Solar intelligent zebra crossing warning floor lamp and traffic signal cooperative warning control method

Abstract

The invention relates to the technical field of intelligent traffic, and discloses a solar intelligent zebra crossing warning ground lamp and traffic signal cooperative warning control method, which comprises the following steps of: calculating an energy entropy index of each distributed ground lamp node to quantify the energy reliability; when no pedestrian trigger exists, the node enters a standby sleep and interception state; capturing and executing cascade wakeup by using the micro-vibration signal, and activating a downstream node; carrying out multi-modal data acquisition and heterogeneous fusion on the awakened node; calculating an intrusion probability index and judging a local warning level; when local judgment conflicts exist, weighting consensus is carried out on the result according to an energy entropy index, and a final global warning decision is generated; and performing cooperative warning and adaptive brightness control based on energy entropy according to the global decision. According to the method, the warning accuracy is improved by adopting multi-mode perception and energy entropy weighted consensus decision, and low-power-consumption and high-robustness operation of the distributed network is realized with the assistance of cascade wake-up and self-adaptive brightness control.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a method for coordinated warning control of solar-powered smart zebra crossing warning lights and traffic signals. Background Technology

[0002] With the acceleration of urbanization and the increasing traffic flow, ensuring pedestrian safety, especially at night and in inclement weather, has become a crucial issue in smart city traffic management. Traditional zebra crossings rely primarily on static markings and traffic lights for guidance, but at intersections without signal control or during signal light transitions, the risk of conflict between pedestrians and vehicles still exists. To address this, illuminated pavement tiles or smart road studs with active warning functions have emerged, aiming to provide drivers and pedestrians with more intuitive and timely risk warning information through dynamic light changes.

[0003] In existing technologies, some intelligent crosswalk systems employ sensors to detect the presence of pedestrians. These systems typically deploy infrared sensors, microwave radar, or utilize roadside cameras for video analysis in the crosswalk area. When a pedestrian is detected entering a designated area, ground lights or road studs are triggered to flash or change color to warn passing vehicles. In some distributed solutions, multiple illuminated ground lights are linked together via a wireless network to form a collaborative warning matrix, enhancing the coverage and visual impact of the warning effect.

[0004] While existing technologies have improved pedestrian safety to some extent, several shortcomings remain: First, the accuracy and reliability of warning decisions need improvement. Most existing technologies rely on single-type sensors for judgment, but any single sensor has inherent physical limitations. For example, infrared sensors are susceptible to ambient temperature and rain / fog, while visual sensors experience performance degradation under drastic changes in lighting or low light conditions at night, leading to potential false alarms or missed alarms. For distributed systems, the decision-making mechanism is typically simple, lacking a mechanism to evaluate the reliability of information reported by each node. A malfunctioning or temporarily obscured node can easily trigger an erroneous response from the entire system. Second, the system consumes a significant amount of energy, posing a serious challenge for independent nodes powered by solar energy. To ensure immediate response, existing sensor modules, especially radar or visual sensors, often need to operate continuously or poll at high frequencies, resulting in substantial static power consumption. During prolonged periods of overcast skies or in winter with insufficient sunlight, nodes are highly susceptible to power depletion and failure. Finally, the overall robustness and lifespan of distributed networks are not ideal. Existing distributed systems lack dynamic balancing mechanisms for task allocation and energy management. Regardless of a node's battery level, lighting conditions, or health status, all nodes are assigned the same warning tasks. This one-size-fits-all approach can cause some nodes with insufficient energy reserves to fail prematurely due to frequent high-power warning actions, creating a system bottleneck and reducing the effectiveness and long-term reliability of the entire warning network. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for coordinated warning control of solar-powered smart zebra crossing warning lights and traffic signals. This method solves the problem that existing smart zebra crossing warning systems, which rely on a single sensor and simple decision-making logic, struggle to balance the accuracy of warning decisions, system energy consumption, and the robustness of distributed networks.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for coordinated warning control of solar-powered smart zebra crossing warning lights and traffic signals, the method comprising the following steps: Step S1: Calculate the energy entropy index for multiple distributed ground light nodes deployed in the zebra crossing area. The energy entropy index is used to quantify the energy reliability of each distributed ground light node. Step S2: When no pedestrians trigger the distributed ground light node, it enters a standby listening and hierarchical sleep state. Step S3: When a pedestrian triggers the event, perform micro-vibration signal capture and cascade wake-up. Step S4: Perform multimodal data acquisition and heterogeneous fusion on the awakened nodes; Step S5: Calculate the intrusion probability index based on the fused data and generate the local warning level judgment result for each node; Step S6: Perform energy entropy weighted consensus decision. When there is a conflict between the local warning level determination results of different nodes, the local warning level determination results are weighted according to the energy entropy index calculated by each node in step S1 to generate a final global warning decision that is consistent across the entire network. Step S7: Based on the final global warning decision, execute collaborative warning and adaptive brightness control.

[0007] Preferably, in step S1, the step of calculating the energy entropy index includes: Collect the state of charge of the energy storage batteries, the standard deviation of load voltage fluctuation, and the historical shading coefficient of the distributed ground light nodes; The energy entropy index is obtained by weighting the state of charge of the energy storage battery, the standard deviation of the load voltage fluctuation, and the historical shading coefficient based on a preset mathematical model.

[0008] Preferably, in step S3, the steps of capturing micro-vibration signals and cascading wake-up include: By using multiple distributed ground light nodes to collaboratively measure the arrival time difference of vibration waves triggered by pedestrian footsteps, the real-time position and moving speed vector of the pedestrian can be calculated. Based on the moving speed vector, the system predicts the future movement trajectory of the pedestrian, determines a wake-up target set consisting of multiple downstream distributed ground light nodes, and sends a directional wake-up command to the nodes within the wake-up target set.

[0009] Preferably, in step S4, the steps of multimodal data acquisition and heterogeneous fusion of the awakened nodes include: Collect radar radial velocity and micro-vibration data of the awakened distributed ground light nodes; And by using the real-time location coordinates of pedestrians, the region of interest containing pedestrians is extracted from the video stream of the roadside visual sensor; The radar radial velocity, micro-vibration data, and visual data within the region of interest are spatiotemporally aligned to construct a multimodal heterogeneous fusion feature vector.

[0010] Preferably, step S5, the step of generating the local warning level determination result for each node, includes: Based on the multimodal heterogeneous fusion feature vector, the intrusion probability index, which quantifies the pedestrian's intrusion intent, is calculated using a linear weighted model. In conjunction with the real-time traffic signal phase status, the intrusion probability index is compared with preset warning thresholds and blocking thresholds to generate a local warning level determination result.

[0011] Preferably, in step S6, the step of performing energy entropy-weighted consensus decision-making includes: Based on the energy entropy index, a dynamic decision weight is calculated for each distributed ground lamp node; The local warning level determination result of each distributed ground light node is multiplied and summed with the corresponding dynamic decision weight to obtain the weighted consensus value; The weighted consensus value is then transformed into a final global alert decision through quantization mapping.

[0012] Preferably, in step S7, the step of executing the collaborative alert includes: When the final global warning decision is a blocking decision, control the relevant distributed ground light nodes to execute a high-frequency red strobe mode; When the final global warning decision is an early warning decision, the relevant distributed ground light nodes are controlled to execute a low-frequency red breathing flashing mode.

[0013] Preferably, the method further includes: When the final global alert decision is a safety or guidance decision, control the relevant distributed ground light nodes to execute the passage guidance mode; The flow rate of the traffic guidance mode is adaptively matched with the real-time movement speed vector of pedestrians.

[0014] Preferably, in step S7, the step of performing adaptive brightness control includes: Calculate the average energy entropy of all distributed ground light nodes; The luminous intensity of each distributed ground light node is adaptively adjusted based on the difference between the energy entropy of each distributed ground light node and the average energy entropy.

[0015] Preferably, in step S2, the steps of entering the standby listening and hierarchical sleep state include: When there is no pedestrian activity, the distributed ground light nodes enter a deep sleep state, while the micro-vibration sensor module remains in an event-driven listening state. The micro-vibration sensor module only generates an interrupt signal to wake up the distributed ground light node when the detected instantaneous vibration acceleration amplitude and vibration frequency simultaneously meet the hardware-preset dynamic trigger threshold.

[0016] This invention provides a method for coordinated warning control of solar-powered smart zebra crossing warning lights and traffic signals. It has the following beneficial effects: 1. This invention overcomes the perception limitations of single sensors caused by environmental interference or physical obstruction by combining micro-vibration sensors, radar, and visual data through multimodal data acquisition and heterogeneous fusion. Furthermore, by performing energy entropy-weighted consensus decision-making, nodes with more reliable energy states can be given higher decision weights when decision conflicts occur. This suppresses misjudgments caused by individual node failures or signal quality degradation at the system level, ensuring the accuracy of the final global warning decision.

[0017] 2. This invention employs a standby listening and hierarchical sleep strategy, allowing the system to enter a deep sleep state at the microampere level during periods of no events. Simultaneously, by capturing micro-vibration signals and cascading wake-up, it precisely activates only nodes along the pedestrian's future path, avoiding unnecessary energy consumption across the entire network. Combined with adaptive brightness control, it dynamically reduces the brightness output of low-energy nodes, achieving multi-level refined energy management, suitable for solar-powered scenarios.

[0018] 3. This invention uses the energy entropy index, which quantifies the energy reliability of nodes, as a core parameter for network operation. This index is used not only for weighted warning decisions but also for adjusting the luminous intensity of each node. This mechanism enables nodes with sufficient energy reserves to proactively undertake more warning tasks, while nodes with insufficient energy enter energy-saving mode to maintain basic functions. This achieves dynamic load balancing across the entire network, preventing premature failure of some nodes due to excessive consumption and improving the long-term operational stability of the entire system. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the logical framework for calculating the intrusion probability index and determining the local warning level of the present invention. Figure 3 This is a schematic diagram illustrating the data flow and logical operations of the energy entropy weighted consensus decision-making process of the present invention; Figure 4 This is a schematic diagram illustrating the instruction issuance and power adjustment process of the collaborative warning execution and adaptive brightness control flow of the present invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see the appendix Figure 1 , Figure 1This is a schematic diagram of a method flow according to an embodiment of the present invention. The present invention provides a pedestrian intrusion intelligent warning method based on multimodal perception and energy entropy consensus, which may include the following steps: S1. Execute system initialization and energy entropy benchmark construction. Each distributed ground lamp node calculates an energy entropy index that quantifies energy reliability by collecting its own energy storage battery state of charge, load voltage fluctuation standard deviation, and historical shading coefficient. This energy entropy index is uploaded to the roadside edge computing unit to construct a network routing table that sets the reciprocal of the energy entropy as the routing cost.

[0022] S2. Enter standby listening and hierarchical sleep state. When there is no pedestrian activity, the distributed ground light nodes enter a deep sleep state through power gating technology. At the same time, the micro-vibration sensor module and the low-power communication module maintain low-power listening. The former uses a hardware comparator to judge the amplitude and frequency of the vibration signal, and generates an interrupt wake-up signal when the conditions are met; the latter uses a periodic intermittent reception mode to listen for and filter over-the-air wake-up commands.

[0023] S3. Perform micro-vibration signal capture and cascade wake-up. The system utilizes multi-point time difference of arrival (TDOA) measurement technology to calculate the real-time position and velocity vector of the vibration source triggered by pedestrian footsteps. Based on this vector, the system predicts the pedestrian trajectory, determines the downstream wake-up target set, activates the nodes within this set through directional commands, and pre-sets a high-speed data link based on the energy entropy index.

[0024] S4. Perform multimodal data acquisition and heterogeneous fusion. The awakened nodes synchronously acquire and upload radar and micro-vibration data. The roadside edge computing unit uses pedestrian position coordinates and a pre-calibrated homography matrix to extract the region of interest from the video stream. By setting a synchronization time window, the radar, micro-vibration, and visual data are spatiotemporally aligned, and kinematic and biological behavioral features are extracted, ultimately integrating them into a multimodal fusion feature vector.

[0025] S5. Calculate the intrusion probability index and determine the local warning level. A linear weighted model is used to calculate the probability index of the fused feature vector, which integrates kinematic velocity, visual posture confidence, and micro-vibration intensity to generate a probability index that quantifies the pedestrian's intrusion intention. Combined with the traffic signal status, this index is compared with a preset threshold to generate a local warning level indicating a safe, warning, or blocked state.

[0026] S6. Execute energy entropy-weighted consensus decision. When local alert levels of various nodes conflict, consensus decision is activated. This decision mechanism uses the energy entropy of each node as the decision weight, calculates a weighted average of the local judgment results, and generates a final global alert decision that is consistent across the entire network through quantization mapping.

[0027] S7. Execute coordinated warnings and adaptive brightness control. The roadside edge computing unit issues global warning decisions, and each node executes coordinated warning modes such as high-frequency flashing, low-frequency breathing flashing, or flowing light guidance according to the instructions. Among them, the speed of the flowing lights in the traffic guidance mode matches the real-time speed of pedestrians. At the same time, the brightness of each node is adaptively adjusted according to the difference between its own energy entropy and the average energy entropy of the entire network to balance network energy consumption.

[0028] To establish a distributed control foundation capable of dynamically adapting to environmental changes, the system initialization and energy entropy benchmark construction step S1 in this embodiment specifically includes the following sub-steps S101 to S104: In sub-step S101, hardware self-test and network topology discovery are performed. After power-on reset, the microcontroller of each distributed ground lamp node first starts a self-test program to verify the electrical connection status of the multimodal sensor group, dual-mode communication module, and power management unit. After the self-test passes, the distributed ground lamp nodes communicate via low-power communication frequency bands. A network registration request frame is sent to the roadside edge computing unit. This request frame contains the node's physical unique identifier and factory firmware version. After receiving the request, the roadside edge computing unit assigns a network logical address and broadcasts a network-wide synchronization clock signal through the downlink channel to ensure that all distributed ground light nodes are on the same time reference as the roadside edge computing unit. The time synchronization accuracy meets the microsecond-level requirements for subsequent calculation of the propagation time difference of micro-vibration waveforms.

[0029] In sub-step S102, energy state parameters and historical environmental data are collected. To accurately quantify the energy health of the nodes, the microcontroller samples the solar cells and energy storage batteries using its internal high-precision analog-to-digital converter (ADC). The distributed ground light nodes employ a calibration method combining state-of-charge monitoring and open-circuit voltage methods to determine the current state of charge of the energy storage batteries. The system performs calculations. Simultaneously, the microcontroller reads power generation logs from memory, representing the power output over several past complete solar cycles. Furthermore, the microcontroller continuously monitors the voltage output at the load end at a high-frequency sampling rate, acquiring voltage fluctuation sequences to assess the degree of battery internal resistance aging and the stability of power supply under transient loads.

[0030] In sub-step S103, the energy entropy index is quantitatively calculated. Based on the collected multi-dimensional energy parameters, the microcontroller uses an embedded mathematical model to calculate the energy entropy in real time. This metric is a comprehensive reliability score used to quantify the energy reliability of a node as both a network routing node and a computing node. Energy Entropy The specific calculation model is as follows:

[0031] In the formula, For the first Distributed ground light nodes in The energy entropy value at any given time serves as a comprehensive quantitative indicator of the node's current energy health status and energy supply reliability. This is the index number of the distributed ground light node; The current moment; For nodes At any moment The state of charge of the energy storage battery is normalized to the range of [0, 1], which directly reflects the remaining power level of the battery. To perform analysis on nodes within the most recent statistical time window. The load output voltage of the first Sub-sample value; This represents the total number of sampling points within the aforementioned statistical time window; Within the same statistical time window, nodes of The arithmetic mean of the voltage samples; The nominal reference voltage set for the system serves as a benchmark value for normalizing the degree of voltage fluctuation. For example, it can be set as the stable output voltage when the battery is fully charged. For nodes The historical shading coefficient is determined by comparing the deviation ratio of the actual power generation of the node over multiple past solar cycles with the theoretical maximum power generation. It is used to quantify the degree of long-term sunlight impact on the node's installation location. The larger the value, the more severe the shading. , , These are the normalized weighting coefficients for the three components: state of charge, voltage stability, and historical shading coefficient. These coefficients are used to adjust the importance of each factor in the final energy entropy calculation and satisfy the following conditions: =1 constraint condition.

[0032] In sub-step S104, the virtual battery cluster model is constructed and the routing table is updated. Each distributed ground lamp node calculates its own energy entropy. Then, it is encapsulated in a heartbeat packet and transmitted via a low-power communication band. The data is broadcast to the roadside edge computing unit. The roadside edge computing unit receives and parses the energy entropy data from all nodes, mapping it in its local memory to generate a global virtual battery cluster (VBC) model. The roadside edge computing unit then uses the energy entropy data... Update the routing table and set the routing cost. Set as the reciprocal of energy entropy In subsequent high-bandwidth data transmission and cascading wake-up processes, the routing algorithm prioritizes nodes with higher energy entropy as relay nodes, establishing a mapping relationship between the physical energy state of distributed ground lamp nodes and the network logical topology weights.

[0033] In order to ensure the system has a millisecond-level response capability to sudden intrusions while minimizing energy consumption during off-peak hours, step S2 is specifically broken down into the following sub-steps S201 to S203: In sub-step S201, the deep sleep strategy is determined and executed. The microcontroller of the distributed ground light node continuously parses downlink channel commands or local traffic signal timing from the roadside edge computing unit. When it is determined that the current traffic phase is a motor vehicle passage state, i.e., a green light state or a red light state without pedestrian requests, and no triggering event is detected within the preset silent time window, the microcontroller sends a command to the power management unit to cut off the power supply circuit for high-energy-consuming modules. The distributed ground light node performs a power gating operation, physically disconnecting the power rails of the high-bandwidth communication module, Doppler radar module, and LED light-emitting driver unit. At this time, the distributed ground light node enters a deep sleep state with microampere-level current consumption, retaining only the power supply to the basic circuits that maintain the system clock.

[0034] In sub-step S202, the event-driven listening state of the micro-vibration sensor module is maintained. In deep sleep mode, the micro-vibration sensor module does not stop working but is configured for ultra-low power event-triggered mode. The micro-vibration sensor module integrates analog front-end circuitry and a threshold comparator, and its sensing unit continuously monitors the mechanical wave amplitude of the road surface medium at a low sampling rate. The microcontroller itself enters a stop mode or deep sleep mode, stopping the CPU clock operation and retaining only the real-time clock, static random access memory data retention function, and external interrupt controller response function. The output pin of the micro-vibration sensor module is connected to the microcontroller's external interrupt wake-up pin. When the vibration signal characteristics detected by the physical circuit meet the hardware preset conditions, a high-level or falling-edge pulse is generated, waking the microcontroller from the stop mode. This process does not require CPU polling, achieving energy decoupling between physical sensing and computational processing.

[0035] To address the hardware triggering logic of the micro-vibration sensor module in sub-step S202, and to filter environmental background noise and avoid frequent false wake-ups, the system sets a dynamic trigger threshold determination model at the sensor hardware layer. Only when the detected instantaneous vibration acceleration amplitude is reached... With vibration frequency A wake-up interrupt signal is generated only when the following hardware logic constraints are met simultaneously: ; In the formula, For the micro-vibration sensor module at any time The amplitude of the instantaneous vibration acceleration detected; The preset basic vibration trigger threshold is calibrated during system initialization based on the damping characteristics of the road surface material (such as asphalt or concrete). This is the environmental noise compensation coefficient, which is automatically adjusted by the analog front-end circuit inside the sensor based on the average noise floor over a period of time, in order to dynamically adapt to the background vibration differences at different times (such as the daytime when traffic is heavy and the nighttime when it is quiet). For the vibration signal at time Real-time main frequency components; The lower limit of the characteristic frequency of typical Rayleigh waves or body waves generated by pedestrian steps is used to filter out extremely low-frequency interferences such as crustal micro-movements or structural stress changes at the hardware level. The upper limit of the characteristic frequency of a typical Rayleigh wave or volume wave generated by a pedestrian's footsteps is used to filter out high-frequency mechanical noise interference caused by passing vehicles or construction at the hardware level.

[0036] In sub-step S203, the low-power communication channel maintains air wake-up monitoring. In parallel with micro-vibration monitoring, the low-power communication module maintains channel activity detection or preamble sampling mechanisms. This module is configured for periodic intermittent reception, activating the RF receiver to monitor carrier signals in the air during a preset time slice, and remaining in sleep mode at other times. This monitoring mechanism is used to receive cascaded wake-up commands sent from upstream adjacent distributed ground light nodes or roadside edge computing units. To prevent false wake-ups and ensure link directionality, the low-power communication module is configured with address filtering at the physical layer. It triggers an RF interrupt signal to wake up the microcontroller only when the received data packet's preamble and target physical address match the local machine, or belong to a specific network-wide broadcast group address. Through the dual monitoring of the physical medium monitoring of the micro-vibration sensor module and the electromagnetic medium monitoring of the low-power communication module, the system constructs a dual-modal asynchronous response mechanism for road surface signals and airborne control signals.

[0037] This embodiment achieves a precise match between communication energy consumption and real-time sensing by converting the physical movement trajectory of pedestrians into dynamic routes for the communication network. Step S3 is specifically broken down into the following sub-steps S301 to S304: In sub-step S301, micro-vibration waveform capture and time difference measurement are performed. This is done when the distributed ground light nodes... After being awakened by the hardware interrupt signal from the micro-vibration sensor module, the microcontroller immediately locks the local clock and records the precise arrival timestamp of the current vibration signal. Distributed ground light nodes Then, it broadcasts its physical coordinates via a low-power communication module. and timestamp The initial trigger information packet. Adjacent distributed ground light nodes. In listening mode, upon receiving the information packet, it combines it with the arrival timestamp of the same vibration event it has captured. The relative time difference of the vibration wave arriving at the two nodes was calculated. This time difference data eliminates the influence of absolute time errors in the system and directly reflects the path differences of vibration waves propagating to different nodes in the road surface medium.

[0038] In sub-step S302, the pedestrian movement velocity vector is calculated. The roadside edge computing unit or a decision master node with high energy entropy collects the physical coordinates and corresponding relative time differences reported by at least three adjacent distributed ground light nodes. The system calls a preset multi-point positioning algorithm, and based on the nominal propagation speed of Rayleigh waves in a specific road surface medium, it uses the nonlinear least squares method to iteratively calculate the real-time position coordinates of the vibration source, i.e., the pedestrian. Based on this, the system calculates the displacement difference of the position coordinates within continuous sampling time intervals to obtain the moving velocity vector describing the pedestrian's motion state. This vector contains the pedestrian's movement speed and instantaneous direction of motion, providing a kinematic basis for subsequent trajectory prediction.

[0039] In sub-step S303, trajectory prediction and wake-up target set determination are performed. The microcontroller then uses the calculated movement velocity vector... and preset prediction time window The trajectory area of ​​the pedestrian during the system response delay time can be calculated. To achieve precise targeted wake-up and reduce unnecessary energy consumption, the system constructs a set of target distributed ground light nodes for this cascaded wake-up based on the trajectory region. Its construction logic satisfies the following set definition:

[0040] In the formula, This refers to the set of distributed ground light nodes targeted for this cascaded wake-up. This is a candidate distributed ground light node; For distributed ground light nodes Physical coordinates in the preset world coordinate system; This is a trajectory region prediction function that calculates and returns a spatial region covering the future location of the pedestrian based on the input pedestrian state parameters. For at any time The calculated real-time physical coordinates of the pedestrian; For at any time The calculated real-time velocity vector of the pedestrian includes the pedestrian's instantaneous speed and direction of movement; This is the preset prediction time window used for trajectory extrapolation.

[0041] This step ensures that only nodes located on the pedestrian's future movement path are included in the wake-up list, enabling the control range to dynamically follow the pedestrian's trajectory.

[0042] In sub-step S304, directed cascading wake-up and high-speed route pre-configuration are performed. The initial triggering node or decision master node communicates with the set via a low-power communication module. The target node sends a directed data packet containing a wake-up command. The downstream distributed ground light nodes are then awakened. Upon receiving the instruction, the high-bandwidth communication module is immediately activated. When constructing the data link back to the roadside edge computing unit, the distributed ground lamp nodes execute a routing strategy based on the energy entropy index determined in step S1, prioritizing neighboring nodes with higher energy entropy as the next-hop relay. Through this process, the system pre-establishes a high-speed data transmission channel in the area where pedestrians are about to arrive, with a physically dynamic topology and link quality optimized based on energy state, providing network assurance for the subsequent real-time transmission of multimodal large amounts of data.

[0043] After completing the cascaded wake-up and high-speed link construction for a specific region, this embodiment focuses on solving the data synchronization and feature alignment problem between sensors with different physical mechanisms. Step S4 is specifically refined into the following sub-steps S401 to S404: In sub-step S401, the synchronization sensing and data uplink transmission of the wake-up nodes are performed. The set of downstream distributed ground lamp nodes in the wake-up state... Each distributed ground light node immediately powers its built-in Doppler radar module. This Doppler radar module is configured in narrow-beam continuous wave or frequency-modulated continuous wave mode to detect targets in the direction of the zebra crossing entrance and obtain their radial velocity values. Simultaneously, the micro-vibration sensor module switches to a high-precision continuous sampling mode, acquiring raw vibration sequences containing complete waveform characteristics. The distributed ground light nodes utilize the established high-bandwidth communication links to transmit data including radial velocity values. Vibration waveform characteristics and local precise timestamp The perceived data packets are uploaded to the roadside edge computing unit in real time.

[0044] In sub-step S402, visual region of interest extraction based on location index is performed. After receiving the perception data packet from the distributed ground light nodes, the roadside edge computing unit uses the real-time physical coordinates of the pedestrian calculated in the previous step. As a spatial index, the video stream from the roadside vision sensor is locally cropped. The roadside edge computing unit uses a perspective transformation matrix to map the physical coordinates of the road surface to the pixel coordinate system of the video stream, thereby extracting the region of interest containing only pedestrians. This mapping transformation follows the coordinate transformation model below: ; In the formula, In response time The homogeneous coordinates of the pedestrian in the image pixel coordinate system; it is a three-dimensional column vector, usually represented as... ,in, For pedestrians at all times The x-coordinate in the image pixel coordinate system; For pedestrians at all times The ordinate in the image pixel coordinate system; It is a time variable; is the scale factor in the homogeneous coordinate system; It is a homography matrix that uniquely describes the geometric relationship of the two-dimensional projection transformation from the world coordinate system (i.e., the physical plane of the road surface paved by the distributed ground light nodes) to the imaging plane of the roadside vision sensor. This matrix is ​​pre-calculated through the camera calibration process and stored in the roadside edge computing unit. In response time The homogeneous coordinate vector of the pedestrian in the world coordinate system; where and These are the x and y coordinates of the pedestrian in the road surface physical coordinate system, calculated from the micro-vibration signal. This coordinate vector is a mathematical expression of the pedestrian's physical position.

[0045] This step allows the system to process image data only within the mapped area, reducing the computational load on the visual algorithm and eliminating interference from non-target vehicles or pedestrians in the background.

[0046] In sub-step S403, temporal alignment and synchronization verification of heterogeneous data are performed. Due to the inconsistency between the data acquisition frame rate of the visual sensor and the sampling rate of the distributed ground light node sensors, and the existence of network transmission jitter, the system performs time alignment before fusion. The roadside edge computing unit maintains a global time axis with millisecond-level precision and sets a synchronization time window. The system iterates through and searches the timestamps of the visual data frames. Timestamp of ground light data packet Only if the absolute value of the time difference between the two is less than or equal to the synchronization time window When two sets of data are determined to represent the same event at the same time, the system uses linear interpolation to complete the data for missing sampling points, ensuring strict correspondence between the multimodal data and the time dimension.

[0047] In sub-step S404, a multimodal heterogeneous fusion feature vector is constructed. The roadside edge computing unit performs feature extraction and normalization encapsulation on the spatiotemporally aligned data. For the visual region of interest, the system uses a pose estimation model to extract key skeleton nodes of the pedestrian and calculates the pose confidence. This parameter reflects whether the pedestrian is leaning forward or turning their head towards oncoming traffic in a preparatory crossing posture. For micro-vibration data, the system performs spectral analysis to extract the frequency domain energy distribution; for radar data, it extracts the radial velocity component. Finally, the system generates a fusion feature vector for subsequent decision-making. Its composition is as follows:

[0048] In the formula, This is a multimodal heterogeneous fusion feature vector; To normalize the radar radial velocity characteristics, the pedestrian radial velocity measured by Doppler radar is vector-projected onto the direction perpendicular to the zebra crossing and then normalized according to the preset maximum pedestrian speed. The visual pose confidence score is obtained by analyzing the relative positions and orientations of key skeletal nodes of the pedestrian's body by a pose estimation algorithm (such as a lightweight version of OpenPose) running on the roadside edge computing unit. The higher the score, the more the pedestrian's current pose (such as leaning forward significantly, turning their head to observe the direction of oncoming vehicles, etc.) matches the typical behavioral intention of crossing the road. The energy spectral density of the micro-vibration is obtained by performing a fast Fourier transform on the original vibration waveform signal collected by the micro-vibration sensor, or by integrating its envelope in the time domain. This parameter is used to quantify the physical authenticity and force of pedestrian walking behavior. As a physical verification method, it can effectively combat visual dummy misjudgment caused by factors such as light and shadow and reflection. Location spatial weight; it is a scalar weight value assigned based on the pedestrian's current physical location (usually determined by the preset coordinates of the first or strongest triggered distributed ground light node). This value is positively correlated with the inherent danger level of the pedestrian's location. For example, when the pedestrian is in the central area of ​​the zebra crossing, the weight value is set to a high value, while when the pedestrian is at the edge of the waiting area, the weight value is set to a low value, in order to distinguish the potential threat level of different locations in risk assessment.

[0049] The fused feature vector It describes the comprehensive state of the target object in three dimensions: physical motion, biological posture, and geographical location.

[0050] Please see the appendix Figure 2 , Figure 2A logic block diagram of the intrusion probability index calculation and local warning level determination process according to an embodiment of the present invention is shown. In order to convert heterogeneous sensor data into quantifiable risk indicators and generate control commands accordingly, step S5 is specifically refined into the following sub-steps S501 to S504: In sub-step S501, the roadside edge computing unit invokes a pre-built algorithm engine to process the fused feature vector generated in the previous stage. The system calculates a normalized intrusion probability index to comprehensively characterize the intensity of the pedestrian's intent to forcibly cross the road at the current moment. The calculation model for this index is constructed as follows: ; In the formula, The calculated intrusion probability index is a comprehensive risk scalar. The weighting coefficients for the kinematic velocity factor; Weighting coefficients for visual pose confidence; Let be the weighting coefficients of the micro-vibration intensity characteristic factor, and let the weighting coefficients satisfy . + + =1; The normalized kinematic velocity factor characterizes the intensity of the pedestrian's motion in the crossing direction; The pose confidence score, obtained through the visual pose estimation algorithm, represents the pedestrian's behavioral intention. This is a characteristic factor of micro-vibration intensity.

[0051] The system can dynamically adjust the weighting coefficients based on the current traffic signal status: for example, when the traffic light is red, the kinematic weights are increased. The value of [value] is adjusted to enhance sensitivity to fast-moving targets; when the traffic light is green, the visual pose weight can be increased. The value of is chosen to focus on identifying pedestrians' hesitation or observation behavior.

[0052] In sub-step S502, the system adjusts the kinematic velocity factor. Vector projection calculations are performed to distinguish between lateral loitering and longitudinal crossing movements of pedestrians. The system uses radial velocity vectors measured by Doppler radar. The unit direction vector relative to the preset physical direction of the zebra crossing Perform a dot product operation. This calculation method only yields a valid kinematic factor when a pedestrian generates a displacement velocity towards the vehicle lane, effectively filtering out parallel movement interference within the waiting area. The calculation formula is as follows:

[0053] In the formula, This represents the normalized kinematic velocity factor. The radial velocity vector of the target (pedestrian) as measured by the Doppler radar module; A predefined unit direction vector that describes the physical direction of the zebra crossing; This is a vector dot product operator used to calculate the magnitude of the projected component of the velocity vector in the zebra crossing direction; This is the preset maximum theoretical speed for pedestrians.

[0054] In sub-step S503, a time-domain integral transform of the micro-vibration intensity characteristic factor is performed to introduce physical entity verification and enhance the system's robustness against visual deception. The micro-vibration intensity characteristic factor is not calculated based on the amplitude at a single moment, but rather by integrating the amplitude envelope of the vibration signal acquired by the micro-vibration sensor within the most recent time sliding window to reflect the energy accumulation effect of the vibration signal. The calculation formula is as follows: ; In the formula, It is a characteristic factor of micro-vibration intensity; The current moment; The width or duration of the time sliding window used for integration calculation; To be within a time sliding window, at any time The amplitude envelope of the micro-vibration signal; In the time interval [ , The integral variable within the brackets represents time. The preset effective step energy threshold; For time variables The differential.

[0055] In sub-step S504, the system generates a local warning level determination result for a specific distributed ground light node. The roadside edge computing unit acquires the real-time phase status of the traffic signal controller. And based on the preset warning threshold With blocking threshold The intrusion probability index (PCIPCI) is classified and determined.

[0056] The specific judgment logic is as follows: like The light is red and If a pedestrian is determined to have a very high probability of entering or being in a dangerous area, a blocking decision is generated. ; like The light is red and The system determines whether a pedestrian is in a state of flirting or inattentiveness, and generates an early warning. ; In the remaining cases, including or When the light is green, it is considered a safe state, and a safety judgment is generated. .

[0057] The result of the local warning level assessment It is then stored in the queue to be sent, serving as the basic input data for the next stage of distributed consensus decision-making.

[0058] Please see the appendix Figure 3 , Figure 3 This diagram illustrates the data flow and logical operation of the energy entropy weighted consensus decision-making process according to an embodiment of the present invention. To eliminate the risk of misjudgment caused by a decrease in the sensor signal-to-noise ratio due to battery voltage fluctuations in a single distributed ground light node, and to establish consistent recognition of the warning status across the entire network, step S6 is specifically refined into the following sub-steps S601 to S604: In sub-step S601, the trigger determination of the distributed consensus mechanism is executed. The roadside edge computing unit continuously monitors the local warning level determination results reported by each distributed ground light node in the wake-up node set. When the system detects that any of the following conditions are met, the consensus decision-making procedure is immediately activated: First, there is a discrete conflict in the local warning level determination results reported by different distributed ground light nodes within the set; Secondly, the intrusion probability index calculated in the previous steps falls within the preset fuzzy critical range.

[0059] This triggering mechanism ensures that the system only consumes computing resources for consensus verification when there is uncertainty in the decision, and directly executes instructions when the decision is clear and consistent.

[0060] In sub-step S602, dynamic decision weight allocation based on energy entropy is performed. Once the consensus mechanism is activated, the roadside edge computing unit reads the latest energy entropy index of each distributed ground light node in the set of distributed ground light nodes participating in this decision. Based on the principle that the more stable the energy state, the more reliable the data, the system uses an exponential normalization function to calculate the dynamic decision weight of each distributed ground light node. The weight calculation model is shown below: ; In the formula, For dynamic decision weights; This is the index number of the distributed ground light node; is the base of the natural logarithm; The weight sensitivity adjustment factor is a preset positive real number. >0), The larger the value, the more obvious the weight advantage of nodes with high energy entropy; For the first The energy entropy index of each distributed ground light node is calculated in step S1 and comprehensively reflects the energy reliability characteristics of the node, such as the current battery state of charge, power supply voltage stability, and historical light shading. The set of distributed ground light nodes participating in this consensus decision consists of nodes that were awakened when the consensus mechanism was triggered and reported a valid local warning level determination result; For traversing the set The index number of all nodes in the array.

[0061] In sub-step S603, a global weighted average calculation is performed. The roadside edge calculation unit calculates the local warning level determination results of each distributed ground light node. Its corresponding dynamic decision weights By performing multiplication and accumulation, a weighted consensus value in continuous numerical form is obtained. : ; In the formula, The weighted consensus value is a continuous value obtained by weighting and averaging energy entropy. This value integrates the judgment results of all participating decision-making nodes and their respective energy reliability. For the first The local warning level determination results reported by each distributed ground lamp node are discrete integer values, where 0 represents a safe state, 1 represents a warning state, and 2 represents a blocking state. For the first The dynamic decision weights of each distributed ground light node are calculated using the first formula above. This is the index number of the distributed ground light node; This refers to the set of distributed ground light nodes that participated in this consensus decision.

[0062] The calculation process treats the local warning level determination result as the original observation result of each physical location sensor, and regards the dynamic decision weight as the confidence level of the observation result. Through weighted filtering, the influence of singular values ​​generated by individual faulty nodes or occluded nodes is eliminated from the statistical level.

[0063] In sub-step S604, the quantization mapping for the final global decision is performed. To transform the continuous weighted consensus values ​​into executable, specific control instructions, the system uses a quantization mapping function. Generate the final global alert decision. The quantization process employs multi-stage hysteresis comparator logic to prevent frequent transitions in the alert state at critical points. Its quantization mapping formula is as follows: ; In the formula, The final global warning decision is the final control command obtained after quantization mapping. Its value is a discrete integer 0, 1 or 2, which correspond to the safety / guidance mode, warning mode and blocking mode, respectively. This is a quantization mapping function used to map continuous weighted consensus values ​​to discrete decision levels; The weighted consensus value is the input to the quantization function; The quantitative decision threshold for the blocking mode is a preset real value used to determine whether to enter the highest level of blocking mode. The quantitative judgment threshold for the warning mode is a preset real value used to distinguish between the safety mode and the warning mode, and it satisfies the following conditions: < .

[0064] When the weighted consensus value leans towards a certain level, it indicates that most high-confidence nodes after weighting have determined that there is a risky behavior of the corresponding level, and the system ultimately locks in a global alert decision for that level. This final global alert decision is then encapsulated into a highest-priority control command and sent to all relevant execution nodes via a high-speed communication link.

[0065] Please see the appendix Figure 4 , Figure 4 A schematic diagram illustrating the instruction issuance and power adjustment process of the collaborative warning execution and adaptive brightness control flow according to an embodiment of the present invention is shown. After completing the global warning decision based on energy entropy, the focus of this embodiment is to transform the decision into a highly collaborative execution action and achieve a dynamic balance between energy and warning effect across the entire network. Step S7 is specifically refined into the following sub-steps S701 to S703: In sub-step S701, the collaborative command for the warning mode is issued and executed. Based on the final global warning decision obtained in step S6, the roadside edge computing unit determines the specific warning mode that the distributed ground light nodes should execute. The command is issued to all awakened distributed ground light nodes via a high-speed data transmission link. Based on the value of the final global warning decision, the warning modes are divided into the following three categories: If... That is, the blocking decision involves the distributed ground light nodes executing a high-frequency red strobe mode at a frequency of 4 Hz to 8 Hz to generate a strong visual impact for blocking; if That is, the distributed ground light nodes execute a low-frequency red breathing flashing pattern at a frequency of 1 Hz to 2 Hz to continuously but relatively gently alert pedestrians; if In other words, for safety or guidance decisions, the distributed ground light nodes execute either a constant green light or a flowing guidance mode.

[0066] In sub-step S702, speed adaptive matching of the traffic guidance mode is performed. When the traffic signal phase is green and the final global warning decision is 0, the system enters the pedestrian guidance mode. At this time, the on / off states of the distributed ground light nodes are sequentially transmitted along the zebra crossing's direction of travel, forming a flowing visual effect. To improve the comfort and accuracy of the guidance, the flowing speed... It is not fixed, but rather based on the pedestrian's real-time moving velocity vector calculated in the previous steps. The modulus value is used for real-time adaptive matching, and its calculation formula is as follows:

[0067] In the formula, The visual flow speed of the ground light flowing water guidance mode; The preset guidance coefficient is a constant greater than 1, used to make the water flow speed slightly faster than the actual speed of pedestrians, so as to play a slight acceleration and guidance role; This is the magnitude of the pedestrian's real-time moving velocity vector calculated in the previous step, i.e., the pedestrian's instantaneous moving rate.

[0068] In sub-step S703, adaptive brightness allocation based on energy entropy is performed. To extend the battery life of low-energy nodes while ensuring that high-energy nodes can output sufficient brightness to guarantee warning effectiveness, the system implements a refined brightness allocation strategy. The roadside edge computing unit first calculates the current average energy entropy of the virtual battery cluster. :

[0069] In the formula, The average energy entropy of all distributed ground lamp nodes in the virtual battery cluster represents the overall energy health of the entire ground lamp network system. This refers to the collection of all distributed ground light nodes deployed within the monitored area. For set The number of elements in the middle, that is, the total number of distributed ground light nodes; This is the index number of the distributed ground light node; For the first The energy entropy index of a distributed ground light node.

[0070] Subsequently, the roadside edge computing unit determines the global reference brightness based on the average energy entropy. And combined with the energy entropy of each distributed ground light node. Dynamic adjustments are made to issue a unique brightness control command to each distributed ground light node. The calculation model for this brightness control command is as follows:

[0071] In the formula, To be issued to the first The target brightness value of each distributed ground light node is normalized and its value ranges from [0, 1]. ]; This is the index number of the distributed ground light node; The maximum physical light intensity output limit for the distributed ground light node is a preset system parameter. To ensure the minimum safety warning brightness is clearly visible under all circumstances, a preset system parameter is used; The global baseline brightness is determined by the roadside edge calculation unit based on the current ambient illuminance or the overall average energy entropy, and is a normalized value. is the brightness adjustment sensitivity coefficient, a positive real number, used to control the correlation strength between the node brightness and its energy entropy deviation; For the first Energy entropy index of a distributed ground light node; denoted as average energy entropy.

[0072] When the energy entropy of a distributed ground light node is greater than the average energy entropy, its output brightness... It will be brighter than the reference brightness, thus fulfilling more warning functions; When the energy entropy is less than the average energy entropy, its output brightness will be limited to below the reference brightness, thereby saving energy.

[0073] By linking the brightness output with the node's energy status, the system achieves refined and balanced scheduling of the entire network's energy load, effectively preventing low-energy nodes from being prematurely exhausted due to localized high-energy-consumption operations.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for solar-powered zebra crossing warning ground light and traffic signal cooperative warning control, characterized in that, The method comprises the following steps: Step S1, calculating an energy entropy index for a plurality of distributed street lamp nodes deployed in a zebra crossing area, the energy entropy index being used to quantify the energy reliability of each distributed street lamp node; Step S2, the distributed street lamp nodes enter a standby listening and hierarchical sleep state when there is no pedestrian triggering; Step S3, when there is a pedestrian triggering, micro-vibration signal capture and cascade wake-up are performed; Step S4, multi-modal data acquisition and heterogeneous fusion are performed on the woken-up nodes; Step S5, a crossing probability index is calculated based on the fused data, and a local warning level judgment result of each node is generated; Step S6, an energy entropy weighted consensus decision is executed, when there is a conflict in the local warning level judgment results of different nodes, the local warning level judgment results are weighted calculated according to the energy entropy indexes calculated in step S1 to generate a final global warning decision consistent with the whole network; Step S7, according to the final global warning decision, a cooperative warning and adaptive brightness control is performed.

2. The solar-powered smart zebra crossing warning ground light and traffic signal cooperative warning control method according to claim 1, characterized in that, In step S1, the step of calculating the energy entropy index comprises: The state of charge of the energy storage battery, the load voltage fluctuation standard deviation and the historical shielding coefficient of the distributed street lamp node are collected; And the state of charge of the energy storage battery, the load voltage fluctuation standard deviation and the historical shielding coefficient are weighted calculated based on a preset mathematical model to obtain the energy entropy index.

3. The solar powered smart zebra crossing warning ground light in coordination with traffic signal warning control method as claimed in claim 2 wherein, In step S3, the step of performing micro-vibration signal capture and cascade wake-up comprises: The arrival time difference of the vibration wave triggered by the pedestrian's footsteps is measured by a plurality of distributed street lamp nodes to solve the real-time position and moving speed vector of the pedestrian; And the future moving track of the pedestrian is predicted according to the moving speed vector to determine a wake-up target set composed of a plurality of downstream distributed street lamp nodes, and a directional wake-up instruction is sent to the nodes in the wake-up target set.

4. The solar-powered smart zebra crossing warning ground light and traffic signal cooperative warning control method according to claim 3, characterized in that, In step S4, the step of performing multi-modal data acquisition and heterogeneous fusion on the woken-up nodes comprises: The radar radial velocity and micro-vibration data of the woken-up distributed street lamp node are collected; And the real-time position coordinates of the pedestrian are used to extract the region of interest containing the pedestrian in the video stream of the roadside vision sensor; The radar radial velocity, micro-vibration data and vision data in the region of interest are spatio-temporally aligned to construct a multi-modal heterogeneous fusion feature vector.

5. The solar powered smart zebra crossing warning ground light in coordination with traffic signal warning control method as claimed in claim 4 wherein, In step S5, the step of generating a local warning level judgment result of each node comprises: Based on the multi-modal heterogeneous fusion feature vector, a crossing probability index quantifying the crossing intention of the pedestrian is calculated through a linear weighting model; And the crossing probability index is compared with the preset warning threshold and blocking threshold combined with the real-time traffic signal phase state to generate a local warning level judgment result.

6. The solar powered smart zebra crossing warning ground light in coordination with traffic signal warning control method as claimed in claim 3 wherein, In step S6, the step of performing energy entropy weighted consensus decision comprises: Based on the energy entropy index, a dynamic decision weight is calculated for each distributed street lamp node; The local warning level judgment result of each distributed street lamp node is multiplied by the corresponding dynamic decision weight to obtain a weighted consensus value; And the weighted consensus value is converted into a final global warning decision through quantization mapping.

7. The solar powered smart zebra crossing warning ground light in coordination with traffic signal warning control method as claimed in claim 6 wherein, In step S7, the step of performing the coordinated warning includes: when the final global warning decision is a blocking decision, controlling the relevant distributed floor lamp nodes to perform a high-frequency red strobe mode; when the final global warning decision is a pre-warning decision, controlling the relevant distributed floor lamp nodes to perform a low-frequency red breathing flash mode.

8. The solar-powered smart zebra crossing warning ground light and traffic signal cooperative warning control method according to claim 7, characterized in that, The method further includes: when the final global warning decision is a safety or guidance decision, controlling the relevant distributed floor lamp nodes to perform a passage guidance mode; the flow rate of the passage guidance mode is adaptively matched with the real-time moving speed vector of the pedestrian.

9. The solar powered smart zebra crossing warning ground light in coordination with traffic signal warning control method as claimed in claim 1 wherein, In step S7, the step of performing adaptive brightness control includes: calculating the average energy entropy of all distributed floor lamp nodes; and based on the difference between the energy entropy of each distributed floor lamp node and the average energy entropy, adaptively adjusting the luminous brightness of each distributed floor lamp node.

10. The solar powered smart zebra crossing warning ground light in coordination with traffic signal warning control method as claimed in claim 1 wherein, In step S2, the step of entering the standby listening and hierarchical sleep state includes: when there is no pedestrian activity, the distributed floor lamp nodes enter a deep sleep state, and the micro-vibration sensor module is maintained in an event-driven listening state; the micro-vibration sensor module only generates an interrupt signal to wake up the distributed floor lamp nodes when the monitored instantaneous vibration acceleration amplitude and vibration frequency simultaneously meet the hardware preset dynamic trigger threshold.

Images