Rail traffic signal lamp sensing method based on multi-sensor cross authentication

Abstract

The invention discloses a rail traffic signal lamp sensing method based on multi-sensor cross authentication, and relates to the technical field of visual sensor identification. A transponder unit sends signal machine state information to an information receiving module; the vehicle-mounted sensing unit shoots a signal machine through a three-dimensional coordinate system established based on a tunnel portal to obtain a multi-dimensional feature vector, and sends the multi-dimensional feature vector to the information receiving module; the track circuit unit sends track circuit code sequence information to the information receiving module; and the information receiving module compares the signal machine state information with the multi-dimensional feature vector and the track circuit code sequence information respectively to confirm that the traffic signal lamp is correctly identified at the moment. According to the invention, accurate identification of the state of the signal lamp is realized through a dual comparison authentication mode.

Description

Technical Field

[0001] This invention relates to the field of visual sensor recognition technology, and more specifically to a method for sensing rail transit signal lights based on multi-sensor cross-certification. Background Technology

[0002] With the rapid development of high-speed rail in my country, the operating mileage is becoming longer and the track environment is becoming more complex. Prolonged operation of high-speed trains in complex environments leads to driver fatigue, affecting their ability to see and interpret traffic signals, and posing potential safety hazards. The rail transit signaling system is a crucial system for ensuring safe train operation and improving transportation efficiency.

[0003] Traditional rail transit signaling systems use either track circuits and transponders or wireless communication and transponders. In human-controlled mode, the identification of signals mainly relies on the driver's observation; in machine-controlled mode, there may be inconsistencies between signal information and signal controller information.

[0004] Currently, methods for obtaining signal status typically include: Signal status reading based on vehicle-to-ground wireless communication: This method utilizes vehicle-to-ground wireless communication technology to transmit signal status information to the train via transponder messages. The transponder is a point-type device; when the train passes the transponder location, it receives the transponder message information through the onboard BTM, thereby obtaining the signal status information ahead. Track circuits achieve dual functions through the on / off switching and encoding of rail current: firstly, real-time detection of track section occupancy status (if the train is present, the wheelset short-circuits the rail, cutting off the current); secondly, transmitting train control commands (such as speed codes and signal status) to the train. Signal recognition based on image recognition: This method uses an onboard camera to capture signal images and identifies the signal status through image processing and target recognition technology. However, this method is greatly affected by lighting conditions and weather factors, has a low recognition rate, and requires high computing resources. In summary, most current methods use a single sensor, which is easily affected by equipment safety and availability.

[0005] Therefore, accurately identifying the status of traffic lights is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a rail transit signal light sensing method based on multi-sensor cross-certification, which realizes accurate identification of signal light status through multiple sensors.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for sensing traffic signals based on multi-sensor cross-certification is provided on an ATO (Automatic Train Operation) system. The method includes: an onboard sensing unit, a transponder unit, a track circuit unit, and an information receiving module. The transponder unit sends signal status information to the information receiving module. The onboard sensing unit captures images of the signal using a three-dimensional coordinate system established at the tunnel entrance, acquires multi-dimensional feature vectors, and sends these feature vectors to the information receiving module. The track circuit unit sends track circuit code sequence information to the information receiving module. The information receiving module compares the signal status information with the multi-dimensional feature vectors and the track circuit code sequence information to confirm that the traffic signal is correctly identified.

[0009] Preferably, the vehicle-mounted perception unit includes a low-light camera, a telephoto camera, a close-focus camera, a lidar, an image processing subunit, and a feature fusion unit;

[0010] The image processing subunit collects light position information using the low-light camera, telephoto camera, and close-focus camera, adjusts the shooting weights of each camera with preset weights, and performs feature fusion on the images captured by each camera to obtain image fusion features.

[0011] The fusion feature unit fuses the point cloud data collected by the lidar with the image fusion features to obtain a multi-dimensional feature vector.

[0012] Preferably, when the fusion feature unit extracts color features from multi-dimensional feature vectors, it uses a color space conversion method to convert the RGB color space to the HSV color space.

[0013] Preferably, the color space conversion method specifically includes:

[0014] S1, normalizes each component value of RGB to the range of 0 to 1;

[0015]

[0016] Where R, G, and B are the red, green, and blue components, respectively; and R′, G′, and B′ are the normalized red, green, and blue components, respectively.

[0017] S2, calculate the maximum, minimum and difference values ​​of the normalized RGB components;

[0018] C max =max(R′,G′,B′);

[0019] C min =min(R′,G′,B′);

[0020] Δ=C max -Cmin ;

[0021] Among them, C max C min Δ and Δ represent the maximum, minimum, and difference values ​​of the normalized RGB components, respectively.

[0022] S3 calculates brightness, saturation, and hue;

[0023] V = C max ;

[0024] If C max ≠0, C max =0, S=0;

[0025] If C max =R′, If C max =G′,

[0026] If C max =B′, When Δ = 0, H = 0;

[0027] Where H < 0, 360° needs to be added to make it fall within the range of 0 to 360°, V is lightness, S is saturation, and H is hue.

[0028] As can be seen from the above technical solution, compared with the prior art, this invention discloses a rail transit signal light sensing method based on multi-sensor cross-certification. By using multi-sensor information redundancy verification, it avoids the risk of information parsing errors or false alarms, improving the accuracy of signal light position recognition. The original ATO system is enhanced by adding an onboard sensing unit, improving the ATO system's sensing capabilities. The onboard sensing unit adopts a long-focus camera + short-focus camera + LiDAR sensor scheme, improving the signal light color recognition algorithm model, enhancing the sensing system's ability to perceive under poor lighting conditions, and improving signal light position recognition capabilities. Through vehicle-to-ground communication, the signal light is transmitted to the newly added ATO sensing unit in advance, allowing for pre-focusing of the camera and enhancing the sensing range. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0030] Figure 1 The structural flowchart provided for this invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] This invention discloses a rail transit signal light sensing method based on multi-sensor cross-certification, installed on an ATO system, including: an onboard sensing unit, a transponder unit, a track circuit unit, and an information receiving module. The transponder unit sends signal light status information to the information receiving module; the onboard sensing unit captures images of the signal light using a three-dimensional coordinate system established based on the tunnel entrance, acquires multi-dimensional feature vectors, and sends these feature vectors to the information receiving module; the track circuit unit sends track circuit code sequence information to the information receiving module; the information receiving module compares the signal light status information with the multi-dimensional feature vectors and the track circuit code sequence information to confirm that the traffic signal light is correctly identified.

[0033] In one specific embodiment, the multi-dimensional feature vector and the signal state information are actually both signal state, and the track circuit code sequence information is actually track circuit code. The correspondence between the track circuit code and the signal state should be as shown in Table 1 below:

[0034] Table 1 Correspondence Table

[0035] Signal status Green light Green and yellow lights Yellow light Double-spring lamp Double Yellow Flash red light Track circuit coding L size (L3 / L2 / L) LU code U code UU code UUS code H code

[0036] When a transponder message containing signal status information is received, this information is compared with the received track circuit code. If the correspondence in the table above is satisfied, the cross-verification of track circuit and transponder information is completed. If not, the manual confirmation process is initiated.

[0037] When the received transponder message contains signal status information, the signal status information is compared with the multi-dimensional feature vector identified by the vehicle-mounted sensing unit. If they match, the cross-verification between the vehicle-mounted sensing unit and the transponder unit is completed. If they do not match, the manual confirmation process is initiated.

[0038] In one specific embodiment, the vehicle-mounted perception unit includes a low-light camera, a telephoto camera, a close-focus camera, a lidar, an image processing subunit, and a feature fusion unit.

[0039] The image processing subunit collects light position information using a low-light camera, a telephoto camera, and a close-focus camera, adjusts the shooting weights of each camera with preset weights, and performs feature fusion on the images captured by each camera to obtain image fusion features.

[0040] The feature unit is fused, and the LiDAR point cloud data is projected into image data. Then, edge features are extracted from the projected image of the point cloud and the image acquired by the camera. Finally, the two feature vectors are concatenated to form a multi-dimensional feature vector, and the fusion is completed.

[0041] In one specific embodiment, when the fusion feature unit extracts color features from multi-dimensional feature vectors, it uses a color space conversion method to convert the RGB color space to the HSV color space.

[0042] In one specific embodiment, the on-board sensing unit needs to identify the status information of the signal lights along the route ahead of the train and identify the status of the signal lights in real time;

[0043] This invention employs the YOLO deep learning algorithm framework to identify the presence of traffic lights in images. After identifying the traffic lights, the status of the light positions is further determined. Due to factors such as distance or poor visibility, the color features of traffic light positions are often difficult to extract accurately. This invention improves upon this feature extraction method by utilizing a color space conversion method. The RGB color space is converted to the HSV color space. The RGB color space contains three components: red, green, and blue. The HSV color space contains three components: hue, saturation, and value. The color space conversion method specifically includes:

[0044] S1, normalizes each component value of RGB to the range of 0 to 1;

[0045]

[0046] Where R, G, and B are the red, green, and blue components, respectively; and R′, G′, and B′ are the normalized red, green, and blue components, respectively.

[0047] S2, calculate the maximum, minimum and difference values ​​of the normalized RGB components;

[0048] C max =max(R′,G′,B′);

[0049] C min =min(R′,G′,B′);

[0050] Δ=C max -Cmin ;

[0051] Among them, C max C min Δ and Δ represent the maximum, minimum, and difference values ​​of the normalized RGB components, respectively.

[0052] S3 calculates brightness, saturation, and hue;

[0053] V = C max ;

[0054] If C max ≠0, C max =0, S=0;

[0055] If C max =R′, If C max =G′,

[0056] If C max =B′, When Δ = 0, H = 0;

[0057] Where H < 0, 360° needs to be added to make it fall within the range of 0 to 360°, V is lightness, S is saturation, and H is hue.

[0058] Final value range: H∈[0°,360°); S∈[0,1] (can be converted to percentage, e.g., S×100%); V∈[0,1] (can be converted to percentage similarly).

[0059] Because the luminance component V in the HSV model is related to the brightness of a color, the impact of lighting on color recognition can be reduced by adjusting the luminance component when processing images affected by illumination. This characteristic allows the HSV model to maintain good color recognition performance even in environments with significant lighting variations, greatly mitigating the impact of insufficient lighting.

[0060] Furthermore, compared to the RGB model, the components in the HSV model have greater independence. This means that in the HSV model, hue, saturation, and brightness can be processed separately without interfering with each other. This characteristic allows for more precise control and processing of color information in image recognition, thereby improving recognition accuracy.

[0061] An onboard perception unit is added to the original ATO system. The perception unit's sensors include a low-light camera, a telephoto camera, a close-focus camera, and a LiDAR.

[0062] The signals are transmitted via transponders installed on the track and the train control onboard system (the ATP system currently widely used on high-speed trains).

[0063] Before approaching the signal, the camera is pre-focused to the signal position based on the signal information, ready to take a picture. The camera uses pre-acquired tunnel entrance data to establish a three-dimensional coordinate system. Then, the low-light camera, telephoto camera, and close-focus camera in the sensing unit sensor adjust the shooting weight of each camera according to preset weights, and the images captured by the cameras are fused together.

[0064] The principle of this invention is as follows Figure 1 As shown, specifically:

[0065] Main thread:

[0066] Vehicle-mounted sensing subprocess

[0067] Transponder Unit Subprocess

[0068] Track circuit sub-flow:

[0069] 1. Initialize each process and update the data.

[0070] 2. The main process begins to distribute data to each subprocess. The vehicle-mounted perception unit subprocess receives image data and lidar data, the transponder unit subprocess receives transponder message information, and the track circuit unit subprocess receives track circuit information.

[0071] 3. The onboard sensing unit subprocess uses machine learning methods to identify the status information of the signal lights on the train's running track and sends it to the main process.

[0072] 4. The transponder unit subprocess parses the transponder message information, determines the line signal information, and sends it to the main process.

[0073] 5. The track circuit unit subprocess parses the track circuit code sequence information and sends it to the main process.

[0074] 6. The main process performs a consistency comparison on the information from the three sub-processes. If they match, a normal status is displayed.

[0075] If there is a discrepancy, display the discrepancy information and request manual confirmation.

[0076] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0077] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for sensing rail transit signal lights based on multi-sensor cross-certification, implemented on an ATO (Automatic Transit) system, characterized in that, include: The system includes an onboard sensing unit, a transponder unit, a track circuit unit, and an information receiving module, wherein the transponder unit sends signal status information to the information receiving module. The vehicle-mounted sensing unit captures images of the traffic signal using a three-dimensional coordinate system established at the tunnel entrance, acquires multi-dimensional feature vectors, and sends these feature vectors to the information receiving module. The track circuit unit sends track circuit code sequence information to the information receiving module. The information receiving module compares the traffic signal status information with the multi-dimensional feature vectors and the track circuit code sequence information to confirm that the traffic light is correctly identified.

2. The rail transit signal light sensing method based on multi-sensor cross-certification according to claim 1, characterized in that, The vehicle-mounted perception unit includes a low-light camera, a telephoto camera, a close-focus camera, a lidar, an image processing subunit, and a feature fusion unit. The image processing subunit collects light position information using the low-light camera, telephoto camera, and close-focus camera, adjusts the shooting weights of each camera with preset weights, and performs feature fusion on the images captured by each camera to obtain image fusion features. The fusion feature unit fuses the point cloud data collected by the lidar with the image fusion features to obtain a multi-dimensional feature vector.

3. The rail transit signal light sensing method based on multi-sensor cross-certification according to claim 2, characterized in that, When extracting color features from multi-dimensional feature vectors, the fusion feature unit uses a color space conversion method to convert the RGB color space to the HSV color space.

4. The rail transit signal light sensing method based on multi-sensor cross-certification according to claim 3, characterized in that, The color space conversion method specifically includes: S1, normalizes each component value of RGB to the range of 0 to 1; Where R, G, and B are the red, green, and blue components, respectively; and R′, G′, and B′ are the normalized red, green, and blue components, respectively. S2, calculate the maximum, minimum and difference values ​​of the normalized RGB components; C max =max(R′,G′,B′); C min =min(R′,G′,B′); Δ=C max -C min ; Among them, C max C min Δ and Δ represent the maximum, minimum, and difference values ​​of the normalized RGB components, respectively. S3 calculates brightness, saturation, and hue; V=C max ; If C max ≠0, C max =0, S=0; If C max =R′, If C max =G′, If C max =B′, When Δ = 0, H = 0; Where H < 0, 360° needs to be added to make it fall within the range of 0 to 360°, V is lightness, S is saturation, and H is hue.

Images