Intelligent control system based on intelligent signboard and method thereof

Abstract

The application discloses a kind of intelligent control system and method based on wisdom signboard, it is related to intelligent traffic and internet of things control technical field, including the following steps: establish unified time scale baseline, collect exposure sequence, saturation threshold sequence and trajectory residual error, generate exposure-trajectory coupling spectrum, for calibration rain and snow trigger time window;Under the constraint of exposure-trajectory coupling spectrum, calculate phase residual error mapping, identify exposure shock root cause, extract saturated source area and trajectory breakpoint set, generate fault anchor point.The present application identifies rain and snow interference by time scale baseline and coupling spectrum, constructs fault anchor point and plays back optical path to evaluate risk, fuses multi-source signal to generate robust trajectory, uses conjugate correction and feedforward constraint to suppress oscillation, combines time reversal closed-loop instruction and multi-strategy joint debugging, realizes adaptive optimization of exposure and induced information, improves the stability and control precision of wisdom signboard under extreme weather.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and Internet of Things control technology, specifically to an intelligent control system and method based on smart signs. Background Technology

[0002] "Smart control based on smart signs" refers to an intelligent control mode that integrates multiple functions such as sensing, data processing, communication, and intelligent decision-making into traffic signs to achieve dynamic collection, real-time analysis, and coordinated regulation of traffic information. Its core idea is to use smart signs as information hubs, uploading data collected by road monitoring equipment (such as traffic flow detectors, weather sensors, and cameras) to a central control platform via the Internet of Things (IoT). Artificial intelligence algorithms then predict and analyze traffic flow, road conditions, and environmental changes, and the optimized control commands are sent in real-time to signs, traffic lights, or guidance screens at intersections, enabling adaptive updates of traffic guidance information and dynamic adjustment of traffic order. This control mechanism can not only automatically adjust the displayed content and control strategies according to real-time road conditions, but also implement emergency responses in abnormal situations such as sudden events and severe weather, thereby achieving intelligent management of road traffic, congestion relief, and safety improvement.

[0003] The existing technology has the following shortcomings:

[0004] In existing technologies, intelligent transportation systems largely rely on visual perception units installed on smart signs to obtain real-time dynamic information such as traffic flow, vehicle speed, and road occupancy, supporting AI control units in predicting traffic conditions and scheduling signals. However, when encountering extreme weather conditions such as sudden rain or snow, the intensity and reflectivity of external light can change drastically in a short period, causing the automatic exposure mechanism of the visual perception unit to frequently restart exposure parameters. This process can easily lead to perception saturation in image sensors, resulting in the loss of critical vehicle trajectories and interruption of image continuity. Because the AI ​​control unit cannot obtain continuous and reliable target tracking data during this period, its traffic flow prediction model and signal optimization algorithm will lose their reference benchmark, leading to a severe misalignment between the output of guidance information and the actual traffic flow status, creating localized perception blind spots and causing dynamic loss of control problems.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a smart control system and method based on smart signs to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a smart control method based on smart signs, comprising the following steps:

[0008] Establish a unified time-scale baseline, collect exposure sequences, saturation threshold sequences and trajectory residuals, and generate an exposure-trajectory coupling spectrum to calibrate the rain and snow trigger time window;

[0009] Under the constraint of exposure-trajectory coupling spectrum, the phase residual mapping is calculated, the root cause of exposure oscillation is identified, the saturation source region and the set of trajectory breakpoints are extracted, and the fault anchor point is generated.

[0010] Based on the fault anchor point, perform counterfactual playback to reconstruct the rain and snow incident path, reflection path and scattering path, calculate the exposure threshold and trajectory continuous surface, and output the misjudgment risk coefficient.

[0011] Based on the misjudgment risk coefficient, a cross-modal compensation chain is constructed, which integrates millimeter-wave echo signals, thermal imaging signals and ground magnetic detection signals to generate a robust trajectory proxy to restore trajectory continuity.

[0012] The robust trajectory proxy is input into the exposure-trajectory joint correction stage to calculate the phase conjugate correction vector and establish a feedforward constraint field to limit exposure bounce and suppress oscillation regeneration.

[0013] Based on the feedforward constraint field to generate time inversion closed-loop instruction set, combined with frequency misdirection traction, time-scale traction and amplitude limiting write-back strategy, the exposure allocation and induced information update are adaptively rearranged to achieve stable dynamic control of the visual perception system in rain and snow environments.

[0014] Preferably, the exposure-trajectory coupling spectrum generation steps are as follows:

[0015] A nanosecond-level unified time baseline is established, and time base signals are synchronized to multiple sensing nodes through a high-stability time base source and fiber optic synchronization network. The sampling timing register is cleared and time code is written to achieve consistent control of the timestamps of all sensing data.

[0016] Under a unified time-stamped baseline, exposure sequence, saturation threshold sequence and trajectory residual data are collected. The image frame exposure parameters, the proportion of extreme values ​​of image grayscale distribution and the target trajectory offset distance are recorded respectively, and three sets of synchronized datasets are generated according to a unified timestamp.

[0017] Normalization and energy correlation analysis were performed on three sets of synchronous datasets to calculate the synchronization intensity of exposure change and trajectory offset. The coupling intensity curve was generated by sliding over a continuous time period, and an exposure-trajectory coupling spectrum reflecting the correlation characteristics of environmental interference and motion anomaly was formed.

[0018] Cluster analysis is performed on the energy density anomaly regions in the coupled spectrum, and raindrop density and humidity change data are integrated to determine the start time of rain and snow disturbance. The time boundary is extended to generate a rain and snow trigger time window, which serves as the input basis for subsequent control processes.

[0019] Preferably, the steps for generating fault anchor points are as follows:

[0020] Under the constraint of exposure-trajectory coupling spectrum, the energy surge segment is extracted, the phase sequence of exposure response and target trajectory between consecutive frames is calculated, and a phase residual mapping map is generated.

[0021] Based on the residual peak time point in the phase residual mapping map, combined with image brightness distribution, contrast change and environmental humidity data, the specific interference causes of exposure oscillation are identified.

[0022] Perform regional grayscale traversal on the corresponding image frame, extract the coordinates of the brightness saturation source region, and combine the analysis of the interruption point with the vehicle trajectory sequence to form a set of saturation source regions and trajectory breakpoints;

[0023] Based on the matching relationship between phase residual amplitude, saturation region location and trajectory interruption timing, effective fault anchor points are generated by fusion, providing spatial positioning reference for subsequent control.

[0024] The preferred steps for outputting the misjudgment risk coefficient are as follows:

[0025] Based on the generated fault anchor points, anchor points that meet the conditions of time continuity, trajectory interruption duration and spatial location are selected, and image sequences, trajectory data and exposure records of the time periods before and after them are extracted to form a counterfactual playback dataset.

[0026] Light propagation analysis was performed on the image sequence to reconstruct the incident, reflected, and scattered paths of rain and snow in sequence, thereby obtaining the propagation distribution of light disturbance in the image space.

[0027] Pixel energy stacking density is calculated based on light propagation results, and an exposure threshold distribution map is generated to describe the degree of exposure overload in areas of abnormal lighting.

[0028] By combining trajectory data, trajectory completion is performed in the interrupted interval to construct a continuous trajectory surface, and high-risk and low-risk paths are marked according to the exposure threshold distribution map;

[0029] By combining the confidence level of the trajectory surface, the exposure overload ratio, and the trajectory interruption duration, a misjudgment risk coefficient is calculated, which is used as a basis for subsequent perception compensation and control decisions.

[0030] The preferred robust trajectory proxy generation steps are as follows:

[0031] The activation priority of the non-visual perception channel is determined by using the misjudgment risk coefficient as the control variable, and the fusion weight of each channel is assigned. A time-aligned index table is generated according to the weight.

[0032] Trajectory features are extracted from millimeter-wave echo, thermal imaging, and ground magnetic detection by time alignment, and the coordinates of each channel are mapped to a unified ground coordinate system.

[0033] The data points with the highest confidence are selected frame by frame according to the channel weight to form a time series point set. The time series point set is then used for path fitting and the fusion confidence level is labeled. A robust trajectory proxy is output to replace the visually interrupted trajectory.

[0034] Preferably, the feedforward constraint field establishment process is as follows:

[0035] By aligning the robust trajectory proxy with the exposure parameter sequence under a unified time scale, a joint reference structure is established for trajectory position, direction of motion, velocity, and rate of change of exposure response.

[0036] The phase difference between target motion and exposure response is calculated based on the joint reference structure, forming an exposure-trajectory phase difference mapping surface. A phase conjugate correction vector is constructed according to the ideal trajectory response path to correct the exposure adjustment direction and response timing.

[0037] A feedforward constraint field is generated based on the phase conjugate correction vector. An exposure restriction zone is constructed in advance during the rebound risk period. By limiting the exposure change amplitude, simulating the gain direction, and adjusting the start time, smooth adjustment and oscillation suppression of exposure control are achieved.

[0038] Preferably, when generating the feedforward constraint field, the exposure time variation is limited to within 120% of the previous time slice, the analog gain direction reversal is prohibited within consecutive frames, and the exposure adjustment start time is delayed to fifty milliseconds after the target trajectory enters the stable section, so as to avoid image flickering and brightness jump caused by control response overshoot.

[0039] Preferably, based on the feedforward constraint field to generate a time-reversal closed-loop instruction set, and combined with frequency misalignment, time-stamping, and amplitude-limiting write-back strategies, the adaptive rearrangement steps for exposure allocation and induced information updates are as follows:

[0040] Based on the feedforward constraint field, a time-inverted conformal closed-loop instruction set is generated. Each control instruction in the instruction set includes the parameter adjustment direction, start time, action time window and maximum change amplitude. The trajectory fitting confidence is used as a weighting factor to complete error compensation.

[0041] Based on the closed-loop instruction set, the golden ratio frequency misalignment traction and double mirror time scale traction strategy are implemented to rearrange the frequency and generate symmetrical buffers for all control timing nodes, so as to improve the elastic adjustment capability and anti-interference stability of the exposure control curve.

[0042] During the instruction execution phase, a pulse-level amplitude limiting write-back mechanism is applied to limit the parameter adjustment range of each frame, and the update rhythm and content priority of the induced information are synchronously controlled based on the amplitude limiting results to achieve dynamic consistency recovery of the sensing link.

[0043] A smart control system based on smart signs includes a time-scaled coupling analysis module, a fault anchor point identification module, a counterfactual evaluation module, a cross-modal compensation module, a feedforward correction control module, and a closed-loop control execution module.

[0044] The time-scaled coupling analysis module establishes a unified time-scaled baseline, collects exposure sequences, saturation threshold sequences, and trajectory residuals, and generates an exposure-trajectory coupling spectrum for calibrating the rain and snow trigger time window;

[0045] The fault anchor point identification module calculates the phase residual mapping under the constraint of exposure-trajectory coupling spectrum, identifies the root cause of exposure oscillation, extracts the saturation source region and trajectory breakpoint set, and generates fault anchor points.

[0046] The counterfactual assessment module performs counterfactual playback based on fault anchor points, reconstructs the incident, reflection, and scattering paths of rain and snow, calculates the exposure threshold and trajectory continuous surface, and outputs the misjudgment risk coefficient.

[0047] The cross-modal compensation module constructs a cross-modal compensation chain based on the misjudgment risk coefficient, integrates millimeter-wave echo signals, thermal imaging signals and ground magnetic detection signals, and generates a robust trajectory proxy to restore trajectory continuity;

[0048] The feedforward correction control module inputs the robust trajectory proxy into the exposure-trajectory joint correction stage, calculates the phase conjugate correction vector, and establishes a feedforward constraint field to limit exposure bounce and suppress oscillation regeneration.

[0049] The closed-loop control execution module generates a time-reversal closed-loop instruction set based on the feedforward constraint field, and combines frequency misdirection traction, time-scale traction and amplitude limiting write-back strategies to adaptively rearrange exposure allocation and induced information updates.

[0050] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0051] This invention achieves precise identification of interference trigger windows by introducing nanosecond-level time-scaled baselines and exposure-trajectory coupling spectra. Furthermore, based on phase residual mapping and fault anchor point construction, it establishes causal tracing capabilities for interference sources. Combining counterfactual playback to reconstruct rain and snow optical paths and trajectory continuity surfaces improves the accuracy of risk assessment. On this basis, by constructing a cross-modal compensation chain and fusing millimeter-wave echo signals, thermal imaging contrast signals, and ground magnetic detection signals, a robust trajectory proxy is generated, enabling continuous tracking of key moving targets. Further introduction of phase conjugate correction and feedforward constraint mechanisms actively suppresses bounce and oscillation regeneration in exposure control, improving the dynamic stability of the visual link. Finally, through a time-reversal conformal closed-loop control instruction set and superimposed multiple control strategies, it achieves adaptive rearrangement of exposure control parameters and guidance information output content, enabling smart signs to maintain high stability, high precision, and high robustness in traffic perception and guidance under complex weather conditions, thereby improving the intelligent response capability and safety assurance level of road traffic systems. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0053] Figure 1 This is a flowchart of a smart control method based on smart signs according to the present invention.

[0054] Figure 2 This is a schematic diagram of a module of a smart control system based on smart signs according to the present invention. Detailed Implementation

[0055] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0056] This invention provides, for example Figure 1 The illustrated intelligent control method based on smart signs includes the following steps:

[0057] Establish a nanosecond-level unified time-scale baseline, collect exposure sequences, saturation threshold sequences and trajectory residuals, generate an exposure-trajectory coupling spectrum, and calibrate the rain and snow triggering time window based on the exposure-trajectory coupling spectrum;

[0058] To achieve accurate identification and dynamic compensation for visual perception failures in rainy and snowy environments, it is necessary to first establish a high-precision time reference, collect key feature data, construct a coupling spectrum, and calibrate the trigger window. Each step focuses on time synchronization and data consistency, ensuring that illumination, exposure, and target motion information are dynamically matched under the same time reference, thus providing a reliable data foundation for subsequent correction and control. The specific steps are as follows:

[0059] In the initial stage of operation, to ensure consistent time reference for data collected from different sensing units, a nanosecond-level unified time-baseline needs to be established. This baseline uses a high-stability rubidium atomic clock as its core time source, forming a two-layer timing calibration structure through a phase-locked loop and a high-frequency temperature-compensated crystal oscillator. The baseline signal is transmitted via an optical fiber synchronization network to each front-end sensing node, including a high-resolution camera assembly installed within the smart signage body, an embedded light intensity monitor, a raindrop density sensor, and a vehicle trajectory analysis terminal. Upon receiving the time-base signal, each node clears its local sampling timing register and writes the current baseline time code, ensuring that the timestamp accuracy of subsequent sampling data is controlled within ±5 nanoseconds. This operation eliminates the time drift phenomenon existing between the camera acquisition end and the trajectory analysis end in the prior art, ensuring complete alignment of exposure changes, environmental disturbances, and target motion trajectories in the time domain. To verify the stability of the time-baseline, the system triggers a self-calibration pulse every 60 seconds, detecting fiber optic propagation delay and automatically adjusting the clock offset, ensuring that the entire network remains synchronized under conditions of temperature fluctuations and electromagnetic interference.

[0060] After establishing a unified time-stamped baseline, the multi-source data synchronous acquisition phase is immediately initiated. Exposure sequence acquisition is performed by the exposure control circuit in the imaging acquisition component. This circuit records the electronic integration time, charge conversion voltage, and output gain of the photosensitive array in real time, and writes the exposure parameters of each frame image along with a timestamp into the data buffer. The generation of the saturation threshold sequence depends on the extreme value distribution statistics of pixel grayscale in each frame image. Therefore, the system performs a regionalized grayscale scan on each image, marking the number of pixels with grayscale values ​​exceeding 255×0.98 as overexposed and the number of pixels with grayscale values ​​below 255×0.02 as underexposed, and calculates the frame-level saturation threshold through linear weighting. Trajectory residual extraction is based on the spatial difference between the vehicle motion trajectory coordinates identified by the front-end target recognition unit and the predicted trajectory of the previous frame. Specifically, the system uses the inter-frame time interval as the calculation unit to calculate the distance difference between the current centroid coordinates of the target vehicle and the corresponding point of the predicted trajectory in the previous frame, obtaining the trajectory offset value. All exposure parameters, saturation thresholds, and trajectory offset values ​​are arranged into a sequence with a unified timestamp, forming three sets of synchronized datasets: the exposure sequence, the saturation threshold sequence, and the trajectory residual sequence. This synchronized acquisition process ensures that illumination response, brightness saturation, and motion changes correspond precisely at the same time reference, eliminating the feature misalignment problem caused by the delay in multi-source signal acquisition in traditional methods.

[0061] After obtaining three sets of time-aligned data sequences, the generation stage of the exposure-trajectory coupling spectrum begins. First, the exposure time, charge, and illumination intensity variation curves for each frame in the exposure sequence are normalized to allow for comparison of exposure responses across different frames on a uniform scale. Then, the normalized exposure sequence and the saturation threshold sequence are cross-paired according to their time indices to calculate the brightness saturation change rate corresponding to exposure fluctuations. Next, energy correlation analysis is performed between this brightness change rate and the trajectory residual amplitude at the corresponding time. By comparing the synchronicity of the exposure response amplitude and the trajectory offset amplitude, a correlation strength model between illumination disturbances and motion deviations is established. By sliding the calculation of this correlation strength across continuous time slices, a coupling strength curve varying over time is obtained. When the exposure change amplitude increases significantly and the trajectory offset increases synchronously within a certain time period, that time period is marked as a high coupling region. Furthermore, the energy density, duration, and illumination change rate of the high coupling region are fused to generate a three-dimensional spectral distribution map, i.e., the exposure-trajectory coupling spectrum. This spectrum not only reflects the intensity of the influence of ambient light changes on the stability of the target trajectory but also reveals the coupling law between illumination anomalies and motion anomalies in the time domain. The generation of this spectrum can visually present the exposure oscillations and trajectory instability areas caused by rain and snow interference, providing a reliable basis for subsequent identification.

[0062] After obtaining the complete exposure-trajectory coupled spectrum, the rain / snow trigger time window is calibrated. First, time periods with significantly higher energy density than the baseline energy in the spectrum are detected, and cluster analysis is performed on continuous high-energy segments, identifying energy surge areas as potential environmental interference candidate areas. Then, synchronous data from a raindrop density sensor and an air humidity detector are introduced to compare the rate of humidity change and raindrop reflection intensity within the candidate areas. When the humidity rises by more than 5% within 10 seconds and the raindrop reflection intensity significantly increases, while the coupled spectrum energy remains more than three times higher than the baseline energy, the start time of the rain / snow trigger event is confirmed. Next, the system performs linear extension of the preceding and following boundaries of this high-energy segment, ensuring the trigger window covers the entire illumination disturbance and trajectory fluctuation process. This joint verification method accurately calibrates the start and end times of rain / snow events, ensuring the completeness of environmental interference identification. After calibration, the trigger time window is recorded as input conditions for subsequent dynamic control processes, used to trigger exposure adjustment, trajectory correction, and information update operations. This method enables the system to provide early warnings before rain and snow arrive, and maintain the continuity of visual acquisition and the stability of trajectory analysis when disturbances occur, effectively solving the problems of response lag and misidentification that occur in traditional visual control methods under complex weather conditions.

[0063] Under the constraint of exposure-trajectory coupling spectrum, the phase residual mapping is calculated to identify the root cause of exposure oscillation, extract the saturation source region and trajectory breakpoint set, and generate accurate fault anchor points;

[0064] After constructing the exposure-trajectory coupling spectrum and determining the rain / snow triggering time window, to further track the true causes and spatial extent of visual perception failure, it is necessary to extract phase residual features from the spectrum, identify sources of illumination interference, calibrate brightness saturation regions and trajectory interruption nodes, and finally output spatial positioning anchor points that can be used for perception correction. The specific steps are as follows:

[0065] This step is based on the generated exposure-trajectory coupling spectrum, which records the synchronization intensity of exposure response and target motion at each time point. Based on this, time segments in the spectrum where the energy value is significantly higher than three times the mean and continues for more than one second are selected as the focus of exposure oscillation analysis. For these time segments, exposure time, charge accumulation, and average image brightness are extracted frame by frame from the original image, and combined with the target vehicle trajectory center coordinates extracted in the same frame to form a composite dataset with a strict time index. Subsequently, the phase sequence of exposure response is formed by calculating the direction of change and acceleration / deceleration rate of exposure parameters between consecutive frames; simultaneously, the trend of change in trajectory center displacement velocity is calculated to construct the phase sequence of motion response. By synchronously comparing the exposure phase sequence and the trajectory phase sequence, their relative residual values ​​at each time point are extracted. This phase residual reflects the degree of lag and interference intensity of illumination fluctuations on motion recognition results in the perception chain. All residual values ​​are plotted as a two-dimensional surface diagram with time as the horizontal axis, which is the phase residual mapping map, used to indicate whether illumination changes may be the main cause of trajectory anomalies within a certain time window.

[0066] After constructing the phase residual mapping map, the residual peak positions and their corresponding image frame contents are analyzed one by one. For each residual extreme point, the corresponding image data is extracted and converted into a grayscale matrix and contrast distribution map. First, the grayscale matrix is ​​searched for regions where the brightness value is close to saturation. Saturation is defined as a pixel grayscale value exceeding 250 and the area of ​​the region exceeding 5% of the total image area. If the region is found to be concentrated in the image center, with sharp edges and high brightness, it can be initially judged as overexposure caused by specular reflection. Second, the contrast distribution of the frame is observed to see if there is a general decreasing trend. If the contrast of the entire image decreases by more than 20%, and the exposure parameters (such as integration time) are sharply lengthened at the same time, this phenomenon usually indicates that there is uniform light field disturbance caused by dense raindrops or snow in the environment. In addition, the humidity value collected by the environmental sensor and the density of droplets in the air are further compared. If the humidity jumps by more than 4% at this time point and the sampling density of the raindrop counter increases by more than 30% per unit time, it further confirms that the exposure anomaly is caused by rain or snow. By employing frame-by-frame verification, feature extraction, and cross-validation with environmental data, the peak value of the phase residual can be matched one-to-one with the specific physical cause, thus clarifying the direct triggering mechanism of exposure oscillation. Unlike existing methods that infer the cause of interference solely based on image contrast degradation, this method combines changes in optical parameters with motion trajectory offset, supplemented by quantitative analysis of environmental factors, to achieve multi-dimensional localization and classification of the cause.

[0067] After confirming the specific root cause of the lighting anomaly, to calibrate the actual spatial regions and temporal points of interference in the perception chain, it is necessary to simultaneously identify the brightness saturation regions and the target trajectory interruption locations in the image. The specific execution method is as follows: For all image frames corresponding to exposure abrupt change time points, perform pixel-level traversal, dividing them into 64×64 image block regions. Within each region, count the percentage of pixels with a grayscale value greater than 248. If the percentage exceeds 20%, the region is determined to be a brightness saturation source. The spatial positions of all image block regions meeting this criterion are recorded as two-dimensional coordinate points, forming a set of saturation source region coordinates. On the other hand, in the vehicle motion trajectory time series, if the vehicle's centroid displacement is less than the image resolution threshold (e.g., 1 pixel) for any three consecutive frames, or if there is a discontinuous abrupt change in instantaneous velocity calculation (e.g., jumping from a positive value to a negative value), the target trajectory is considered interrupted. Based on this, the changes in exposure parameters, saturation, and contrast within each second before and after the interruption are statistically analyzed to verify whether there is a corresponding exposure anomaly or image saturation phenomenon. Finally, all trajectory breakpoints meeting the conditions are unified into a trajectory breakpoint set, and spatially overlapped with the saturation source region coordinate set. This two-way verification operation can not only confirm the actual location of the interruption in the perception chain, but also effectively eliminate false interruptions caused by non-environmental factors such as target occlusion and algorithm recognition fluctuations.

[0068] After obtaining the set of coordinates of the saturated source region and the set of trajectory breakpoints, these data are further fused to generate spatial positioning anchor points for control response. The merging process employs three matching criteria: First, the anchor point must be located within a time period where the residual value in the phase residual map exceeds a threshold (e.g., a 30° phase difference); second, the distance between the anchor point's spatial coordinates and the center point of the saturated source region must be less than a set error range (e.g., 5 pixels); third, the difference between the anchor point's timestamp and the trajectory breakpoint's timestamp must not exceed 200 milliseconds. Spatiotemporal points meeting these three conditions are marked as valid anchor points. Each anchor point includes exposure control parameter variation data, a saturated region location index, target loss start and end times, phase residual amplitude, and its spatial distribution curve. All valid anchor points are sequentially stored in a readable database as the basic input for subsequent counterfactual reconstruction, trajectory repair, and control command correction.

[0069] Based on the fault anchor point, perform counterfactual playback to reconstruct the rain and snow incident path, reflection path and scattering path, calculate the exposure threshold and trajectory continuous surface, and output the misjudgment risk coefficient.

[0070] After completing the exposure oscillation identification and fault anchor point generation, in order to further accurately identify the optical environment mechanism behind the perception interruption and assess the risk of trajectory recognition error caused by the interruption of the perception chain, the backtracking reconstruction and risk quantification of the causal chain between light environment interference and target motion anomaly were realized through multi-path counterfactual light field reconstruction, target trajectory continuity restoration and judgment intensity assessment based on fault anchor points.

[0071] The generated fault anchor points already contain key attributes such as timestamps, spatial coordinates, exposure parameter perturbation information, image saturation positions, and trajectory interruption intervals. To initiate the counterfactual playback operation, points meeting the following conditions must first be selected from the anchor point set as the reconstruction starting point: 1) located in the middle of a continuous fault segment in the time series; 2) the trajectory interruption time is greater than 500 milliseconds; and 3) spatially located within 30% of the image's central region, ensuring sufficient preceding and following image data and high-quality trajectory description. After selecting the starting anchor point, image sequences, trajectory data, and exposure control records within two seconds before and after that time point are extracted to form the original dataset for reconstruction. This dataset will serve as the initial boundary conditions for the counterfactual simulation, carrying information on the source of illumination changes and target behavior characteristics in a rain and snow disturbance environment.

[0072] After extracting the initial dataset, the light propagation path leading to image saturation is reconstructed by analyzing the expansion trend of illuminated areas and changes in grayscale distribution frame by frame. This operation consists of three sub-steps:

[0073] Incident path reconstruction: In each frame, the edges of regions with abrupt changes in grayscale are selected, and the direction of the light source causing the brightness increase in that region is deduced by combining the integration time of the photosensitive element recorded by the exposure controller. By analyzing the direction and angle shift of pixel brightness increase, the perturbation effect of raindrops or snowflakes on the incident angle of the light source in the external environment is determined.

[0074] Reflection path tracing: For areas with high-intensity bright spots in the image, extract their edge shape and halo diffusion trajectory, and invert the material surfaces that may reflect based on the principle of specular reflection, such as road surface water, vehicle surface or sign film, and mark the possible reflection areas in the form of spatial coordinates.

[0075] Scattering path analysis: In areas where the overall brightness of the image background increases and the contrast decreases, optical diffusion characteristics are analyzed to determine whether there is large-area scattering from water droplets or snow particles in the air. Based on the degree of brightness blurring and the extent of edge softening, the main scattering directions and intensity regions are calculated.

[0076] The incident path, reflection path, and scattering path will be integrated into a single light propagation diagram and mapped to the actual image coordinate system for subsequent pixel energy correction and trajectory continuity evaluation.

[0077] After obtaining the three types of light propagation paths, the maximum exposure response intensity at each location under actual light field perturbation is calculated by quantifying the energy superposition density per unit area of ​​the image pixel. Specifically, for each pixel location, its brightness change curve is summarized over time and compared with the normal response range of that pixel under standard lighting conditions. If the actual brightness increase slope is more than three times that under standard conditions, the point is determined to have exceeded the effective exposure control range. Energy weights are assigned based on the density of the intersection region between the incident and reflected paths at that point to correct the upper limit of the exposure threshold for the region where the pixel is located.

[0078] Finally, the entire image is divided into a 128×128 pixel grid. Within each grid, the deviation of the average exposure energy per unit time from the exposure threshold is calculated, generating an exposure threshold distribution map. This map visually represents the overload level of the visual sensing device at different spatial locations under abnormal lighting conditions, helping to identify which areas lose their effective recognition capability due to lighting disturbances.

[0079] After understanding the light energy load in each region of the image, and combining the trajectory data before and after the anchor point, a completion and reconstruction operation is performed on the broken sections of the target vehicle's trajectory. The operation process is as follows: between the last valid position before the trajectory interruption and the first stable position after the trajectory is restored, based on the target vehicle's speed, acceleration, turning radius, and path direction trend, a sequence of continuous positions that it may pass through is estimated by fitting. At the same time, the exposure threshold distribution map is used to determine whether the fitted path crosses a high-risk area. If the average exposure energy of the crossed area exceeds the upper limit of the threshold, the fitted path segment is marked as a low-confidence area; otherwise, it is marked as a high-confidence interval.

[0080] This method visualizes the completed trajectory segment as a continuous surface, with a confidence level for each surface segment. The resulting continuous trajectory surface describes the full picture of the possible actual motion path when target recognition is interrupted, and also indicates the distribution of the probability of visual perception recovery. Unlike existing techniques that rely on simple linear interpolation to reconstruct trajectories, this method not only considers the dynamics of vehicle motion but also incorporates the intensity of illumination disturbances, effectively improving the environmental adaptability and reliability of trajectory reconstruction.

[0081] After obtaining the exposure threshold distribution map and the trajectory continuous surface, the risk level of perceptual misjudgment is calculated by combining the spatial intersection areas of the two. Specifically, all low-confidence segments are extracted from the trajectory continuous surface, and their proportion of the total reconstructed trajectory is calculated. Simultaneously, the degree of exposure overload and the time span within each segment are evaluated, ultimately generating a set of risk level indicators. The risk coefficient corresponding to each anchor point includes the following three components: first, the proportion of trajectory interruption duration; second, the proportion of high-exposure area crossing; and third, the lowest confidence value of the trajectory surface.

[0082] This misjudgment risk coefficient will serve as a key criterion for cross-modal compensation in subsequent steps, determining whether to replace and reconstruct the current target perception result. By introducing this coefficient, the intelligent perception platform, when faced with unstructured ambient light disturbances, no longer relies on a single image quality score, but instead integrates motion behavior and environmental changes, establishing a quantifiable cognitive error index system.

[0083] Based on the misjudgment risk coefficient, a cross-modal compensation chain is constructed to enhance the joint weight of millimeter-wave echo signals, thermal imaging contrast signals and ground magnetic detection signals, and to generate a robust trajectory proxy to restore trajectory continuity;

[0084] After quantifying the risk of misjudgment for visually perceived abnormal segments, to enhance the integrity and anti-interference capability of the target trajectory, a cross-modal data compensation mechanism needs to be constructed based on the numerical results of the misjudgment risk coefficient. This mechanism utilizes millimeter-wave echo, thermal imaging comparison images, and ground magnetic detection data for multi-source information joint analysis to generate a robust trajectory proxy for repairing missing segments of the visual trajectory. The specific steps are as follows:

[0085] In the previous steps, the misjudgment risk coefficient corresponding to each fault anchor point has been calculated, including three core indicators: trajectory interruption duration, illumination disturbance intensity, exposure saturation area ratio, and trajectory fitting confidence. In this step, the risk coefficient is used as the main control variable to establish the priority ranking of sensing signals. The specific method is as follows:

[0086] When the misjudgment risk coefficient is lower than the set threshold (e.g., 0.3), it indicates that the visual failure has a minor impact. Only the thermal imaging contrast channel is enabled because it can penetrate rain and fog and identify heat sources without relying on ambient light.

[0087] When the misjudgment risk coefficient is in the medium range (e.g., 0.3 to 0.6), the millimeter wave echo channel is activated and processed in conjunction with the thermal imaging channel. The millimeter wave can penetrate non-metallic materials to identify the vehicle edge structure under conditions such as night, strong light, and obstruction.

[0088] When the misjudgment risk coefficient exceeds the high-risk threshold (e.g., greater than 0.6), it indicates that visual perception is severely interrupted and accompanied by optical saturation. The ground magnetic detection channel needs to be activated simultaneously. This channel relies on the disturbance of magnetic flux by the vehicle's metal structure to identify targets. It is particularly suitable for solving the problem of visual and thermal imaging failure when the target is completely stationary or in a slow-moving state.

[0089] After selecting different sensing channels based on risk levels, a fusion weight is defined for each channel. Millimeter-wave channel weights prioritize signal reflection intensity and distance clarity; thermal imaging weights are allocated based on the edge clarity of the heat source image and the significance of the center temperature difference; and geomagnetic channel weights are determined based on the relative rate of change of magnetic flux disturbance and signal continuity. Within each time slice, the system automatically calculates the weight values ​​for the three channels according to the above strategies and writes them into a time alignment index table, forming a dynamic channel response configuration matrix, laying the foundation for the next step of data fusion.

[0090] Based on the weighted activation results, the feature data extraction stage begins. First, the coordinates of continuous reflection points of the target in the time series are extracted from the millimeter-wave echo data, and the target echo intensity, arrival time, and relative speed are labeled. Millimeter-wave reflection features are particularly suitable for identifying vehicle contour edges and movement directions, and are a key basis for judging acceleration / deceleration behavior and lane changes.

[0091] Secondly, the temperature distribution changes between frames are extracted from the thermal imaging data. The foreground image of the heat source is extracted through differential processing, and the direction of heat source movement is identified using a boundary fitting algorithm. The center point and edge contour of the target heat source are recorded. In the case of multiple heat sources, the continuity of the target's thermal trajectory before and after is introduced as a constraint to ensure that the matched entity is the same.

[0092] Then, the sampling results of each magnetic sensor are read from the ground magnetic flux data to analyze the vehicle's entry, departure, and stationary states. When a vehicle enters above a magnetic sensor, the magnetic flux value will show a significant dip; by analyzing the magnetic flux descent time, amplitude, and recovery pattern, the sequence of the vehicle's passage and the changes in wheel axle spacing can be deduced, thus obtaining the target's movement trajectory on the fixed sensor array.

[0093] All three types of data have independent timestamps and spatial locations, but differ in their acquisition origin, resolution, and refresh rate. To achieve unified processing, spatiotemporal alignment is necessary. Time alignment uses a nanosecond-level timescale alignment benchmark, aligning the data from each channel to a frame-level temporal granularity within ±2 seconds of the visual anchor point. Spatial coordinates are uniformly represented using an image field-of-view projection model, mapping millimeter-wave polar coordinates, thermal imaging thermal coordinates, and geomagnetic array coordinates to a real-world coordinate system, converting them into a two-dimensional top-down plane coordinate system with the signboard as the origin. In this way, the three sensing results for each time slice will have a unified display basis on the same trajectory map.

[0094] After achieving unified coordinate alignment, trajectory fusion is performed frame by frame. Within each time slice, the channel data with the highest confidence level is selected as the primary data source for the trajectory points of that frame, based on the channel weight matrix. For example, if the thermal imaging image in a frame loses its heat source outline due to strong interference, but the millimeter-wave echo reflection is stable and the echo amplitude is higher than a set threshold, then the millimeter-wave coordinates are selected as the primary trajectory points for that frame. If the thermal imaging boundary is clear and the temperature difference structure is good, then thermal imaging data is the preferred source. Geomagnetic data is mainly used to supplement data in low-speed segments, stationary segments, or other time periods where channels are missing, filling in trajectory points in segments without a primary data source.

[0095] After all time-series trajectory points are selected, trajectory path continuity fitting is performed. The fitting method comprehensively considers the target's velocity direction, acceleration, and path trend before the visual interruption, and uses non-visual trajectory points as control points to construct a Bézier curve or polynomial interpolation path. Simultaneously, during the fitting process, a fusion confidence level is calculated for each path segment. If a trajectory segment is continuously provided by both thermal imaging and millimeter-wave data, the confidence level is high; if it consists entirely of geomagnetic points, the confidence level is low.

[0096] The final generated trajectory proxy includes: a time-series point set, two-dimensional spatial coordinates, channel source records, and fusion confidence level. This proxy path will be written into the trajectory repair buffer area and replace the original visual trajectory interruption area, achieving trajectory continuity restoration. This compensation path will serve as a core input in the subsequent exposure-trajectory joint correction stage during the phase conjugate vector generation process.

[0097] The robust trajectory proxy is input into the exposure-trajectory joint correction stage to calculate the phase conjugate correction vector and establish a feedforward constraint field to limit the exposure bounce slope and suppress the regeneration of exposure oscillation.

[0098] After obtaining a robust trajectory proxy, to address the issues of excessive bounce slope and oscillation regeneration in exposure control, the trajectory proxy needs to be introduced into the joint exposure-trajectory correction stage. This is achieved by establishing a high-resolution phase conjugate correction vector field and generating a time-precursor-based feedforward constraint mechanism, thus enabling predictive constraint and active smoothing control of exposure adjustment behavior. The specific steps are as follows:

[0099] The robust trajectory proxy generated in the previous stage is a set of time-series trajectory points derived from non-visual channel fusion. Each point contains the target's absolute position in the two-dimensional plane, the weight assignment from each sensor channel, the trajectory fitting confidence, and timestamp information. Meanwhile, the visual sensor unit has recorded a continuous sequence of exposure parameters, including the exposure time, charge accumulation, analog gain value, and dynamic adjustment frequency of the exposure controller for each frame.

[0100] In this step, these two types of data need to be aligned with high precision in time. Specifically, based on a unified nanosecond-level timescale, a robust trajectory proxy position index is established at each time point, and a binding relationship is established between the index and the corresponding image frame exposure parameters. To address microsecond-level synchronization deviations between devices, an interpolation alignment strategy is introduced to ensure that the exposure parameters of each image frame can match the nearest neighbor trajectory point, and the distance difference and time offset are calculated as trajectory accuracy evaluation factors.

[0101] The established joint reference structure is a two-dimensional table structure, with each row corresponding to a time point. Columns include indicators such as target spatial location, trajectory motion direction change, trajectory velocity, exposure response change rate, and the difference between the exposure parameters and those of the previous frame. This provides complete data support for subsequent judgment of the coupling between exposure dynamics and target motion trends. This structure will be used to derive the evolution trend of the phase difference between trajectory and exposure changes, and to provide the original variables for constructing the phase conjugate vector.

[0102] After achieving time synchronization and data fusion, the phase difference between target motion and exposure response is calculated. The phase difference here refers to the lag and fluctuation amplitude of the response on the time axis between changes in target motion direction and adjustments to image exposure time. Therefore, at each pair of time points, the angle change between the target trajectory velocity vector and the derivative vector of the exposure parameters is calculated. If their directions are consistent, the control response is good; if a phase shift exceeds 90 degrees, there is a risk of lag; if the shift exceeds 180 degrees, it indicates that the system has entered the reverse oscillation zone.

[0103] The phase differences of all time points are plotted as a continuous distribution map, forming an exposure-track phase difference mapping surface. A phase reversal segment is selected within this surface, and an ideal track response path is constructed as a reference line. The vector difference between the actual exposure response path and this reference line constitutes the phase conjugate correction vector. This vector includes three components: direction correction, response time delay, and a suggested response speed value. These components guide the exposure controller in adjusting the direction, timing, and magnitude of exposure parameter changes within that time period.

[0104] Compared to existing technologies that rely solely on passive adjustment strategies based on changes in image brightness, the phase conjugate vector constructed in this method takes the target's actual behavioral trajectory as its core and the target's dynamic change trend as the dominant factor. This method possesses stronger foresight and adaptive adjustment capabilities, and can effectively identify the causes of unstable exposure behavior and intervene in advance.

[0105] After constructing the phase conjugate correction vector, the process moves to the generation stage of the feedforward constraint mechanism. This mechanism is used to proactively limit the controller's response range based on the expected deviation between the target trajectory prediction and the exposure adjustment behavior before the controller adjusts the exposure parameters, thereby preventing image flickering, brightness jumps, or exposure oscillation regeneration problems caused by over-adjustment or over-adjustment.

[0106] The specific implementation method is as follows: First, select several time slices with the highest risk of backslip from the phase conjugate vector set. These are typically characterized by the exposure time adjustment direction and trajectory direction changing in opposite directions simultaneously, and the exposure parameter jump exceeding 20% ​​of the previous frame. Then, backtrack 300 to 500 milliseconds from these time slices, and construct an exposure restriction zone in advance within the backtracking section based on the target acceleration and predicted direction.

[0107] Adjustments to control parameters within the exposure restriction zone will be subject to three conditions: first, the exposure time change must not exceed 1.2 times that of the previous time slice; second, the analog gain must not undergo a directional change between two consecutive frames (i.e., "rise-fall-rise" type changes are not allowed); and third, the adjustment start time should be delayed until 50 milliseconds after the trajectory enters the stable segment. All adjustments that meet the above conditions will be permitted; those that exceed these conditions will be locked as "high-risk adjustments" and their execution will be delayed.

[0108] Throughout the execution of the feedforward constraint, the controller determines whether to enter the suppression state based on the presence of a high conjugate correction vector value within the current time window. If such a value exists, the controller employs an adjustment mechanism to smooth out conflict points. For example, when the target trajectory features continuous curves while the visual exposure control still adjusts the exposure value based on straight-line prediction, the feedforward mechanism will reduce the exposure adjustment magnitude in advance to prevent image saturation caused by changes in orientation.

[0109] The resulting feedforward constraint field will span the high-risk segment of the entire target trajectory. Its role is not only limited to reducing the instability of the current frame exposure adjustment, but more importantly, it prevents feedback overshoot caused by control lag and maintains the dynamic stability of the visual channel in a continuous interference environment.

[0110] Based on the feedforward constraint field generation time inversion conformal closed-loop instruction set, superimposed with the golden ratio frequency misalignment traction, dual mirror time scale traction and pulse-level amplitude limiting write-back strategy, the exposure allocation and induced information update are adaptively rearranged to achieve stable dynamic control of the visual perception system in rain and snow environments.

[0111] Based on the established feedforward constraint field using the trajectory phase conjugate vector, the rebound slope of the exposure control parameters is actively limited, and oscillation regeneration is suppressed. To transform the effect of this constraint field into an adaptive reconstruction capability of the actual control parameters and induced content in the vision link, a closed-loop instruction set with time reversal characteristics needs to be constructed, and three linked control strategies need to be superimposed—golden ratio frequency misalignment traction, dual-mirror time-scale traction, and pulse-level amplitude limiting write-back control—ultimately achieving steady-state adaptive dynamic control of exposure allocation and induced output within the perception link. The specific steps are as follows:

[0112] The feedforward constraint field clearly identifies the key interference periods in the visual perception link, including abrupt jumps in exposure parameters, rapid declines in trajectory confidence, and interruptions in the trajectory source signal. For these interference characteristics, a closed-loop control sequence needs to be constructed in reverse to form control logic capable of time-reverse evolution.

[0113] The specific steps include: taking each disturbance point as the center, tracing back 500 to 1000 milliseconds of trajectory evolution and exposure change history, extracting all exposure response parameters (including exposure duration, charge accumulation rate, and analog gain control quantity) and trajectory velocity, direction, and acceleration parameters in this segment; generating a target control sequence based on the ideal response state, that is, the coordinated change trajectory that the trajectory and exposure should exhibit under the assumption of no disturbance. This ideal state serves as the control target, forming a comparison error with the actual state, constituting the segment to be compensated.

[0114] Subsequently, using this error term as the core variable, a set of reverse control commands is generated. Each set of control commands includes the parameter adjustment direction, start time, action time window, and maximum allowable change range. Throughout the generation process, the trajectory fitting confidence level is consistently used as a weighting factor to ensure that the control commands more accurately match the trajectory behavior change trend.

[0115] The closed-loop instruction set is generated according to the control logic of "predicting the future - adjusting the present". All adjustment behaviors are implemented in advance before the disturbance occurs, forming a reverse scheduling path in time. Unlike the passive response method in traditional exposure control that relies on image frame feedback to start the adjustment mechanism, this closed-loop method is preventive, autonomous and targeted, and has clear non-obviousness and structural innovation.

[0116] After the closed-loop instruction set is constructed, the problems of concentrated frequency distribution of control parameters, steep execution curves, and insufficient system stability margin need to be addressed. To this end, this step introduces two continuous traction strategies to enable the control commands to have a flexible distribution structure and symmetrical adjustment capability during actual execution, thereby improving the flexibility of the control curve and enhancing anti-interference redundancy.

[0117] The first strategy is the golden ratio frequency redistribution. Its operation involves redistributing the frequencies of all time points in the closed-loop instruction set at a ratio of 1:0.618. This means that high-frequency response actions and low-frequency stabilization actions are interleaved at non-integer intervals. Specifically, in three consecutive exposure adjustment instructions, the interval between the first and second instructions is set to the main frame period, while the interval between the second and third instructions is set to the main frame period multiplied by 0.618, forming a staggered frequency structure. This structure effectively reduces the risk of feedback accumulation caused by concentrated parameter changes and lowers the frequency of pulse jitter in the link execution.

[0118] The second strategy is dual-mirror time-scale traction. The method involves generating a set of parameter adjustment curves in the time domains before and after the control command set, forming a buffer curve with a decreasing amplitude and opposite direction to the main control curve in terms of control trend. For example, if the exposure value in the main adjustment curve increases from 60 microseconds to 100 microseconds at a certain time, the first segment of the mirror curve will execute a transition curve that gradually increases to 60 microseconds around 50 microseconds, and the second segment will gradually decrease to 80 microseconds after 100 microseconds, remaining stable. This "dual cushioning" effectively prevents overshoot or instability caused by sudden jumps in the main adjustment curve.

[0119] The synergistic effect of these two strategies ensures that control commands not only respond promptly but also possess robust parameter variation buffering capabilities. In high-speed scenarios, extreme weather conditions, or continuous interference, they effectively mitigate internal feedback shocks, achieving both stability and continuity in the visual link control rhythm.

[0120] In the final stage of executing closed-loop instructions, to ensure that all adjustment behaviors do not cause abnormal jumps in exposure parameters due to sudden variables, it is necessary to constrain and control the maximum change amplitude during instruction execution through a pulse-level limiting strategy, and to synchronously adjust the update rhythm and content sorting of induced information according to the adjustment execution results, so as to achieve the final consistency restoration at the information level.

[0121] The clipping strategy works as follows: For each frame's exposure adjustment, the instantaneous change in exposure time between that frame and the previous frame is detected. If this change exceeds a set threshold (e.g., the exposure time per frame must not exceed a 20% increase), the change is marked as a pulse feature point, triggering the clipping write-back mechanism. The clipping mechanism reconstructs the parameter change path using linear interpolation within a range of three frames before and after this point, ensuring its slope matches the system's maximum allowed response gradient.

[0122] Simultaneously, based on the adjusted results after amplitude limiting, the perception stability index is reassessed and fed back to the guidance information distribution unit. Before the perception link stability is restored, the refresh of guidance information content in that area is paused to avoid incorrect guidance prompts due to image flickering or misidentification. Once the stability index exceeds the confidence threshold, content updates resume, and the order and priority of guidance prompts are rearranged based on the path direction predicted by the trajectory agent and the traffic flow density distribution. For example, when visual perception on the main road is interrupted, guidance content is migrated to the auxiliary road area in real time, and the traffic flow trend is determined by thermal imaging trajectory, prioritizing the display of speed limits, lane changes, and slow-down instructions along the path.

[0123] By combining the above-mentioned amplitude and rearrangement strategies, the system can not only achieve precise steady-state transition at the exposure control level, but also ensure the continuity, accuracy and dynamic consistency of the guidance content at the information output level, forming a complete response closed loop from the physical control layer to the traffic guidance layer.

[0124] This invention achieves precise identification of interference trigger windows by introducing nanosecond-level time-scaled baselines and exposure-trajectory coupling spectra. Furthermore, based on phase residual mapping and fault anchor point construction, it establishes causal tracing capabilities for interference sources. Combining counterfactual playback to reconstruct rain and snow optical paths and trajectory continuity surfaces improves the accuracy of risk assessment. On this basis, by constructing a cross-modal compensation chain and fusing millimeter-wave echo signals, thermal imaging contrast signals, and ground magnetic detection signals, a robust trajectory proxy is generated, enabling continuous tracking of key moving targets. Further introduction of phase conjugate correction and feedforward constraint mechanisms actively suppresses bounce and oscillation regeneration in exposure control, improving the dynamic stability of the visual link. Finally, through a time-reversal conformal closed-loop control instruction set and superimposed multiple control strategies, it achieves adaptive rearrangement of exposure control parameters and guidance information output content, enabling smart signs to maintain high stability, high precision, and high robustness in traffic perception and guidance under complex weather conditions, thereby improving the intelligent response capability and safety assurance level of road traffic systems.

[0125] The invention provides, for example Figure 2 The intelligent control system based on smart signs shown includes a time-scaled coupling analysis module, a fault anchor point identification module, a counterfactual evaluation module, a cross-modal compensation module, a feedforward correction control module, and a closed-loop control execution module.

[0126] The time-scaled coupling analysis module establishes a unified time-scaled baseline, collects exposure sequences, saturation threshold sequences, and trajectory residuals, and generates an exposure-trajectory coupling spectrum for calibrating the rain and snow trigger time window;

[0127] The fault anchor point identification module calculates the phase residual mapping under the constraint of exposure-trajectory coupling spectrum, identifies the root cause of exposure oscillation, extracts the saturation source region and trajectory breakpoint set, and generates fault anchor points.

[0128] The counterfactual assessment module performs counterfactual playback based on fault anchor points, reconstructs the incident, reflection, and scattering paths of rain and snow, calculates the exposure threshold and trajectory continuous surface, and outputs the misjudgment risk coefficient.

[0129] The cross-modal compensation module constructs a cross-modal compensation chain based on the misjudgment risk coefficient, integrates millimeter-wave echo signals, thermal imaging signals and ground magnetic detection signals, and generates a robust trajectory proxy to restore trajectory continuity;

[0130] The feedforward correction control module inputs the robust trajectory proxy into the exposure-trajectory joint correction stage, calculates the phase conjugate correction vector, and establishes a feedforward constraint field to limit exposure bounce and suppress oscillation regeneration.

[0131] The closed-loop control execution module generates a time-reversal closed-loop instruction set based on the feedforward constraint field. It combines frequency misdirection traction, time-scale traction, and amplitude limiting write-back strategies to adaptively rearrange exposure allocation and induced information updates, thereby achieving stable dynamic control of the visual perception system under rain and snow conditions.

[0132] The present invention provides a smart control method based on smart signs, which is implemented through the above-mentioned smart control system based on smart signs. For details of the specific method and process of the smart control system based on smart signs, please refer to the above-mentioned embodiment of the smart control method based on smart signs, which will not be repeated here.

[0133] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A smart control method based on smart signs, characterized in that, Includes the following steps: Establish a unified time-scale baseline, collect exposure sequences, saturation threshold sequences and trajectory residuals, and generate an exposure-trajectory coupling spectrum to calibrate the rain and snow trigger time window; Under the constraint of exposure-trajectory coupling spectrum, the phase residual mapping is calculated, the root cause of exposure oscillation is identified, the saturation source region and the set of trajectory breakpoints are extracted, and the fault anchor point is generated. Based on the fault anchor point, perform counterfactual playback to reconstruct the rain and snow incident path, reflection path and scattering path, calculate the exposure threshold and trajectory continuous surface, and output the misjudgment risk coefficient. Based on the misjudgment risk coefficient, a cross-modal compensation chain is constructed, which integrates millimeter-wave echo signals, thermal imaging signals and ground magnetic detection signals to generate a robust trajectory proxy to restore trajectory continuity. The robust trajectory proxy is input into the exposure-trajectory joint correction stage to calculate the phase conjugate correction vector and establish a feedforward constraint field to limit exposure bounce and suppress oscillation regeneration. Based on the feedforward constraint field to generate a time inversion closed-loop instruction set, and combined with frequency misdirection, time-stamping and amplitude limiting write-back strategies, the exposure allocation and induced information update are adaptively rearranged.

2. The intelligent control method based on intelligent signage according to claim 1, characterized in that, The steps for generating the exposure-trajectory coupled spectrum are as follows: A nanosecond-level unified time standard baseline is established, and the time base signal is synchronized to multiple sensing nodes through a high-stability time base source and fiber optic synchronization network. The sampling timing register is cleared and the time code is written. Under a unified time-stamped baseline, exposure sequence, saturation threshold sequence and trajectory residual data are collected. The image frame exposure parameters, the proportion of extreme values ​​of image grayscale distribution and the target trajectory offset distance are recorded respectively, and three sets of synchronized datasets are generated according to a unified timestamp. Normalization and energy correlation analysis were performed on three sets of synchronous datasets to calculate the synchronization intensity of exposure change and trajectory offset. The coupling intensity curve was generated by sliding over a continuous time period, and an exposure-trajectory coupling spectrum reflecting the correlation characteristics of environmental interference and motion anomaly was formed. Cluster analysis is performed on the energy density anomaly regions in the coupled spectrum, and raindrop density and humidity change data are integrated to determine the start time of rain and snow disturbance. The time boundary is extended to generate a rain and snow trigger time window, which serves as the input basis for subsequent control processes.

3. The intelligent control method based on intelligent signage according to claim 2, characterized in that, The steps for generating fault anchors are as follows: Under the constraint of exposure-trajectory coupling spectrum, the energy surge segment is extracted, the phase sequence of exposure response and target trajectory between consecutive frames is calculated, and a phase residual mapping map is generated. Based on the residual peak time point in the phase residual mapping map, combined with image brightness distribution, contrast change and environmental humidity data, the specific interference causes of exposure oscillation are identified. Perform regional grayscale traversal on the corresponding image frame, extract the coordinates of the brightness saturation source region, and combine the analysis of the interruption point with the vehicle trajectory sequence to form a set of saturation source regions and trajectory breakpoints; Based on the matching relationship between phase residual amplitude, saturation region location and trajectory interruption timing, effective fault anchor points are generated by fusion, providing spatial positioning reference for subsequent control.

4. The intelligent control method based on intelligent signage according to claim 3, characterized in that, The steps for outputting the risk factor for misjudgment are as follows: Based on the generated fault anchor points, anchor points that meet the conditions of time continuity, trajectory interruption duration and spatial location are selected, and image sequences, trajectory data and exposure records of the time periods before and after them are extracted to form a counterfactual playback dataset. Light propagation analysis was performed on the image sequence to reconstruct the incident, reflected, and scattered paths of rain and snow in sequence, thereby obtaining the propagation distribution of light disturbance in the image space. Pixel energy stacking density is calculated based on light propagation results, and an exposure threshold distribution map is generated to describe the degree of exposure overload in areas of abnormal lighting. By combining trajectory data, trajectory completion is performed in the interrupted interval to construct a continuous trajectory surface, and high-risk and low-risk paths are marked according to the exposure threshold distribution map; By combining the confidence level of the trajectory surface, the exposure overload ratio, and the trajectory interruption duration, a misjudgment risk coefficient is calculated, which is used as a basis for subsequent perception compensation and control decisions.

5. The intelligent control method based on intelligent signage according to claim 4, characterized in that, The steps for generating a robust trajectory proxy are as follows: The activation priority of the non-visual perception channel is determined by using the misjudgment risk coefficient as the control variable, and the fusion weight of each channel is assigned. A time-aligned index table is generated according to the weight. Trajectory features are extracted from millimeter-wave echo, thermal imaging, and ground magnetic detection by time alignment, and the coordinates of each channel are mapped to a unified ground coordinate system. The data points with the highest confidence are selected frame by frame according to the channel weight to form a time series point set. The time series point set is then used for path fitting and the fusion confidence level is labeled. A robust trajectory proxy is output to replace the visually interrupted trajectory.

6. The intelligent control method based on intelligent signage according to claim 5, characterized in that, The process of establishing the feedforward constraint field is as follows: By aligning the robust trajectory proxy with the exposure parameter sequence under a unified time scale, a joint reference structure is established for trajectory position, direction of motion, velocity, and rate of change of exposure response. The phase difference between target motion and exposure response is calculated based on the joint reference structure, forming an exposure-trajectory phase difference mapping surface. A phase conjugate correction vector is constructed according to the ideal trajectory response path to correct the exposure adjustment direction and response timing. A feedforward constraint field is generated based on the phase conjugate correction vector. An exposure restriction zone is constructed in advance during the rebound risk period. By limiting the exposure change amplitude, simulating the gain direction, and adjusting the start time, smooth adjustment and oscillation suppression of exposure control are achieved.

7. The intelligent control method based on intelligent signage according to claim 6, characterized in that, When generating the feedforward constraint field, the exposure time variation is limited to within 120% of the previous time slice, the analog gain direction reversal is prohibited within consecutive frames, and the exposure adjustment start time is delayed to fifty milliseconds after the target trajectory enters the stable section, in order to avoid image flickering and brightness jumps caused by control response overshoot.

8. The intelligent control method based on intelligent signage according to claim 6, characterized in that, Based on the feedforward constraint field generating time-inversion closed-loop instruction set, and combined with frequency misdirection, time-scale direction, and amplitude limiting write-back strategies, the adaptive rearrangement steps for exposure allocation and induced information updates are as follows: A time-inversion conformal closed-loop instruction set is generated based on a feedforward constraint field, and error compensation is completed using trajectory fitting confidence as a weighting factor. Based on the closed-loop instruction set, the golden ratio frequency misalignment traction and double mirror time scale traction strategy are implemented to rearrange the frequency and generate symmetrical buffers for all control timing nodes, so as to improve the elastic adjustment capability and anti-interference stability of the exposure control curve. During the instruction execution phase, a pulse-level amplitude limiting write-back mechanism is applied to limit the parameter adjustment range of each frame, and the update rhythm and content priority of the induced information are synchronously controlled based on the amplitude limiting results to achieve dynamic consistency recovery of the sensing link.

9. A smart control system based on smart signs, used to implement the smart control method based on smart signs as described in any one of claims 1-8, characterized in that, It includes a time-scaled coupling analysis module, a fault anchor point identification module, a counterfactual evaluation module, a cross-modal compensation module, a feedforward correction control module, and a closed-loop control execution module. The time-scaled coupling analysis module establishes a unified time-scaled baseline, collects exposure sequences, saturation threshold sequences, and trajectory residuals, and generates an exposure-trajectory coupling spectrum for calibrating the rain and snow trigger time window; The fault anchor point identification module calculates the phase residual mapping under the constraint of exposure-trajectory coupling spectrum, identifies the root cause of exposure oscillation, extracts the saturation source region and trajectory breakpoint set, and generates fault anchor points. The counterfactual assessment module performs counterfactual playback based on fault anchor points, reconstructs the incident, reflection, and scattering paths of rain and snow, calculates the exposure threshold and trajectory continuous surface, and outputs the misjudgment risk coefficient. The cross-modal compensation module constructs a cross-modal compensation chain based on the misjudgment risk coefficient, integrates millimeter-wave echo signals, thermal imaging signals and ground magnetic detection signals, and generates a robust trajectory proxy to restore trajectory continuity; The feedforward correction control module inputs the robust trajectory proxy into the exposure-trajectory joint correction stage, calculates the phase conjugate correction vector, and establishes a feedforward constraint field to limit exposure bounce and suppress oscillation regeneration. The closed-loop control execution module generates a time-reversal closed-loop instruction set based on the feedforward constraint field, and combines frequency misdirection traction, time-scale traction and amplitude limiting write-back strategies to adaptively rearrange exposure allocation and induced information updates.

Images