Road section risk early warning method based on Leiyu fusion perception

Abstract

The invention relates to a road section risk early warning method based on thunder-vision fusion perception, and the method comprises the following steps: S1, obtaining the high-precision space-time position information of a vehicle in a tunnel through the multi-source information fusion of a millimeter-wave radar and machine vision; s2, mapping the vehicle position under the global coordinate system to a coordinate system expanded along a road center line, and representing the vehicle position as a longitudinal projection position and a transverse offset along the road direction; s3, under the coordinate system, taking anti-collision time as a core index, combining lane judgment and relative speed, constructing a dynamic longitudinal risk grading model, and performing linkage triggering with a warning system; s4, constructing a regional risk integral model; and S5, speed limiting information is issued to all driving-in vehicles in real time through the variable speed limiting screen, and the rear-end collision risk is actively prevented and controlled. According to the method, the sudden risk response time can be shortened, the speed limit control response precision can be improved, the rear-end collision rate of the diversion and convergence areas can be reduced, false alarm interference can be reduced, and the tunnel driving safety and the information guiding efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent traffic safety technology, and more specifically, to a road section risk warning method based on radar and visual fusion perception. Background Art

[0002] Underground tunnel interchanges, a vital component of modern transportation networks, present significant safety risks due to the unique spatial structure of their internal merging and diverging areas. In tunnels with restricted longitudinal alignment, the entrance to the merging area often experiences visual obstruction caused by sudden changes in road curvature, reducing the driver's effective sight range to less than 100 meters. At the diverging area exit, double obstruction from the tunnel walls and anti-glare features can result in lateral blind spots exceeding 30 degrees. This three-dimensional sense of oppression not only impairs the driver's situational awareness of surrounding vehicles but also poses multiple safety risks in areas with intersecting traffic.

[0003] In the area of ​​traffic safety control in tunnel merge / divergence zones, existing technologies primarily rely on single-sensor monitoring and static warning mechanisms, which present significant technical flaws. Traditional millimeter-wave radar systems have a sector-shaped detection range due to their installation angle, and detection accuracy is easily affected by reflective objects. Pure vision solutions, on the other hand, suffer from image overexposure and noise interference in tunnels with alternating light and dark areas. Current collision warning systems often use a TTC algorithm based on Cartesian coordinates, but this fails to account for heading deviations caused by road curvature. More significantly, fixed speed limit signs cannot dynamically respond to real-time risk changes. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a road section risk warning method based on radar-visual fusion perception, which significantly improves the speed limit control response accuracy, effectively reduces the incidence of rear-end collisions in merge and diverge areas, and significantly reduces false alarm interference, thereby comprehensively improving tunnel driving safety and information guidance efficiency.

[0005] The technical solution adopted by the present invention to solve the technical problem is to construct a road section risk warning method based on radar and visual fusion perception, including the following steps:

[0006] S1. Obtain high-precision spatiotemporal position information of vehicles in tunnels through multi-source information fusion of millimeter-wave radar and machine vision;

[0007] S2. Mapping the vehicle position in the global coordinate system to a Frenet coordinate system developed along the centerline of the road, and expressing the vehicle position as a longitudinal projection position and a lateral offset along the road direction;

[0008] S3. In the Frenet coordinate system, using Inverse Time-to-Collision (iTTC) as the core indicator, combined with lane determination and relative speed, a dynamic longitudinal risk classification model is constructed and triggered in conjunction with the warning system.

[0009] S4. Using the merge / diverge area as the analysis window in the Frenet coordinate system, and combining the spatial distribution and impact intensity of all high-risk vehicle pairs, a regional risk integral model is constructed.

[0010] S5. Design a variable speed limit control algorithm linked to risk levels based on the real-time calculated regional risk score, and publish speed limit information to all entering vehicles in real time through the LED variable speed limit screen to achieve active prevention and control of rear-end collision risks.

[0011] According to the above solution, in step S1, the method for obtaining high-precision spatiotemporal position information of vehicles in a tunnel includes the following steps:

[0012] S101. Establish millimeter wave radar coordinate system C r and the visual image coordinate system C v The spatial mapping relationship between them is achieved by using the rotation matrix R∈R 3×3 With the translation vector T∈R 3 Complete the coordinate transformation:

[0013]

[0014] Among them, (x r ,y r , z r ) is the spatial position of the vehicle in the radar coordinate system; (x c ,y c , z c ) is the target position in the unified coordinate system; R and T are obtained by optimizing the calibration plate and external parameters;

[0015] S102: Using a confidence weighting strategy based on measurement variance to fuse the location information:

[0016]

[0017] Among them, x r ,y r is the radar detection result; x v ,y v is the visual inspection result; The radar and visual measurement variance are fused to output the vehicle position information (x, y) in a unified world coordinate system.

[0018] S103. To reduce the error of a single sensor under high light ratio, occlusion or target loss, a complementary fusion method of radar Doppler velocity measurement and visual optical flow velocity is used:

[0019] v=α·v r +(1-α)·v v

[0020] Among them, v r is the radar speed; v v is the visual speed, which is calculated by extracting the displacement Δx / Δt of the target feature points from the previous and next frame images; α is the weighting factor, which is dynamically adjusted according to the visual visibility and is set as:

[0021]

[0022] According to the above scheme, radar arrays are set up at intervals of 500m on the side walls of the tunnel in the longitudinal direction, and are all set up on the top of the tunnel to cover the entire lane; the operating frequency of the radar array is 76-81GHz, and the detection distance is greater than 500m; the radar array is coaxially provided with a machine camera, and the sensor of the machine camera is not less than 2 million pixels, and the effective visual distance is greater than 300m.

[0023] According to the above scheme, the method for mapping the vehicle position (x, y) in the global coordinate system to the Frenet coordinate system developed along the centerline of the road includes the following steps:

[0024] S201. Assume that the road centerline is a discrete point set The parameterization is constructed by cubic spline interpolation or piecewise linear fitting, and the curve C(s) is:

[0025] C(s) = [x(s), y(s)], s∈[0, L], where s is the arc length of the reference line and L is the length of the entire tunnel;

[0026] S202: Assume the current position of the vehicle is P = (x, y). Project this point to the nearest point on the curve C(s) to obtain the longitudinal coordinate s and the lateral offset l. The calculation formula is as follows:

[0027]

[0028] l=(PC(s * ))·n(s * )

[0029] Among them, n(s * ): The curve is in s * The unit normal vector at , is calculated by the curve tangent t(s):

[0030]

[0031] S203. Perform rapid optimization near the known reference point to find the nearest projection point, and use Newton's method to iteratively solve the optimal s:

[0032]

[0033] The iteration termination condition is set to Δs < 10 between two iterations. -3 m, converges within 3 to 5 iterations;

[0034] S204. Each vehicle finally outputs a Frenet coordinate pair:

[0035] (s i , l i , v i ), i = 1, 2, ..., N

[0036] Among them, s i is the longitudinal position along the road direction; l i is the lateral offset of the lane centerline; v i It is the radar fusion speed value output by S1.

[0037] According to the above scheme, in step S3, the method for constructing a dynamic longitudinal risk grading model includes the following steps:

[0038] S301. Determine whether the vehicles belong to the same lane by comparing the difference in Frenet lateral offset l:

[0039] |l i -l j |<δ

[0040] Among them, l i 、l j is the lateral offset between vehicles i and j; δ is the lane tolerance threshold;

[0041] S302: Calculate iTTC after determining the front and rear vehicles in the same lane

[0042]

[0043] Among them, s j -s i is the longitudinal distance between vehicles, v i -v j is the relative velocity, ∈=0.1; when v i <v j When , it means that the speed of the rear vehicle is slower than that of the front vehicle, and there is no risk of rear-end collision. The Inverse Time-to-Collision (iTTC) at this time is not calculated.

[0044] According to the above scheme, in step S4, the method for constructing the regional risk score model includes the following steps:

[0045] S401: The 400-meter-long tunnel junction and confluence area is treated as an independent analysis unit, and the risk score is refreshed every 30 seconds.

[0046] S402: For all high-risk vehicle pairs (i, j) identified in the area, construct the following risk score model:

[0047]

[0048] Among them, iTTC ij Calculated from step S3, d ij is the longitudinal distance between the two vehicles, and λ is the attenuation parameter, which is used to reflect the impact of distance on risk propagation;

[0049] S403. Using historical traffic conflict data through cluster analysis, the following recommended values ​​for regional risk score classification are obtained, calculated every 30 seconds:

[0050]

[0051] When a level one risk is detected, the buffer zone's ring warning light is a single green ring that is constantly on; when a level two risk is detected, the ring warning light is in a yellow half-ring scanning state; when a level three risk is detected, the ring warning light turns red and flashes in a full ring; the light strip controller directly reads the level value from the risk integration module, switches the lighting mode through PLC control, and displays the text level on the LED display.

[0052] According to the above scheme, the annular warning light is longitudinally arranged in the buffer zone just above the lane at the top of the tunnel; the width of the annular light is 30-50 cm, and the interval between adjacent annular lights is 5 meters, forming a 200-500 meter continuous warning buffer zone;

[0053] The ring light is centrally controlled by the central control system, and reads the overall risk level of the road section output in step S4 in real time and switches the lighting mode: when it is a level one risk, the light is green and the single ring is always on, with a brightness of 600cd / m 2 to 800cd / m 2 When the risk level is level 2, the light moves toward the main road in a yellow half-circle scanning pattern; when the risk level is level 3, the light flashes in a red full-circle pulse with a brightness of 2500cd / m 2 ; The system refreshes the lighting status every 30 seconds.

[0054] According to the above solution, in step S5, the method for actively preventing and controlling the risk of rear-end collision accidents includes the following steps:

[0055] S501. Combining the smoothness and responsiveness of the logarithmic function, the following variable speed limit calculation formula is designed:

[0056]

[0057] Parameter settings:

[0058] v base : Base speed limit,

[0059] Δv: maximum drop,

[0060] R0: safety risk threshold,

[0061] Δ: risk growth adjustment parameter,

[0062] V min : Minimum speed limit,

[0063] Among them, if R≤R0, then V limit =V base , if R<R0, the speed limit decreases continuously as the risk score increases;

[0064] S502: A variable speed limit LED screen is used to announce the speed limit. The control center refreshes the speed limit every 30 seconds and updates it in real time on the LED screen at the top of the tunnel. A speed reduction transition strategy is also adopted: if the speed limit is reduced by more than 10 km / h, the system will be implemented in two stages to ensure a smooth response from the driver.

[0065] The implementation of the road section risk warning method based on radar-visual fusion perception of the present invention has the following beneficial effects:

[0066] This invention achieves significant improvements in multi-dimensional traffic safety indicators by constructing a tunnel risk warning and dynamic speed limit control system based on radar-visual fusion perception. First, the radar-visual fusion positioning algorithm effectively integrates millimeter-wave radar and visual information. Second, the Frenet coordinate transformation model is introduced to overcome the influence of tunnel curvature and, combined with the Inverse Time-to-Collision (iTTC) indicator, enables real-time dynamic assessment of vehicle longitudinal conflict risk. Furthermore, the risk integral model integrates the relative risk relationship between vehicles with the distance attenuation mechanism to construct a risk level assessment system for the entire road section. This system, in conjunction with the ring light strip and LED variable speed limit screen, responds in a coordinated manner to achieve a closed-loop mechanism of "warning-prompt-control." In actual measurements, the system can effectively shorten the response time to sudden risks, significantly improve the response accuracy of speed limit control, effectively reduce the incidence of rear-end collisions in merge and diverge areas, and significantly reduce false alarm interference, comprehensively improving tunnel driving safety and information guidance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0068] Figure 1 This is a flow chart of the road section risk warning method based on radar and visual fusion perception of the present invention;

[0069] Figure 2 It is a schematic diagram of the internal layout of the tunnel of the present invention;

[0070] Figure 3 It is a plan view of the tunnel merging area of ​​the present invention. DETAILED DESCRIPTION

[0071] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0072] like Figure 1 As shown, the road section risk warning method based on radar and visual fusion perception of the present invention includes the following steps:

[0073] S1. Extract vehicle location information through radar and visual fusion

[0074] By fusing multi-source information from millimeter-wave radar and machine vision, we can obtain high-precision spatiotemporal position information for vehicles in tunnels, laying the data foundation for subsequent coordinate transformation and risk assessment. Due to the dramatic changes in lighting and frequent structural occlusions in tunnel environments, a single sensor may have blind spots or errors. Therefore, a confidence-weighted radar-visual fusion method is used to improve robustness and accuracy. Specifically, the following steps are involved:

[0075] S101. Sensor Coordinate Unification: Joint Calibration and Coordinate Transformation

[0076] Establish millimeter wave radar coordinate system C r and the visual image coordinate system C v The spatial mapping relationship between them is achieved by using the rotation matrix R∈R 3×3 With the translation vector T∈R 3 Complete the coordinate transformation:

[0077]

[0078] Among them, (x r ,y r , z r ) is the spatial position of the vehicle in the radar coordinate system;

[0079] (x c ,y c , z c ) is the target position in the unified coordinate system;

[0080] R and T are obtained by optimizing the calibration plate and external parameters.

[0081] S102. Position Data Fusion: Confidence Weighted Algorithm

[0082] Since radar has high accuracy in the depth direction but has deviations in the lateral direction, and vision has accurate lateral positioning in the image plane but is affected by occlusion, fusing the two measurements can improve overall robustness. A confidence weighting strategy based on measurement variance is used to fuse the position information:

[0083]

[0084] Among them, x r ,y r is the radar detection result; x v ,y v is the visual detection result (through image recognition + projection); is the variance of radar and visual measurement, which is usually calculated from historical data or online residuals. The variance of radar measurement in the tunnel is approximately (Unit m 2 ), visual variance depends on scene changes. The fusion result outputs the vehicle position information (x, y) in a unified world coordinate system for use by the coordinate conversion module in step S2.

[0085] S103, Speed ​​Data Fusion: Complementary Filter Model

[0086] The accuracy of vehicle speed estimation in tunnels directly affects the Inverse Time-to-Collision (iTTC) calculation. Therefore, a complementary fusion method of radar Doppler velocity measurement and visual optical flow velocity is adopted to reduce the error of a single sensor under high light ratio, occlusion or target loss.

[0087] v=α·v r +(1-α)·v v

[0088] Among them, v r is the radar speed (unit: km / h); v v is the visual speed, which is calculated by extracting the displacement Δx / Δt of the target feature points from the previous and next frame images; α is the weighting factor, which is dynamically adjusted according to the visual visibility and is usually set to:

[0089]

[0090] The fused velocity value v will be used to calculate the longitudinal time to collision (iTTC) and the regional conflict integral in step S3.

[0091] Radar arrays are placed at 500m intervals longitudinally along the tunnel sidewalls and installed at the tunnel roof, providing full lane coverage. Technical specifications include an operating frequency of 76-81GHz (in anti-interference mode) and a detection range greater than 500m (with an accuracy of ±0.1m). A robotic camera, with a sensor of at least 2 megapixels and an effective viewing range greater than 300m, is coaxially positioned with the radar.

[0092] S2, Frenet coordinate transformation

[0093] In tunnel diverging and merging areas, due to the curvature of the road, assessing inter-vehicle distances and risk directly in a global Cartesian coordinate system will introduce directional deviation errors. Therefore, the vehicle position (x, y) in the global coordinate system is mapped to a Frenet coordinate system along the road centerline. The vehicle position is represented as a longitudinal projection s and a lateral offset l along the road direction to improve the accuracy and spatial compatibility of risk assessment. This involves the following steps:

[0094] S201. Reference Line Construction: Road Geometry Modeling

[0095] Assume that the road centerline is a discrete point set The parameterization is constructed by cubic spline interpolation or piecewise linear fitting, and the curve C(s) is:

[0096] C(s)=[x(s), y(s)], s∈[0, L]

[0097] Where s is the arc length of the reference line, and L is the length of the entire tunnel. In addition, the recommended spacing between curve points is 2 to 5 meters to balance fitting accuracy and calculation efficiency.

[0098] S202, Vehicle Projection: Frenet Coordinate Solution

[0099] Assume the vehicle's current position is P = (x, y). Project this point to the nearest point on the curve C(s) to obtain the longitudinal coordinate s and the lateral offset l. The calculation formula is as follows:

[0100]

[0101] l=(PC(s * ))·n(s * )

[0102] Among them, n(s * ): The curve is in s * The unit normal vector at , usually calculated by the curve tangent t(s):

[0103]

[0104] S203, Projection Optimization: Newton Iteration Method

[0105] Perform fast optimization near the known reference point to find the nearest projection point, and use Newton's method to iteratively solve the optimal s:

[0106]

[0107] The iteration termination condition is set to Δs < 10 between two iterations. -3 It usually converges within 3 to 5 iterations.

[0108] S204, output results

[0109] Each vehicle finally outputs a Frenet coordinate pair:

[0110] (s i , l i , v i )i=1,2,...,N

[0111] Among them, s i is the longitudinal position along the road direction (m); l i is the lateral offset of the lane centerline; v i is the radar-visual fusion speed value output by S1. This coordinate is used for lane determination and longitudinal risk calculation in the subsequent S3.

[0112] S3. Vertical risk dynamic assessment

[0113] In tunnel merge / diverge areas, the longitudinal distance between vehicles decreases rapidly, enabling real-time assessment of potential rear-end collision risks. Using the Frenet coordinate system, a dynamic longitudinal risk classification model is constructed using the time-to-collision (iTTC) as the core indicator, combined with lane determination and relative speed, and then linked to the warning system. This includes the following steps:

[0114] S301, screening of vehicles in the same lane

[0115] To avoid misjudgment, risk assessment is limited to vehicles in the same lane. By comparing the difference in Frenet lateral offset l between vehicles, it is determined whether they belong to the same lane:

[0116] |l i -l j |<δ

[0117] Among them: i 、l j is the lateral offset between vehicles i and j; δ is the lane tolerance threshold, which is set to 1.75m (half of the reference lane width of 3.5m).

[0118] S302: Calculate iTTC after determining the front and rear vehicles in the same lane

[0119]

[0120] Among them, s j -s i is the longitudinal distance between vehicles, v i -v j is the relative velocity, ∈=0.1 (unit: m / s), which is used to avoid abnormal values ​​generated by division by zero or a very small number. i <v j When , it means that the speed of the rear vehicle is slower than that of the front vehicle, and there is no risk of rear-end collision, so iTTC is not calculated at this time.

[0121] S4. Regional risk score calculation and classification

[0122] In high-density tunnel traffic environments, the iTTC risk levels between individual vehicles alone cannot fully reflect the overall risk trend of a specific area. Using the merge / diverge area as the analysis window in the Frenet coordinate system, we combined the spatial distribution and impact intensity of all high-risk vehicle pairs to construct a regional risk integration model to achieve a quantifiable and dynamically updateable risk level assessment. This includes the following steps:

[0123] S401. Analysis segment definition

[0124] The 400-meter diverging and merging area in the tunnel is taken as an independent analysis unit, and the risk score is refreshed every 30 seconds.

[0125] S402. Risk Integral Modeling: Exponential Decay Integral Function

[0126] For all high-risk vehicle pairs (i, j) identified in the region, the following risk score model is constructed

[0127]

[0128] Among them, iTTC ij Calculated from step S3, d ij is the longitudinal distance between the two vehicles, and λ is the attenuation parameter, which is used to reflect the impact of distance on risk propagation. According to the actual situation of this method, λ = 30 (unit: m) (based on the statistics of the impact range of traffic accidents, 30 m is the main response area).

[0129] S403. Risk score threshold division

[0130] To achieve stable classification and adapt to different traffic scenarios, historical traffic conflict data (including rear-end collisions) was used through cluster analysis to fit the following regional risk score classification recommendations, calculated every 30 seconds:

[0131]

[0132] When a Level 1 risk is detected, the buffer zone's ring warning light illuminates a single, solid green ring. When a Level 2 risk is detected, the ring warning light flashes a half-circle of yellow. When a Level 3 risk is detected, the ring warning light flashes a full red ring. The light strip controller reads the risk score value directly from the risk score module, switches lighting modes through PLC control, and displays the level in text on the LED display.

[0133] A set of high-brightness ring warning light arrays are set up in the buffer zone. Each ring light is 30-50cm wide, fits the top of the tunnel, and is encapsulated with an aluminum alloy shell and a high-brightness LED module. It has IP65 protection and can adapt to the harsh environment of the tunnel, such as heat, humidity, and dust. The lamps are arranged longitudinally along the top of the tunnel and fixed directly above the lane. The adjacent ring warning lights are 5 meters apart, forming a 200-500 meter continuous warning buffer zone, which is coordinated with the speed limit display screen and the risk level LED screen to ensure multi-channel risk reminders within the line of sight. All ring lights are centrally dispatched by the central control system, and the overall risk level L of the road section output by step S4 is read in real time and the lighting mode is switched: when the risk is level one (safe), the light is a single green ring that is always on, and the brightness is 600cd / m 2 to 800cd / m 2 When the risk level is level 2 (warning), the light moves in a yellow half-circle scanning pattern toward the main road; when the risk level is level 3 (high risk), the light flashes in a red full-circle pulse with a brightness of at least 2500 cd / m 2 The system refreshes the lighting status every 30 seconds to ensure real-time communication of changes in risk levels. This warning light array provides at least 15 seconds of continuous dynamic warning when a vehicle approaches at 60 km / h, significantly enhancing the driver's risk perception in limited-viewing environments and effectively reducing delays in responding to emergencies.

[0134] S5. Dynamic speed limit algorithm design

[0135] To improve traffic safety response capabilities in complex tunnel scenarios, a variable speed limit control algorithm linked to risk levels is designed based on the real-time calculated regional risk score R. Speed ​​limit information is also released to all entering vehicles in real time via an LED variable speed limit screen, enabling proactive prevention and control of rear-end collision risks. This includes the following steps:

[0136] S501. Combining the smoothness and responsiveness of the logarithmic function, the following variable speed limit calculation formula is designed:

[0137]

[0138] Parameter settings:

[0139] v base : Base speed limit (tunnel design speed 60km / h)

[0140] Δv: Maximum drop (20km / h is recommended for tunnels)

[0141] R0: Security risk threshold (the graded thresholds correspond to level 1 risk, level 2 risk, and level 3 risk respectively)

[0142] Δ: Risk growth adjustment parameter (take 100 to smooth the impact of risk score growth)

[0143] V min : Minimum speed limit (30km / h is recommended for tunnels)

[0144] Where: If R≤R0, then V limit =V base ;

[0145] If R>R0, the speed limit decreases continuously as the risk score increases.

[0146] S502. Release Mechanism and Frequency Control

[0147] The speed limit is displayed on a variable LED screen, updated every 30 seconds by the control center and updated in real time on the LED screen at the top of the tunnel. A speed reduction transition strategy is also implemented: if the speed limit decreases by more than 10 km / h, the system will gradually implement the reduction in two stages to ensure a smooth driver response.

[0148] To ensure timely and accurate transmission of dynamic speed limit information within the tunnel to drivers, this system installs an LED display at the top of the tunnel at the starting point of the variable speed limit zone. This displays the risk level and speed limit information for the road section, combining the risk level and speed limit values ​​output in steps S4 and S5. The variable speed limit LED display is located at least 5.5 meters above the road surface and utilizes a dual-color, high-brightness LED display module with a character height of at least 320 mm. It is capable of displaying information in all weather conditions. This ensures drivers have sufficient reaction time at a speed of 60 km / h. The display includes the current dynamic speed limit and the corresponding risk level. The control logic is managed by a central control system, which receives the speed limit value output in step S5 and transmits it to each LED controller via a communication link. By default, the display is updated every 30 seconds, and any change in the risk level will be immediately refreshed. The display color is controlled based on the risk level: Level 1 (normal traffic) displays a green speed limit of 60 km / h; Level 2 (warning level) displays a yellow speed limit of 45 km / h; and Level 3 (high risk) displays a flashing red speed limit of 30 km / h. In addition, it is equipped with an automatic brightness adjustment function in night environments to effectively prevent glare from affecting the driver's vision.

[0149] like Figure 2-3The figure shows the layout of the radar-visual fusion perception system within the tunnel, including the installation locations of the millimeter-wave radar and machine vision cameras, the distribution of the circular warning light array, and the positional relationship of the LED dynamic speed limit display. The radar and camera are paired at the tunnel roof, at a vertical height of at least 5.5 meters, with a fixed pitch angle toward the road surface, forming a perception zone covering the entire lane. High-brightness circular lights are arranged vertically at 5-meter intervals across the top of the buffer zone, forming a continuous 200-500 meter warning zone. The circular warning lights illuminate green, yellow, and red at risk levels 1, 2, and 3, respectively. The LED display, located at the beginning of the speed limit zone, displays the road section's risk level and the variable speed limit. Facing the direction of oncoming traffic, it ensures drivers have ample reading distance and reaction time at normal speeds.

[0150] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A road section risk warning method based on radar and visual fusion perception, characterized in that: The following steps are involved: S1. Obtain high-precision spatiotemporal position information of vehicles in tunnels through multi-source information fusion of millimeter-wave radar and machine vision; S2. Mapping the vehicle position in the global coordinate system to a Frenet coordinate system developed along the centerline of the road, and expressing the vehicle position as a longitudinal projection position and a lateral offset along the road direction; S3. In the Frenet coordinate system, using anti-collision time as the core indicator, combined with lane determination and relative speed, a dynamic longitudinal risk classification model is constructed and triggered in conjunction with the warning system; S4. Using the merge / diverge area as the analysis window in the Frenet coordinate system, and combining the spatial distribution and impact intensity of all high-risk vehicle pairs, a regional risk integral model is constructed. S5. Design a variable speed limit control algorithm linked to risk levels based on the real-time calculated regional risk score, and publish speed limit information to all entering vehicles in real time through the LED variable speed limit screen to achieve active prevention and control of rear-end collision risks.

2. The road section risk warning method based on radar and visual fusion perception according to claim 1 is characterized in that: In step S1, the method for obtaining high-precision spatiotemporal position information of a vehicle in a tunnel comprises the following steps: S101. Establish millimeter wave radar coordinate system C r and the visual image coordinate system C v The spatial mapping relationship between them is achieved by using the rotation matrix R∈R 3×3 With the translation vector T∈R 3 Complete the coordinate transformation: Among them, (x r ,y r , z r ) is the spatial position of the vehicle in the radar coordinate system; (x c ,y c , z c ) is the target position in the unified coordinate system; R and T are obtained by optimizing the calibration plate and external parameters; S102: Using a confidence weighting strategy based on measurement variance to fuse the location information: Among them, x r ,y r is the radar detection result; x v ,y v is the visual inspection result; The radar and visual measurement variance are fused to output the vehicle position information (x, y) in a unified world coordinate system. S103. To reduce the error of a single sensor under high light ratio, occlusion or target loss, a complementary fusion method of radar Doppler velocity measurement and visual optical flow velocity is used: v=α·v r +(1-α)·v v Among them, v r is the radar speed; v v is the visual speed, which is calculated by extracting the displacement Δx / Δt of the target feature points from the previous and next frame images; α is the weighting factor, which is dynamically adjusted according to the visual visibility and is set as:

3. The road section risk warning method based on radar and visual fusion perception according to claim 2 is characterized in that: In step S1, radar arrays are set up on the tunnel sidewalls at longitudinal intervals of 500m, and are all set up at the top of the tunnel to cover the entire lane; the operating frequency of the radar array is 76-81GHz, and the detection range is greater than 500m; the radar array is coaxially provided with a machine camera, and the sensor of the machine camera has no less than 2 million pixels and an effective visual range of greater than 300m.

4. The road section risk warning method based on radar and visual fusion perception according to claim 3 is characterized in that: In step S2, the method of mapping the vehicle position (x, y) in the global coordinate system to the Frenet coordinate system developed along the road centerline includes the following steps: S201. Assume that the road centerline is a discrete point set The parameterization is constructed by cubic spline interpolation or piecewise linear fitting, and the curve C(s) is: C(s)=[x(s),y(s)],s∈[0,L] Where s is the arc length of the reference line, and L is the length of the entire tunnel; S202: Assume the current position of the vehicle is P = (x, y). Project the point to the nearest point on the curve C(s) to obtain the longitudinal coordinate s and the lateral offset l. The calculation formula is as follows: l=(P-C(s * ))·n(s * ) Among them, n(s * ): The curve is in s * The unit normal vector at , is calculated by the curve tangent t(s): S203. Perform rapid optimization near the known reference point to find the nearest projection point, and use Newton's method to iteratively solve the optimal s: The iteration termination condition is set to Δs < 10 between two iterations. -3 m, converges within 3 to 5 iterations; S204. Each vehicle finally outputs a Frenet coordinate pair: (s i ,l i ,v i ),i=1,2,...,N Among them, s i is the longitudinal position along the road direction; l i is the lateral offset of the lane centerline; v i It is the radar fusion speed value output by S1.

5. The road section risk warning method based on radar and visual fusion perception according to claim 4 is characterized in that: In step S3, the method for constructing a dynamic longitudinal risk grading model includes the following steps: S301. Determine whether the vehicles belong to the same lane by comparing the difference in Frenet lateral offset l: |l i -l j |<d Among them, l i 、l j is the lateral offset between vehicles i and j; δ is the lane tolerance threshold; S302: Calculate iTTC after determining the front and rear vehicles in the same lane Among them, s j -s i is the longitudinal distance between vehicles, v i -v j is the relative velocity, ∈=0.1; when v i <v j When , it means that the speed of the rear vehicle is slower than that of the front vehicle, and there is no risk of rear-end collision, so the anti-collision time at this time is not calculated.

6. The road section risk warning method based on radar and visual fusion perception according to claim 5 is characterized in that: In step S4, the method for constructing the regional risk score model includes the following steps: S401: The 400-meter-long tunnel divergence and merging area is treated as an independent analysis unit, and the risk score is refreshed every 30 seconds. S402: For all high-risk vehicle pairs (i, j) identified in the area, construct the following risk score model: Among them, iTTC ij Calculated from step S3, d ij is the longitudinal distance between the two vehicles, and λ is the attenuation parameter, which is used to reflect the impact of distance on risk propagation; S403. Using historical traffic conflict data through cluster analysis, the following recommended values ​​for regional risk score classification are obtained, calculated every 30 seconds: When a level one risk is detected, the buffer zone's ring warning light is a single green ring that is constantly on; when a level two risk is detected, the ring warning light is in a yellow half-ring scanning state; when a level three risk is detected, the ring warning light turns red and flashes in a full ring; the light strip controller directly reads the level value from the risk integration module, switches the lighting mode through PLC control, and displays the text level on the LED display.

7. The road section risk warning method based on radar and visual fusion perception according to claim 6 is characterized in that: The annular warning light is longitudinally arranged in the buffer zone just above the lane at the top of the tunnel; the width of the annular light is 30-50 cm, and the interval between adjacent annular lights is 5 meters, forming a 200-500 meter continuous warning buffer zone; The ring light is centrally controlled by the central control system, and reads the overall risk level of the road section output in step S4 in real time and switches the lighting mode: when it is a level one risk, the light is green and the single ring is always on, with a brightness of 600cd / m 2 to 800cd / m 2 When the risk level is level 2, the light moves toward the main road in a yellow half-circle scanning pattern; when the risk level is level 3, the light flashes in a red full-circle pulse with a brightness of 2500cd / m 2 ; The system refreshes the lighting status every 30 seconds.

8. The road section risk warning method based on radar and visual fusion perception according to claim 7 is characterized in that: In step S5, the method for actively preventing and controlling the risk of rear-end collisions includes the following steps: S501. Combining the smoothness and responsiveness of the logarithmic function, the following variable speed limit calculation formula is designed: Parameter settings: v base : Base speed limit, Δv: maximum drop, R0: safety risk threshold, Δ: risk growth adjustment parameter, V min : Minimum speed limit, Among them, if R≤R0, then V limit =V base , if R>R0, the speed limit decreases continuously as the risk score increases; S502: A variable speed limit LED screen is used to announce the speed limit. The control center refreshes the speed limit every 30 seconds and updates it in real time on the LED screen at the top of the tunnel. A speed reduction transition strategy is also adopted: if the speed limit is reduced by more than 10 km / h, the system will be implemented in two stages to ensure a smooth response from the driver.

Images