the discharge hopper in three-dimensional space . The three-dimensional space coordinate calculation formula is:

[0112]

[0113]

[0114]

[0115] Where is the pixel coordinate, is the depth value of the corresponding pixel, in meters. After camera calibration, is the focal length of the camera, is the position of the optical center of the camera, in pixels.

[0116] The constructed projection equation calculation formula is:

[0117]

[0118] wherein is a two-dimensional image point, is a three-dimensional space point.

[0119] The technical process of railway bulk cargo unloading hopper recognition and positioning disclosed in the above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs.

[0120] When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0121] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0122] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as a limitation of the present invention itself. Various changes can be made in its form and details without departing from the spirit and scope of the present invention defined by the appended claims.

Claims

1. A method for identifying and locating a railway bulk cargo unloading hopper based on machine vision, characterized in that: The steps include: Collect depth map samples of the target discharge hopper; Performing preprocessing on the depth map samples to generate a sample data set; Extracting geometric features of the discharge hopper from the sample data set; Taking the geometric features of the discharge hopper as input, a convolutional neural network is used to perform target recognition and positioning training for the discharge hopper to obtain a discharge hopper recognition model; The real-time depth map of the unloading hopper captured by the camera is sent to the unloading hopper recognition model for real-time recognition and positioning of the unloading hopper. The unloading hopper recognition model outputs the three-dimensional spatial information and existence probability of the unloading hopper, and transmits the result to the control system to guide the robotic arm to perform the unloading operation.

2. The method for identifying and locating a railway bulk cargo unloading hopper based on machine vision according to claim 1, characterized in that: Preprocessing the depth map samples includes: For each pixel in the depth map sample Perform mean filtering, the calculation formula is: ； Where k is the mean filter size, Represents pixels The horizontal and vertical axes are shifted by m and n pixel values ​​respectively.

3. The method for identifying and locating a railway bulk cargo unloading hopper based on machine vision according to claim 1 or 2, characterized in that: Performing preprocessing on the depth map samples further includes: Use Gaussian kernel for smoothing, the calculation formula is: ； In the formula, is the Gaussian kernel at point The value at is the standard deviation of the Gaussian kernel.

4. The method for identifying and locating a railway bulk cargo unloading hopper based on machine vision according to claim 1, characterized in that: The geometric features of the discharge hopper are extracted from the sample data set, specifically including: The edge region is determined from the sample data set, and the calculation formula is as follows: ； ； ； In the formula, and are the gradients of the image in the x and y directions respectively, and G represents the edge strength, thereby determining the edge area in the image.

5. The method for identifying and locating a railway bulk cargo unloading hopper based on machine vision according to claim 4, characterized in that: Extracting geometric features of the discharge hopper from the sample data set also includes: Use K-means clustering to perform regional segmentation, extract areas with similar depth values, and extract the area where the discharge hopper is located. The calculation formula is as follows: ； ； In the formula, is the distance from the depth map pixel to each cluster center, is the cluster center, is the set of pixels belonging to cluster k, is the number of pixels in cluster k.

6. The method for identifying and locating a railway bulk cargo unloading hopper based on machine vision according to claim 1, characterized in that: In the process of training the convolutional neural network CNN, semi-supervised learning is used to optimize the convolutional neural network CNN. The loss function of semi-supervised learning is: ； In the formula, is the classification loss, is the regression loss, is the depth map reconstruction loss, is the self-supervision loss, All hyperparameters are set through cross-validation and engineering experience.

7. The method for identifying and locating a railway bulk cargo unloading hopper based on machine vision according to claim 1, characterized in that: The three-dimensional spatial information includes the three-dimensional coordinates of the center point of the discharge hopper, the three-dimensional coordinates of the center point The calculation formula is: ； ； ； In the formula, is the pixel coordinate, is the depth value of the corresponding pixel, , is the focal length of the camera, , is the optical center position of the camera.

8. The method for identifying and locating a railway bulk cargo unloading hopper based on machine vision according to claim 7, characterized in that: Construct the projection equation and transform the three-dimensional coordinates of the center point Three-dimensional space point Projection as 2D image points : ； In the formula, is a two-dimensional image point, is a point in three-dimensional space, and the superscript T indicates transposition.

9. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for identifying and locating a railway bulk cargo unloading hopper based on machine vision as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one executable instruction. When the executable instruction is executed on the electronic device, the electronic device executes the railway bulk cargo unloading hopper identification and positioning method based on machine vision as described in any one of claims 1 to 8.

Images