Railway track obstacle early warning method and device

Abstract

The invention relates to the technical field of image recognition, in particular to a railway track obstacle early warning method and device. The method can be used for real-time monitoring of the railway track, intrusive objects such as people, animals and vehicles are detected in real time through the railway monitoring video, and the influence of track obstacles on railway transportation can be reduced in severe weather. According to the scheme, a Retinex method is used for carrying out defogging and noise reduction processing on a railway monitoring video in severe weather, then an improved SSD algorithm is used for detecting front obstacles, the front obstacles are detected, the positions and the target sizes of the front obstacles are refined, information is provided for a driver through vehicle-mounted equipment, the driver is reminded to pay attention to or brake, and the safety of the driver is improved. Therefore, the safe and efficient operation of the railway and the safety of passengers are effectively ensured, and high reliability and usability are achieved.

Description

Technical Field

[0001] The present disclosure relates to the field of image recognition technology, and in particular to a railway track obstacle early warning method and device. Background Art

[0003] Common bad weather near the ground, such as heavy fog, rain, snow, etc., directly affects the visibility within the visual range of the railway track, which is very unfavorable for the safe and smooth operation of the train. In order to avoid danger to the greatest extent and reduce the threat of railway obstacle warning due to weather reasons, video surveillance is usually installed as an auxiliary response measure. Due to the poor quality of the original image interfered by fog and other factors, the object is very blurred. If the received video image is directly detected, the accuracy of the railway track obstacle warning will be low, resulting in a high false alarm of the warning system, which will bring a lot of unnecessary work to the train drivers. Therefore, it is of great research significance to defog and reduce noise on the collected images, improve the clarity of the pictures, and then identify obstacles that may affect the operation of the train.

[0004] In the existing technology, the detected objects are identified by computer assistance, and in order to identify obstacles, the detected objects are evaluated by computer assistance to see whether they actually become obstacles. A route map method is used to evaluate objects, wherein the route map contains a large number of objects to be identified and the positions of the objects at or on the route. The position information is determined for the identified object, and then the position information is compared with the route map. If its position matches or overlaps with the position of the corresponding object to be identified in the route map, the following evaluation is triggered: there is an obstacle on the route. Uncertainties may arise here: an object is not detected, or an object that is an obstacle is not evaluated as an obstacle, etc. Therefore, this existing obstacle identification technology has potential risks.

[0005] In summary, there is an urgent need for a technical solution to provide early warning of railway track obstacles in severe weather. Summary of the invention

[0006] In view of the above problems, the present disclosure provides a railway track obstacle early warning method and device, which are used to solve the problem of early warning of railway track obstacles in severe weather.

[0007] In a first aspect, a railway track obstacle early warning method is provided, the method comprising:

[0008] The Retinex method is used to enhance the collected railway track obstacle image dataset to obtain an enhanced image.

[0009] Based on the SSD model, the VGG16 backbone network is replaced with the MobileNetV3 network to obtain the MobileNetV3-SSD model;

[0010] According to the historical enhanced images, the MobileNetV3-SSD model is trained to obtain the trained MobileNetV3-SSD model;

[0011] Input the enhanced image collected in real time into the trained MobileNetV3-SSD model for recognition;

[0012] When an obstacle on the railway track is identified, an early warning signal is issued.

[0013] Furthermore, the Retinex method is used to enhance the collected railway track obstacle image dataset, including:

[0014] Use a low-pass filter to perform Gaussian blur processing on the original image to retain the color information in the image, reduce the noise in the image, and obtain a color restored image;

[0015] The original image is filtered by a high-pass filter to enhance the brightness information in the image, highlight the details and edges in the image, and obtain a brightness-adjusted image;

[0016] The color restored and brightness adjusted images are rescaled to obtain the final enhanced image dataset.

[0017] Furthermore, the original image is Gaussian blurred using a low-pass filter, including:

[0018] The following formula is used to obtain the illumination image L(x,y) for the original image I(x,y). The illumination image L(x,y) is the color restored image:

[0019] δ is the standard deviation.

[0020] Furthermore, the enhanced image collected in real time is input into the trained MobileNetV3-SSD model for recognition, including:

[0021] The image-enhanced dataset is input into the trained MobileNetV3-SSD model. Through a forward propagation, the image is analyzed and the bounding box and corresponding confidence of each possible obstacle in the image are predicted. The bounding box represents the obstacle position predicted by the MobileNetV3-SSD model, and the confidence represents the degree of confidence of the MobileNetV3-SSD model in its prediction.

[0022] The image detected by the MobileNetV3-SSD model contains multiple bounding boxes and multiple confidence levels. The non-maximum suppression method is used to post-process the detection data; the detection results are improved by removing bounding boxes with overlap higher than the overlap threshold and confidence lower than the confidence threshold.

[0023] Furthermore, when an obstacle on the railway track is identified, an early warning signal is issued, including:

[0024] The detection data of the MobileNetV3-SSD model is transmitted to the on-board ATP to remind the driver of the type, size and location of obstacles ahead.

[0025] In a second aspect, a railway track obstacle warning device includes: an image enhancement unit, a model building unit, a model training unit, a real-time monitoring unit and a warning unit;

[0026] An image enhancement unit, used for performing image enhancement on a collected railway track obstacle image data set by using a Retinex method to obtain an enhanced image;

[0027] A model building unit is used to replace the VGG16 backbone network with the MobileNetV3 network on the basis of the SSD model to obtain the MobileNetV3-SSD model;

[0028] A model training unit is used to train a MobileNetV3-SSD model based on historical enhanced images to obtain a trained MobileNetV3-SSD model;

[0029] The real-time monitoring unit is used to input the enhanced images collected in real time into the trained MobileNetV3-SSD model for recognition;

[0030] The early warning unit is used to send out an early warning signal when an obstacle on the railway track is identified.

[0031] Furthermore, the Retinex method is used to enhance the collected railway track obstacle image dataset, including:

[0032] Use a low-pass filter to perform Gaussian blur processing on the original image to retain the color information in the image, reduce the noise in the image, and obtain a color restored image;

[0033] The original image is filtered by a high-pass filter to enhance the brightness information in the image, highlight the details and edges in the image, and obtain a brightness-adjusted image;

[0034] The color restored and brightness adjusted images are rescaled to obtain the final enhanced image dataset.

[0035] Furthermore, the enhanced image collected in real time is input into the trained MobileNetV3-SSD model for recognition, including:

[0036] The image-enhanced dataset is input into the trained MobileNetV3-SSD model. Through a forward propagation, the image is analyzed and the bounding box and corresponding confidence of each possible obstacle in the image are predicted. The bounding box represents the obstacle position predicted by the MobileNetV3-SSD model, and the confidence represents the degree of confidence of the MobileNetV3-SSD model in its prediction.

[0037] The image detected by the MobileNetV3-SSD model contains multiple bounding boxes and multiple confidence levels. The non-maximum suppression method is used to post-process the detection data; the detection results are improved by removing bounding boxes with overlap higher than the overlap threshold and confidence lower than the confidence threshold.

[0038] In a third aspect, an electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0039] a memory storing a computer program;

[0040] The processor is used to implement the above-mentioned railway track obstacle early warning method when executing the computer program stored in the memory.

[0041] In a fourth aspect, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned railway track obstacle warning method is implemented.

[0042] The present disclosure includes at least the following beneficial effects:

[0043] The present invention is based on the Retinex-MobileNetV3-SSD algorithm, which uses the Retinex method to eliminate the blurring effect of fog and the like on the collected images, and then detects the type, size and position of obstacles in front of the track through the MobileNetV3-SSD algorithm. The detection data is transmitted to the on-board ATP through the existing intelligent perception system, prompting the driver and issuing an early warning. It can more effectively detect and locate intrusions in front of the railway track and prevent the occurrence of railway traffic accidents, and has high usability and practicality.

[0044] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or be understood by implementing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures indicated in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 This is a flowchart of the early warning method according to an embodiment of the present disclosure;

[0047] Figure 2 This is a schematic diagram of the SSD network model structure of an embodiment of the present disclosure;

[0048] Figure 3 This is a schematic diagram of the network structure of the SSD algorithm according to an embodiment of the present disclosure;

[0049] Figure 4 This is a schematic diagram of the MobileNetV3-SSD algorithm model structure of the embodiment of the present disclosure;

[0050] Figure 5 This is a schematic diagram of the Retinex-MobileNetV3-SSD algorithm flow in the embodiment of the present disclosure;

[0051] Figure 6 This is a schematic diagram of the Retinex image preprocessing process according to an embodiment of the present disclosure;

[0052] Figure 7 This is a schematic diagram of the structure of the early warning device according to an embodiment of the present disclosure;

[0053] Figure 8 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0055] The existing method is a method in which a vehicle uses a sensor device to detect an object in front of the vehicle along the driving direction. This method determines whether the detected object in front is an obstacle based on a route map. There is a potential risk that the object is not detected or not evaluated as an obstacle by the route map, and this method cannot determine the specific type and size of the obstacle in front.

[0056] In response to this defect, the present invention proposes a railway track obstacle warning method in severe weather based on the Retinex-MobileNetV3-SSD algorithm. The method can be used for real-time monitoring of railway tracks, and detects intrusive objects such as people, animals, and vehicles in real time through railway monitoring videos. It can also reduce the impact of track obstacles on railway transportation in severe weather. The solution uses the Retinex method to perform defogging and noise reduction on railway monitoring videos in severe weather, and then uses the improved SSD algorithm to detect obstacles ahead, detect obstacles ahead and refine their positions and target sizes, and provide this information to the driver through the on-board equipment to remind the driver to pay attention or brake to prevent accidents, thereby effectively ensuring the safe and efficient operation of the railway and the safety of passengers, and has high reliability and availability.

[0057] like Figure 1 As shown, a railway track obstacle early warning method, the method comprising:

[0058] S101, using a Retinex method to perform image enhancement on a collected railway track obstacle image dataset to obtain an enhanced image;

[0059] S102, based on the SSD model, the VGG16 backbone network is replaced with the MobileNetV3 network to obtain the MobileNetV3-SSD model;

[0060] S103, training a MobileNetV3-SSD model according to the historical enhanced image to obtain a trained MobileNetV3-SSD model;

[0061] S104, inputting the enhanced image collected in real time into the trained MobileNetV3-SSD model for recognition;

[0062] S105, when a railway track obstacle is identified, a warning signal is issued.

[0063] The specific implementation is as follows:

[0064] Retinex is a computational theory of color constancy perception. The theory holds that the color of an object is determined by its ability to reflect long-wave, medium-wave, and short-wave light, rather than by the absolute value of the intensity of the reflected light. The color of an object is not affected by illumination non-uniformity and is consistent. Retinex is an image enhancement method based on the human visual system. Its core idea is to adjust the contrast and brightness of an image while retaining image detail information. The algorithm assumes that the original image I(x,y) is the product of the illumination image L(x,y) and the reflection image R(x,y), that is, I(x,y) = L(x,y)*R(x,y). By estimating the illumination L(x,y) and decomposing the reflection image R(x,y), the effect of uneven illumination is eliminated and the visual effect of the image is improved. Among them, low-pass filtering the image I(x,y) can approximately obtain L(x,y): δ is the standard deviation. The image processed by the Retinex method can significantly enhance the target details of the original image, reduce the target color distortion, and provide good support for subsequent target detection.

[0065] like Figure 2 As shown in the figure, the SSD (Single Shot MultiBox Detector) network architecture can simultaneously predict the location and category of the target in a single forward propagation. The SSD network structure optimizes the multi-scale target detection performance by modifying the VGG16 network, adding additional convolutional layers, and adopting a feature pyramid structure. Among them, VGG16 is a classic convolutional neural network consisting of 13 convolutional layers (Conv1_1 to Conv5_3) and 3 fully connected layers (FC6, FC7, FC8). In the SSD network, the fully connected layers of VGG16 are usually converted to convolutional layers to adapt to the construction of the feature pyramid.

[0066] The SSD algorithm is a single-stage multi-box prediction target detection algorithm. It first obtains candidate boxes through CNN, and then performs classification and regression. SSD is a powerful target detection algorithm with good real-time performance and high detection accuracy. It uses VGG16 as the basic model, changes the fully connected layer after Fc5 in the VGG16 network to a convolutional layer, and then adds Conv8, Conv9, Conv10, and Conv11 after Fc7 to form a complete SSD network.

[0067] The SSD algorithm scales the original input image to a fixed size, extracts the target feature map by the SSD network, and obtains multi-scale feature maps with sizes of 38×38, 19×19, 10×10, 5×5, 3×3, and 1×1 pixels; by setting different aspect ratios and different numbers of prior frames for feature maps of different sizes, a series of bounding box sets can be generated. The given prior frame will perform IoU matching according to the annotation frame, and the calculation formula is: Where: S A Represents the area of ​​the prior box; S B Indicates the area of ​​the actual annotation box; F IoU It represents the ratio of the intersection and union of the areas of the predicted box and the true box. When the IoU value between the true box and the prior box is greater than the threshold of 0.5, the prior box is retained, and the category of the prior box is the category of the true box. All prior boxes are regressed, and a 3×3 convolution kernel is used to obtain the two predicted values ​​of the category confidence and position offset value of the prior box; finally, the bounding box is adjusted by the non-maximum suppression algorithm to obtain the final positioning bounding box.

[0068] The algorithm network structure is as follows Figure 3 As shown in Figure 1, feature pyramid is an effective multi-scale feature representation method that uses feature maps at different levels to detect objects of different sizes. In SSD, starting from Conv4_3 and ending with the newly added Conv layer, the feature maps of each layer are used for object detection. The feature maps of the shallower layers have higher resolution and are suitable for detecting small objects; while the feature maps of the deeper layers have lower resolution but richer semantic information and are suitable for detecting large objects. On each selected feature map, SSD applies a series of preset default boxes and predicts the bounding box offset, confidence (indicating whether the box contains the object), and the category of the object for each default box. These predictions are implemented through convolutional layers, and each position on each feature map outputs a series of prediction values ​​related to the default box. In the training phase, SSD uses a multi-task loss function to optimize the position regression and classification tasks at the same time. In the inference phase, the network outputs the prediction results of each default box, and removes redundant detection results through post-processing steps such as non-maximum suppression (NMS), and finally obtains the precise position and category of each object.

[0069] Based on the original SSD model, this solution replaces the VGG16 backbone network with the MobileNetV3 network to build a lightweight target detection model MobileNetV3-SSD, reduce the number of model parameters, and improve the detection efficiency and real-time performance of the algorithm. The MobileNetV3-SSD algorithm extracts feature layers of different sizes from the input image through the MobileNetV3 network, divides each layer into different grids according to the size, obtains different prior frames, covers targets of different sizes, and then adjusts the bounding box through the non-maximum suppression algorithm to obtain the final positioning bounding box. Figure 4 As shown in the figure, the MobileNetV3-SSD model removes the VGG16 backbone network in the original SSD model and replaces it with the MobileNetV3 network. The MobileNetV3 network is used as a feature extractor to provide multi-scale feature maps for the subsequent SSD detection layer. Due to the high efficiency of MobileNetV3, the MobileNetV3-SSD model can significantly reduce the computational complexity and model size while maintaining high detection accuracy, which is particularly suitable for resource-constrained mobile devices and embedded systems. In the railway system, this lightweight and efficient target detection model has broad application prospects.

[0070] The MobileNetV3-SSD algorithm has good robustness and real-time performance, high detection speed and accuracy, and can achieve more comprehensive and accurate target positioning and classification in actual scenarios, which can well meet the real-time obstacle detection of railway tracks. The MobileNetV3-SSD algorithm is used for railway track obstacle detection and early warning. The current intelligent perception system is used to transmit the detection data of the MobileNetV3-SSD algorithm to the on-board ATP to remind the driver of the type, size and location of the obstacles ahead. This can provide good support for preventing intrusions around the railway and provide certain guarantees for railway operation safety.

[0071] This solution uses the Retinex-MobileNetV3-SSD algorithm. Figure 5 Combining the advantages of Retinex method denoising and the excellent detection performance of MobileNetV3-SSD algorithm, the obtained railway track images containing obstacles in bad weather are detected and processed. First, the Retinex method is used to perform denoising preprocessing on the received image dataset. The specific steps are as follows: Figure 6, the original image is processed by Gaussian blurring using a low-pass filter, the color information in the image is retained, the noise in the image is reduced, and then the original image is filtered by a high-pass filter to enhance the brightness information in the image, highlight the details and edges in the image, and finally the image after color restoration and brightness adjustment is proportionally adjusted to obtain the final enhanced image dataset, reducing the interference of bad weather on railway track obstacle detection. Then the preprocessed dataset is input into the trained MobileNetV3-SSD model, and the image is analyzed through a forward propagation, and the bounding box and corresponding confidence of each possible obstacle in the image are predicted, where the bounding box represents the obstacle position predicted by the MobileNetV3-SSD model, and the confidence represents the degree of confidence of the MobileNetV3-SSD model in its prediction. The image detected by the MobileNetV3-SSD model contains multiple bounding boxes and multiple confidences. Since the same obstacle may be detected multiple times, it is necessary to use the non-maximum suppression method to post-process the detection data. By removing bounding boxes with high overlap and low confidence, the best detection results are retained, thereby solving the problem of multiple detections. After the data set is processed by non-maximum suppression, an accurate obstacle detection bounding box will be obtained, and the location of the obstacle can be intuitively obtained. The MobileNetV3-SSD model is used to detect and locate the target after the image is denoised by the Retinex method, which can achieve efficient and accurate detection of intrusion targets ahead of the train route in bad weather, so as to send relevant warning information to the on-board ATP, which can prompt the relevant drivers to pay attention or take relevant measures such as braking in time, achieving the purpose of improving the safety and reliability of railway operation.

[0072] The existing railway track warning method evaluates whether there are obstacles based on the line map, but cannot determine the type, size and exact location of the obstacles, and cannot avoid responding to the blurred image targets in bad weather, which poses a great potential risk. Compared with the existing technology, the present invention uses the Retinex-MobileNetV3-SSD algorithm to eliminate the blurring effect of fog and other factors on the collected images, and then detects the type, size and location of obstacles in front of the track through the MobileNetV3-SSD algorithm, and transmits this detection data to the on-board ATP through the existing intelligent perception system, prompting the driver and issuing an early warning, which can more effectively detect and locate intrusions in front of the railway track, prevent the occurrence of railway traffic accidents, and has high usability and practicality.

[0073] like Figure 7As shown, a railway track obstacle warning device includes: an image enhancement unit 701, a model building unit 702, a model training unit 703, a real-time monitoring unit 704 and a warning unit 705;

[0074] An image enhancement unit 701 is used to enhance the collected railway track obstacle image data set by using a Retinex method to obtain an enhanced image;

[0075] A model building unit 702 is used to replace the VGG16 backbone network with the MobileNetV3 network on the basis of the SSD model to obtain a MobileNetV3-SSD model;

[0076] The model training unit 703 is used to train the MobileNetV3-SSD model according to the historical enhanced image to obtain a trained MobileNetV3-SSD model;

[0077] A real-time monitoring unit 704 is used to input the enhanced image collected in real time into the trained MobileNetV3-SSD model for recognition;

[0078] The early warning unit 705 is used to send out an early warning signal when an obstacle on the railway track is identified.

[0079] like Figure 8 As shown, the present disclosure provides an electronic device, including a processor 801, a communication interface 802, a memory 803 and a communication bus 804, wherein the processor 801, the communication interface 802 and the memory 803 communicate with each other through the communication bus 804;

[0080] Memory 803, storing computer programs;

[0081] The processor 801 is configured to implement the above method when executing the computer program stored in the memory 803 .

[0082] The present disclosure provides a computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0083] The computer-readable storage medium may be included in the device / apparatus described in the above embodiment; or it may exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0084] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0085] Although the present disclosure 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 replacements for some of the technical features therein; and these modifications or replacements 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 disclosure.

Claims

1. A railway track obstacle early warning method, characterized in that: The method comprises: The Retinex method is used to enhance the collected railway track obstacle image dataset to obtain an enhanced image. Based on the SSD model, the VGG16 backbone network is replaced with the MobileNetV3 network to obtain the MobileNetV3-SSD model; According to the historical enhanced images, the MobileNetV3-SSD model is trained to obtain the trained MobileNetV3-SSD model; Input the enhanced image collected in real time into the trained MobileNetV3-SSD model for recognition; When an obstacle on the railway track is identified, an early warning signal is issued.

2. A railway track obstacle early warning method according to claim 1, characterized in that: The Retinex method is used to enhance the collected railway track obstacle image dataset, including: Use a low-pass filter to perform Gaussian blur processing on the original image to retain the color information in the image, reduce the noise in the image, and obtain a color restored image; The original image is filtered by a high-pass filter to enhance the brightness information in the image, highlight the details and edges in the image, and obtain a brightness-adjusted image; The color restored and brightness adjusted images are rescaled to obtain the final enhanced image dataset.

3. A railway track obstacle early warning method according to claim 2, characterized in that: Use a low-pass filter to Gaussian blur the original image, including: The following formula is used to obtain the illumination image L(x,y) for the original image I(x,y). The illumination image L(x,y) is the color restored image: δ is the standard deviation.

4. A railway track obstacle early warning method according to claim 1, characterized in that: The enhanced image collected in real time is input into the trained MobileNetV3-SSD model for recognition, including: The image-enhanced dataset is input into the trained MobileNetV3-SSD model. Through a forward propagation, the image is analyzed and the bounding box and corresponding confidence of each possible obstacle in the image are predicted. The bounding box represents the obstacle position predicted by the MobileNetV3-SSD model, and the confidence represents the degree of confidence of the MobileNetV3-SSD model in its prediction. The image detected by the MobileNetV3-SSD model contains multiple bounding boxes and multiple confidence levels. The non-maximum suppression method is used to post-process the detection data; the detection results are improved by removing bounding boxes with overlap higher than the overlap threshold and confidence lower than the confidence threshold.

5. A railway track obstacle early warning method according to claim 1, characterized in that: When an obstacle on the railway track is identified, an early warning signal is issued, including: The detection data of the MobileNetV3-SSD model is transmitted to the on-board ATP to remind the driver of the type, size and location of obstacles ahead.

6. A railway track obstacle warning device, characterized in that: include: Image enhancement unit, model building unit, model training unit, real-time monitoring unit and early warning unit; An image enhancement unit, used for performing image enhancement on a collected railway track obstacle image data set by using a Retinex method to obtain an enhanced image; A model building unit is used to replace the VGG16 backbone network with the MobileNetV3 network on the basis of the SSD model to obtain the MobileNetV3-SSD model; A model training unit is used to train a MobileNetV3-SSD model based on historical enhanced images to obtain a trained MobileNetV3-SSD model; The real-time monitoring unit is used to input the enhanced images collected in real time into the trained MobileNetV3-SSD model for recognition; The early warning unit is used to send out an early warning signal when an obstacle on the railway track is identified.

7. A railway track obstacle warning device according to claim 6, characterized in that: The Retinex method is used to enhance the collected railway track obstacle image dataset, including: Use a low-pass filter to perform Gaussian blur processing on the original image to retain the color information in the image, reduce the noise in the image, and obtain a color restored image; The original image is filtered by a high-pass filter to enhance the brightness information in the image, highlight the details and edges in the image, and obtain a brightness-adjusted image; The color restored and brightness adjusted images are rescaled to obtain the final enhanced image dataset.

8. The railway track obstacle warning device according to claim 6, characterized in that: The enhanced image collected in real time is input into the trained MobileNetV3-SSD model for recognition, including: The image-enhanced dataset is input into the trained MobileNetV3-SSD model. Through a forward propagation, the image is analyzed and the bounding box and corresponding confidence of each possible obstacle in the image are predicted. The bounding box represents the obstacle position predicted by the MobileNetV3-SSD model, and the confidence represents the degree of confidence of the MobileNetV3-SSD model in its prediction. The image detected by the MobileNetV3-SSD model contains multiple bounding boxes and multiple confidence levels. The non-maximum suppression method is used to post-process the detection data; the detection results are improved by removing bounding boxes with overlap higher than the overlap threshold and confidence lower than the confidence threshold.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; a memory storing a computer program; The processor is used to implement a railway track obstacle early warning method as described in any one of claims 1 to 5 when executing a computer program stored in a memory.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a railway track obstacle early warning method according to any one of claims 1 to 5 is implemented.

Images