Multi-dimensional intelligent detection method for video quality

Abstract

This invention belongs to the field of video processing technology and is used to solve the problem that existing technologies cannot extract the features that truly represent the essence of interference from luminance time-series data and establish a discrimination mechanism that can adapt to environmental changes. Specifically, it is a multi-dimensional intelligent detection method for video quality, including the following steps: acquiring luminance information of multiple consecutive frames in a video stream, constructing a luminance time-series sequence, extracting the main frequency component and its phase information from the luminance time-series sequence, calculating the phase entropy value based on the phase information of M consecutive time windows; dynamically adjusting the judgment threshold for luminance anomaly detection based on the deviation of the phase entropy value from a preset benchmark, and performing out-of-bounds detection on the average luminance of the current frame based on the judgment threshold; and verifying the preliminary anomaly detection results with the motion information of the current frame. This invention, by introducing phase entropy as a quantitative indicator of the periodic intensity of luminance fluctuations, can effectively identify periodic interference disguised within the normal luminance average range.

Description

Technical Field

[0001] This invention belongs to the field of video processing technology, specifically a multi-dimensional intelligent detection method for video quality. Background Technology

[0002] Video quality inspection is a fundamental supporting technology in fields such as video surveillance, content review, and judicial evidence collection. With the popularization of video acquisition equipment and the surge in video data, how to automatically and accurately identify screen anomalies has become a key link in ensuring data availability. In current engineering practice, detection methods for extreme image quality problems such as black screens and white screens are relatively mature. Effective identification can usually be achieved by analyzing the brightness statistical characteristics of a single frame image and combining it with multi-frame verification.

[0003] However, in real-world scenarios, abnormal video display is not limited to extreme brightness deviations. A more subtle form of interference manifests as periodic fluctuations in brightness, such as screen flickering caused by light source strobe, power supply ripple interference, or malicious attacks. In this case, the average brightness of a single frame may still be within the normal range, but the alternating brightness changes between consecutive frames severely disrupt the visual continuity and information integrity of the image. Existing detection mechanisms, focusing only on the absolute brightness value of a single frame and ignoring the fluctuation patterns of brightness over time, struggle to detect such "disguised" abnormalities below normal statistical values. More importantly, inherent regular movements in the scene, such as a fan turning or swaying tree branches, also exhibit periodic characteristics in brightness timing. This makes detection based solely on fluctuation intensity highly susceptible to misjudgment, failing to distinguish genuine interference from normal scene content.

[0004] Therefore, it is evident that how to extract the features that truly represent the essence of interference from brightness time-series data in complex dynamic scenes, and how to establish a discrimination mechanism that can adapt to environmental changes, has become a bottleneck that existing video quality detection technologies urgently need to overcome. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-dimensional intelligent detection method for video quality, which solves the problem that existing technologies cannot extract the features that truly represent the essence of interference from brightness time-series data, and establish a discrimination mechanism that can adapt to environmental changes;

[0006] The technical problem to be solved by this invention is: how to provide a multi-dimensional intelligent detection method for video quality that can extract the features that truly represent the nature of interference from luminance time-series data and establish a discrimination mechanism that can adapt to environmental changes.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A multi-dimensional intelligent method for video quality detection includes the following steps:

[0009] Obtain the brightness information of multiple consecutive frames in the video stream and construct a brightness temporal sequence;

[0010] Spectral analysis is performed on the luminance time series to extract the dominant frequency component and its phase information. The phase entropy value is calculated based on the phase information of M consecutive time windows, where M is a preset positive integer and the value of M is related to the frame rate of the video stream, so that the duration covered by the M time windows is within the preset duration range. The phase entropy value is used to characterize the periodic intensity of luminance fluctuations.

[0011] The threshold for detecting abnormal brightness is dynamically adjusted based on the degree of deviation between the phase entropy value and the preset benchmark.

[0012] Based on the adjusted judgment threshold, the average brightness of the current frame is checked for out-of-bounds detection, and the periodic brightness interference is jointly judged by combining the phase entropy value to obtain the preliminary anomaly detection results.

[0013] The final video quality detection result is generated by verifying the preliminary anomaly detection results with the motion information of the current frame.

[0014] The present invention has the following beneficial effects:

[0015] 1. By introducing phase entropy as a quantitative indicator of the periodic intensity of brightness fluctuations, this invention overcomes the limitations of existing technologies that rely solely on single-frame brightness statistics, effectively identifying periodic interference disguised within the normal brightness average range. When malicious actors use a stroboscopic light source synchronized with the camera frame rate to launch an attack, the average brightness of each frame may remain at a normal level, but the brightness fluctuations between consecutive frames exhibit highly regular phase synchronization characteristics. This invention performs spectral analysis on the brightness time series and calculates the phase entropy of multiple consecutive time windows. When the phase entropy is lower than a preset threshold, it can be determined that periodic interference exists, thus solving the problem of missed detection of covert interference methods such as stroboscopic attacks by traditional detection mechanisms, ensuring that video images can be promptly perceived and alerted when subjected to such interference.

[0016] 2. By constructing a dynamic modulation mechanism between phase entropy and brightness threshold, the detection sensitivity is adaptively adjusted according to environmental fluctuations. A dynamic modulation factor is generated based on the deviation of phase entropy from a completely random state, and the basic black screen threshold and white screen threshold are asymmetrically adjusted: when the phase entropy is low (i.e., there is periodic interference), the dynamic black screen threshold is increased and the dynamic white screen threshold is decreased, so that anomalies can be effectively captured even when the average brightness does not exceed the extreme limit. This data-driven threshold modulation method avoids the problem of insufficient adaptability of fixed thresholds under different lighting scenarios, enabling the detection system to automatically optimize the sensitivity configuration according to environmental changes and improve the detection robustness in complex scenarios.

[0017] 3. By introducing an orthogonal verification mechanism of motion information and phase entropy, the causes of periodic brightness interference can be accurately distinguished, effectively eliminating misjudgments caused by regular motion within the scene. When initially determined to be periodic brightness interference, the absolute difference between the current frame and the previous frame is further calculated as a motion energy indicator: if the motion energy is higher than the preset motion threshold, it indicates that the periodic fluctuation originates from normal motion such as fan rotation and swaying branches, and no alarm is triggered; if the motion energy is not higher than the preset motion threshold, then the existence of real interference is confirmed. This verification mechanism based on orthogonal information solves the problem that existing technologies cannot distinguish between interference and normal motion, significantly improving the reliability of detection results and the accuracy of operation and maintenance alarms.

[0018] 4. By constructing a closed-loop feedback mechanism that includes statistical updates, threshold adaptation, and parameter adjustment, the detection system possesses online learning and environmental adaptability capabilities. When a condition is determined to be normal and the motion is stationary, the statistical mean and standard deviation of the normal phase entropy are updated using an exponentially weighted moving average method, and the preset phase entropy threshold is dynamically adjusted based on the updated statistics. Simultaneously, based on the statistical analysis of historical events where periodic interference has been confirmed or filtered, the modulation factor's adjustment coefficient is dynamically adjusted, enabling the system to adapt to the intensity of periodic interference in different scenarios. Furthermore, by sensing the long-term trend of phase entropy, the system distinguishes between slow environmental changes and abrupt changes and adopts differentiated threshold adjustment strategies, avoiding detection performance drift caused by natural factors such as seasonal changes and weather variations, thus ensuring the long-term stable operation of the system. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an overall flowchart of the present invention;

[0021] Figure 2 This is a flowchart illustrating the closed-loop optimization of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In the engineering practice of video surveillance systems, the quality stability of video footage essentially depends on the continuity and consistency of the light signals captured by the image sensor in the time domain. A video stream that meets quality standards should exhibit natural random fluctuations in brightness information over time. These fluctuations originate from minute changes in lighting in the scene, the reflective properties of object surfaces, and the noise distribution within the sensor itself. However, when the video footage suffers external interference or equipment malfunction, these natural random fluctuations are disrupted, replaced by periodic fluctuations with a fixed frequency and stable phase. Based on this physical consensus, existing video quality detection technologies typically use the statistical characteristics of brightness in a single frame as the criterion, identifying extreme abnormal states such as black screens and white screens by setting fixed brightness thresholds.

[0024] The limitation of existing technologies lies in their assumption that abnormal brightness necessarily manifests as persistent extreme brightness values, neglecting the crucial information dimension of brightness fluctuation patterns over time. Specifically, when malicious actors use a strobe light source synchronized with the camera's frame rate for interference, the average brightness of each frame may fall within the normal range due to alternating bright and dark periods, thus bypassing brightness threshold-based detection mechanisms. However, the brightness fluctuations between consecutive frames exhibit highly regular phase synchronization characteristics, a pattern fundamentally different from the random fluctuations in normal scenes. Simultaneously, regular movements inherent in the monitored scene, such as fan blades rotating or tree branches swaying, also produce periodic fluctuations in brightness over time, but these are caused by genuine changes in scene content, not by abnormal image quality itself. Existing technologies lack the ability to distinguish between these two different causes of periodic fluctuations, potentially causing the system to miss detections of genuine interference while falsely reporting normal movement.

[0025] For example, in an interrogation room in a detention center, if a suspect's accomplice uses a strobe light synchronized with the frame rate of a surveillance camera to illuminate the lens, the average brightness of each frame may remain at a normal level, but the actual video image will exhibit severe flickering, making key facial details unrecognizable. Existing detection systems, focusing only on single-frame brightness statistics, cannot detect this periodic flickering, leading to an unnoticed decline in evidence quality. Similarly, when a fan in a monitored environment is operating normally, its periodic obstruction will create regular fluctuations in brightness over time. If the system judges solely based on the intensity of these fluctuations, it is highly likely to misinterpret them as image quality interference, triggering unnecessary alarms. The root cause of these missed detections and false alarms in these two scenarios lies in the current technology's failure to effectively extract the characteristic dimension from brightness time-series data that can distinguish the nature of interference—namely, the phase consistency characteristic of the fluctuations.

[0026] A deeper problem lies in the fact that even if the system can identify the existence of periodic fluctuations, if it cannot distinguish whether the cause is real interference or normal motion, subsequent alarm decisions and processing strategies will lack specificity. For real periodic interference, the system needs to issue timely alarms and may trigger anti-flicker processing; while for normal, regular motion, the system should exclude it from the alarm range. Existing technologies lack orthogonal verification mechanisms for motion information, making it difficult to establish accurate discrimination logic in complex dynamic scenarios, thus leading to a decrease in the reliability of detection results.

[0027] If the aforementioned problems are not addressed, video quality inspection systems will remain in a state of "blindness" for a long time: on the one hand, they will be powerless against periodic interference disguised within the normal brightness range, resulting in damage to the integrity of key evidence; on the other hand, repeated false alarms for normal, regular motion will cause maintenance personnel to lose trust in alarm information. This lack of detection capability will systematically affect the reliability and evidentiary value of video surveillance systems, especially in application scenarios with stringent video quality requirements, such as judicial interrogations and financial security, potentially leading to irreparable consequences. Therefore, how to extract phase features that characterize the nature of interference from brightness time-series data in complex dynamic scenarios, and establish an orthogonal verification mechanism that can distinguish between real interference and normal motion, has become a core problem that existing video quality inspection technologies urgently need to overcome.

[0028] Example 1: As Figure 1 As shown, the multi-dimensional intelligent detection method for video quality includes the following steps:

[0029] Step S1: Obtain the brightness information of multiple consecutive frames in the video stream and construct a brightness temporal sequence;

[0030] The brightness information is specifically the average brightness value of the Y channel of each frame; the brightness temporal sequence is specifically the average brightness value sequence of N consecutive frames within a sliding window of length N, where N is preset according to the video frame rate.

[0031] Before executing step S1, the initial configuration of the detection parameters needs to be completed in advance. In this embodiment, the initial values ​​of each preset parameter are not arbitrarily set, but are obtained based on statistical analysis of historical normal video data. Specifically, the system retrieves no less than 100 hours of historical surveillance video from the storage medium. These videos are manually labeled and confirmed to be normal scenes without any image quality abnormalities. If there is no available historical data when the system is first deployed, the default statistical values ​​can be used first. , This default value is set based on statistical experience in general monitoring scenarios. After running for a period of time and accumulating sufficient normal samples, it will be switched to the actual statistical value. A brightness temporal analysis, as described later, is performed frame-by-frame on this batch of historical videos to extract the phase entropy value of each sliding window, forming a normal phase entropy sample set. This sample set is then statistically analyzed, and its mean is calculated. and standard deviation This will further reduce the preset phase entropy threshold. The initial value is set to This value is determined based on the three-sigma criterion, ensuring coverage of 99.7% of the normal fluctuation range under the assumption of a normal distribution. Simultaneously, a preset reference entropy value is used. Set to ln36, corresponding to the maximum theoretical entropy value under 36 discrete phase states; the preset adjustment coefficient β is initialized to 0.3; the preset low entropy threshold is... Set to 1.0; preset motion threshold. Based on experience, the value is set to 5; a preset forgetting factor is also included. Set the value to 0.95; the preset first duration (i.e., the actual duration corresponding to the sliding window length N) is set to 1.2 seconds. The specific value of N is obtained by calculating N=FPS×1.2 and rounding it down based on the video frame rate FPS.

[0032] Step S1 aims to extract the basic luminance data required for subsequent analysis from the raw video stream. When the system receives a real-time video stream or reads a historical video file, it first decodes each frame to obtain its YUV color space representation. Since the Y channel directly represents pixel luminance, and its value ranges from 0 to 255, this embodiment uses Y channel data as the source of luminance information. For the current t-th frame image, let its width be W pixels and its height be H pixels, then all pixel positions are traversed. The average brightness of the frame is obtained by summing the Y channel values ​​and dividing by the total number of pixels. The calculation method is as follows This value reflects the overall brightness and darkness of the current frame.

[0033] To capture the continuous trajectory of brightness changes over time, discrete frame brightness values ​​need to be organized into a temporal sequence. This embodiment uses a sliding window mechanism to maintain a fixed-length queue, named the brightness temporal sequence. The length N of this sequence is dynamically determined based on the frame rate of the video stream to ensure it covers a time span of approximately 1.2 seconds. For example, for a video stream of 25 frames per second, N is 30; for a video stream of 30 frames per second, N is 36. The average brightness value of each new frame... After calculation, the system inserts it at the end of the queue; if the current queue length has reached N, the oldest historical data at the head of the queue is removed, keeping the queue length constant at N. After this operation, the luminance time-series sequence always contains the luminance average of the most recent N consecutive frames, denoted as . This sequence forms the basis for subsequent spectral analysis.

[0034] At the same time, in order to support subsequent image entropy value determination, this step also calculates the image entropy value of the current frame. Specifically, for the Y channel data of the current frame, a normalized histogram of 256 gray levels is calculated, and the pixel proportion of the i-th gray level is denoted as... Then the formula for calculating image entropy is: This value measures the richness of image content. In extreme cases such as complete black or complete white, the entropy value approaches 0; while in normal images containing complex textures, the entropy value is usually greater than 2.0. This feature will be used in subsequent joint determination to help distinguish between real anomalies and normal scene changes.

[0035] It's important to note that the length N of the sliding window is not fixed. When the frame rate of the video stream changes, the system automatically detects the frame rate change event, recalculates the value of N, clears the current window, and refills the data, ensuring that the duration covered by the luminance timing sequence remains stable within the preset 1.2-second range. This adaptive mechanism allows this method to be compatible with video sources of different frame rates without requiring separate configuration for specific frame rates.

[0036] Step S2: Perform spectral analysis on the luminance time series, extract the main frequency component and its phase information, and calculate the phase entropy value based on the phase information of M consecutive time windows, where M is a preset positive integer and the value of M is related to the frame rate of the video stream, so that the duration covered by the M time windows is within the preset duration range; the phase entropy value is used to characterize the periodic intensity of luminance fluctuations.

[0037] The extraction process of the dominant frequency component and its phase information includes:

[0038] Perform a Fast Fourier Transform on the brightness time sequence within the current window to obtain the complex spectrum of each frequency component;

[0039] Determine the frequency component with the largest amplitude as the dominant frequency component, and calculate its corresponding actual frequency.

[0040] Extract the phase value of the dominant frequency component and use it as the dominant frequency phase of the brightness fluctuation in the current window.

[0041] The phase entropy value is calculated based on the phase information from multiple consecutive time windows, specifically including:

[0042] Map the main frequency phase values ​​of M consecutive time windows to a preset angle interval, with each angle interval corresponding to a discretized phase state;

[0043] Statistically analyze the frequency distribution of M phase values ​​in each discretized phase state;

[0044] The information entropy is calculated based on the distribution frequency to obtain the phase entropy value at the current moment, where the phase entropy value is negatively correlated with the periodic intensity of brightness fluctuations.

[0045] After constructing a brightness time series of length N in step S1, step S2 performs spectral analysis on the sequence to uncover the deep characteristics of brightness fluctuations in the frequency domain. This embodiment uses Fast Fourier Transform to convert the time-domain signal to the frequency domain, thereby extracting the dominant frequency component and its phase information that characterizes the fluctuation pattern, and further calculating the phase entropy value through the phase consistency of multiple time windows.

[0046] Specifically, for the luminance time sequence corresponding to the current t-th frame First, the input signal is used for a Fast Fourier Transform. Since the sequence length is N, the transform yields N complex spectral components. , Where k=0 corresponds to the DC component; it should be noted that if all values ​​in the brightness time sequence are equal (i.e., the image is completely still and the brightness is constant), then the amplitude of all AC components after the Fast Fourier Transform is zero, and in this case, the effective main frequency component cannot be extracted. To avoid subsequent calculation errors, the system checks whether the maximum amplitude of the AC component is less than a preset minimum threshold (e.g., ...). If the value is less than this threshold, it is determined that the current window has no periodic fluctuations, and the phase entropy of the current window is directly calculated. The value is assigned to the preset reference entropy value. and will increase the main frequency Record it as zero, and skip the subsequent phase entropy calculation steps. Each complex number... It can be represented as the real part. and the virtual part Its magnitude phase The range of values ​​is To eliminate the influence of DC components, the component at k=0 is ignored. The frequency component with the largest amplitude is then searched within the range of k=1 to N-1, and its index is recorded. The actual physical frequency of this component is determined by the video frame rate (FPS). For example, if the maximum amplitude of the luminance sequence of a 25 frames per second video appears after FFT, it is because... =3, then the main frequency This means the brightness fluctuates at a frequency of 2.5 times per second. Simultaneously, the phase value of this component is extracted. This serves as the dominant frequency phase of the brightness fluctuation in the current window.

[0047] To assess the periodic intensity of brightness fluctuations, relying solely on the phase information of a single window is insufficient; the temporal stability of the phase needs to be examined. Therefore, this embodiment maintains a phase history queue of length M to store the dominant frequency phase values ​​of the most recent M time windows. The value of M is not arbitrarily set but is related to the video stream frame rate, ensuring that the total duration covered by the M windows falls within a preset duration range; specifically, let the target coverage duration be... In this embodiment, If the second is M, then the formula for calculating M is: ;in This indicates rounding down. Taking a frame rate of 25 frames per second and N=30 as an example, the calculation is as follows: The corresponding actual coverage duration is 6 × 30 / 25 = 7.2 seconds, which is still within the preset duration range of 5 to 10 seconds. If the calculated result exceeds this range, the boundary value is taken (i.e., 5 if less than 5, and 10 if greater than 10). Specifically, each time window corresponds to N frames, and its actual duration is N / FPS seconds. Therefore, the total duration of M windows is... In this embodiment, the preset duration range is 5 to 10 seconds to ensure that a sufficiently long fluctuation pattern can be captured while avoiding a decrease in response speed to scene changes due to too many windows. The M value is dynamically calculated based on the video frame rate. Or round down, where A middle value, such as 7.5 seconds, can be taken. For example, for a video of 25 frames per second and N=30, seconds, if If the duration is between 5 and 10 seconds, then M should be an integer between 5 and 8. In this embodiment, M is set to 6 by default, corresponding to a total duration of 7.2 seconds. This setting ensures that the calculation of phase entropy is statistically significant and can reflect recent fluctuation characteristics.

[0048] Obtain the main frequency phase values ​​of M consecutive windows Next, the phase entropy is calculated. First, the phase value is changed from... Mapped to The range facilitates uniform discretization; if the original phase ,but ;otherwise Ensure all phases fall within... The interval. This embodiment will Divide the phase into K equal angular intervals, where K is 36, and each interval covers 10°. For each phase value, determine the interval index it falls into based on its magnitude. Count the frequency of M phase values ​​in each interval. Then calculate the probability of occurrence in each interval. Phase entropy The information entropy of this probability distribution is defined as: Using the natural logarithm, its maximum value is when uniformly distributed. The physical meaning of phase entropy is that if brightness fluctuations have a stable periodicity, the dominant frequency phase of each window will be highly consistent, concentrated in a few intervals, leading to... A concentrated distribution of phase results in a smaller entropy value; conversely, if the fluctuations are random and the phase is evenly distributed across intervals, the entropy value is larger. Therefore, phase entropy is negatively correlated with the intensity of periodicity; the smaller the entropy value, the stronger the periodicity.

[0049] It is worth noting that FFT analysis requires the input sequence to be stationary and without missing frames. In this embodiment, the sliding window mechanism ensures continuous updates of the sequence. If the number of frames in the window is less than N due to frame drops during transmission, processing is paused until the window is filled. Furthermore, for scene transitions or violent motion that may occur in actual videos, the brightness time series may exhibit non-stationary characteristics, but this method distinguishes these characteristics through subsequent motion verification steps; here, it only handles feature extraction. At this point, step S2 outputs the dominant frequency at the current moment. Phase And the key feature—phase entropy These data will serve as the core basis for dynamic threshold adjustment and joint determination in subsequent steps.

[0050] Step S3: Dynamically adjust the judgment threshold for brightness anomaly detection based on the degree of deviation between the phase entropy value and the preset benchmark;

[0051] The threshold for detecting abnormal brightness is dynamically adjusted based on the degree of deviation between the phase entropy value and a preset benchmark, specifically including:

[0052] Calculate the normalized deviation between the phase entropy value and the preset reference entropy value;

[0053] Based on the normalized deviation and the preset adjustment coefficient, a dynamic modulation factor is generated;

[0054] Multiply the preset base black screen threshold by the dynamic modulation factor to obtain the dynamic black screen threshold; divide the preset base white screen threshold by the dynamic modulation factor to obtain the dynamic white screen threshold.

[0055] In step S2, the phase entropy value at the current moment is calculated. Subsequently, step S3 dynamically adjusts the brightness threshold upon which subsequent anomaly detection depends based on this feature. Traditional methods use a fixed threshold, which cannot adapt to scenarios with fluctuating illumination or periodic interference; this embodiment modulates the threshold in real time using phase entropy, enabling the detection sensitivity to adaptively change with the brightness fluctuation pattern.

[0056] To achieve this modulation, a reference standard for the phase entropy must first be defined. Theoretically, if brightness fluctuations are completely random and the phase is uniformly distributed across discrete intervals, the phase entropy reaches its maximum value, which depends only on the number of discretization intervals, K. In this embodiment, K is set to 36, therefore a preset reference entropy value is used. This value represents the theoretical upper limit, requires no statistical learning, and has universal applicability. However, floating-point calculations may lead to... Slightly exceeds the theoretical maximum value Therefore, before calculating the deviation, we first need to... Truncation: Based on this, the normalized deviation is calculated. Its physical meaning is the percentage decrease in the current phase entropy relative to a completely random state, calculated using the following formula: As defined, the range of values ​​for d(t) is... .when near hour, A value close to 0 indicates that the brightness fluctuation has no obvious periodicity; when When it approaches 0, A value approaching 1 indicates the presence of extremely strong periodic disturbances. Therefore, It can intuitively reflect the periodic intensity.

[0057] Based on the normalized deviation, a dynamic modulation factor is generated. The modulation factor's role is to appropriately relax the brightness threshold when periodic interference is detected, ensuring effective anomaly detection even if the average brightness does not exceed extreme limits; while maintaining the original threshold when there is no periodicity. This embodiment employs a linear mapping relationship: ;in This is a preset adjustment coefficient used to control the modulation amplitude. The value of needs to be determined by balancing sensitivity and false alarm risk: An excessively large threshold can lead to overly relaxed thresholds, potentially introducing false positives. If the value is too small, it becomes insensitive to periodic disturbances. Experiments have verified that... A value of 0.3 achieves a good balance in most monitoring scenarios. To avoid extreme values ​​in the modulation factor, [the following is omitted as the original text is incomplete and requires further context]. Limit the amplitude so that its value range is restricted to... This means that the maximum allowable threshold is relaxed by 50%.

[0058] After obtaining the modulation factor, it is applied to the base brightness threshold. Base black screen threshold. and basic white screen threshold These are empirical values ​​pre-set based on camera characteristics and ambient lighting; in this embodiment, we take... , These correspond to extremely dark and extremely bright states, respectively. The dynamically adjusted threshold is calculated as follows: Dynamic black screen threshold: Dynamic white screen threshold: ;because , The value is higher than the baseline, which means that higher brightness is allowed and the screen can still be considered black (i.e., the black screen condition is relaxed). The reduction from the baseline value means that a lower brightness is acceptable to determine a white screen (i.e., the white screen condition is relaxed). The logic behind this asymmetric adjustment is that periodic interference often manifests as fluctuating brightness, at which point the boundary between a black screen and a white screen becomes blurred. Appropriately relaxing the threshold helps to capture anomalies while avoiding missed detections due to the mean not being extremely out of bounds.

[0059] For example: Assume the current phase entropy ,but .Pick ,calculate Therefore, the dynamic black screen threshold... Dynamic white screen threshold Compared to the original thresholds of 5 and 240, the black screen threshold increased slightly, while the white screen threshold decreased, indicating that the detection sensitivity was enhanced in the presence of periodic interference.

[0060] It is worth noting that the adjustment coefficient The initial value is not static; it will be dynamically adjusted in subsequent steps based on historical events where periodic disturbances are confirmed or filtered, to form a closed-loop adaptive mechanism. In this step, an initial fixed value of 0.3 is used to ensure system stability during the startup phase.

[0061] At this point, step S3 outputs two dynamic thresholds. and These will serve as direct criteria for brightness overshoot detection in step S4. The entire adjustment process is data-driven, requiring no manual intervention, enabling the detection system to automatically optimize threshold configuration in response to environmental fluctuations.

[0062] Step S4: Based on the adjusted judgment threshold, perform boundary detection on the average brightness of the current frame, and combine the phase entropy value to jointly judge the periodic brightness interference to obtain preliminary anomaly detection results;

[0063] The joint determination process for preliminary anomaly detection results includes:

[0064] When the average brightness of the current frame is lower than the dynamic black screen threshold and the image entropy value of the current frame is lower than the preset low entropy threshold, it is determined to be a black screen abnormality.

[0065] When the average brightness of the current frame is higher than the dynamic white screen threshold and the image entropy value of the current frame is lower than the preset low entropy threshold, it is determined to be a white screen abnormality.

[0066] When no black screen or white screen abnormality is triggered, but the phase entropy value is lower than the preset phase entropy threshold, it is determined to be periodic brightness interference.

[0067] When both black screen and white screen anomalies are triggered simultaneously and the phase entropy value is lower than the preset phase entropy threshold, it is determined to be a mixed anomaly.

[0068] In step S3, the dynamic black screen threshold of the current frame is obtained. and dynamic white screen threshold Meanwhile, steps S1 and S2 have respectively provided the average brightness value of the current frame. Image entropy and phase entropy Step S4 performs a joint determination based on these inputs to generate preliminary anomaly detection results. The core of this determination lies in combining the brightness features in the spatial domain with the fluctuation features in the temporal domain to distinguish different types of image quality anomalies.

[0069] First, brightness exceeding limits is detected. This detection uses a dynamic threshold as a baseline, while also incorporating image entropy as an auxiliary criterion to avoid misjudging normal low-brightness scenes (such as turning off lights at night) as a black screen. The specific judgment logic is as follows: If... and If so, the current frame is determined to be a black screen error; if and If this condition is met, the current frame is determined to be a white screen anomaly. A preset low-entropy threshold is used. The value is set to 1.0, based on extensive experimental statistics: for an image containing any valid content, whether the image is dark or bright, its image entropy value is usually not lower than 1.5; while the entropy value of a completely black or completely white image approaches 0. Therefore, [the value is set to 1.0]. Setting it to 1.0 can effectively filter false alarms caused by normal lighting changes while ensuring the detection rate. If neither of the above two conditions is met, the out-of-bounds type is marked as "no out-of-bounds".

[0070] Next, periodic interference detection is performed. This detection is independent of brightness exceedance and is based on the magnitude of the phase entropy value. If the exceedance type is "no exceedance," and If so, it is determined to be periodic brightness interference, and the current main frequency is recorded. As an interference frequency. Preset phase entropy threshold. It is a dynamically updated parameter, whose initial value is determined by statistical analysis of phase entropy from historical normal video footage: for example, extracting phase entropy samples from 100 hours of normal surveillance video and calculating their mean. and standard deviation ,make During system operation, It will continuously update through a closed-loop feedback mechanism to adapt to environmental changes. It should be noted that when... When the value is below this threshold, it indicates that the brightness fluctuations exhibit significant periodicity. Even if the average brightness value does not exceed the dynamic threshold range, it can still be determined that there is interference.

[0071] When brightness over-limit detection and periodic interference detection are triggered simultaneously, the following occurs: Meeting the conditions for a black or white screen and ,at the same time If the image is in an extreme brightness state (such as being obscured or overexposed) and is accompanied by periodic fluctuations, it is considered a mixed anomaly. A mixed anomaly usually means that the image is in an extreme brightness state (such as being obscured or overexposed) and is superimposed with periodic fluctuations, which may indicate a more serious equipment failure (such as camera damage accompanied by power interference).

[0072] For example, suppose in a certain monitoring scenario, there is a dynamic blackout threshold. Dynamic white screen threshold Preset low entropy threshold Current phase entropy threshold If the average brightness of the current frame Image entropy Phase entropy Then it satisfies and ,at the same time Therefore, it was determined to be a mixed anomaly (black screen + periodic interference). If in another scenario... , , Therefore, and This is determined to be periodic brightness interference, and the main frequency is recorded. If... , , Therefore, and ,but It was only identified as a white screen anomaly, without any periodic interference.

[0073] After completing the above logical judgment, step S4 outputs the preliminary anomaly detection result. This result is structured data and contains at least the following fields: frame index, anomaly type (black screen, white screen, periodic interference, mixed anomaly, or normal), and associated parameters (if it is periodic interference or mixed anomaly, the main frequency is added). This result will serve as input for motion verification in step S5 to further eliminate periodic misjudgments caused by normal motion. The entire judgment process is entirely data-driven, and all thresholds have clear physical meaning or statistical basis, ensuring the objectivity and repeatability of the detection results.

[0074] Step S5: Verify the preliminary anomaly detection results with the motion information of the current frame to generate the final video quality detection results;

[0075] Verification processing is performed based on the preliminary anomaly detection results and the motion information of the current frame, specifically including:

[0076] Calculate the absolute difference between the current frame and the previous frame to obtain the motion energy index;

[0077] When the preliminary judgment result is periodic brightness interference and the motion energy index is higher than the preset motion threshold, the preliminary judgment result is marked as a normal motion scene and no alarm is triggered.

[0078] When the initial judgment result is periodic brightness interference and the motion energy index is not higher than the preset motion threshold, the existence of periodic brightness interference is confirmed, and alarm preparation is triggered.

[0079] When the initial judgment result is a black screen, white screen, or mixed abnormality, motion verification is not performed, and the alarm preparation is initiated directly.

[0080] After obtaining the preliminary anomaly detection results in step S4, step S5 introduces motion information to verify the results, so as to eliminate misjudgments caused by periodic brightness fluctuations due to normal regular movements in the scene (such as fan rotation, swaying tree branches), and ensure that the final output detection results accurately reflect the real image quality anomalies.

[0081] First, the motion energy index of the current frame needs to be calculated. Motion information extraction is based on inter-frame difference, that is, comparing the differences in the luma channel between the current frame and the previous frame. For the... Frame and the The Y-channel image of each frame is used to calculate the absolute difference pixel by pixel to obtain the inter-frame difference image. Motion energy index. Defined as the arithmetic mean of the absolute differences of all pixels, its calculation formula is: Where W and H are the width and height of the image, respectively. Indicates the current frame's position The brightness value. This indicator reflects the degree of drastic change in the overall image, and its value ranges from 0 to 255. When the image is completely still, Approaching 0; when significant motion exists, It will increase accordingly.

[0082] Preset motion threshold This threshold is used to distinguish between normal movement and static images. The threshold is set based on statistical analysis of numerous monitoring scenarios: in most indoor and outdoor fixed camera monitoring scenarios, the average inter-frame difference caused by normal movement (such as people walking or objects moving) is usually above 10, while image flicker caused by camera shake or light source flicker, although resulting in brightness changes, does not change the spatial structure, and its average inter-frame difference is often lower. To balance sensitivity and anti-interference capability, this embodiment will... The empirical value is set to 5 (it should be noted that this empirical value can be scaled and adapted according to the bit depth of the video. For 8-bit videos (grayscale range 0-255), 5 is an empirically valid value; for 10-bit videos, it can be scaled proportionally to 20). This value means that when the average brightness change per pixel is less than 5 grayscale levels, the image is considered approximately static; exceeding this value indicates significant motion. In practical applications, this threshold can also be adjusted offline based on the specific scenario.

[0083] Based on the different types of preliminary anomaly detection results, execute differentiated verification logic:

[0084] If the initial assessment indicates a black screen, white screen, or mixed anomaly, motion verification will not be performed. This is because exceeding the brightness limit already constitutes a serious image quality anomaly, and the presence or absence of motion in the image does not affect the authenticity of the anomaly. For example, when the screen is completely obscured, resulting in a black screen, the fact that an object is moving in the image (which is practically impossible) still holds true. Therefore, such anomalies directly enter the alarm preparation stage, which involves recording the anomaly type, timestamp, and related parameters of the current frame, and waiting for confirmation in subsequent consecutive frames to trigger an alarm.

[0085] If the initial assessment indicates periodic brightness interference, motion verification is needed to eliminate interference from regular motion within the scene. Specifically, this involves analyzing the motion energy index of the current frame. With preset motion threshold Comparison:

[0086] when When this occurs, it indicates that there is obvious motion in the image, and the periodic brightness fluctuations are likely due to the regular movements of moving objects (such as fan blades rotating or curtains swaying). At this point, the preliminary judgment result is marked as a "normal motion scene," without triggering any alarms, and this event is used as a normal sample for subsequent closed-loop feedback parameter adjustments. This process avoids misreporting normal motion as interference.

[0087] when When the image is completely still, but the brightness fluctuates periodically, it indicates a genuine anomaly such as external light source flickering, power interference, or malicious attack. In this case, confirming the presence of periodic brightness interference triggers an alarm preparation, and the anomaly type and frequency are recorded. Information such as the time of occurrence.

[0088] After the above verification process, step S5 outputs the final video quality detection result. This result is structured data, including frame index, final judgment type (normal, black screen anomaly, white screen anomaly, periodic brightness interference, mixed anomaly), anomaly start and end time (if it is an anomaly), and associated interference frequency, etc. For anomalies that trigger alarm preparation, the system will activate a continuous frame confirmation mechanism (this mechanism can be combined with closed-loop optimization or implemented independently). An alarm event will only be officially output when the same anomaly type occurs consecutively a preset number of times (e.g., 3 frames), in order to avoid false alarms caused by accidental fluctuations in a single frame.

[0089] For example, suppose there is a running fan in a monitored scene. Its periodic occlusion causes a significant low phase entropy in the brightness time sequence, which is initially identified as periodic brightness interference. Calculate the motion energy index of the current frame. The result is greater than the preset threshold of 5, therefore it is marked as a normal motion scene and no alarm is triggered. In another scenario, criminals use a strobe light to illuminate the camera; the overall image is still, but the brightness fluctuates strongly over time. Calculation... If the value is less than the threshold of 5, then periodic brightness interference is confirmed, alarm preparation is triggered, and the interference frequency is recorded.

[0090] By introducing motion information verification, step S5 effectively improves the accuracy of periodic interference detection, enabling the system to accurately distinguish between real interference and normal motion in complex dynamic scenarios, laying a reliable foundation for subsequent closed-loop feedback and alarm output.

[0091] like Figure 2 As shown, in addition to the above real-time detection process, this embodiment also includes a parallel or asynchronous closed-loop feedback optimization process. This process utilizes the final detection result output in step S5 and the intermediate data accumulated in steps S1 to S5 to continuously and adaptively adjust the detection parameters, thereby improving the system's adaptability to environmental changes and its long-term operational stability. Specifically, it includes:

[0092] When the final determination is normal and the exercise energy index is not higher than the preset exercise threshold, the statistical mean and standard deviation of the normal phase entropy are updated using an exponentially weighted moving average method: ; ;in This is the current phase entropy value. This is the statistical mean of normal phase entropy. For normal phase entropy variance, The preset forgetting factor is used; the specific update order is as follows: first, the current phase entropy is used. Update the mean , get new Then based on the updated Calculate the variance update term and update the variance This order ensures that the variance calculation is consistent with the latest mean.

[0093] The preset phase entropy threshold is recalculated based on the updated statistics, so that the threshold is adaptively adjusted as the scene changes.

[0094] The adjustment coefficient of the dynamic modulation factor is dynamically adjusted based on historical events in which periodic interference is confirmed or filtered.

[0095] The system performs backtracking on the batch-stored historical video files, executes the detection steps frame by frame, records all triggered abnormal events and their associated parameters, and determines the initial value of the preset phase entropy threshold offline based on the phase entropy statistics of multiple historical normal videos.

[0096] Specifically, the closed-loop feedback optimization process first relies on the dynamic tracking of the phase entropy statistical characteristics under normal scene conditions. When the final detection result output in step S5 is "normal," and the motion energy index of the current frame... Below the preset exercise threshold At this time, the system determines that the frame is a valid normal sample that can be used for statistical learning. The phase entropy value at this moment... The normal phase entropy is included in the statistical update. To balance the stability of historical information with the responsiveness to recent changes, this embodiment uses an exponentially weighted moving average to recursively update the statistical mean and variance of the normal phase entropy. Let the currently stored mean of the normal phase entropy be... The variance is The updated formula is as follows: ; ;in To predetermine the forgetting factor, this embodiment uses 0.95. This value means that the weight of historical data decays exponentially over time, while recent data contributes more to the statistics, thus enabling smooth tracking of environmental changes. and The initial value is obtained from the statistics of historical normal videos before step S1, and the specific method has been explained in the initialization description of step S1.

[0097] Based on the updated statistics, the system recalculates the preset phase entropy threshold. This allows it to adaptively adjust to changes in the scene. The adjustment formula is: This formula is based on the normal distribution assumption, ensuring that under normal circumstances, approximately 99.7% of the phase entropy values ​​are above the threshold, thus keeping the false alarm rate of periodic interference at a low level. If the calculated... If the value is below the preset lower limit of 0.5, then 0.5 is used; 0.5 is an empirical threshold, and the phase entropy below this value can be considered as strong periodic interference. Further reducing the threshold will have limited effect on improving detection sensitivity and may introduce false alarms; if the value is above the preset upper limit of ln36, then ln36 is used to avoid the threshold deviating too much from the reasonable range.

[0098] In addition, dynamic modulation factor The adjustment coefficient in It also dynamically adjusts based on the statistical results of periodic interference events. The system maintains two counters: Record the number of periodic interference events confirmed by motion verification. Record the number of suspected periodic interference events filtered by motion verification (i.e., events initially identified as periodic interference but marked as normal motion scenes by motion verification). Calculate the confirmation rate every 100 frames processed. .like and This indicates the current situation. Setting settings that might lead to missed detections can be appropriately increased. (For example, increase by 0.05 each time, not exceeding 0.5) The upper limit of 0.5 corresponds to a 50% relaxation of the maximum threshold, avoiding a large number of false detections due to excessive relaxation; the lower limit of 0.1 ensures that a minimum modulation capability is retained even in the absence of periodic interference. and ,illustrate This could lead to too many false alarms; the error rate could be appropriately reduced. (For example, decrease by 0.05 each time, but not below 0.1). In this way, the adjustment coefficient can adaptively match the periodic interference intensity of different scenarios.

[0099] For batch-stored historical video files, this method supports offline backtracking. The system executes steps S1 to S5 frame by frame, recording all triggered abnormal events and their associated parameters (such as abnormality type, timestamp, main frequency, etc.), and storing the results in a log file. Simultaneously, the initial value of the preset phase entropy threshold can be determined offline using the phase entropy statistics of multiple historical normal video segments; the specific method is described in the initialization section of step S1.

[0100] Building upon the aforementioned online adaptive update, this embodiment also introduces a mechanism for sensing the long-term trend of phase entropy to address the impact of slow or abrupt environmental changes on the detection threshold. This mechanism is executed after each statistical update of normal samples, and the specific steps are as follows:

[0101] Obtain the short-term mean of the phase entropy values ​​within the most recent second time period, which is shorter than the duration of historical data used to calculate the statistical mean and standard deviation;

[0102] The rate of change of brightness is calculated based on the average brightness of the current frame and the average brightness of the previous frame.

[0103] The short-term mean is compared with the statistical mean. When the short-term mean is consistently lower than the statistical mean and no periodic brightness interference or mixing anomaly is triggered, the preset phase entropy threshold or the update strategy of the statistical mean for subsequent frames is dynamically adjusted based on the comparison result of the brightness change rate and the preset change rate threshold.

[0104] Based on the comparison between the brightness change rate and the preset change rate threshold, a strategy for dynamically adjusting the preset phase entropy threshold or statistical mean of subsequent frames is implemented, specifically including:

[0105] When the rate of change of brightness is less than the first rate of change threshold, it is determined that the environment is changing slowly. Then, the statistical mean is corrected towards the short-term mean according to the preset step size, and the preset phase entropy threshold is recalculated based on the corrected statistical mean.

[0106] When the rate of change of brightness exceeds the second rate of change threshold, it is determined to be an environmental mutation. The update of the statistical mean and standard deviation is then paused, and the preset phase entropy threshold is temporarily increased according to a preset ratio. After the rate of change of brightness recovers to below the second rate of change threshold and remains below it for a third duration, the update is resumed and the preset phase entropy threshold is restored.

[0107] First, obtain the short-term mean of the phase entropy values ​​within the most recent second time period. The second time period should be shorter than the duration of historical data used to calculate the statistical mean and standard deviation (i.e., the effective duration covered by the exponentially weighted moving average). In this embodiment, the second time period is set to 5 minutes, by maintaining a length of... The circular queue stores all phase entropy values ​​within the last 5 minutes (if the video frame rate is 25fps, then...). Each time a new phase entropy is obtained, it is enqueued, and the oldest data at the head of the queue is removed. Simultaneously, all values ​​in the queue are accumulated and divided by . Calculate the arithmetic mean This calculation is performed once per frame to ensure... It reflects the recent phase entropy level in real time.

[0108] Simultaneously, the rate of change of brightness is calculated based on the average brightness of the current frame and the average brightness of the previous frame. ;in It is a very small positive number (e.g. This is used to avoid the denominator being zero. For simplicity, this embodiment directly stipulates that when... hour, This indicates that the rate of change has reached its upper limit. This indicator is used to quantify the degree of drastic instantaneous change in ambient light.

[0109] short-term mean Compared with long-term statistical mean Comparison. When Persistently below (For example, 10 consecutive frames satisfy) If no periodic brightness interference or mixing anomalies are triggered (i.e., steps S4 and S5 are deemed normal), it indicates that a slow change in the environment may have occurred (such as seasonal changes or gradual changes in lighting patterns), leading to a drift in the phase entropy baseline. In this case, different adjustment strategies are implemented based on the magnitude of the brightness change rate:

[0110] like If the change rate is less than the first change rate threshold (0.01 in this embodiment, meaning the brightness change rate is less than 1%), it is determined to be a slow environmental change. At this time, the system proceeds according to a preset step size (e.g., each time...). Towards Direction correction The statistical mean is gradually adjusted to allow long-term statistics to drift slowly with the environment, and the adjusted mean is used as the basis for further adjustments. Recalculate the preset phase entropy threshold ; Set the correction step size to That is, each frame will Towards Adjust the direction by 10%. If the condition is met for multiple consecutive frames, then approximate frame by frame until... and If the difference is less than a preset threshold (e.g., 0.01), then directly set... This smoothing correction avoids abrupt changes caused by fluctuations in a single frame.

[0111] like If the change rate exceeds the second change rate threshold (0.1 in this embodiment, meaning the brightness change rate is greater than 10%), it is determined to be an environmental abrupt change (such as a sudden change in illumination caused by cloud cover). At this time, the system immediately suspends the updating of the statistical mean and standard deviation, and temporarily increases the preset phase entropy threshold by a preset ratio (e.g., 1.2 times). ,Right now The improved threshold is then temporarily used in subsequent detections. Once the brightness change rate recovers to below the second change rate threshold and remains below it for a third duration (30 seconds in this embodiment), the system resumes normal updates to the statistical mean and standard deviation, and restores the preset phase entropy threshold to the value calculated based on the updated statistics.

[0112] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0113] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0114] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-dimensional intelligent detection method for video quality, characterized in that, Includes the following steps: Obtain the brightness information of multiple consecutive frames in the video stream and construct a brightness temporal sequence; Spectral analysis is performed on the brightness time sequence to extract the main frequency component and its phase information. The phase entropy value is calculated based on the phase information of M consecutive time windows, where M is a preset positive integer and the value of M is related to the frame rate of the video stream, so that the duration covered by the M time windows is within a preset duration range. The phase entropy value is used to characterize the periodic intensity of brightness fluctuations. The threshold for detecting abnormal brightness is dynamically adjusted based on the degree of deviation between the phase entropy value and the preset reference. Based on the adjusted judgment threshold, the average brightness of the current frame is checked for out-of-bounds detection, and the periodic brightness interference is jointly judged in combination with the phase entropy value to obtain preliminary anomaly detection results. The preliminary anomaly detection results are verified against the motion information of the current frame to generate the final video quality detection results.

2. The video quality multi-dimensional intelligent detection method according to claim 1, characterized in that, The brightness information is specifically the average brightness value of the Y channel of each frame of the image; the brightness time sequence is specifically a sequence of average brightness values ​​of N consecutive frames within a sliding window of length N, where N is preset according to the video frame rate.

3. The video quality multi-dimensional intelligent detection method according to claim 1, characterized in that, The extraction process of the dominant frequency component and its phase information includes: Perform a Fast Fourier Transform on the brightness time sequence within the current window to obtain the complex spectrum of each frequency component; Determine the frequency component with the largest amplitude as the dominant frequency component, and calculate its corresponding actual frequency. Extract the phase value of the dominant frequency component and use it as the dominant frequency phase of the brightness fluctuation in the current window.

4. The video quality multi-dimensional intelligent detection method according to claim 1, characterized in that, The phase entropy value is calculated based on the phase information from multiple consecutive time windows, specifically including: Map the main frequency phase values ​​of M consecutive time windows to a preset angle interval, with each angle interval corresponding to a discretized phase state; Statistically analyze the frequency distribution of M phase values ​​in each discretized phase state; The information entropy is calculated based on the distribution frequency to obtain the phase entropy value at the current moment, wherein the phase entropy value is negatively correlated with the periodic intensity of the brightness fluctuation.

5. The video quality multi-dimensional intelligent detection method according to claim 1, characterized in that, The threshold for detecting abnormal brightness is dynamically adjusted based on the degree of deviation between the phase entropy value and a preset reference, specifically including: Calculate the normalized deviation between the phase entropy value and the preset reference entropy value; Based on the normalized deviation and the preset adjustment coefficient, a dynamic modulation factor is generated; Multiply the preset base black screen threshold by the dynamic modulation factor to obtain the dynamic black screen threshold; divide the preset base white screen threshold by the dynamic modulation factor to obtain the dynamic white screen threshold.

6. The video quality multi-dimensional intelligent detection method according to claim 5, characterized in that, The joint determination process for preliminary anomaly detection results includes: When the average brightness of the current frame is lower than the dynamic black screen threshold and the image entropy value of the current frame is lower than the preset low entropy threshold, it is determined to be a black screen abnormality. When the average brightness of the current frame is higher than the dynamic white screen threshold and the image entropy value of the current frame is lower than the preset low entropy threshold, it is determined to be a white screen abnormality. When no black screen or white screen abnormality is triggered, but the phase entropy value is lower than the preset phase entropy threshold, it is determined to be periodic brightness interference. When a black screen or white screen anomaly is triggered simultaneously and the phase entropy value is lower than the preset phase entropy threshold, it is determined to be a mixed anomaly.

7. The video quality multi-dimensional intelligent detection method according to claim 1, characterized in that, The verification process is performed based on the preliminary anomaly detection results and the motion information of the current frame, specifically including: Calculate the absolute difference between the current frame and the previous frame to obtain the motion energy index; When the preliminary judgment result is periodic brightness interference and the motion energy index is higher than the preset motion threshold, the preliminary judgment result is marked as a normal motion scene and no alarm is triggered. When the initial determination result is periodic brightness interference and the motion energy index is not higher than the preset motion threshold, the existence of periodic brightness interference is confirmed, and alarm preparation is triggered. When the initial judgment result is a black screen, white screen, or mixed abnormality, motion verification is not performed, and the alarm preparation is initiated directly.

8. The video quality multi-dimensional intelligent detection method according to claim 1, characterized in that, It also includes closed-loop feedback optimization steps: When the final determination is normal and the exercise energy index is below the preset exercise threshold, the statistical mean and standard deviation of the normal phase entropy are updated using an exponentially weighted moving average method: ; ;in This is the current phase entropy value. This is the statistical mean of normal phase entropy. For normal phase entropy variance, Preset forgetting factor; The preset phase entropy threshold is recalculated based on the updated statistics, so that the threshold is adaptively adjusted as the scene changes. The adjustment coefficient of the dynamic modulation factor is dynamically adjusted based on historical events in which periodic interference is confirmed or filtered. The system performs backtracking processing on batch-stored historical video files, executes detection steps frame by frame, records all triggered abnormal events and their associated parameters, and determines the initial value of the preset phase entropy threshold offline based on the phase entropy statistics of multiple historical normal videos.

9. The video quality multi-dimensional intelligent detection method according to claim 8, characterized in that, After updating the statistical mean and standard deviation of the normal phase entropy, the following is also included: Obtain the short-term mean of the phase entropy values ​​within the most recent second time period, where the second time period is less than the duration of historical data used to calculate the statistical mean and standard deviation; The rate of change of brightness is calculated based on the average brightness of the current frame and the average brightness of the previous frame. The short-term mean is compared with the statistical mean. When the short-term mean is consistently lower than the statistical mean and no periodic brightness interference or mixing anomaly is triggered, the preset phase entropy threshold or the update strategy of the statistical mean for subsequent frames is dynamically adjusted based on the comparison result of the brightness change rate and the preset change rate threshold.

10. The video quality multi-dimensional intelligent detection method according to claim 9, characterized in that, The strategy of dynamically adjusting the preset phase entropy threshold or the statistical mean of subsequent frames based on the comparison result of the brightness change rate and the preset change rate threshold specifically includes: When the brightness change rate is less than the first change rate threshold, it is determined that the environment is changing slowly. Then, the statistical mean is corrected towards the short-term mean according to the preset step size, and the preset phase entropy threshold is recalculated based on the corrected statistical mean. When the brightness change rate is greater than the second change rate threshold, it is determined to be an environmental change. The update of the statistical mean and standard deviation is then paused, and the preset phase entropy threshold is temporarily increased according to a preset ratio. After the brightness change rate recovers to below the second change rate threshold and remains below it for a third duration, the update is resumed and the preset phase entropy threshold is restored.

Images