A method and system for intelligent traffic signal control

Abstract

This invention discloses an intelligent traffic signal control method and system, relating to the field of intelligent traffic control technology. The method includes: acquiring environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data; identifying target rainfall interference states based on the visual image data and the millimeter-wave radar data; identifying vehicle driving events based on the induction coil sensing data, including vehicle stop events, vehicle slow-moving events, and vehicle normal-moving events; and adjusting traffic signals based on the target rainfall interference states and the vehicle driving events. This invention can combine visual image data, millimeter-wave radar data, and induction coil sensing data to adjust traffic signals, thereby achieving traffic signal control and improving the accuracy of road condition recognition and the reliability of signal control.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic control technology, and in particular to an intelligent traffic signal control method and system. Background Technology

[0002] In urban traffic management, intelligent traffic signal control systems collect traffic information in real time using various sensors deployed at intersections, such as visual capture devices, millimeter-wave radar, and ground induction coils. Based on this information, they dynamically adjust signal timing to improve road efficiency. Under normal weather conditions, such as sunny days or light rain, these systems perform stably and effectively handle routine traffic pressures such as morning and evening rush hours. However, when a local area suddenly experiences severe rainfall or other adverse environmental conditions, a water film forms on the lens surface of the visual capture device, creating dense rain lines in the image. Simultaneously, the beam of the millimeter-wave radar is attenuated due to raindrop scattering, resulting in a significant decrease in the data quality of these key information collection devices. This environmental interference causes the system to frequently misidentify blurred rain images as vehicles or incorrectly merge multiple closely moving vehicles, leading to serious deviations in traffic condition perception. Meanwhile, the system struggles to distinguish between sensor data distortion caused by environmental factors and actual traffic events, leading to significant data conflicts between different types of sensors. The traffic flow prediction mechanism cannot effectively analyze the current road conditions, resulting in inappropriate signal control strategies. This not only exacerbates traffic congestion but also severely impacts the ability to respond quickly to emergencies, resulting in low accuracy in road condition identification and low reliability in signal control.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0004] The main objective of this invention is to propose an intelligent traffic signal control method and system that can combine visual image data, millimeter-wave radar data, and induction coil sensing data to adjust traffic signals, thereby achieving traffic signal control and improving the accuracy of road condition recognition and the reliability of signal control.

[0005] On one hand, embodiments of the present invention provide an intelligent traffic signal control method, comprising the following steps: Acquire environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data; Based on the visual image data and the millimeter-wave radar data, identify the target's rainfall interference status; Based on the sensing data from the induction coil, vehicle driving events are identified, including vehicle stationary events, vehicle slow-moving events, and vehicle normal-moving events. Traffic signals are adjusted based on the target rainfall disturbance status and the vehicle driving events.

[0006] In some embodiments, identifying the target rainfall interference state based on the visual image data and the millimeter-wave radar data includes: The blurriness of the visual image data is calculated by using the Laplacian operator to obtain the image blurriness. Rain line features are obtained by performing Hough transform on the visual image data, and the rain line features include rain line length and rain line angle. Based on the millimeter-wave radar data, the signal scattering attenuation rate is calculated. The signal scattering attenuation rate is used to reflect the degree of signal strength attenuation caused by raindrops scattering the radar echo signal. The target rainfall interference state is identified based on the image blur, the rain line features, and the signal scattering attenuation rate.

[0007] In some embodiments, identifying vehicle driving events based on the sensing data from the induction coil includes: Based on the sensing data from the induction coil, vehicle driving characteristics are calculated, including lane occupancy rate, vehicle speed, and vehicle frequency. Based on the vehicle driving characteristics and the preset event feature patterns, the feature matching degree is calculated using the cosine similarity algorithm. The preset event feature patterns are used to reflect the driving characteristics under different vehicle driving events. The vehicle driving event is identified based on the feature matching degree and the preset matching degree threshold.

[0008] In some embodiments, calculating vehicle driving characteristics based on the sensing data from the induction coil includes: Vehicle pass-through count, vehicle entry time series, and vehicle exit time series are extracted from the sensing data of the induction coil. The vehicle entry time series includes the time when each vehicle enters the induction coil, the vehicle exit time series includes the time when each vehicle leaves the induction coil, and the vehicle pass-through count is used to represent the number of vehicles leaving the induction coil. Calculate the average dwell time of the vehicle based on the vehicle entry time sequence and the vehicle exit time sequence; The average idle time of the induction coil is calculated based on the vehicle entry time sequence and the vehicle exit time sequence. The average idle time of the induction coil is used to represent the average time during which no vehicle resides on the induction coil. The lane occupancy rate is calculated based on the average vehicle dwell time and the average idle time of the induction coil. Based on the vehicle's entry time sequence into the coil, calculate the first time interval between the vehicle passing through adjacent induction coils; The vehicle's passing speed is calculated based on the distance between adjacent induction coils and the first time interval; The vehicle passage frequency is calculated based on the vehicle passage count.

[0009] In some embodiments, identifying vehicle driving events based on the sensing data from the induction coil includes: The current vehicle dwell time and vehicle entry time sequence are extracted from the sensing data of the induction coil. The current vehicle dwell time is used to represent the duration during which the vehicle enters the induction coil and does not leave. Based on the vehicle entry time sequence, calculate the second time interval between adjacent vehicles entering the same induction coil; If the current vehicle dwell time is greater than a preset stagnation time threshold, then the vehicle driving event is determined to be a vehicle stagnation event. If the current vehicle dwell time is less than a preset stagnation time threshold, then determine whether the second time interval is greater than a preset slow passage time threshold. If the second time interval is greater than the preset slow passage time threshold, the vehicle driving event is determined to be a slow passage event; otherwise, the vehicle driving event is determined to be a normal passage event.

[0010] In some embodiments, identifying the target rainfall interference state based on the visual image data and the millimeter-wave radar data includes: Acquire a reference road image, which represents a road image under no-rain conditions; Based on the reference road image and the visual image data, identify the first rainfall disturbance state; Based on the millimeter-wave radar data, identify the second rainfall interference state; The target rainfall disturbance state is generated based on the first rainfall disturbance state and the second rainfall disturbance state.

[0011] In some embodiments, identifying a first rainfall disturbance state based on the reference road image and the visual image data includes: Droplet density and droplet size data are obtained using a laser scattering droplet density sensor. Based on the droplet density data and the droplet size data, the extinction coefficient is calculated. The extinction coefficient is used to represent the degree to which light is weakened due to absorption and scattering when it propagates in the medium. Based on the extinction coefficient, the image parameter offset is predicted, which includes the degree of contrast reduction and color offset; The image parameter offset is applied to the reference road image to obtain the desired blurred image; Extract the current road image from the visual image data; The expected blurred image is compared with the current road image to identify the first rainfall interference state.

[0012] In some embodiments, identifying the second rainfall interference state based on the millimeter-wave radar data includes: Electromagnetic wave attenuation data is obtained by a multi-band electromagnetic wave attenuation measuring instrument. The electromagnetic wave attenuation data is used to reflect the degree of attenuation of electromagnetic waves after absorption and scattering by fog droplets. Based on the electromagnetic wave attenuation data, predict the background noise information formed by droplet absorption and scattering; Calculate the average signal attenuation intensity based on the electromagnetic wave attenuation data; Based on the background noise information and the average signal attenuation intensity, the expected radar noise data is generated; The expected radar noise data is compared with the millimeter-wave radar data to identify the second rainfall interference state.

[0013] In some embodiments, adjusting the traffic signal based on the target rainfall disturbance state and the vehicle driving event includes: If the target rainfall interference state is no rainfall interference, then the green light passage time for all lane directions will be reduced to shorten the traffic light cycle and thus reduce the red light waiting time. If the target rainfall interference state is rainfall interference, and the vehicle driving event is a slow vehicle passage event or a normal vehicle passage event, then the green light passage time for all lane directions will be increased to reduce the decision-making pressure on drivers in low visibility conditions. If the target rainfall interference state is rainfall interference and the vehicle driving event is a vehicle stop event, then the green light time in the direction of the lane where the vehicle stop event occurs will be increased to reduce congestion caused by vehicle accidents.

[0014] On the other hand, embodiments of the present invention provide an intelligent traffic signal control system, comprising: The data acquisition module is used to acquire environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data. The rainfall interference status identification module is used to identify the target rainfall interference status based on the visual image data and the millimeter-wave radar data; The vehicle driving event recognition module is used to recognize vehicle driving events based on the sensing data of the induction coil. The vehicle driving events include vehicle stationary events, vehicle slow-moving events, and vehicle normal-moving events. The traffic signal adjustment module is used to adjust the traffic signal according to the target rainfall interference state and the vehicle driving event.

[0015] The embodiments of this application include at least the following beneficial effects: The embodiments of this application first acquire environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data. Then, based on the visual image data and millimeter-wave radar data, the target rainfall interference state is identified. Next, based on the induction coil sensing data, vehicle driving events are identified. Finally, based on the target rainfall interference state and vehicle driving events, traffic signals are adjusted. This allows for the adjustment of traffic signals by combining visual image data, millimeter-wave radar data, and induction coil sensing data to achieve traffic signal control, thereby improving the accuracy of road condition identification and the reliability of signal control.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description and the drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0018] Figure 1 This is a flowchart of an intelligent traffic signal control method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent traffic signal control system according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0020] In related technologies, intelligent traffic signal control systems are a core component of modern urban traffic management. Their core function is to improve road efficiency by sensing real-time traffic flow and dynamically adjusting control strategies. These systems typically rely on three types of data acquisition devices—visual capture devices, ground induction coils, and millimeter-wave radar—to acquire real-time road network information, thereby dynamically adjusting traffic light timings and even achieving regional coordinated control. However, in actual operation, when local areas experience sudden heavy rainfall or other extreme weather events, existing systems face severe challenges across the entire chain of information acquisition, data processing, and decision output: the data quality of core acquisition devices deteriorates significantly due to environmental interference, leading to serious deviations in the system's perception of traffic conditions. This results in the inability to distinguish between sensor distortion and real traffic events, and the system is prone to outputting incorrect control strategies, exacerbating congestion and significantly weakening its ability to respond to emergencies.

[0021] Under normal weather conditions, the system's operational efficiency has been fully validated. The system deploys various data acquisition devices at key intersections and road sections throughout the city, working collaboratively to acquire real-time traffic data: high-resolution visual capture devices output vehicle image information, completing vehicle type recognition, speed estimation, and queue length calculation; underground induction coils accurately count vehicle passages and confirm vehicle presence; millimeter-wave radar supplements this by detecting vehicles in scenarios where visual equipment is interfered with. All raw data is transmitted in real-time to the main processing unit at the traffic control center. After in-depth analysis, it is input into a predictive model trained on historical traffic patterns and real-time data to predict short-term traffic conditions, such as identifying traffic flow growth trends and pinpointing potential congestion points. Based on the prediction results, the system flexibly adjusts the green light duration, signal cycle, and phase sequence at individual intersections to maximize the capacity of each intersection. When multiple intersections experience simultaneous traffic congestion in the same direction, the system also activates regional "green wave" control, adjusting the phase difference of traffic lights along the route to allow vehicles to pass through multiple intersections at the recommended speed, reducing the number of stops and starts and improving the overall traffic experience. Under normal weather conditions such as sunny days and light rain, the system can reliably cope with routine traffic pressures such as morning and evening rush hours, ensuring the smooth operation of the urban road network.

[0022] However, sudden, localized heavy rainfall can completely disrupt this stable operation. A sudden afternoon downpour in the core area of ​​a city's main road fully exposed the shortcomings of the existing system: the rainfall came on rapidly, causing a sharp drop in visibility and flooding of roads in a short period of time, posing the first challenge to the system. As the rain intensified, various data acquisition devices were severely affected, with the visual capture device being the most significantly interfered with: raindrops formed a water film on the lens surface, the image was filled with dense rain lines, and water mist in the air further reduced the clarity, directly leading to a significant decline in the quality of core data acquisition. The originally clear outlines of vehicles became difficult to discern, and their movement trajectories became intermittent.

[0023] Since visual equipment is the core basis for the system to classify vehicle types, estimate speeds, and calculate queue lengths, a decline in data quality directly leads to biases in the system's perception of traffic conditions. The system's built-in image processing algorithm is only optimized for normal weather, and its performance drops sharply in extreme heavy rainfall, frequently misidentifying blurred images caused by rain as vehicles, or misidentifying multiple closely following vehicles as a single oversized vehicle. Simultaneously, due to the lack of clear visual features, the actual speed of vehicles cannot be accurately estimated, and queue length calculations also frequently show significant deviations. Meanwhile, millimeter-wave radar is also affected: raindrops in heavy rainfall scatter and attenuate radar waves, resulting in noisy and chaotic echo signals, generating false targets and missing real vehicles. Only underground induction coils, whose working principle is not directly affected by rain, can still provide relatively accurate vehicle counts. In this situation, the system cannot distinguish whether "abnormal data is caused by environmental interference or real traffic events," and cannot determine whether the detected anomalies are real road conditions or false data generated by sensors in harsh environments.

[0024] Deteriorating data quality and information uncertainty further exacerbate data conflicts from multiple sensor sources: for example, a vision device might report sparse traffic or even an empty road in a certain direction, while an induction coil might show continuous vehicle traffic and high lane occupancy; millimeter-wave radar might detect multiple irregular, fast-moving targets, but the vision device cannot provide corresponding visual evidence. This contradictory information flow paralyzes the traffic prediction mechanism that supports system decision-making—existing prediction models rely on high-quality, consistent multi-source data for training, but their analytical capabilities are significantly reduced when faced with highly uncertain and conflicting data, making it impossible to integrate fragmented and contradictory information to form an accurate judgment of the current road conditions.

[0025] Based on erroneous traffic condition assessments, the system output completely incorrect signal control strategies: if the system misjudges low traffic volume and high vehicle speed in a certain direction, it shortens the green light duration for that direction, failing to extend the green light to address the actual situation of slow-moving vehicles in the rain; if it misjudges rain noise as a large number of vehicles, it unnecessarily extends the green light, causing excessively long waiting times for vehicles in other directions. This erroneous adjustment not only failed to alleviate congestion but also rapidly exacerbated regional traffic pressure. The traffic flow, which had already slowed down in the rain, quickly formed long queues, and traffic efficiency plummeted. At the same time, abnormal data in the core area also rendered the regional "green wave" strategy ineffective, causing traffic lights at adjacent intersections to fail to synchronize with the affected area, ultimately leading to chaos in the traffic order of the entire road segment.

[0026] Meanwhile, due to congestion and low visibility, a minor rear-end collision occurred on the main road: two cars scraped each other due to obstructed vision and slippery road conditions. The accident itself was not serious, but the vehicles involved remained in the lane, further obstructing traffic. With the system already misjudging congestion due to environmental interference, this real-life incident amplified the complexity of the situation. At this point, the main processing unit in the control center received multiple alerts: reports of persistently high occupancy rates from induction coils, abnormally blurry images from vision devices, and reports of irregular targets from millimeter-wave radar. Manual monitoring also confirmed congestion and an accident based on limited information, but the system still could not effectively distinguish which data were distortions caused by heavy rainfall and which were signals of genuine congestion or accidents. The emergency response module, which originally relied on clear event identification and location, was overwhelmed by a large amount of contradictory and unreliable data. Although the system attempted to activate conventional diversion strategies to adjust the timing of surrounding intersections, these measures not only failed to address the root cause of the congestion due to misjudgments of the actual situation in the core congestion area, but also spread the congestion to surrounding areas. At the same time, the system could not accurately identify or prioritize the rear-end collision, as the accident vehicle might be obscured by rain or misjudged as other traffic anomalies. With multiple problems overlapping, the system was unable to generate an effective global coordination and traffic management plan, and the regional traffic situation deteriorated rapidly, evolving from localized congestion to widespread paralysis. The accuracy of road condition identification was low, and the reliability of signal control was also low.

[0027] The embodiments of this application will be explained in detail below with reference to the accompanying drawings: Figure 1 This is an optional flowchart of an intelligent traffic signal control method provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S104.

[0028] Step S101: Acquire environmental sensing data, which includes visual image data, millimeter-wave radar data, and induction coil sensing data. Step S102: Identify the target's rainfall interference status based on visual image data and millimeter-wave radar data; Step S103: Based on the sensing data of the induction coil, identify vehicle driving events, including vehicle stationary events, vehicle slow-moving events, and vehicle normal-moving events. Step S104: Adjust traffic signals based on the target rainfall interference status and vehicle driving events.

[0029] Steps S101 to S104 as shown in the embodiments of this application can combine visual image data, millimeter-wave radar data and induction coil sensing data to adjust traffic signals, thereby achieving traffic signal control and improving the accuracy of road condition recognition and the reliability of signal control.

[0030] In some embodiments, steps S101-S104 may involve acquiring environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data. Environmental sensing data can be acquired synchronously using multiple types of sensors deployed at the intersection. Specifically, during system initialization, the traffic signal control host establishes a connection with the high-definition camera via standard network communication protocols, such as Transmission Control Protocol (TCP) and Internet Protocol (IP), receiving compressed or raw visual image data at a rate of 25 frames per second. The visual image data is primarily collected by high-definition network cameras installed at the intersection. These devices are typically mounted on horizontal bars 5 to 6 meters above the ground, viewing the road surface at a 30 to 45-degree angle, and transmitting video streams of 1920×1080 resolution or higher in real time via Ethernet. Simultaneously, the host unit connects to the millimeter-wave radar via a serial communication interface or Ethernet, receiving radar data packets containing target distance, speed, angle, and signal-to-noise ratio. The millimeter-wave radar data is acquired by radar sensors operating in the 24 GHz or 77 GHz frequency band. These devices are mounted on gantry frames or dedicated poles, using electromagnetic waves to detect vehicles within a range of 50 to 200 meters, and output point cloud data or target tracking information via a controller area network bus or Ethernet. Furthermore, the host unit also collects real-time state change sequences of the induction coil via a digital input interface or a dedicated vehicle detector communication protocol, including precise timestamps of vehicle entry and exit from the coil. For acquiring the induction coil sensing data, the induction coil is made of polyethylene-insulated copper core wire with a diameter of 1.0 mm to 1.5 mm, and the coil inductance is approximately 50 microhenries to 200 microhenries. It is connected to the vehicle detector via a twisted-pair shielded cable. The detector contains an oscillation circuit with a frequency ranging from 20 kHz to 100 kHz. When a vehicle enters the coil area, the metal of the vehicle body causes a change in the coil inductance, resulting in a shift in the oscillation frequency. The detector converts this frequency change into a digital level signal through a frequency discriminator circuit. The system scans this signal at a sampling period of 1 to 10 milliseconds through a high-speed digital input port, recording the absolute timestamps of the rising edge (vehicle entry) and falling edge (vehicle departure), with a time accuracy of milliseconds, forming a time series of vehicle entry and exit from the coil. The induction coil sensing data comes from a ring-shaped induction coil buried 2 to 3 centimeters below the road surface. The coil is typically 2 meters long and 0.5 meters wide and is connected to a vehicle detector on the roadside via a feeder. When a metal vehicle body passes by, it causes a change in the coil inductance, and the detector outputs a vehicle presence signal, a passage pulse, or digitized traffic parameters based on this.

[0031] Then, based on visual image data and millimeter-wave radar data, the target's rain interference status is identified. A relatively direct feature threshold judgment method can be used. For example, the system can perform grayscale processing on the visual image data, calculate the Laplacian operator response variance of the entire image or a specific region of interest, and when the variance value is lower than a preset threshold (e.g., 50 to 100 for an 8-bit grayscale image), it is determined that the image has significant blurring, which may be caused by rain. At the same time, the system monitors the signal-to-noise ratio (SNR) parameter in the millimeter-wave radar data. When the average SNR is lower than a certain threshold (e.g., 10 dB) or a large number of low-speed, weakly reflective suspected raindrop targets are detected, it is determined that the radar is affected by rain interference.

[0032] Based on the induction coil sensing data, vehicle movement events are identified, including vehicle stationary events, slow-moving events, and normal-moving events. To more accurately identify real traffic conditions, the system also identifies vehicle movement events based on induction coil sensing data. The system calculates the ratio of the average time a coil is occupied by a vehicle to the average time it is idle per unit time (i.e., lane occupancy rate), and the time interval between vehicles passing adjacent coils to estimate speed. When a vehicle's dwell time on a coil exceeds twice the normal passage time (e.g., more than 5 seconds), and the interval between subsequent vehicles entering the coil significantly increases, it is initially identified as a vehicle stationary event; when the vehicle speed is significantly lower than the road speed limit (e.g., below 30 km / h) and the distance between vehicles is small, it is identified as a slow-moving event; otherwise, it is identified as normal-moving traffic. It is understandable that vehicle driving events describe the dynamic characteristics of traffic flow. In this embodiment, there are three types of states: vehicle stagnation events, which are when a vehicle stays above the induction coil for an abnormally long time, usually corresponding to traffic accidents or severe congestion; vehicle slow passage events, which are when a vehicle passes through an intersection at a speed far below the road's design speed, commonly seen on slippery roads in rainy weather or in minor congestion; and vehicle normal passage events, which are when a vehicle passes smoothly at or near the design speed.

[0033] Finally, traffic signals are adjusted based on the target rainfall interference status and vehicle movement events. Based on the above basic identification results, the system performs traffic signal adjustments. When it is determined that there is no rainfall interference and traffic is normal, the system executes the standard signal timing scheme; when it is determined that there is rainfall interference and slow vehicle passage is detected, the system appropriately extends the green light time to cope with vehicle start delays; when a vehicle stagnation event is detected, the system further extends the green light duration in that direction or triggers a special phase to clear the intersection.

[0034] Through the above technical solution, this embodiment realizes multi-source data fusion perception under severe weather conditions. It uses induction coil data to compensate for the performance degradation of vision and radar during rainfall, which to some extent alleviates signal control errors caused by sensor misjudgment and improves the accuracy of road condition recognition and the reliability of signal control.

[0035] In some embodiments, step S102, identifying the target rainfall interference state based on visual image data and millimeter-wave radar data, may include, but is not limited to, the following steps: The blur level of the visual image data is obtained by calculating the blur level using the Laplacian operator. Rain line features are identified by performing Hough transform on visual image data. The rain line features include rain line length and rain line angle. Based on millimeter-wave radar data, the signal scattering attenuation rate is calculated. The signal scattering attenuation rate is used to reflect the degree of signal strength attenuation caused by raindrops scattering the radar echo signal. The target rainfall interference status is identified based on image blur, rain line characteristics, and signal scattering attenuation rate.

[0036] In some embodiments, the blurriness of the visual image data can be calculated first using the Laplacian operator to obtain the image blurriness. The received color visual image can be converted into a grayscale image. A 3×3 Laplacian convolution kernel is applied for spatial filtering. The matrix form of this convolution kernel is: the center element is 8, the four neighboring elements are -1, and the four corner elements are 0; or a 5×5 kernel including the four neighboring diagonals is used, with the center element being 24 and the surrounding elements being -1. The system highlights the edge and detail information in the image by calculating the second derivative of the difference between the grayscale values ​​of each pixel in the image and its eight neighboring pixels. For each pixel in the image, its Laplacian response value is the grayscale value of that point multiplied by 8, and then subtracted from the sum of the grayscale values ​​of its four neighboring pixels. After the calculation is completed, the system uses the sum of the grayscale variance or absolute values ​​of the filtered image as a quantitative indicator of the image blurriness. Specifically, the system calculates the variance of the Laplacian response values ​​of all pixels. Under clear weather conditions, sharp image edges produce a strong Laplacian response with a high variance, typically greater than 150. However, under rainy conditions, due to light scattering by water droplets and defocusing caused by lens glare, image edges become smoother, the Laplacian response weakens, and the variance drops below 100. By setting an adaptive threshold (e.g., based on the mean of historical clear weather data minus three standard deviations, or a fixed threshold of 80), the system can accurately determine whether an image is blurred due to rain.

[0037] Then, rain line feature recognition is performed on the visual image data using Hough transform to obtain rain line features, including rain line length and angle. During implementation, the image can be preprocessed, including noise removal using a 5×5 pixel Gaussian filter kernel with a standard deviation of 1.0, and edge information extraction using the Canny edge detection algorithm, with high and low thresholds set to 50 and 150 respectively. Since raindrops appear as approximately straight lines in the image during their fall, probabilistic Hough transform can be used to detect straight line segments in the image, with parameters set as follows: accumulator threshold of 50, minimum line length of 30 pixels, and maximum gap of 10 pixels. By setting angle constraints (e.g., only detecting lines with an angle within ±15 degrees of the vertical direction, i.e., the near-vertical angle typically observed in rain lines, ranging from 75 to 105 degrees) and length constraints (e.g., length between 30 and 150 pixels), the system filters out features suspected to be rain lines. By statistically analyzing the number and average length of rain lines per unit area, we can obtain quantitative parameters for rain line density (lines per square kilometer) and average length (pixels). These parameters directly reflect the intensity of rainfall. For example, when the rain line density exceeds 10 lines per square kilometer and the average length exceeds 50 pixels, it is considered significant rainfall interference.

[0038] Based on millimeter-wave radar data, the signal scattering attenuation rate is calculated. This rate reflects the degree of signal strength attenuation caused by raindrop scattering of the radar echo signal. In practice, the system records the radar's transmit power (in dBmW) and receive power (in dBmW) during transmission and reception, respectively. By comparing the current received power with a clear-sky reference power, the system subtracts the current received power from the clear-sky reference power and divides by the clear-sky reference power to obtain the signal scattering attenuation rate. For example, under clear-sky conditions, the system records the received power at a standard reflector (radar cross-section of 10 square meters) at a distance of 100 meters as -40 dBmW as a reference. During rainfall, the received power at the same distance and with the same reflector drops to -55 dBmW, resulting in a calculated signal scattering attenuation rate of 37.5%. When the attenuation rate exceeds 30%, it indicates strong rainfall interference along the radar beam path. By analyzing the attenuation distribution at different range gates (e.g., every 10 meters), the spatial distribution characteristics of the rainfall can also be determined.

[0039] Finally, the target's rainfall interference status is identified based on image blur, rain line characteristics, and signal scattering attenuation rate. In a specific embodiment, a three-dimensional feature space can be constructed, with the three dimensions being normalized image blur (variance value divided by the clear-day baseline variance, ranging from 0 to 1), rain line density (normalized to 0 to 1), and radar signal attenuation rate (normalized to 0 to 1). The system uses a preset decision tree logic to map the current feature vector to different interference levels. The preset rules of the decision tree are as follows: if all three features are less than 0.3, it is determined to be no rainfall interference; if the image blur is greater than 0.7 and the rain line density is greater than 0.6, but the radar attenuation is less than 0.4, it is determined to be mild visual interference (possibly just lens water film); if the radar attenuation is greater than 0.8 and the image quality is severely degraded (blurriness greater than 0.8), it is determined to be severe comprehensive interference. Through cross-validation of multi-dimensional features, misjudgment by a single sensor is effectively avoided. For example, if the lens is obscured by dirt, it will only cause image blur but will not cause radar attenuation, thus being identified by the system as non-rainfall interference.

[0040] Through the above technical solution, this embodiment, based on the fusion of Laplace operator, Hough transform, and radar scattering attenuation rate, can accurately characterize the impact of rainfall on sensors from three complementary dimensions: image clarity, rain line morphology, and electromagnetic wave propagation characteristics. This multimodal fusion mechanism not only improves the accuracy of rainfall recognition but also distinguishes different types of interference sources, providing a reliable environmental perception basis for subsequent traffic signal adjustments, significantly outperforming single-threshold judgment methods.

[0041] In some embodiments, step S103, identifying vehicle driving events based on induction coil sensing data, may include, but is not limited to, the following steps: Step S201: Calculate vehicle driving characteristics based on the sensing data from the induction coil. Vehicle driving characteristics include lane occupancy rate, vehicle speed, and vehicle frequency. Step S202: Based on the vehicle driving characteristics and the preset event feature pattern, calculate the feature matching degree using the cosine similarity algorithm. The preset event feature pattern is used to reflect the driving characteristics under different vehicle driving events. Step S203: Identify vehicle driving events based on feature matching degree and preset matching degree threshold.

[0042] In some embodiments, vehicle driving characteristics, including lane occupancy, vehicle speed, and vehicle frequency, can be calculated first based on induction coil sensing data. Simultaneously, a cosine similarity algorithm is used to calculate the feature matching degree based on the vehicle driving characteristics and preset event feature patterns. The preset event feature patterns reflect the driving characteristics under different vehicle driving events. Vehicle driving events are then identified based on the feature matching degree and a preset matching degree threshold. In practical applications, the preset event feature patterns are standardized feature descriptions established based on statistical analysis of historical traffic data. To construct these patterns, the system needs to collect a large amount of sample data under typical operating conditions. For example, under clear weather conditions, the system records induction coil data under three typical states: For vehicle stagnation events, at least 50 long-term stagnation samples caused by traffic accidents or severe congestion are collected, and the lane occupancy rate (usually greater than 80%), vehicle speed (close to 0 km / h), and vehicle frequency (less than 5 vehicles per minute) of each sample are calculated. The average of each dimension is taken to form a stagnation mode vector, denoted as vector A (0.9, 0.0, 0.1); For vehicle slow-moving events, samples of slow-moving queues at intersections are collected, and the occupancy rate is calculated to be moderate (40% to 70%), the speed to be low (10 to 20 km / h), and the frequency to be moderate (10 to 20 vehicles per minute), forming a slow-moving mode vector B (0.55, 0.15, 0.25); For normal traffic events, data under smooth traffic conditions are collected, with a low occupancy rate (less than 40%), a high speed (30 to 50 km / h), and a high frequency (greater than 25 vehicles per minute), forming a normal mode vector C (0.2, 0.8, 0.8). Each vector element is normalized, and its numerical range is mapped to the interval between 0 and 1.

[0043] In practical implementation, the three dimensions of driving characteristics mentioned above can be extracted from the induction coil sensing data. Using a 30-second time window, the lane occupancy rate, average passing speed, and passing frequency within that window are calculated to construct a three-dimensional feature vector. The cosine similarity between this real-time feature vector and three preset event feature pattern vectors is then calculated. The cosine similarity formula is the dot product of the two vectors divided by the product of their magnitudes; the result ranges from -1 to 1, with values ​​closer to 1 indicating higher similarity. The specific calculation process is as follows: Let the real-time feature vector be V, and the stationary pattern vector be A. Then, the similarity S is equal to the dot product of vector V and vector A divided by the magnitude of vector V multiplied by the magnitude of vector A. For example, if the real-time feature vector is calculated to be (0.85, 0.05, 0.12), its cosine similarity with the stationary event pattern vector (0.9, 0.0, 0.1) is 0.92; its similarity with the slow-moving event pattern vector is 0.45; and its similarity with the normal-moving event pattern vector is 0.12, then the current event is determined to be a vehicle stationary event. A preset matching threshold can be set, typically between 0.70 and 0.85. Only when the highest similarity exceeds this threshold is the recognition result confirmed; otherwise, it is marked as unrecognized or in a transitional state.

[0044] Through the above technical solution, this embodiment, based on multi-dimensional features and cosine similarity algorithm, can comprehensively consider the relationship between lane occupancy rate, speed and frequency, rather than relying on a single threshold, thereby effectively distinguishing abnormal events in mixed traffic flows composed of different vehicle types, significantly improving the accuracy of vehicle driving event identification and adaptability to complex traffic scenarios.

[0045] In some embodiments, in step S201, calculating the vehicle driving characteristics based on the sensing data from the induction coil may include, but is not limited to, the following steps: Vehicle pass-through count, vehicle entry time series, and vehicle exit time series are extracted from the induction coil sensing data. The vehicle entry time series includes the time when each vehicle enters the induction coil, the vehicle exit time series includes the time when each vehicle leaves the induction coil, and the vehicle pass-through count is used to represent the number of vehicles leaving the induction coil. Calculate the average dwell time of vehicles based on the vehicle entry time series and vehicle exit time series; Based on the time series of vehicles entering and leaving the induction coil, the average idle time of the induction coil is calculated. The average idle time of the induction coil is used to represent the average time during which no vehicle stays on the induction coil. Lane occupancy rate is calculated based on average vehicle dwell time and average idle time of induction coils; Calculate the first time interval between vehicles passing through adjacent induction coils based on the vehicle's entry time sequence; The vehicle speed is calculated based on the distance between adjacent induction coils and the first time interval. Calculate the vehicle passing frequency based on the number of vehicles passing through.

[0046] In some embodiments, vehicle pass-through counts, vehicle entry time series, and vehicle exit time series can be extracted from the induction coil sensing data. The vehicle entry time series includes the time each vehicle enters the induction coil, and the vehicle exit time series includes the time each vehicle leaves the induction coil. The vehicle pass-through count indicates the number of vehicles leaving the induction coil. In practical applications, the acquisition of the vehicle entry and exit time series depends on the hardware deployment and signal acquisition mechanism of the induction coil. For example, the detector of the induction coil contains an oscillation circuit with a frequency of 20 kHz to 100 kHz. When a vehicle enters the coil area, the metal of the vehicle body causes a change in the coil inductance, resulting in a shift in the oscillation frequency. The detector converts this frequency change into a digital level signal (high level indicates a vehicle is present, low level indicates no vehicle is present) through a frequency discrimination circuit. The system scans this signal through a high-speed digital input port with a sampling period of 1 millisecond to 10 milliseconds, recording the absolute timestamps of the rising edge (vehicle entry) and falling edge (vehicle exit), achieving millisecond-level time accuracy, thus forming the vehicle entry and exit time series.

[0047] Then, based on the vehicle entry and exit time sequences, the average vehicle dwell time is calculated. When calculating the average dwell time, the system iterates through the vehicle exit time sequence. For each exit timestamp, it searches for the corresponding entry timestamp in the vehicle entry time sequence (usually using a first-in, first-out pairing principle, i.e., the earliest entering vehicle corresponds to the earliest exiting vehicle). The difference is calculated to obtain the dwell time per vehicle, and the arithmetic mean of all successfully paired dwell times is taken. For example, if 10 vehicles pass through within a 30-second statistical window, with dwell times of 1.2 seconds, 1.5 seconds, 8.0 seconds (stalled), and 1.3 seconds, the average dwell time is the sum of these values ​​divided by 10.

[0048] Based on the vehicle entry and exit time sequences, the average idle time of the induction coil is calculated. This average idle time represents the average time no vehicle remains on the induction coil. When calculating the average idle time, the system analyzes the interval between the vehicle exit time sequence and the next entry time stamp. Specifically, for the exit time stamp of the i-th vehicle and the entry time stamp of the (i+1)-th vehicle, the difference is calculated to obtain the individual idle time. Averaging these values ​​over all interval events yields the average idle time of the induction coil.

[0049] Lane occupancy rate is calculated based on the average vehicle dwell time and the average idle time of the induction coil. The lane occupancy rate can be obtained by dividing the average vehicle dwell time by the sum of the average vehicle dwell time and the average idle time of the induction coil. For example, if the average dwell time is 2.5 seconds and the average idle time is 1.5 seconds, then the lane occupancy rate is 2.5 / 4.0 = 62.5%.

[0050] Based on the vehicle's entry time sequence, the first time interval between the vehicle's passage through adjacent induction coils is calculated. Then, the vehicle's speed is calculated based on the distance between the adjacent induction coils and the first time interval. To calculate the vehicle's speed, the system utilizes two induction coils buried at a certain distance between them in the same lane, typically 3 to 5 meters. Based on the vehicle's entry time sequence, the time it takes to pass through the first coil and the time it takes to pass through the second coil are recorded and subtracted to obtain the first time interval. If the distance between adjacent induction coils is known (e.g., 4.0 meters), the distance between the adjacent induction coils can be divided by the first time interval to obtain the vehicle's speed.

[0051] Finally, the vehicle passing frequency is calculated based on the vehicle passing count. This can be obtained by dividing the vehicle passing count by the statistical period length. For example, if 15 vehicles pass through in 60 seconds, the passing frequency is 0.25 vehicles per second or 15 vehicles per minute.

[0052] Through the above technical solution, this embodiment can accurately extract multi-dimensional driving features with physical meaning from the original induction coil timestamp data, providing a reliable data foundation for subsequent pattern matching and event recognition, and ensuring the accuracy and consistency of feature calculation.

[0053] In some embodiments, step S103, identifying vehicle driving events based on induction coil sensing data, may include, but is not limited to, the following steps: Extract the current vehicle dwell time and vehicle entry time sequence from the induction coil sensing data. The current vehicle dwell time is used to represent the duration during which the vehicle enters the induction coil and does not leave. Calculate the second time interval between adjacent vehicles entering the same induction coil based on the vehicle entry time sequence; If the current vehicle dwell time is greater than the preset stagnation time threshold, then the vehicle driving event is determined to be a vehicle stagnation event. If the current vehicle dwell time is less than the preset stagnation time threshold, then determine whether the second time interval is greater than the preset slow passage time threshold. If the second time interval is greater than the preset slow passage time threshold, the vehicle driving event is determined to be a slow passage event; otherwise, the vehicle driving event is determined to be a normal passage event.

[0054] In some embodiments, the current vehicle dwell time and vehicle entry time sequence can be extracted from the induction coil sensing data. The current vehicle dwell time represents the duration during which a vehicle enters the induction coil and remains there. The current vehicle dwell time is a dynamic parameter calculated in real time by monitoring the difference between the timestamp of the latest vehicle entering the coil and the current system time.

[0055] Then, based on the time sequence of vehicles entering the coil, a second time interval is calculated for adjacent vehicles entering the same induction coil. This second time interval is obtained by comparing the timestamp difference between the two most recent adjacent vehicles entering the coil, reflecting the time distribution characteristics of vehicle arrivals.

[0056] The system then determines the vehicle movement event based on the current vehicle dwell time and the second time interval. If the current vehicle dwell time is greater than a preset stagnation time threshold, the vehicle movement event is determined to be a vehicle stagnation event. If the current vehicle dwell time is less than the preset stagnation time threshold, the system checks if the second time interval is greater than a preset slow passage time threshold. If the second time interval is greater than the preset slow passage time threshold, the vehicle movement event is determined to be a vehicle slow passage event. Otherwise, the vehicle movement event is determined to be a normal passage event. In specific implementation, the system adopts a hierarchical judgment logic: First, it checks if the current vehicle dwell time is greater than a preset stagnation time threshold (e.g., 8 seconds). When the system detects that a vehicle has entered the loop and has not left for 8 seconds, it is immediately determined to be a vehicle stagnation event, regardless of whether subsequent vehicles arrive. If the current vehicle dwell time is less than this threshold, the system further checks if the second time interval is greater than a preset slow passage time threshold (e.g., 4 seconds). For example, if the second time interval is 6 seconds (greater than the threshold of 4 seconds), it indicates that vehicles are arriving sparsely. Even if there are no vehicles currently stopped, it is still considered a slow traffic event, which may correspond to a decrease in vehicle arrival rate due to congestion ahead. If the second time interval is 1.5 seconds (less than the threshold), it is considered a normal traffic event.

[0057] In practical applications, the determination of the aforementioned preset stall time threshold and preset slow passage time threshold needs to be combined with the specific road design standards and historical traffic flow characteristics. For example, the preset stall time threshold can be obtained as follows: collect all induction loop records from the intersection over the past month, filter out time periods confirmed to have resulted in traffic accidents or vehicle malfunctions (confirmed through traffic police reports or manual annotation via video surveillance), extract the dwell time data of vehicles on the loops during these time periods, and take the 95th percentile as the threshold (e.g., 8.5 seconds) to ensure that the vast majority of genuine stall events are captured, while allowing a small number of stalls caused by yielding to pedestrians or brief waiting to not trigger stall event determination. The preset slow passage time threshold reflects traffic density and can be obtained by analyzing the vehicle arrival intervals under historical smooth traffic conditions, taking the 85th percentile plus a margin of 1 to 2 seconds (e.g., 4.5 seconds). When the actual interval exceeds this value, it indicates a significant decrease in arrival rate, potentially indicating congestion ahead.

[0058] Through the above technical solution, this embodiment can achieve event recognition through simple comparison operations, with a computational complexity of constant order. It is suitable for signal control scenarios with extremely high real-time requirements, and can effectively distinguish between three states: stagnation, slow passage, and normal passage, providing the system with a lightweight event recognition method.

[0059] In some embodiments, step S102, identifying the target rainfall interference state based on visual image data and millimeter-wave radar data, may include, but is not limited to, the following steps: Step S301: Obtain a reference road image, which is used to represent a road image under no rainfall conditions; Step S302: Identify the first rainfall disturbance state based on the reference road image and visual image data; Step S303: Identify the second rainfall interference state based on millimeter-wave radar data; Step S304: Generate the target rainfall interference state based on the first rainfall interference state and the second rainfall interference state.

[0060] In some embodiments, a reference road image can be acquired first, representing a road image under conditions of no rainfall. In a specific implementation, the system selects a time of clear weather, good lighting conditions (e.g., midday, with light intensity between 30,000 and 80,000 lux), and dry road surface, and uses the same high-definition camera to capture a high-definition image of the intersection as a reference. During acquisition, the camera's exposure time is set to 1 to 5 milliseconds, the aperture to F8 to F11, and the ISO to 100 to ensure image sharpness and depth of field. This image should clearly show static features such as lane lines, stop lines, road markings, and surrounding buildings, and is stored in the system's non-volatile memory in a lossless compressed bitmap or raw sensor data format. The system periodically (e.g., weekly or monthly) updates the reference image under the same lighting conditions (monitored by a light sensor to ensure light intensity differences are less than 10%) to eliminate the effects of seasonal changes in shadow angles and vegetation variations.

[0061] Then, based on the baseline road image and visual image data, the first rainfall interference state is identified. For example, fog droplet density data and fog droplet size data can be used to predict the expected blurred image and compare it with the current road image to identify the first rainfall interference state. Simultaneously, based on millimeter-wave radar data, the second rainfall interference state is identified. For example, electromagnetic wave attenuation data can be used to generate expected radar noise data and compare it with millimeter-wave radar data to identify the second rainfall interference state.

[0062] Finally, the target rainfall interference state is generated based on the first and second rainfall interference states. In one embodiment, a logical AND operation can be used, meaning that the target rainfall interference state is only determined to exist when both visual comparison confirms interference (the first state is true, i.e., the structural similarity index is greater than 0.75 and less than 0.5) and radar comparison also confirms interference (the second state is true, i.e., the noise level match is within ±3 dB and exceeds the baseline by 10 dB). This conservative strategy avoids misjudgment by a single sensor. In another embodiment, a weighted fusion can also be used, where a more refined interference level classification is obtained by weighting the confidence levels of the two states (such as comparison similarity or match). The weight coefficients can be calibrated, for example, the visual confidence weight is 0.6 and the radar confidence weight is 0.4.

[0063] Through the above technical solution, this embodiment achieves the identification of the physical nature of rainfall interference. It can not only determine whether there is rainfall, but also predict the specific impact of rainfall on sensor data, providing a more reliable environmental perception input for the precise adjustment of traffic signals, and significantly improving the robustness and decision-making accuracy of the system under severe weather conditions.

[0064] In some embodiments, step S302, identifying the first rainfall disturbance state based on the reference road image and visual image data, may include, but is not limited to, the following steps: Droplet density and droplet size data are obtained using a laser scattering droplet density sensor. Based on droplet density and droplet size data, the extinction coefficient is calculated. The extinction coefficient is used to represent the degree to which light is weakened due to absorption and scattering when it propagates in a medium. Based on the extinction coefficient, predict the image parameter offset, which includes the degree of contrast reduction and color shift; By applying the image parameter offset to the baseline road image, the desired blurred image is obtained; Extract the current road image from visual image data; The expected blurred image is compared with the current road image to identify the first rainfall interference state.

[0065] In some embodiments, droplet density and droplet size data can be acquired first using a laser scattering droplet density sensor. The laser scattering droplet density sensor is typically mounted on a monitoring pole at an intersection, 3 to 5 meters above the ground. It operates in forward scattering mode, emitting a laser beam of a specific wavelength (such as 650 nm red light or 905 nm near-infrared light) with a laser power of 1 to 10 milliwatts. When the laser passes through a rainy / foggy area, raindrops scatter the light, and the sensor receives the scattered light signal at its receiving end (at an angle of 30 to 45 degrees with the transmitting end). The signal processing unit inside the sensor performs inversion based on Mie scattering theory, measuring the intensity of the scattered light. With incident light intensity The ratio, combined with the calibration coefficient, yields the droplet number density per unit volume. (Unit: droplets per cubic centimeter) and the average equivalent sphere diameter D of the droplets (unit: micrometers). The specific calculation formula is as follows: ,in This is the instrument calibration constant, obtained through calibration using a standard particle generator.

[0066] Then, based on the droplet density and droplet size data, the extinction coefficient is calculated. The extinction coefficient represents the degree to which light is attenuated due to absorption and scattering as it propagates through a medium. During implementation, the system calculates the extinction cross-section of a single droplet based on Mie scattering theory. The calculation formula is: in The extinction efficiency factor is determined based on the Mie parameter (e.g., ...). * The wavelength (of the laser) is obtained by referring to a table or through numerical calculation. Then, the extinction section... Multiply by the droplet density The extinction coefficient is obtained by integrating the droplet size distribution. (Unit: per kilometer). This extinction coefficient quantitatively describes the ability of light to attenuate in the rainy atmosphere. The higher the extinction coefficient, the lower the visibility and the more severe the image degradation. For example, when the extinction coefficient reaches 1.0 per kilometer, it corresponds to moderate rain conditions with a visibility of approximately 3 kilometers.

[0067] Then, based on the extinction coefficient, the image parameter shift is predicted. This shift includes both the degree of contrast reduction and color shift. According to atmospheric optics principles, the decrease in image contrast follows an exponential decay law. This can be determined based on the observation distance between the camera and the observation area (e.g., the opposite stop line). (Typically 20 to 50 meters), calculate the expected decrease in contrast using the formula: ,in For observation distance, This represents the degree of contrast reduction. Simultaneously, since the extinction coefficients of different wavelengths of light differ slightly in rain and fog (Rayleigh scattering is inversely proportional to the fourth power of wavelength, while Mie scattering has a more complex relationship with wavelength), the system also predicts color shift, particularly the relative intensity changes between the red and blue channels. Specifically, the extinction coefficient of blue light (450 nm) is slightly higher than that of red light (650 nm), and the system calculates the color shift based on this difference in extinction coefficients.

[0068] Image parameter offsets are applied to a baseline road image to obtain the desired blurred image. An image degradation model can be used to apply contrast reduction (multiplied by a contrast reduction ratio), gamma correction adjustment (gamma value adjusted to 1.2 to 1.5 to simulate increased brightness in foggy weather), and blur convolution based on a point spread function to the baseline image, generating a simulated image that should be observed under current rainfall conditions. The point spread function uses a Gaussian function whose standard deviation is proportional to the extinction coefficient, with a scaling factor ranging from 0.5 to 2.0 pixels per extinction coefficient. A Gaussian kernel is used to simulate blurring caused by scattering.

[0069] Finally, the current road image is extracted from the visual image data. The expected blurred image is then compared with the current road image to identify the first rainfall interference state. A structural similarity index algorithm can be used to calculate the similarity between the two images. The structural similarity index compares images from three dimensions: brightness, contrast, and structure, with a value ranging from 0 to 1. If the real-time image is highly similar to the expected blurred image (structural similarity index exceeding 0.75) but significantly different from the original baseline image (structural similarity index below 0.5), then the first rainfall interference state is confirmed as genuine, and the degree of interference matches the prediction model. If the real-time image is more blurred than expected or has local differences, it may be accompanied by other interference factors (such as lens contamination). This method predicts image degradation through a physical model, achieving an objective quantitative assessment of rainfall interference.

[0070] By combining the above technical solution with fog droplet density and size data, this embodiment calculates the extinction coefficient, predicts the expected blurred image, and compares it with the current road image to identify the first rainfall interference state. This embodiment achieves rainfall interference state identification from a visual sensor perspective, providing a more reliable environmental perception input for precise adjustment of traffic signals.

[0071] In some embodiments, step S303, identifying the second rainfall interference state based on millimeter-wave radar data, may include, but is not limited to, the following steps: Electromagnetic wave attenuation data is obtained by a multi-band electromagnetic wave attenuation meter. The electromagnetic wave attenuation data is used to reflect the degree of attenuation of electromagnetic waves after absorption and scattering by fog droplets. Based on electromagnetic wave attenuation data, predict background noise information formed by fog droplet absorption and scattering; Calculate the average signal attenuation intensity based on electromagnetic wave attenuation data; Based on background noise information and average signal attenuation intensity, generate expected radar noise data; The expected radar noise data is compared with millimeter-wave radar data to identify the second rainfall interference state.

[0072] In some embodiments, electromagnetic wave attenuation data can be acquired first using a multi-band electromagnetic wave attenuation meter. This data reflects the degree of attenuation of electromagnetic waves after absorption and scattering by fog droplets. For millimeter-wave radar data processing, electromagnetic wave attenuation data can also be acquired using a multi-band electromagnetic wave attenuation meter, reflecting the degree of attenuation of electromagnetic waves after absorption and scattering by fog droplets. This meter is integrated with traffic radar or installed independently, simultaneously emitting electromagnetic waves of at least two different frequencies, such as 24 GHz (K-band) and 77 GHz (E-band). Since the scattering cross-sections of electromagnetic waves of different frequencies interacting with raindrops are different, the 77 GHz band is more sensitive to small raindrops, while the 24 GHz band is more sensitive to large raindrops. By comparing the attenuation differences between the two bands, the size distribution of raindrops can be inverted. The instrument directly outputs the total attenuation value along the path (in decibels per kilometer), obtained by measuring the difference between the transmitted and received power and subtracting path loss.

[0073] Then, based on electromagnetic wave attenuation data, the background noise information formed by fog droplet absorption and scattering is predicted. In radar meteorology, backscattering from rainfall forms rain clutter, the power spectral density of which is related to rainfall intensity. The system uses the ZR relationship (the relationship between reflectivity factor Z and rainfall rate R) based on measured attenuation data. ,in and This is an empirical coefficient, usually Take 200, Using a specific rain clutter model (e.g., 1.6), predict the background noise level that should appear in the radar receiver, including the power spectrum distribution and time-domain statistical characteristics of the noise.

[0074] Then, based on the electromagnetic wave attenuation data, the average signal attenuation intensity is calculated. This can be achieved by weighted averaging of measurements from multiple frequency bands or by selecting the measurement value from the band closest to the traffic radar's operating frequency, thus obtaining the average attenuation intensity characterizing the overall attenuation capability of current rainfall on radar waves. For example, if the traffic radar operates at 77 GHz, the attenuation measurement value from the 77 GHz band is primarily used, with the 24 GHz band used as auxiliary verification.

[0075] Based on background noise information and average signal attenuation intensity, expected radar noise data is generated. Existing radar noise models can be used to generate simulated radar echo data, which includes the expected thermal noise level (-174 dBmW / Hz plus noise figure plus 10 times logarithmic bandwidth) and additional noise fluctuations caused by rainfall (calculated based on rain clutter power).

[0076] Finally, the expected radar noise data is compared with millimeter-wave radar data to identify the second rainfall interference state. The differences between the actual radar data's noise floor, false alarm rate, or background power spectrum and the expected radar noise data can be compared. If the actual noise level matches the predicted value (within ±3 dB) and is significantly higher than the clear-day baseline (more than 10 dB), the second rainfall interference state is confirmed as genuine. Based on the physical principles of multi-band measurements, this method can distinguish rainfall interference from other electromagnetic interference, such as narrowband interference from surrounding equipment.

[0077] Through the above technical solution, this embodiment combines background noise information and average signal attenuation intensity to generate expected radar noise data, which is then compared with millimeter-wave radar data to identify the second rainfall interference state. This embodiment achieves rainfall interference state identification from the perspective of millimeter-wave radar sensors, providing a more reliable environmental perception input for the precise adjustment of traffic signals.

[0078] In some embodiments, step S104, adjusting the traffic signal based on the target rainfall disturbance state and vehicle driving events, may include, but is not limited to, the following steps: If the target rainfall interference status is no rainfall interference, then reduce the green light passage time for all lane directions to shorten the traffic light cycle and thus reduce red light waiting time; If the target rainfall interference state is rainfall interference, and the vehicle driving event is a slow-moving vehicle event or a normal-moving vehicle event, then the green light time for all lane directions will be increased to reduce the decision-making pressure on drivers in low visibility conditions. If the target rainfall interference status is "rain interference" and the vehicle driving event is a vehicle stoppage event, then the green light time in the direction of the lane where the vehicle stoppage event occurs will be increased to reduce congestion caused by vehicle accidents.

[0079] In some embodiments, the target rainfall interference state and vehicle driving events can be determined first. If the target rainfall interference state is without rainfall interference, the green light duration for all lane directions is reduced to shorten the traffic light cycle and thus reduce red light waiting time. For cases without rainfall interference, the system implements an optimized traffic efficiency strategy. For example, if the original standard green light duration is 30 seconds, the system shortens it to 25 seconds, while simultaneously reducing the signal cycle from 120 seconds to 100 seconds, resulting in a corresponding reduction of 5 to 10 seconds in red light waiting time for each direction, improving the overall intersection turnaround rate. In practical applications, the above signal adjustment strategy is implemented through the phase control module inside the signal controller. This module stores a preset timing scheme library and calls the corresponding timing parameters for different combinations of rainfall interference states and vehicle driving events. In specific implementation, the signal controller determines the green light duration for each phase by looking up a preset decision mapping table based on real-time identification results.

[0080] Then, the system combines the target rainfall interference status and vehicle driving events for judgment. If the target rainfall interference status is rainfall interference, and the vehicle driving events are slow-moving or normal-moving events, the green light duration for all lanes is increased to reduce the decision-making pressure on drivers in low visibility conditions. For situations with rainfall interference and slow-moving or normal-moving events, the system implements a safety-first strategy. Because the road surface adhesion coefficient decreases in rainy weather (from 0.7-0.8 in dry conditions to 0.3-0.4 in wet conditions), vehicle braking distance increases significantly, requiring drivers to have longer reaction and decision-making time. In this case, the system extends the green light duration from the standard 30 seconds to 35-40 seconds to ensure vehicles have sufficient time to pass through intersections in low visibility conditions, reducing the risk of rear-end collisions due to poor visibility and insufficient braking.

[0081] If the target rainfall interference status is "rain interference" and the vehicle movement event is a vehicle stagnation event, the green light time for the lane where the stagnation event occurs will be increased to reduce congestion caused by vehicle accidents. For situations with rainfall interference and detected vehicle stagnation events, the system will implement a congestion mitigation priority strategy. In this case, not only will the green light time be extended, but a targeted extension will also be implemented, meaning the green light time will only be extended (e.g., to 45 to 60 seconds) for the lane where the stagnation event occurred (e.g., east-west straight lanes), while other directions will maintain normal or slightly shortened, to quickly clear the queue of vehicles behind the accident vehicles and prevent congestion from spreading. Simultaneously, the system can coordinate with upstream intersection signals, sending coordination commands via wired or wireless communication networks (e.g., fiber optic lines or 4G / 5G private networks) to implement traffic diversion or diversion measures, prioritizing the right-of-way for the direction of the accident.

[0082] Through the above technical solution, this embodiment's signal adjustment strategy based on the coupled analysis of environmental conditions and traffic events can dynamically optimize signal timing according to different weather conditions and traffic situations. It pursues efficiency when there is no rain, takes safety into account when it is rainy, and prioritizes traffic flow during accidents, thus realizing the refinement and intelligence of traffic signal control and effectively mitigating the impact of severe weather and emergencies on traffic flow.

[0083] The beneficial effects of implementing the embodiments of the present invention include: the embodiments of this application first acquire environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data; then, based on the visual image data and millimeter-wave radar data, the target rainfall interference state is identified; then, based on the induction coil sensing data, vehicle driving events are identified; finally, based on the target rainfall interference state and vehicle driving events, traffic signals are adjusted. Thus, it is possible to combine visual image data, millimeter-wave radar data, and induction coil sensing data to adjust traffic signals, thereby achieving traffic signal control and improving the accuracy of road condition identification and the reliability of signal control.

[0084] like Figure 2 As shown, this embodiment of the invention also provides an intelligent traffic signal control system, comprising: The data acquisition module 401 is used to acquire environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data. Rainfall interference status identification module 402 is used to identify the target rainfall interference status based on visual image data and millimeter-wave radar data; The vehicle driving event recognition module 403 is used to recognize vehicle driving events based on the sensing data of the induction coil. Vehicle driving events include vehicle stationary events, vehicle slow-moving events, and vehicle normal-moving events. The traffic signal adjustment module 404 is used to adjust traffic signals based on the target rainfall interference state and vehicle driving events.

[0085] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0086] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

Claims

1. A method for intelligent traffic signal control, characterized in that, Includes the following steps: Acquire environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data; Based on the visual image data and the millimeter-wave radar data, identify the target's rainfall interference status; Based on the sensing data from the induction coil, vehicle driving events are identified, including vehicle stationary events, vehicle slow-moving events, and vehicle normal-moving events. Traffic signals are adjusted based on the target rainfall disturbance status and the vehicle driving events.

2. The method according to claim 1, characterized in that, The step of identifying the target's rainfall interference status based on the visual image data and the millimeter-wave radar data includes: The blurriness of the visual image data is calculated by using the Laplacian operator to obtain the image blurriness. Rain line features are obtained by performing Hough transform on the visual image data, and the rain line features include rain line length and rain line angle. Based on the millimeter-wave radar data, the signal scattering attenuation rate is calculated. The signal scattering attenuation rate is used to reflect the degree of signal strength attenuation caused by raindrops scattering the radar echo signal. The target rainfall interference state is identified based on the image blur, the rain line features, and the signal scattering attenuation rate.

3. The method according to claim 1, characterized in that, The step of identifying vehicle driving events based on the sensing data from the induction coil includes: Based on the sensing data from the induction coil, vehicle driving characteristics are calculated, including lane occupancy rate, vehicle speed, and vehicle frequency. Based on the vehicle driving characteristics and the preset event feature patterns, the feature matching degree is calculated using the cosine similarity algorithm. The preset event feature patterns are used to reflect the driving characteristics under different vehicle driving events. The vehicle driving event is identified based on the feature matching degree and the preset matching degree threshold.

4. The method according to claim 3, characterized in that, The step of calculating vehicle driving characteristics based on the sensing data from the induction coil includes: Vehicle pass-through count, vehicle entry time series, and vehicle exit time series are extracted from the sensing data of the induction coil. The vehicle entry time series includes the time when each vehicle enters the induction coil, the vehicle exit time series includes the time when each vehicle leaves the induction coil, and the vehicle pass-through count is used to represent the number of vehicles leaving the induction coil. Calculate the average dwell time of the vehicle based on the vehicle entry time sequence and the vehicle exit time sequence; The average idle time of the induction coil is calculated based on the vehicle entry time sequence and the vehicle exit time sequence. The average idle time of the induction coil is used to represent the average time during which no vehicle resides on the induction coil. The lane occupancy rate is calculated based on the average vehicle dwell time and the average idle time of the induction coil. Based on the vehicle's entry time sequence into the coil, calculate the first time interval between the vehicle passing through adjacent induction coils; The vehicle's passing speed is calculated based on the distance between adjacent induction coils and the first time interval; The vehicle passage frequency is calculated based on the vehicle passage count.

5. The method according to claim 1, characterized in that, The step of identifying vehicle driving events based on the sensing data from the induction coil includes: The current vehicle dwell time and vehicle entry time sequence are extracted from the sensing data of the induction coil. The current vehicle dwell time is used to represent the duration during which the vehicle enters the induction coil and does not leave. Based on the vehicle entry time sequence, calculate the second time interval between adjacent vehicles entering the same induction coil; If the current vehicle dwell time is greater than a preset stagnation time threshold, then the vehicle driving event is determined to be a vehicle stagnation event. If the current vehicle dwell time is less than a preset stagnation time threshold, then determine whether the second time interval is greater than a preset slow passage time threshold. If the second time interval is greater than the preset slow passage time threshold, the vehicle driving event is determined to be a slow passage event; otherwise, the vehicle driving event is determined to be a normal passage event.

6. The method according to claim 1, characterized in that, The step of identifying the target's rainfall interference status based on the visual image data and the millimeter-wave radar data includes: Acquire a reference road image, which represents a road image under no-rain conditions; Based on the reference road image and the visual image data, identify the first rainfall disturbance state; Based on the millimeter-wave radar data, identify the second rainfall interference state; The target rainfall disturbance state is generated based on the first rainfall disturbance state and the second rainfall disturbance state.

7. The method according to claim 6, characterized in that, The step of identifying the first rainfall disturbance state based on the reference road image and the visual image data includes: Droplet density and droplet size data are obtained using a laser scattering droplet density sensor. Based on the droplet density data and the droplet size data, the extinction coefficient is calculated. The extinction coefficient is used to represent the degree to which light is weakened due to absorption and scattering when it propagates in the medium. Based on the extinction coefficient, the image parameter offset is predicted, which includes the degree of contrast reduction and color offset; The image parameter offset is applied to the reference road image to obtain the desired blurred image; Extract the current road image from the visual image data; The expected blurred image is compared with the current road image to identify the first rainfall interference state.

8. The method according to claim 6, characterized in that, The step of identifying the second rainfall interference state based on the millimeter-wave radar data includes: Electromagnetic wave attenuation data is obtained by a multi-band electromagnetic wave attenuation measuring instrument. The electromagnetic wave attenuation data is used to reflect the degree of attenuation of electromagnetic waves after absorption and scattering by fog droplets. Based on the electromagnetic wave attenuation data, predict the background noise information formed by droplet absorption and scattering; Calculate the average signal attenuation intensity based on the electromagnetic wave attenuation data; Based on the background noise information and the average signal attenuation intensity, the expected radar noise data is generated; The expected radar noise data is compared with the millimeter-wave radar data to identify the second rainfall interference state.

9. The method according to claim 1, characterized in that, The step of adjusting traffic signals based on the target rainfall disturbance state and the vehicle driving events includes: If the target rainfall interference state is no rainfall interference, then the green light passage time for all lane directions will be reduced to shorten the traffic light cycle and thus reduce the red light waiting time. If the target rainfall interference state is rainfall interference, and the vehicle driving event is a slow vehicle passage event or a normal vehicle passage event, then the green light passage time for all lane directions will be increased to reduce the decision-making pressure on drivers in low visibility conditions. If the target rainfall interference state is rainfall interference and the vehicle driving event is a vehicle stop event, then the green light time in the direction of the lane where the vehicle stop event occurs will be increased to reduce congestion caused by vehicle accidents.

10. An intelligent traffic signal control system, characterized in that, include: The data acquisition module is used to acquire environmental sensing data, including visual image data, millimeter-wave radar data, and induction coil sensing data. The rainfall interference status identification module is used to identify the target rainfall interference status based on the visual image data and the millimeter-wave radar data; The vehicle driving event recognition module is used to recognize vehicle driving events based on the sensing data of the induction coil. The vehicle driving events include vehicle stationary events, vehicle slow-moving events, and vehicle normal-moving events. The traffic signal adjustment module is used to adjust the traffic signal according to the target rainfall interference state and the vehicle driving event.

Images