A low-light traffic monitoring method combined with nighttime enhancement algorithm

Abstract

The present invention discloses a low-light traffic monitoring method combined with a nighttime enhancement algorithm, specifically relating to the field of traffic image processing and monitoring. The method includes extracting a brightness disturbance factor group based on an image frame sequence of a traffic scene image, and determining whether the image frame enters an image enhancement path by performing disturbance amplitude judgment, masked area ratio judgment, and interference frame continuous length judgment. By constructing a disturbance-driven label judgment mechanism and a regional structure perception model, combined with a multi-path image enhancement strategy and a behavior consistency callback process, the method realizes the partition recognition and dynamic enhancement repair of structural dislocation, brightness strobing, and boundary blur in low-light traffic images, thereby effectively improving the structural coherence and target discernibility of nighttime monitoring images, thereby solving the problems of poor image continuity and poor structural restoration in current low-light scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic image processing and monitoring, and more particularly to a low-illumination traffic monitoring method combined with a nighttime enhancement algorithm. Background Art

[0002] In nighttime low-light traffic monitoring, due to the low ambient light, image acquisition equipment must rely on high sensitivity settings or long exposure times for fill light. However, this also makes it very easy for noise, reflection artifacts, and motion blur to be superimposed on the image, further amplifying the impact of light source flicker on image quality.

[0003] Especially in scenarios where PWM dimming is frequently used for LED headlights or streetlights, sudden brightness drops or image stripes may appear periodically in the video frame, disrupting the spatial continuity and temporal consistency of the image.

[0004] Existing general image processing technologies are mostly based on local contrast stretching or brightness equalization of single frames. These techniques are unable to detect structural dislocations and content defects caused by light intensity fluctuations. Instead, they mistakenly enhance stroboscopic shadow areas as normal images, resulting in blurred boundaries of traffic targets, interrupted motion trajectories, and even mis-segmentation into multiple independent targets.

[0005] Therefore, the present invention proposes a low-light traffic monitoring method combined with a nighttime enhancement algorithm to solve the problem that image continuity and image structure restoration in current low-light scenes are poor, which is not conducive to monitoring. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a low-light traffic monitoring method combined with a nighttime enhancement algorithm. By constructing a disturbance-driven label judgment mechanism and a regional structure perception model, combined with a multi-path image enhancement strategy and a behavior consistency callback process, it realizes the partition recognition and dynamic enhancement repair of structural dislocation, brightness stroboscopic and boundary blur in low-light traffic images, so as to effectively improve the structural coherence and target discernibility of nighttime monitoring images, thereby solving the problems of poor image continuity and poor structural restoration in current low-light scenes.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a low-light traffic monitoring method combined with a nighttime enhancement algorithm, comprising:

[0008] A brightness disturbance factor group is extracted based on the image frame sequence of the traffic scene image, and the disturbance amplitude, masked area ratio and interference frame continuous length are judged to determine whether the image frame enters the image enhancement path.

[0009] By performing structural feature extraction and joint attribute determination on each region in the image frame, an image processing label map is generated and a multi-path image enhancement strategy is bound;

[0010] Activate the image enhancement path based on the image processing label map, and perform regional structure restoration and brightness consistency restoration in combination with inter-frame neighborhoods;

[0011] Based on the image enhancement path processing results, the behavior preservation factor group of the traffic monitoring object is extracted to determine whether there is dynamic offset of the traffic target structure, and the original path execution strategy is called back based on the scoring results.

[0012] In a preferred embodiment, a traffic scene image is acquired and used as input. A frame segmentation operation is performed according to a timestamp sequence to construct an image frame sequence. The image frame sequence is input into a disturbance analysis process. The average brightness, edge brightness gradient, and frequency domain main peak frequency map of each frame in the image frame sequence are extracted. Inter-frame statistical processing is performed on the extracted results. A brightness texture consistency term, an edge dominant energy term, and a disturbance scale response adjustment term are calculated to generate a brightness disturbance factor group.

[0013] Perform inter-frame difference operation on the brightness perturbation factor group and calculate the perturbation amplitude trend curve to obtain the perturbation trend function of the frame group for dynamic perturbation pattern extraction;

[0014] The perturbation trend function is compared with the static benchmark trend model for similarity. If the perturbation amplitude is greater than the reference threshold, the frame group is determined to constitute a first-level perturbation and enters the preliminary screening candidate set.

[0015] Perform local "extremely dark, extremely bright" pixel ratio statistics on each frame in the initial screening candidate set, and determine whether the proportion of the masked area exceeds the structural masking threshold, thereby constructing the secondary perturbation condition;

[0016] If the proportion of the masked area reaches the preset threshold range, the continuous length of the interference frame is calculated to see whether it exceeds the time window threshold, which is used as the third-level disturbance condition. Finally, a disturbance start label map is generated to identify the image frame to be processed.

[0017] In a preferred embodiment, the image frames with the perturbation start label are divided into regions according to the perturbation start label map, and the edge connectivity, corner point density and directional gradient stability of each region are extracted to form a structural continuity index group;

[0018] Align the structural continuity index group with the perturbation activation label map and perform joint analysis to generate a perturbation-structure joint attribute map. The perturbation-structure joint attribute map is used to determine the processing level of the image region.

[0019] According to the disturbance amplitude and structural continuity index of each region in the disturbance-structure joint attribute graph, an image processing label is assigned to the image region, and the image processing labels include "high disturbance and high structure", "high disturbance and low structure", "low disturbance and high structure" and "low disturbance and low structure"; when the disturbance amplitude is greater than a preset disturbance amplitude threshold, it is judged as "high disturbance", otherwise it is "low disturbance"; when the comprehensive score of the structural continuity index is higher than a preset comprehensive score threshold, it is judged as "high structure", otherwise it is "low structure";

[0020] According to the image processing label, the preset image enhancement path is called, and the image area is bound to the contour migration reconstruction path, illumination compensation path, edge refinement path and basic filtering path respectively, and a path allocation map is established.

[0021] In a preferred embodiment, a contour migration reconstruction path is input to the image block marked as the "high disturbance high structure" image area, and gradient extension calculation is performed based on the edge direction map of the previous and next frames to generate a structure-filled image area;

[0022] The image blocks marked as "high disturbance low structure" are input into the illumination compensation path, and the brightness trend curve of adjacent frames is used to perform local illumination reconstruction to generate the brightness consistency compensation image area;

[0023] The image blocks marked as "low disturbance high structure" are input into the edge refinement path, and the boundary details are stretched by contrast-preserving local lifting filtering to generate clear boundary image areas;

[0024] The image blocks marked as "low disturbance low structure" image regions are input into the basic filtering path, and a stable reference image region is generated using a stable filter bank and participates in full image fusion.

[0025] In a preferred embodiment, after the image enhancement path is executed and the updated image frame sequence is output, a behavior preservation factor group of the traffic monitoring object is extracted from the updated image frame sequence. The behavior preservation factor group includes trajectory continuity, boundary jump rate, and contour stability score. The behavior preservation factor group is used to characterize the degree of influence of the image enhancement result on the dynamic consistency of the traffic target;

[0026] The behavior preservation factor group is compared with the reference behavior factor group of the traffic monitoring object extracted before the image enhancement path is executed, the interference response of the image enhancement to the traffic target structure is solved, and the corresponding behavior deviation score map is generated;

[0027] If the offset score of any area in the behavior offset score map exceeds the preset offset score threshold, the image processing label of the image block corresponding to the area is located and its originally configured image enhancement path is called back;

[0028] The compensation path parameters are called for the located image blocks, the enhancement process is re-executed and the original enhancement results are replaced, the structural consistency correction is completed and the updated enhanced image frame is output.

[0029] In a preferred embodiment, the peak value of the perturbation trend response is Construct a perturbation trend function:

[0030]

[0031] in time window for disturbance analysis; express A frame time index in ; is the brightness texture consistency item; is the edge dominant energy term; is the frequency domain regularization term; is the disturbance scale response adjustment item; Activate the mask item for masking.

[0032] In a preferred embodiment, an image processing label map is generated and label assignment of the image enhancement path is completed based on the disturbance initiation label map and the disturbance-structure joint attribute map constructed by the structural continuity index;

[0033]

[0034] in Image position The image enhancement path label map represents the image processing label map; is a label set, and the label set value is {high disturbance high structure, high disturbance low structure, low disturbance high structure, low disturbance low structure}; is the local disturbance intensity value; Enhancement items for structural features; is the structural complexity term; is the threshold response control function of the path label.

[0035] In a preferred embodiment, an image enhancement model 1 is constructed based on the contour migration reconstruction path:

[0036]

[0037] in Indicates that under the path "Contour Migration Reconstruction", the position The enhanced output pixel value of is the time offset variable; is the upper bound of time offset; is the lower bound of time offset; Indicates the time offset is Frame Middle, location The brightness trend consistency weight; Frame Middle position The original pixel value of Indicates that along the dominant edge direction The directional first derivative of the image intensity; is the local structure credibility factor;

[0038] Building image enhancement model 2 based on illumination compensation path:

[0039]

[0040] in Indicates that under the path "Light Compensation", the position The enhanced output pixel value of Frame Up, position The brightness trend consistency weight; is the illumination difference term;

[0041] Building image enhancement model 3 based on edge refinement path:

[0042]

[0043] in Indicates that under the path "edge refinement", the position The enhanced output pixel value of is the edge enhancement adjustment factor; represents the second-order directional derivative performed along the dominant structural direction; is the local contrast stretching factor;

[0044] Building image enhancement model 4 based on basic filtering path:

[0045]

[0046] in Indicates that under the path "Basic Filtering", the position The enhanced output pixel value of Indicates A local pixel neighborhood window constructed for the center; is a pixel position in the neighborhood window; is the steady-state filter response coefficient; is the total number of neighborhood pixels.

[0047] In a preferred embodiment, a path correction mechanism based on traffic behavior consistency feedback is constructed by using the deviation score of behavior consistency. Indicates Shift scores for behavioral consistency;

[0048]

[0049] in It is the factor for maintaining behavior after reinforcement; To enhance the pre-behavioral factor; A set of dimensions that hold factors for behavior; is the inter-frame change rate of the behavior factor; is the regularization coefficient; Represents the cumulative term of behavioral fluctuations in the time window; is the structural perturbation activation map.

[0050] Technical effects and advantages of the present invention:

[0051] 1. Improve the ability to identify the continuity of nighttime image structure. By constructing a brightness perturbation factor group and performing multi-level judgments on perturbation amplitude, occlusion ratio, and interference persistence, the system can identify segments in the frame sequence that require enhancement, effectively resolving image structure discontinuities and timing jumps caused by stroboscopic and blurring effects.

[0052] 2. Implements structural-based image enhancement path allocation. By combining a perturbation-initiated label map with a structural continuity index, a perturbation-structure attribute map is generated. Four categories of image processing labels are hierarchically constructed. Based on the differences in perturbation and structural distributions, image enhancement paths are automatically matched, improving the adaptability of enhancement strategies.

[0053] 3. Maintain the stability of traffic target boundary and outline information; four paths are used to process various image regions: outline migration reconstruction, illumination compensation, edge refinement, and basic filtering. This achieves partitioned enhancement of target boundaries, illumination consistency, and structural details, while preserving key target features.

[0054] 4. Provide a dynamic feedback mechanism for traffic behavior consistency; construct a behavior preservation factor group by extracting trajectory continuity, boundary jump rate and contour stability, and compare the offset scores of the results before and after enhancement to identify structural interference areas and trigger path callback corrections. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] Refer to the instruction manual Figure 1 According to an embodiment of the present invention, a low-light traffic monitoring method combined with a nighttime enhancement algorithm includes:

[0058] Based on the image frame sequence of the traffic scene image, a group of brightness disturbance factors is extracted. Then, by performing disturbance amplitude judgment, occlusion area ratio judgment, and interference frame continuous length judgment, the image frame is determined to enter the image enhancement path as the entry judgment mechanism of the processing link.

[0059] By performing structural feature extraction and joint attribute determination on each region in the image frame, an image processing label map is generated and a multi-path image enhancement strategy is bound;

[0060] Activate the image enhancement path based on the image processing label map, and perform regional structure restoration and brightness consistency restoration in combination with inter-frame neighborhoods;

[0061] Based on the image enhancement path processing results, the behavior maintenance factor group of the traffic monitoring object is extracted to determine whether there is dynamic offset in the traffic target structure. The original path execution strategy is called back based on the scoring results to complete the closed-loop correction of the enhancement results and ensure the consistency of traffic targets.

[0062] Obtain a traffic scene image and use it as input. Perform a frame segmentation operation according to the timestamp sequence to construct an image frame sequence. Input the image frame sequence into the disturbance analysis process, extract the average brightness, edge brightness gradient, and frequency domain main peak frequency map of each frame in the image frame sequence, and perform inter-frame statistical processing on the extracted results. Calculate the brightness texture consistency term, edge dominant energy term, and disturbance scale response adjustment term to generate a brightness disturbance factor group. The brightness disturbance factor group is used to characterize the significance of the low-light disturbance effect.

[0063] Perform inter-frame difference operation on the brightness perturbation factor group and calculate the perturbation amplitude trend curve to obtain the perturbation trend function of the frame group for dynamic perturbation pattern extraction;

[0064] The perturbation trend function is compared with the static benchmark trend model for similarity. If the perturbation amplitude is greater than the reference threshold, the frame group is determined to constitute a first-level perturbation and enters the preliminary screening candidate set.

[0065] Perform local "extremely dark, extremely bright" pixel ratio statistics on each frame in the initial screening candidate set, and determine whether the proportion of the masked area exceeds the structural masking threshold, thereby constructing the secondary perturbation condition;

[0066] If the proportion of the masked area reaches the preset threshold range, the continuous length of the interference frame is calculated to see whether it exceeds the time window threshold, which is used as the third-level disturbance condition. Finally, a disturbance start label map is generated to identify the image frame to be processed.

[0067] The image frames with perturbation start labels are divided into regions according to the perturbation start label map, and the edge connectivity, corner point density and directional gradient stability of each region are extracted to form a structural continuity index group;

[0068] Align the structural continuity index group with the perturbation activation label map and perform joint analysis to generate a perturbation-structure joint attribute map. The perturbation-structure joint attribute map is used to determine the processing level of the image region.

[0069] According to the disturbance amplitude and structural continuity index of each region in the disturbance-structure joint attribute graph, an image processing label is assigned to the image region, and the image processing labels include "high disturbance and high structure", "high disturbance and low structure", "low disturbance and high structure" and "low disturbance and low structure"; the disturbance amplitude is derived from the disturbance trend function calculated in the brightness disturbance factor group, and when the disturbance amplitude is greater than the preset disturbance amplitude threshold, it is judged as "high disturbance", otherwise it is "low disturbance"; the structural continuity index is derived from the structural continuity index group, and when the comprehensive score of the structural continuity index is higher than the preset comprehensive score threshold, it is judged as "high structure", otherwise it is "low structure";

[0070] According to the image processing label, the preset image enhancement path is called, and the image area is bound to the contour migration reconstruction path, illumination compensation path, edge refinement path and basic filtering path respectively, and a path allocation map is established.

[0071] The image blocks marked as "high-perturbation high-structure" image areas are input into the contour migration reconstruction path, and the gradient extension calculation is performed based on the edge direction maps of the previous and next frames to generate the structure-filled image area;

[0072] The image blocks marked as "high disturbance low structure" are input into the illumination compensation path, and the brightness trend curve of adjacent frames is used to perform local illumination reconstruction to generate the brightness consistency compensation image area;

[0073] The image blocks marked as "low disturbance high structure" are input into the edge refinement path, and the boundary details are stretched by contrast-preserving local lifting filtering to generate clear boundary image areas;

[0074] The image blocks marked as "low disturbance low structure" image regions are input into the basic filtering path, and a stable reference image region is generated using a stable filter bank and participates in full image fusion.

[0075] After the image enhancement path is executed and an updated image frame sequence is output, a behavior preservation factor group of the traffic monitoring object is extracted from the updated image frame sequence. The behavior preservation factor group includes trajectory continuity, boundary jump rate, and contour stability score. The behavior preservation factor group is used to characterize the degree of influence of the image enhancement result on the dynamic consistency of the traffic target;

[0076] The behavior preservation factor group is compared with the reference behavior factor group of the traffic monitoring object extracted before the image enhancement path is executed, the interference response of the image enhancement to the traffic target structure is solved, and the corresponding behavior deviation score map is generated;

[0077] If the offset score of any area in the behavior offset score map exceeds the preset offset score threshold, the image processing label of the image block corresponding to the area is located and its originally configured image enhancement path is called back;

[0078] The compensation path parameters are called for the located image blocks, the enhancement process is re-executed and the original enhancement results are replaced, the structural consistency correction is completed and the updated enhanced image frame is output.

[0079] It should be noted that in the formula structure involved in this solution, dimensionless terms can serve as proportionality or structural adjustment factors. When combined with quantities with units, they only play a numerical scaling role and do not introduce new physical dimensions. Therefore, they will not change or confuse the overall unit system of expression. This combination of "dimensionless terms and units" can be understood as a composite structural expression commonly used in mathematical and physical modeling, conforming to the principle of dimensional consistency and having a clear physical interpretation basis.

[0080] Secondly, in the formula structure of this scheme, if multiple variables with different physical units are involved, including but not limited to time, mass or energy variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable can be formed into a unified structure through function mapping, ratio combination or normalization adjustment. The units and meanings are clear, and the overall expression conforms to the principle of dimensional consistency and the common formula of engineering modeling.

[0081] In this solution, any design constants, weights, adjustment factors, threshold parameters, and proportional coefficients are adjustable control parameters for different application environments. Their values depend on the target device configuration, data input characteristics, and performance optimization goals. During the implementation phase, they are set within a reasonable range through model verification, performance constraints, or engineering calibration. Although these parameters do not have unique preset values, they have clear adjustment logic and calculation paths, and are part of the deterministic setting process in engineering implementation. The purpose of such setting is to ensure that the solution is both universally adaptable, reproducible, and operable, without affecting its technical clarity and feasibility.

[0082] Responding to spikes by perturbing trends Construct a perturbation trend function:

[0083]

[0084] The peak value of the disturbance trend response Used to drive the perturbation start label map; The perturbation analysis time window refers to the continuous frame subinterval in the constructed image frame sequence used to extract the dynamic change characteristics of brightness and structure; express A frame time index in ; is the brightness texture consistency term, which is composed of the average brightness and edge brightness gradient The calculated spatial joint tensor; is the edge-dominant energy term, which is represented by the main peak frequency diagram in the frequency domain. The square integral of the derivative of is obtained; is the frequency domain regularization term. In practical applications, the value of the frequency domain regularization term can be defined as the 5% quantile of the brightness histogram of the traffic scene image, which is equivalent to the lower limit of the image noise level and is estimated by the background brightness noise. is the disturbance scale response adjustment item, which is composed of the brightness fluctuation rate , local exposure ratio , spectrum drift rate The composite disturbance weight field formed; To mask the active mask item, the mask active mask item depends on the extreme dark area ratio Compared with the extremely bright area Whether it exceeds the masking threshold range;

[0085] in ; Represents all pixel positions in the image The covariance calculation operation is performed to quantify the coupling relationship between the overall texture direction and brightness changes;

[0086] in ; is a frequency variable, which is used to describe the periodic characteristics in the frequency domain image; the main peak frequency graph in the frequency domain Represents an image frame The spectrum amplitude function can be obtained by FFT or DCT transformation of the grayscale image of the frame in practical applications; Indicates taking the square of the absolute value of the derivative, which is used to emphasize the area of strong change; Represents the main frequency analysis interval, which is selected as the area with concentrated energy coverage in the spectrum; where the frequency interval is integrated and the frequency disturbance energy is accumulated, The larger the value, the more unstable the edge structure is in the frequency domain, reflecting the intensity of the disturbance of the image.

[0087]

[0088]

[0089]

[0090]

[0091] in for The average brightness of the frame, Indicates the maximum change in absolute value at all pixel positions; is the total number of pixels. In the above formula, the total number of pixels Indicates "region of interest" The total number of pixels contained in ; Indicates that the spatial range traversed by the summation is the "region of interest" in the image; is the brightness threshold; is an indicator function. In the indicator function, when the condition in the brackets is true, the value is 1, otherwise it is 0. It is used to count the number of pixels exceeding the exposure threshold;

[0092]

[0093] in It is an indicator function, which takes the value 1 when the condition in the brackets is true, and 0 otherwise; Trigger threshold for extremely dark areas; is the trigger threshold for extremely bright areas; This means that as long as any of the extremely dark or extremely bright area ratios exceeds the threshold, the image frame is judged to have "occlusion interference";

[0094] exist In the formula Indicates the growth rate of disturbance trend over time, which is used to select the fastest growing disturbance segment and enhance the sensitivity to dynamic disturbance response; the overall structure is finally The strongest perturbation frame in the entire image frame sequence is extracted as the trigger source of the perturbation start graph.

[0095] Based on the perturbation-initiation label map and the perturbation-structure joint attribute map constructed by the structural continuity index, the image processing label map is generated and the label assignment of the image enhancement path is completed;

[0096]

[0097] in Image position The image enhancement path label map represents the image processing label map; is a label set, and the label set value is {high disturbance high structure, high disturbance low structure, low disturbance high structure, low disturbance low structure}; is the local perturbation intensity value, which is derived from the mapping of the perturbation trend function to the corresponding pixel block; It is a structural feature enhancement item, which is derived from "edge connectivity", "directional gradient stability" and "corner density". Edge connectivity is calculated by the ratio of the length of the largest connected edge region in the edge graph extracted by the Sobel operator in the region to the total number of edge pixels. Directional gradient stability is obtained by calculating the main direction concentration (the proportion of the maximum response direction) of the gradient direction histogram in the region. Corner density is the unit area corner response rate obtained by dividing the number of Harris or FAST detected corners in the region by the number of pixels in the region. Finally, the structural feature enhancement item is obtained by normalizing the edge connectivity ratio, the main gradient direction concentration and the unit area corner response rate and then performing weighted superposition. is the structural complexity term, which is obtained by weighted summation of the regional gradient direction difference and the boundary direction cross frequency; is the threshold response control function of the path label, which is generated from the dynamic path scoring function according to the characteristics of each type of path;

[0098] In the formula Indicates the location The product of the disturbance response and the structural retention strength is used as the path activation response value; the denominator Used to penalize areas with high structural complexity to avoid false enhancement; The whole of The operation selects the optimal path label category to form the final label graph.

[0099] Constructing image enhancement model 1 based on contour migration reconstruction path:

[0100]

[0101] in Indicates that under the path "Contour Migration Reconstruction", the position The enhanced output pixel value of is the time offset variable, indicating the current frame The adjacent frame offset, Value range ; is the upper bound of time offset; is the lower bound of time offset; Indicates the time offset is Frame Middle, location The brightness trend consistency weight is used to measure whether the brightness change trend of the frame is consistent with the lighting trajectory of the current frame at the position; Frame Middle position The original pixel value of Indicates that along the dominant edge direction The directional first derivative of the image intensity along the dominant edge direction The directional first-order derivative of the image intensity is used to extract the contour gradient; is the local structure credibility factor, the local structure credibility factor is in the frame The stability of the main axis direction of the structure tensor is calculated to measure whether the structure information can be used for migration reconstruction; in addition, the stability of the main axis direction of the structure tensor can be calculated by calculating the stability of the image area in the frame. It is measured by the ratio of the structural tensor eigenvalues (such as the ratio of the minimum eigenvalue to the maximum eigenvalue). The closer the value is to 1, the more stable the directional structure is.

[0102] Building image enhancement model 2 based on illumination compensation path:

[0103]

[0104] in Indicates that under the path "Light Compensation", the position The enhanced output pixel value of Frame Up, position The brightness trend consistency weight of , which indicates the degree to which the brightness change of the adjacent frame is consistent with the trend of the current frame; is the illumination difference term, which represents the compensation amount under the brightness trend prediction;

[0105] Building image enhancement model 3 based on edge refinement path:

[0106]

[0107] in Indicates that under the path "edge refinement", the position The enhanced output pixel value of is the edge enhancement adjustment factor, which is used to control the boundary enhancement strength; It represents the second-order directional derivative performed along the dominant structure direction and is used to detect the local change rate of thin edges or low-contrast structures in the image. Its essence is to measure the curvature of the image grayscale in the main direction, that is, the "acceleration" of the structural change; is the local contrast stretching factor, which is derived from the brightness variance and texture contrast of the pixel neighborhood. The local contrast stretching factor is used to enhance edge details.

[0108] Building image enhancement model 4 based on basic filtering path:

[0109]

[0110] in Indicates that under the path "Basic Filtering", the position The enhanced output pixel value of Indicates The local pixel neighborhood window built as the center includes 、 wait; is a pixel position in the neighborhood window; is the steady-state filter response coefficient, which is used to adjust the local influence of the current pixel. In practical applications, the local influence of the current pixel can be calculated based on texture sparsity, boundary gradient clarity, etc. is the total number of neighborhood pixels, which is used to normalize the filter output value.

[0111] The path correction mechanism of traffic behavior consistency feedback is constructed by using the deviation score of behavior consistency. Indicates Shift scores for behavioral consistency;

[0112]

[0113] in is the behavior preservation factor after enhancement, which means the behavior preservation factor after image enhancement from pixel position The first behavioral factor group extracted from dimensions (e.g., trajectory, contour, boundary shape, etc.); is the pre-enhancement behavior factor, which represents the behavior description of the traffic target at the same location extracted before image enhancement is performed; It is a dimensional set of behavior retention factors, including multiple behavior description indicators such as trajectory, boundary, and contour; is the inter-frame change rate of the behavior factor, where is the first-order difference of the behavior factor on the time axis, which is used to measure its inter-frame fluctuation degree; is the regularization coefficient, which is defined as the structural change threshold under the dynamic range of the target scale; Represents the cumulative term of behavioral fluctuations in the time window; is the structure perturbation activation map, which is represented at the pixel position Whether the image structure after enhancement has abnormal jumps. Its value can be a continuous value or a binary mask (1 indicates abnormal structure disturbance);

[0114] It should be noted that the "Low-Illumination Traffic Monitoring Method Combined with a Nighttime Enhancement Algorithm" proposed in this invention is based on an in-depth analysis of the background issues of persistently impaired imaging quality in nighttime traffic monitoring scenarios. It addresses the issues of image continuity discontinuity, blurred structural boundaries, and difficulty identifying stroboscopic interference. The resulting method is a partitioned, enhanced image processing mechanism with regional awareness and path allocation capabilities. The overall design concept of this scheme is derived from the four-stage closed-loop logic of "disturbance drive - structure judgment - path activation - behavior feedback." By introducing a multi-stage judgment mechanism and regionally differentiated enhancement paths, it adapts to the actual situation of drastic image quality fluctuations in low-illumination scenarios at night.

[0115] The implementation process of the solution begins with "frame sequence disturbance perception." First, by performing brightness statistics, frequency domain analysis, and edge information extraction on traffic image frames continuously collected along the timeline, a set of disturbance factors is formed to characterize the trends of illumination disturbances and structural changes. The goal of this stage is to screen image segments that are truly affected by low illumination and require enhancement, avoiding the problem of overprocessing caused by indiscriminate enhancement of all images. Therefore, this process constructs a triple nested judgment condition of "disturbance amplitude change trend," "occlusion ratio assessment," and "disturbance continuity length judgment" to identify target frames for enhancement. Ultimately, a disturbance initiation label map is generated to indicate key areas.

[0116] After perturbation identification is complete, the system enters the "regional structure identification" phase. Based on the aforementioned label map, this phase divides the frame into regions to be analyzed and extracts structural indicators such as edge connectivity, corner point distribution density, and main gradient direction stability from the local image. This further constructs a set of structural continuity indicators that reflect the regional image structure preservation ability. By aligning and comparing this set of indicators with the perturbation label map, the system generates a "perturbation-structure joint attribute map," based on which the image regions are accurately divided into four typical labels: high perturbation high structure, high perturbation low structure, low perturbation high structure, and low perturbation low structure. This label division not only indicates the necessity of enhancement, but also implicitly carries the preset logic of the enhancement strategy type.

[0117] Then, in the "path matching enhancement" phase, each image region label is bound to four corresponding image enhancement paths: contour migration reconstruction, illumination compensation, edge refinement, and basic filtering. These four paths are responsible for preserving fine structure, restoring local illumination, enhancing boundary features, and restoring the overall steady state, respectively, to adapt to the enhancement requirements of image regions with different perturbations. The core design of this partition binding is "enhancement and structural compatibility," avoiding the structural misalignment and information overfitting caused by traditional methods that apply a unified algorithm to all regions.

[0118] After the enhancement is complete, the solution enters the "behavior consistency feedback correction" phase, extracting a set of behavioral factors of traffic targets before and after image enhancement, including indicators such as trajectory continuity, boundary jump rate, and contour stability. By comparing before and after, a score map is generated to quantify the degree of interference of the enhancement process on the dynamic structure of the traffic target. If a certain area is found to have a large behavioral deviation after enhancement, the original path is recalled and parameter compensation is performed and re-executed based on the image processing label information located in the score map, thus forming a "closed-loop enhancement-feedback correction" self-regulatory mechanism to ensure that the motion logic and structural characteristics of the traffic target remain consistent before and after enhancement.

[0119] In summary, the four-stage serial structure is adopted in this scheme to implement a highly robust image processing strategy for low-light interference in actual monitoring environments; the disturbance-driven activation mechanism is used to improve response efficiency, structural analysis is used to achieve precise allocation of enhanced paths, path execution is used to complete targeted enhancement processing, and behavioral feedback is used to construct an enhanced correction loop; this design is not a simple combination of algorithms, but is based on in-depth modeling of the relationship between noise sources, structural change characteristics and target behavioral logic of night traffic images, ensuring that the scheme does not destroy the dynamic characteristics of traffic objects while maintaining image quality, and has engineering practicality and adaptability to multiple scenarios.

[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A low-light traffic monitoring method combined with a nighttime enhancement algorithm, characterized in that: include: A brightness disturbance factor group is extracted based on the image frame sequence of the traffic scene image, and the disturbance amplitude, masked area ratio and interference frame continuous length are judged to determine whether the image frame enters the image enhancement path. By performing structural feature extraction and joint attribute determination on each region in the image frame, an image processing label map is generated and a multi-path image enhancement strategy is bound; Activate the image enhancement path based on the image processing label map, and perform regional structure restoration and brightness consistency restoration in combination with inter-frame neighborhoods; Based on the image enhancement path processing results, the behavior preservation factor group of the traffic monitoring object is extracted to determine whether there is dynamic deviation of the traffic target structure, and the original path execution strategy is called back based on the scoring results. Obtain a traffic scene image and use it as input. Perform a frame segmentation operation according to the timestamp sequence to construct an image frame sequence. Input the image frame sequence into the perturbation analysis process, extract the average brightness, edge brightness gradient, and frequency domain main peak frequency map of each frame in the image frame sequence, perform inter-frame statistical processing on the extracted results, calculate the brightness texture consistency term, edge dominant energy term, and perturbation scale response adjustment term, and generate a brightness perturbation factor group. Perform inter-frame difference operation on the brightness perturbation factor group and calculate the perturbation amplitude trend curve to obtain the perturbation trend function of the frame group for dynamic perturbation pattern extraction; The perturbation trend function is compared with the static benchmark trend model for similarity. If the perturbation amplitude is greater than the reference threshold, the frame group is determined to constitute a first-level perturbation and enters the preliminary screening candidate set. Perform statistics on the ratio of extremely dark and extremely bright pixels in each frame of the initial screening candidate set, and determine whether the proportion of the masked area exceeds the structural masking threshold, thereby constructing the secondary perturbation condition; If the proportion of the masked area reaches the preset threshold range, the continuous length of the interference frame is calculated to see if it exceeds the time window threshold, which is used as the third-level disturbance condition. Finally, a disturbance start label map is generated to identify the image frame to be processed. The image frames with perturbation start labels are divided into regions according to the perturbation start label map, and the edge connectivity, corner point density and directional gradient stability of each region are extracted to form a structural continuity index group; Align the structural continuity index group with the perturbation activation label map and perform joint analysis to generate a perturbation-structure joint attribute map. The perturbation-structure joint attribute map is used to determine the processing level of the image region. According to the disturbance amplitude and structural continuity index of each region in the disturbance-structure joint attribute graph, an image processing label is assigned to the image region, and the image processing labels include high disturbance high structure, high disturbance low structure, low disturbance high structure and low disturbance low structure; when the disturbance amplitude is greater than a preset disturbance amplitude threshold, it is judged as high disturbance, otherwise it is low disturbance; when the comprehensive score of the structural continuity index is higher than a preset comprehensive score threshold, it is judged as high structure, otherwise it is low structure; Call the preset image enhancement path according to the image processing label, bind the image area to the contour migration reconstruction path, illumination compensation path, edge refinement path and basic filtering path respectively, and establish a path allocation map; The image blocks marked as high-perturbation and high-structure image areas are input into the contour migration reconstruction path, and the gradient extension calculation is performed based on the edge direction maps of the previous and next frames to generate the structure-filled image area; The image blocks marked as high-disturbance low-structure image areas are input into the illumination compensation path, and the brightness trend curves of adjacent frames are used to perform local illumination reconstruction to generate brightness consistency compensation image areas; The image blocks marked as low-disturbance high-structure image areas are input into the edge refinement path, and the boundary details are stretched by contrast-preserving local lifting filtering to generate clear boundary image areas; The image blocks marked as low-perturbation and low-structure image regions are input into the basic filtering path, and a stable reference image region is generated using a stable filter bank and participates in full image fusion; After the image enhancement path is executed and an updated image frame sequence is output, a behavior preservation factor group of the traffic monitoring object is extracted from the updated image frame sequence. The behavior preservation factor group includes trajectory continuity, boundary jump rate, and contour stability score. The behavior preservation factor group is used to characterize the degree of influence of the image enhancement result on the dynamic consistency of the traffic target; The behavior preservation factor group is compared with the reference behavior factor group of the traffic monitoring object extracted before the image enhancement path is executed, the interference response of the image enhancement to the traffic target structure is solved, and the corresponding behavior deviation score map is generated; If the offset score of any area in the behavior offset score map exceeds the preset offset score threshold, the image processing label of the image block corresponding to the area is located and its originally configured image enhancement path is called back; The compensation path parameters are called for the located image blocks, the enhancement process is re-executed and the original enhancement results are replaced, the structural consistency correction is completed and the updated enhanced image frame is output.

2. The low-light traffic monitoring method combined with a nighttime enhancement algorithm according to claim 1, characterized in that: Responding to spikes by perturbing trends Construct a perturbation trend function: in time window for disturbance analysis; express A frame time index in ; is the brightness texture consistency item; is the edge dominant energy term; is the frequency domain regularization term; is the disturbance scale response adjustment item; Activate the mask item for masking.

3. The low-light traffic monitoring method combined with a nighttime enhancement algorithm according to claim 2, characterized in that: Based on the perturbation-initiation label map and the perturbation-structure joint attribute map constructed by the structural continuity index, the image processing label map is generated and the label assignment of the image enhancement path is completed; in Image position The image enhancement path label map represents the image processing label map; is a label set, and the label set value is {high disturbance high structure, high disturbance low structure, low disturbance high structure, low disturbance low structure}; is the local disturbance intensity value; Enhancement items for structural features; is the structural complexity term; is the threshold response control function of the path label.

4. The low-light traffic monitoring method combined with a nighttime enhancement algorithm according to claim 3, characterized in that: Constructing image enhancement model 1 based on contour migration reconstruction path: in Indicates that under the path contour migration reconstruction, the position The enhanced output pixel value of is the time offset variable; is the upper bound of time offset; is the lower bound of time offset; Indicates the time offset is Frame Middle, location The brightness trend consistency weight; Frame Middle position The original pixel value of Indicates that along the dominant edge direction The directional first derivative of the image intensity; is the local structure credibility factor; Building image enhancement model 2 based on illumination compensation path: in Indicates that under path illumination compensation, the position The enhanced output pixel value of Frame Up, position The brightness trend consistency weight; is the illumination difference term; Building image enhancement model 3 based on edge refinement path: in Indicates that under path edge refinement, the position The enhanced output pixel value of is the edge enhancement adjustment factor; represents the second-order directional derivative performed along the dominant structural direction; is the local contrast stretching factor; Building image enhancement model 4 based on basic filtering path: in Indicates that under path-based filtering, the position The enhanced output pixel value of Indicates A local pixel neighborhood window constructed for the center; is a pixel position in the neighborhood window; is the steady-state filter response coefficient; is the total number of neighborhood pixels.

5. The low-light traffic monitoring method combined with a nighttime enhancement algorithm according to claim 4, characterized in that: The path correction mechanism of traffic behavior consistency feedback is constructed by using the deviation score of behavior consistency. Indicates Shift scores for behavioral consistency; in It is the factor for maintaining behavior after reinforcement; To enhance the pre-behavioral factor; A set of dimensions that hold factors for behavior; is the inter-frame change rate of the behavior factor; is the regularization coefficient; Represents the cumulative term of behavioral fluctuations in the time window; is the structural perturbation activation map.

Images