Rail safe driving determination method and system

Abstract

The invention belongs to the technical field of railway foreign matter invasion, and provides a rail safe driving determination method and system, and the method comprises the steps: obtaining a rail image comprising an invasion foreign matter; according to the track image, determining coordinates of a safety limit area and an invasion limit foreign matter, and based on the safety limit area and the coordinates, determining whether the invasion limit foreign matter is located in the safety limit area; if the invasion and limit foreign matter is located in the safety limit area, judging that the attribute of the invasion and limit foreign matter is planar or three-dimensional; if the attribute of the invasion and limitation foreign matter is three-dimensional, the occupied area and height of the invasion and limitation foreign matter are calculated; judging the degree of influence on driving safety according to the occupied area and the height; the method can improve the accuracy and effectiveness of traffic safety assessment by comprehensively judging the attributes of the invasion limit foreign matter, further calculates the occupied area and height of the foreign matter, and can more comprehensively understand the occupation condition and potential threats of the foreign matter to the track space.

Description

Technical Field

[0001] The present invention belongs to the technical field of railway foreign body intrusion limitation, and in particular relates to a method and system for determining safe track driving. Background Art

[0002] In recent years, the railway has developed rapidly. The environment faced by trains during operation is changeable and complex. Accidents caused by natural disasters such as debris flow and rockfall are more frequent, posing a serious threat to the safety of trains. Therefore, it is necessary to improve the safety protection of railways accordingly.

[0003] At present, the railway has adopted two methods to deal with disasters: setting up guardrails along the railway to prevent disasters from invading the track area and manual patrol inspection. However, these methods are less reliable, and the guardrails may be damaged after a long time of use, while manual inspection of natural disasters is inefficient and not very accurate. As trains run faster and faster, traditional intrusion detection methods can no longer meet the needs of railway development. Summary of the invention

[0004] In view of the problems in the background technology, the present invention proposes a method and system for determining safe track driving.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present disclosure provides a method for determining safe rail driving, which is characterized by comprising: Acquire orbital images including intruding foreign objects; Determine the coordinates of a safety clearance area and an intruding foreign object according to the track image, and determine whether the intruding foreign object is located in the safety clearance area based on the safety clearance area and the coordinates; If the intruding foreign object is located in the safety limit area, determining whether the intruding foreign object is a plane or a three-dimensional object; If the property of the intrusive foreign object is three-dimensional, then the occupied area and height of the intrusive foreign object are calculated; The degree of impact on driving safety is determined based on the occupied area and height.

[0006] Preferably, determining the safe clearance area according to the track image includes: preprocessing the track image; Detecting a track straight line in the track image based on Hough transform; A safe clearance area is obtained according to the track straight line and the preset clearance distance.

[0007] Preferably, the preprocessing of the track image comprises: Processing the track image using a grayscale method based on a weighted average method; Processing the track image using an image denoising method; Binarization processing of the track image using Otsu algorithm; Using Sobel operator to perform gradient filtering on the track image; Extract track edge information based on Canny edge detection algorithm.

[0008] Preferably, determining the coordinates of the intruding foreign object includes: The track image is input into the trained recognition model to obtain the coordinates of the intruding foreign object.

[0009] Preferably, the track image is input into a trained recognition model to obtain the type of the intruding foreign matter; The determination of whether the property of the intruding foreign body is plane or three-dimensional is as follows: According to the type of the intruding foreign object, it is determined whether the intruding foreign object has a plane or a three-dimensional property.

[0010] Preferably, the calculating the occupied area of ​​the intrusive foreign matter includes: Processing the track image based on background difference method and morphological method to obtain the outline of the intruding foreign body and the number of pixels occupied; According to the actual area mapped by each occupied pixel and the number of occupied pixels, the occupied area of ​​the intruding foreign matter is obtained.

[0011] Preferably, the track image is processed based on background difference method and morphological method to obtain the outline of the intruding foreign body; Calculate the height of the intruding foreign object, including: determining boundary points of the contour; Calculate the actual width per unit pixel based on the horizontal coordinate of the boundary point; Based on the vertical coordinate of the boundary point and the actual width of the unit pixel, the height of the intruding foreign object is calculated.

[0012] Preferably, the track image also includes disasters; The track image is input into a trained recognition model to obtain a disaster category; The method further includes: judging the degree of impact on driving safety according to the disaster category.

[0013] Preferably, training the recognition model includes: Acquire a data set processed by the Mixup data enhancement algorithm; wherein the data set includes orbital images of various disasters and / or intrusive foreign objects; Labeling the data set and setting coordinates; Dividing the labeled data set into a training set, a validation set, and a test set; The divided data set is used to train the recognition model to obtain a trained recognition model. In a second aspect, the present disclosure provides a track safety driving determination system, comprising: A track image acquisition module, used to acquire track images including intruding foreign objects; A safety clearance area determination module is used to determine the safety clearance area according to the track image and determine whether the intruding foreign object is located in the safety clearance area; An intrusive foreign object attribute determination module, used to determine whether the attribute of the intrusive foreign object is plane or three-dimensional if the intrusive foreign object is located in the safety limit area; An occupied area and height calculation module, used for calculating the occupied area and height of the intrusive foreign object if the property of the intrusive foreign object is three-dimensional; The safe driving determination module is used to determine the impact of the intrusive foreign object on driving safety based on the occupied area and height.

[0014] Beneficial effects of the present invention: The method of the present invention can significantly improve the accuracy and effectiveness of driving safety assessment by comprehensively judging the properties of the intruding foreign objects (i.e., whether they are three-dimensional or two-dimensional); distinguishing between two-dimensional and three-dimensional foreign objects helps to more accurately assess their impact on train operation; on this basis, further calculating the occupied area and height of the foreign objects can provide a more comprehensive understanding of their occupation of track space and potential threats, because although two-dimensional objects may occupy a large area, they have less impact on driving safety, while three-dimensional objects have a greater impact on driving safety due to their volume characteristics.

[0015] Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 The process of the method for determining safe track driving of the present invention is shown Figure 1 ; Figure 2 The process of the method for determining safe track driving of the present invention is shown Figure 2; Figure 3 A diagram showing the process of dividing the track limit area of ​​the present invention is shown; Figure 4 A schematic diagram of the railway line safety clearance area of ​​the present invention is shown; Figure 5 The track limit area cutting of the present invention is shown; Figure 6 The Mixup transformation effect diagram of the present invention is shown; Figure 7 An image label processing diagram of the present invention is shown; Figure 8 The accuracy, recall, and F1 score curves of the YOLOv5s model of the present invention are shown; Fig. 9 The YOLOv5s track foreign body detection and recognition results of the present invention are shown; Fig.10 The invention shows the background-differential intrusion foreign body recognition; Fig.11 A schematic diagram showing the height recognition of a three-dimensional foreign body according to the present invention is shown; Fig.12 A schematic diagram of an experimental scenario of the present invention is shown; Fig.13 Shown is a schematic diagram of the experimental process of the present invention; Fig.14 A schematic diagram of the experimental results of the present invention is shown; Fig.15 The device framework diagram of the present invention is shown; Fig.16 A schematic diagram of the device structure of the present invention is shown. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Reference Figure 1 , Figure 2 , Fig.12 , Fig.13 and Fig.14 As shown, the present disclosure proposes a method for dynamically identifying track hazards and intrusive foreign objects, which specifically includes the following steps: S1. Acquire the orbital image including the intruding foreign objects; Specifically, a real-time track image is acquired through a camera; wherein the track image includes intrusive foreign objects.

[0020] S2. Determine the coordinates of the safety clearance area and the intruding foreign object according to the track image, and determine whether the intruding foreign object is located in the safety clearance area based on the safety clearance area and the coordinates; Specifically, in step S2, determining the safety clearance area according to the track image includes the following steps: S21, preprocessing track image; S22, detecting a track straight line in the track image based on Hough transform; S23, obtaining a safe clearance area according to the track straight line and the preset clearance distance; Specifically, as a preferred embodiment of the present disclosure, in step S21, the specific steps of preprocessing the track image include: S211, processing the acquired track image by a grayscale method based on a weighted average method; Specifically, track images are usually in color, and each pixel consists of the values ​​of three channels: red (R), green (G), and blue (B). The three channels of red, green, and blue together determine the color and brightness of the image. The weighted average method assigns different weights to each color channel, and then calculates the weighted average as the grayscale value, and its expression is: (1) In formula (1), , and are the weights of the red, green and blue channels respectively, is the grayscale value, and the sum of the weight values ​​is 1. For example, , , .

[0021] S212, processing the track image using an image denoising method; As a preferred embodiment of the present disclosure, the image denoising method is median filtering. Specifically, the track image processed by the grayscale method of the weighted average method is respectively subjected to mean filtering, Gaussian filtering, and median filtering, and the image quality is judged according to three evaluation indicators: peak signal-to-noise ratio (PSNR), structural similarity coefficient (SSIM), and mean square error (MSE). The specific values ​​obtained are shown in Table 1 below.

[0022]

[0023] According to the evaluation index data in Table 1, median filtering is selected as the image denoising method. Median filtering can sort the target pixel to be processed and its adjacent pixels from large to small, and then use the middle pixel value as the standard. Its expression is: (2) In formula (2), is the set of all pixel coordinates within the filter window, is the original image, is the processed image.

[0024] S213, binarizing the track image using the Otsu algorithm; The Otsu algorithm can automatically determine the binarization threshold of an image and divide the image into two parts: foreground and background. The expression of the Otsu algorithm is: (3) (4) In formula (3) and formula (4), is the probability of the first type of pixel (background), is the probability of the second type of pixel (foreground), is the number of pixels of the first category.

[0025] (5) (6) In formula (5) and formula (6), is the average probability of the first type of pixels, is the average probability of the second type of pixels, is the total number of pixels.

[0026] (7) In formula (7), is the between-class variance of two types of pixels.

[0027] By traversing all possible thresholds, find the threshold that maximizes the inter-class variance , and use the threshold to segment the image into foreground and background, thereby achieving the optimal binarization effect.

[0028] S214, using Sobel operator to perform gradient filtering on the track image; The Sobel operator is based on the first-order derivative of the image grayscale. Through a series of convolution calculations, it generates two numerical output images. The gradient information can be divided into horizontal and vertical directions and is used to derive the size and angle of the image gradient.

[0029] The definition of the Sobel operator is as follows: (8) Where A is the input grayscale image, and are the convolution outputs of the image in the horizontal and vertical directions respectively. The calculation method is to weight the sum of the pixels around each pixel in the image to obtain two output values ​​in the horizontal and vertical directions. The expression for calculating the gradient amplitude and direction of the pixel point is: (9) In formula (9), G is the gradient amplitude, is the gradient direction.

[0030] S215, extracting track edge information based on Canny edge detection algorithm; Specifically, the algorithm composition of the Canny edge detection algorithm mainly includes: (1) Non-maximum suppression: After determining the pixel gradient, the direction is converted to 0°, 45°, 90°, and 135° by rounding. Then, at each pixel, the gradient size is compared with the sizes of the two pixels around it. If the gradient at this position is the largest, the point is retained, otherwise the point is suppressed. The expression is: (10) In formula (10), is the low threshold value of gradient intensity, and is the offset value in the gradient direction.

[0031] (2) Double threshold processing: By setting two thresholds, high and low, the image pixels are divided into three categories. Pixels above the high threshold are significant edges, pixels below the low threshold are considered non-edges, and pixels between the two are classified as potential edges. This effectively distinguishes the edge strength in the image. The expression is: (11) In formula (11), is the high threshold of gradient strength, , and They are strong edge, weak edge, and non-edge respectively.

[0032] (3) Edge connection: For pixels that are initially judged to be weak edges, when these weak edge pixels are determined to be surrounded by strong edge pixels, they will be reclassified and regarded as strong edge pixels. If there are no weak edges around them, the pixel point will be regarded as a non-edge.

[0033] The preprocessed image has eliminated most of the useless information. For track detection, Hough transform is used as the detection algorithm. Its core is to map the feature points of the two-dimensional image space to a three-dimensional Hough space, thereby realizing the transformation from specific shapes in the image to discrete points in the Hough space.

[0034] Specifically, refer to Figure 3 As shown, in step S22, the track straight line in the track image is detected based on the Hough transform, specifically including: For line detection, it is assumed that a line on a two-dimensional plane can be represented as: (12) In formula (12), is the slope of the straight line, is the intercept.

[0035] Then, in Hough space, a straight line can be represented by a set of two-tuples Indicates, including the value from the origin and with Axis Angle For each image point , we need to calculate the distance to all possible straight lines and Axis angle .

[0036] Each point will make the corresponding The position is incremented by 1. After traversing all the pixels of the image, the curve in the Hough space can be represented as a series of curves with dense intersections.

[0037] Specifically, refer to Figure 4 and Figure 5 As shown, in step S23, a safety clearance area is obtained according to the track straight line and the preset clearance distance, specifically including: treating the track straight line detected by the Hough transform as a straight line equation and the preset clearance distance, and calculating the safety clearance area of ​​the railway.

[0038] Since the train body is wider than the track gauge, in order to ensure driving safety, it is necessary not only to monitor the track, but also to monitor the area outside the track and within the limit. Assume that the distance between the limit area and the center of the rail is , the safety limit area of ​​the railway can be divided.

[0039] For example, according to the two straight line equations obtained by the Hough transform of the identified rail in step S2, any two points with the same Y coordinate are selected. and , then the number of pixels between two points is , the horizontal coordinate of the boundary of the lower horizontal bounding area can be calculated: (13) In formula (13), and Respectively represent the horizontal coordinates of the left and right boundaries of the bounded area, is the limit distance. In practical applications, it can be set according to specific monitoring needs. If you need to monitor a farther area, you can increase it appropriately. If you only need to monitor the area next to the track, you can reduce It can be understood that the 1435 used in formula (13) is the standard track gauge of my country (1435 mm), but this value can be adjusted accordingly according to the specific track gauge requirements.

[0040] Therefore, the coordinates of any point on the track limit boundary can be calculated by formula (13), and the boundary of the edge area can be obtained according to formula (12), and the image can be cropped accordingly.

[0041] Specifically, in step S2, the coordinates of the intruding foreign object are determined to determine whether the intruding foreign object is located in the safety limit area, including the following steps: S24, inputting the track image into the trained recognition model to obtain the coordinates of the intruding foreign object; However, even if the coordinates of the intruding foreign object fall outside the safety limit area, it does not mean that the entire foreign object is within the safety limit. This is because the coordinates are just a point, while the intruding foreign object is not. Therefore, it is necessary to further determine the properties of the intruding foreign object, that is, whether it is a flat object or a three-dimensional object (i.e., two-dimensional or three-dimensional), in order to fully evaluate its impact on driving safety.

[0042] S3. If the intruding foreign object is located in the safety limit area, determine whether the intruding foreign object is a plane or a three-dimensional object; Specifically, if the intruding foreign object is located in the safety limit area, it is necessary to further determine whether the intruding foreign object is a plane or a three-dimensional object. This is because the degree of impact of plane and three-dimensional intruding foreign objects on driving safety is different. For example: plane objects (such as paper, plastic bags, etc.) are usually light and have little impact on train operation, and will not pose a major threat to the driving stability of the train. Three-dimensional objects (such as stones, trees, etc.) are usually heavy and large in size, and have a greater impact on train operation. They may cause serious collisions and even derailment accidents.

[0043] Specifically, in step S3, judging whether the property of the intruding foreign object is plane or three-dimensional includes the following steps: In step S24, the track image is input into the trained recognition model, and the type of the intruding foreign object can also be obtained. Therefore, the property of the intruding foreign object can be judged as plane or three-dimensional according to the type of the intruding foreign object.

[0044] S4. If the property of the intruding foreign object is three-dimensional, calculate the occupied area and height of the intruding foreign object; Specifically, calculating the occupied area of ​​the intrusive foreign matter includes the following steps: S41, processing the track image based on background difference method and morphological method to obtain the outline of the intruding foreign object and the number of pixels occupied; S42. Obtain the occupied area of ​​the intruding foreign matter according to the actual area mapped by each occupied pixel and the number of occupied pixels.

[0045] Specifically, in step S41, refer to Fig.10 As shown, the track image is processed based on the background difference method, including the following steps: S411. First, select a reference frame as a background image. In some cases, the average or median of multiple frames can be used as the background image to reduce the impact of noise and illumination changes.

[0046] S412, then perform a difference operation on the current frame image and the background image. In the differenced image, the area with a larger pixel value represents the different parts of the current frame and the background image, that is, the area where the track intrusion object is located. By comparing the current frame image with the background image, the area where the moving object is located is found, and this area is considered to be the target area.

[0047] At the moment ,image By background target and prospect goals Composition, ignoring noise In the case of, its expression is: (14) From formula (14), we can get that the moving target The expression is: (15) However, in the actual acquisition process of the image, the noise It will definitely cause some adverse effects on the image. Therefore, in the target area, a differential image and noise Together they form the actual motion target area, namely: (16) From the above formula we can get the difference picture ,Right now: (17) Finally, the binary threshold processing is performed. is the binarization threshold, that is: (18) Among them, the threshold Different choices will have different effects on the danger sources in the image. Too high a threshold will make the danger sources incomplete; too low a threshold will make the image have a lot of noise, making it difficult to better identify the danger sources.

[0048] In order to reduce the false detection pixels (i.e., false noise points) in the binary image, it is also necessary to process the track image including the intruding foreign objects based on the morphological method, which includes the following steps: Use the opening operation to process binary images. The opening operation consists of two steps: first, an erosion operation, and then a dilation operation.

[0049] Erosion refers to the sliding of the structural element on the image. If the pixel of the structural element is the same as the pixel at the relative position in the image, the pixel at this position remains unchanged, otherwise it will become the background color. is the grayscale function, is a structural element, then corrosion can be defined as: (14) The dilation operation is to slide the structure element on the image. If the structure element pixel corresponds to the relative position pixel in the image, the pixel at this position becomes the foreground color. Dilation can be defined as: (15) The definition of the opening operation is as follows: (16) The image after background subtraction and morphological processing is as follows Figure 8 As shown in the figure. After background difference and morphological processing, the shape and position of the intruding foreign body can be more accurately identified. At this point, the intruding foreign body has been marked as a connected area in the image. It can accurately identify the shape and number of pixels occupied by the intruding foreign body, calculate the actual area mapped by each pixel, and thus obtain the occupied area of ​​the intruding foreign body in the two-dimensional image.

[0050] However, the physical volume of the intruding foreign object is manifested as a three-dimensional structure, but the two-dimensional features expressed by the pixel statistics obtained by background difference cannot accurately express the impact of height on driving. For example, a stone and a large leaf may occupy the same pixel in a two-dimensional image, but it is obvious that the leaf will not affect driving safety, while the large stone has a greater impact on driving safety. Therefore, if the recognition model determines that the intruding foreign object has obvious three-dimensional features, such as being classified as a person or a three-dimensional object with a certain height, then the key feature affecting driving safety is determined to be its height. In view of this, it is also necessary to estimate the height of the intruding foreign object to evaluate its impact on the safety of train operation. Therefore, how to extract the height of the intruding foreign object from the two-dimensional image becomes the key.

[0051] Specifically, in step S4, calculating the height of the intruding foreign object includes: S421, determining the boundary points of the contour; S422, calculating the actual width per unit pixel based on the horizontal coordinate of the boundary point; S423, based on the vertical coordinate of the boundary point and the actual width of the unit pixel, calculate the height of the intruding foreign object; illustratively, refer to Fig.11 As shown, take the bottom of the intruding foreign body as the reference to determine its corresponding vertical coordinate , and the left and right horizontal coordinates at the bottom and , and determine the vertical coordinate of the top ; Ignoring the influence of image distortion at the same level, according to the bottom left and right horizontal coordinates and ,calculate (Right now The actual width of the unit pixel at the corresponding pixel level , the formula is Since the intruding foreign body is regarded as a vertical plane, the expression of its height in the vertical pixels is similar to that at the bottom. Therefore, the vertical pixel difference of the foreign body in the image (i.e. ) and the known actual width in pixels , to estimate the height of the intruding foreign body For example, the results of the judgment of the intrusion disaster are shown in Table 2 below.

[0052]

[0053] Ultimately, whether it affects driving is determined based on the actual area and height.

[0054] As a preferred embodiment of the present disclosure, the method for determining safe driving on a track further includes: S6, judging the degree of impact on driving safety according to the disaster category. Specifically, the track image is input into a trained recognition model, and the disaster category is obtained. The track image also includes the disaster.

[0055] Specifically, training the recognition model includes the following steps: S61, obtaining a data set processed by the Mixup data enhancement algorithm; wherein the data set includes orbital images of various disasters and / or intrusive foreign objects; Specifically, the dataset is obtained through collection, network retrieval and data enhancement. The dataset consists of pictures of natural disasters and / or intrusions of foreign objects into orbit, including mudslides, sandstorms, blizzards, falling rocks, floods, fallen trees, etc.

[0056] Reference Figure 6 As shown, the Mixup data enhancement algorithm uses a The random parameters between , weighted average of the pixel values ​​and labels of the two samples. For example, given two samples and , its Mixup sample is expressed as: (13) In formula (13), and Respectively represent the pixel value and label of the Mixup sample, is a A random parameter between .

[0057] S62, labeling the data set and setting coordinates; Specifically, the labelimg tool is used to label the dataset. labelimg is an image tool that supports rectangular annotation. It supports real-time image preview, can annotate images and set coordinates for them. Figure 7 shown.

[0058] S63, dividing the labeled data set into a training set, a validation set, and a test set; As a preferred embodiment of the present disclosure, the division ratio of the training set, the validation set and the test set is 8:1:1, as shown in Table 3 below.

[0059]

[0060] S64, using the divided data set to train the recognition model to obtain a trained recognition model; As a preferred embodiment of the present disclosure, the recognition model is a convolutional neural network. Further preferably, the recognition model is a YOLOv5s network model. The YOLOv5s training network parameters include self-learning adjustment parameters and manually set parameters, i.e., hyperparameters. The setting of hyperparameters needs to be optimized according to the actual track natural disaster scenario. The hyperparameters are shown in Table 4 below.

[0061]

[0062] After the data set is input into the network, the convolutional neural network repeatedly adjusts the parameter values ​​during training to output the best training model, which can be used for track natural disaster identification and detection. The results obtained using the trained recognition model are as follows: Figure 5 shown.

[0063] Based on the confusion matrix, the performance of the network model is analyzed through three indicators: Precision, Recall, and F1. The accuracy and recall of various disaster data are calculated as samples to obtain the YOLOv5s detection accuracy analysis results shown in Table 5. Figure 8 As shown, the precision, recall and F1 score curves are shown.

[0064]

[0065] From Table 5, we can see that the Precision is relatively high, all above 0.97. However, in the recognition of fallen rocks and fallen wood, the Recall is relatively low, 0.33 and 0.60 respectively. The possibility of false negative examples increases, but the prediction reliability of the recognition model is good, and the target objects can be detected more comprehensively. The recognition results are as follows: Fig. 9 shown.

[0066] Based on the same inventive concept, Fig.15 As shown, the present disclosure also proposes a track safety driving determination system, comprising: A track image acquisition module 110, used to acquire a track image including intruding foreign objects; A safety clearance area determination module 120 is used to determine the safety clearance area according to the track image and determine whether the intruding foreign object is located in the safety clearance area; The intrusion foreign body attribute determination module 130 is used to determine whether the attribute of the intrusion foreign body is plane or three-dimensional if the intrusion foreign body is located in the safety limit area; An occupied area and height calculation module 140 is used to calculate the occupied area and height of the intruding foreign object if the property of the intruding foreign object is three-dimensional; The safe driving determination module 150 is used to determine the impact of the intruding foreign object on driving safety based on the occupied area and height.

[0067] Based on the same inventive concept, Fig.16 As shown, the present disclosure also proposes a device, including a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, and the processor executes the above-mentioned method for determining safe track driving when executing the computer instructions.

[0068] Based on the same inventive concept, the present disclosure also proposes a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed, the above-mentioned method for determining safe track driving can be implemented.

[0069] Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0070] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0071] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining safe rail driving, characterized in that: include: Acquire orbital images including intruding foreign objects; Determine the coordinates of a safety clearance area and an intruding foreign object according to the track image, and determine whether the intruding foreign object is located in the safety clearance area based on the safety clearance area and the coordinates; If the intruding foreign object is located in the safety limit area, determining whether the intruding foreign object is a plane or a three-dimensional object; If the property of the intrusive foreign object is three-dimensional, then the occupied area and height of the intrusive foreign object are calculated; The degree of impact on driving safety is determined based on the occupied area and height.

2. The method for determining safe rail driving according to claim 1, characterized in that: The determining of the safety clearance area according to the track image includes: preprocessing the track image; Detecting a track straight line in the track image based on Hough transform; A safe clearance area is obtained according to the track straight line and the preset clearance distance.

3. The method for determining safe rail driving according to claim 2, characterized in that: The preprocessing of the track image comprises: Processing the track image using a grayscale method based on a weighted average method; Processing the track image using an image denoising method; Binarizing the track image using the Otsu algorithm; Using Sobel operator to perform gradient filtering on the track image; Extract track edge information based on Canny edge detection algorithm.

4. The method for determining safe rail driving according to claim 1, characterized in that: Determine the coordinates of intruding foreign objects, including: The track image is input into the trained recognition model to obtain the coordinates of the intruding foreign object.

5. The method for determining safe rail driving according to claim 4, characterized in that: The track image is input into the trained recognition model to obtain the type of the intruding foreign matter; The determination of whether the property of the intruding foreign body is plane or three-dimensional is as follows: According to the type of the intruding foreign object, it is determined whether the intruding foreign object has a plane or a three-dimensional property.

6. The method for determining safe rail driving according to claim 1, characterized in that: The calculating the occupied area of ​​the intruding foreign object comprises: Processing the track image based on background difference method and morphological method to obtain the outline of the intruding foreign body and the number of pixels occupied; According to the actual area mapped by each occupied pixel and the number of occupied pixels, the occupied area of ​​the intruding foreign matter is obtained.

7. The method for determining safe rail driving according to claim 6, characterized in that: The track image is processed based on the background difference method and the morphological method to obtain the outline of the intruding foreign body; Calculate the height of the intruding foreign object, including: determining boundary points of the contour; Calculate the actual width per unit pixel based on the horizontal coordinate of the boundary point; Based on the vertical coordinate of the boundary point and the actual width of the unit pixel, the height of the intruding foreign object is calculated.

8. The method for determining safe rail driving according to claim 5, characterized in that: The orbital image also includes hazards; The track image is input into a trained recognition model to obtain a disaster category; The method further includes: judging the degree of impact on driving safety according to the disaster category.

9. The method for determining safe track driving according to claim 8, characterized in that: Training the recognition model, including: Acquire a data set processed by the Mixup data enhancement algorithm; wherein the data set includes orbital images of various disasters and / or intrusive foreign objects; Labeling the data set and setting coordinates; Dividing the labeled data set into a training set, a validation set, and a test set; The divided data set is used to train the recognition model to obtain a trained recognition model.

10. A track safety driving determination system, characterized in that: include: A track image acquisition module, used to acquire track images including intruding foreign objects; A safety clearance area determination module is used to determine the safety clearance area according to the track image and determine whether the intruding foreign object is located in the safety clearance area; An intrusion foreign body property determination module, used to determine whether the property of the intrusion foreign body is plane or three-dimensional if the intrusion foreign body is located in the safety limit area; An occupied area and height calculation module, used for calculating the occupied area and height of the intrusive foreign object if the property of the intrusive foreign object is three-dimensional; The safe driving determination module is used to determine the impact of the intrusive foreign object on driving safety based on the occupied area and height.

Images