A CG animation video detection method based on large model and multi-feature data processing.

Abstract

This invention discloses a CG animation video detection method based on large-model multi-feature data processing, belonging to the field of computer vision technology, including the following steps: S100, constructing a specular reflection-motion coupling observation layer, acquiring geometric feature parameters of the reflected light path in a dynamic scene, generating a reflection phase polarity baseline, and establishing a dual-path topological fingerprint based on the baseline as a detection reference standard; S200, running a time-frequency polarity consistency discriminator, parsing the dual-path topological fingerprint, identifying the phase reversal window, calculating the coherence deviation of the corresponding frequency band, and generating a misleading risk cone with boundary features. This invention, by constructing a reflection-motion observation layer and a topological fingerprint reference, combined with time-frequency consistency discrimination, counterfactual playback, and optical path reconstruction, achieves phase reversal recognition and correction, and effectively improves the accuracy and robustness of CG video authenticity judgment through optical field interference and frequency band modulation of the broken closed-loop coherent path.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, in particular to a CG animation video detection method based on large model multi-feature data processing. BACKGROUND

[0002] "CG animation video detection based on large model multi-feature data processing" refers to using a large-scale artificial intelligence model with strong learning and reasoning ability to perform multi-dimensional and multi-modal feature extraction and analysis on CG (computer graphics) animation videos, including texture details of image frames, time sequence consistency of motion trajectories, rendering characteristics of light and material, matching degree of sound and picture, etc. Through multi-feature comprehensive modeling and fusion processing, an intelligent detection mechanism is established to distinguish real videos from CG synthesized videos, identify abnormal rendering segments or forged content, so as to realize efficient discrimination and verification of the authenticity, integrity and quality of CG animation videos.

[0003] The prior art has the following disadvantages:

[0004] In the prior art, the detection of CG animation videos usually relies on optical features, spectral features and coherence discrimination mechanisms. However, when mirror materials or water objects move violently in the scene, their reflection paths will produce abnormal phase reversal phenomena due to sudden changes in incident angle and motion speed. Such phase reversal not only destroys the original optical consistency, but also generates false frequency band features that are highly similar to real video signals. In this case, the detection system is easily misled when performing coherence-based discrimination, mistaking false reflection features for credible signals, resulting in serious deviation of the overall discrimination result, thereby directly affecting the stability and reliability of the CG animation video detection process.

[0005] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a CG animation video detection method based on large model multi-feature data processing to solve the problems in the background art.

[0007] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a CG animation video detection method based on large model multi-feature data processing, comprising the following steps:

[0008] S100, construct a mirror reflection-motion coupling observation layer, collect geometric feature parameters of the reflection light path in a dynamic scene, generate a reflection phase polarity baseline, and establish a double-path topology fingerprint based on the reflection phase polarity baseline as a detection reference standard;

[0009] S200, running the frequency polarity consistency discriminator, analyzing the dual-path topology fingerprint, identifying the phase inversion window, calculating the coherent deviation of the corresponding frequency band, and generating a misleading risk cone with boundary characteristics;

[0010] S300, according to the misleading risk cone, starting the counterfactual playback chain, constructing the low-reflection shadow light path sequence to replace the abnormal segment, reconstructing the mapping relationship between the incident light path and the reflected light path, calculating the steady-state consistency score, and determining the phase correction budget;

[0011] S400, according to the phase correction budget, setting a double-mirror time scale, rearranging the time delay and correcting the difference for the reconstructed cross-frame video sequence, outputting the intervention time window, and determining the light path control target point set;

[0012] S500, based on the light path control target point set, synthesizing the time-varying virtual impedance light field, using the topological barrier energy landscape to suppress the reflected backflow signal, generating an inverse phase micro-pattern spectrum in the intervention time window, and outputting the execution rhythm parameters;

[0013] S600, according to the rhythm parameters, implementing time reversal light field traction in the intervention time window, injecting inverse phase micro-patterns into the light path, triggering frequency band migration combined with shadow light blocking array and programmable liquid surface micro-rhythm array, realizing the dynamic regulation and control of coherent misleading path breaking and closing, thereby completing the phase inversion correction and real signal recognition of CG animation video.

[0014] Preferably, step S100 comprises:

[0015] Extracting image frames from the video sequence, identifying the reflection region with mirror characteristics, and collecting incident light direction, reflection point spatial displacement parameters and rotation angle;

[0016] Based on the collected reflection path geometric parameters, generate a reflection phase polarity baseline, and calculate the phase polarity state identifier through optical phase offset direction;

[0017] According to the reflection phase polarity baseline, select the main reflection path and the auxiliary reflection path, construct a dual-path topology fingerprint map, including timestamp, path coordinate pair, phase state pair and energy difference;

[0018] Set static threshold and dynamic offset tolerance index, construct the topology map as a partition map, realize the division of stable zone, warning zone and abnormal zone, and use it for node-level matching and abnormal reflection identification in the subsequent detection process.

[0019] Preferably, step S200 comprises:

[0020] Take the reflection phase polarity baseline as the time axis reference, map the dual-path topology fingerprint frame by frame, and compare the phase polarity difference to form a frame-level polarity difference sequence;

[0021] Extract the time period of continuous difference in the polar difference sequence as the analysis window, obtain four types of physical characteristic parameters of brightness value change trend, high light spot position drift trajectory, edge texture gradient distribution, and reflection energy normalization value, and perform frequency domain analysis to identify the time window with phase inversion risk;

[0022] Calculate the path trajectory curvature, spatial offset, phase polarity change, and energy mutation features based on the identified time window, construct a three-dimensional error vector, and generate a misdirection risk cone;

[0023] Compare the misdirection risk cone with the double-path topology fingerprint, mark the risk level of the intersecting and overlapping path segments, and use the result as the basis for starting the counterfactual playback chain and the priority ranking of the light path intervention.

[0024] Preferably, step S300 comprises:

[0025] According to the time frame, spatial boundary, and frequency offset information of the misdirection risk cone, extract the geometric and optical parameters of the original reflection path in the abnormal area;

[0026] Construct a low-reflection shadow light path sequence in the undisturbed area and simulate a stable light path;

[0027] Match and replace the abnormal path according to the spatial position, direction angle, and phase change to complete the path reconstruction;

[0028] Based on the phase continuity, spatial stability, and frequency consistency of the replaced path, generate a stable consistency score;

[0029] According to the score result, output the intervention time range, light path correction area, and phase offset correction amount to form a phase correction budget.

[0030] Preferably, step S400 comprises:

[0031] Based on the time window in the phase correction budget, construct a double-mirror time section with the center frame as the symmetric axis, and extract the geometric and optical parameters of the reflection light path in each frame;

[0032] Based on the phase behavior difference between the mirror frames, perform a time delay rearrangement operation on the cross-frame video sequence, and correct the reflection path continuity through trajectory interpolation and reflection angle smoothing;

[0033] After completing the time axis alignment, identify the stable frame interval and extract the aggregation area of the reflection points in space, and generate a light path control target point set combining the phase, direction, and coordinate information.

[0034] Preferably, each target point in the light path control target point set comprises a time frame number, a pixel position in an image frame, an incident direction vector, a reflection direction vector, and a phase polarity value, for accurately marking the time position and direction parameters of subsequent light field intervention.

[0035] Preferably, step S500 comprises:

[0036] Based on the light path control target point set, a local coordinate system with the reflection direction as the center is constructed, an impedance value is set, and a laser source is driven to form a dynamic interference field, thereby generating a time-varying virtual impedance light field with variable phase control capability;

[0037] According to the reflection energy change of the target point in the continuous frame, a topological barrier energy landscape is constructed, a backflow path is identified, and a reverse phase disturbance wave is injected, thereby forming a topological barrier structure with energy suppression and phase gap characteristics;

[0038] In combination with the spatial distribution and time sequence of the reflection path in the topological barrier structure, an inverse phase micro-pattern spectrum is constructed, the injection time, injection direction, injection intensity, and phase inversion value corresponding to each pattern are extracted, and rhythm parameters are synchronously output.

[0039] Preferably, step S600 comprises:

[0040] According to the rhythm parameters, an inverse phase micro-pattern is injected into the light path, and a directional destructive interference zone is formed on the reflection path to achieve phase inversion suppression;

[0041] In combination with the shadow shading array, a non-target path is spatially shielded to enhance the propagation accuracy and interference effect of the inverse phase pattern in the target path;

[0042] Through the programmable liquid surface micro-ripple array, frequency migration is induced in the phase drift area, so that the abnormal reflection path loses coherence in the frequency domain and is disconnected;

[0043] A full-frame closed-loop control structure is established, and the authenticity of the reflection path is determined according to the consistency verification results in time and space, and identification and correction are completed.

[0044] In the above technical solutions, the present application provides technical effects and advantages:

[0045] The application can capture the geometric changes of the reflection path in the dynamic scene by constructing a mirror reflection-motion coupling observation layer, form a reflection phase polarity baseline and a double-path topological fingerprint, and provide a high-fidelity reference for subsequent identification; further, the phase inversion window and frequency band coherence deviation information are accurately extracted through time-frequency polarity consistency identification, and the false optical features that are easy to be misjudged are effectively identified; by introducing the counterfactual playback chain and the steady-state consistency scoring mechanism, the real light path mapping relationship is reconstructed and the correction demand is quantized, so that the detection process has a high recovery ability; in addition, based on the phase correction budget, the time axis alignment and the light path control target point set are generated, and a controllable intervention window is constructed to ensure the accurate positioning of the intervention operation; on this basis, the synthetic time-varying virtual impedance light field and the injected inverse phase micro-pattern are combined with the shadow shading array and the liquid surface micro-ripple array, multi-dimensional interference and regulation from physical reflection behavior to frequency band signal are realized, and finally the breaking and closed-loop control of the coherent misleading path are realized, effectively avoiding the misjudgment problem caused by high similarity false reflection signals in the prior art, and improving the recognition accuracy, anti-interference ability and engineering practicability of real and non-real content in the CG synthetic video. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0047] Figure 1 The method flowchart of the CG animation video detection method based on large model multi-feature data processing of the present application. DETAILED DESCRIPTION

[0048] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive concept to those skilled in the art.

[0049] The present application provides a CG animation video detection method based on large model multi-feature data processing as shown in Figure 1 The method flowchart of the CG animation video detection method based on large model multi-feature data processing of the present application.

[0050] S100, constructing a mirror reflection-motion coupling observation layer, collecting dynamic geometric characteristic parameters of the reflection light path in the dynamic scene, generating a reflection phase polarity baseline according to the collected dynamic geometric characteristic parameters, and establishing a double-path topological fingerprint according to the generated reflection phase polarity baseline, for forming a reference standard for subsequent detection process;

[0051] To realize high-precision identification of the reflection path mutation and phase polarity change of the mirror reflection target in the CG animation video in a dynamic scene, a method for constructing a mirror reflection-motion coupling observation layer is provided, which includes the following steps:

[0052] A dynamic observation framework of the reflection light path is constructed. In this step, each frame image is extracted from the video sequence, and the target area with mirror characteristics is identified by traversing the pixel area. The identification criteria of the mirror area are that the surface brightness value is higher than the set threshold, the edge area has obvious mirror reflection highlights, and there is a significant reflection moving track in the consecutive frames. The identified mirror area is labeled as a candidate reflection source area. Then, the incident light direction of the area in the current frame is obtained by using light modeling, and the spatial displacement parameters and rotation change angle of the area in the three-dimensional coordinate system are obtained by analyzing the motion vector of the area in the adjacent frames. At the same time, a three-dimensional geometric reconstruction method is used to model the whole process of the reflection path from the light source to the mirror and then to the observation point in space, and record the coordinates of the starting point, reflection point and end point of the path in each time frame, as well as the corresponding light propagation direction. The geometric description of each reflection path needs to include the path length, reflection angle, incident angle, surface normal vector, mirror deflection angle and its sequence position on the time axis, and finally form a complete dynamic geometric feature set.

[0053] Based on the dynamic geometric feature parameters of the reflection path, a reflection phase polarity baseline is generated. The generation of this baseline takes time as the index unit, and binds the amount of optical phase change involved in each frame of reflection path with the reflection geometric parameters. The calculation method is as follows: first, the amplitude information of the light at the reflection point and the change trend of the electric field direction of the incident light are extracted, and the possible phase shift direction at this position is calculated according to the electromagnetic wave propagation law; then, according to the positive and negative direction of the phase change, the polarity state identifier is given, where the positive phase is marked as "+1" and the negative phase is marked as "-1", and a phase polarity curve is constructed with time sequence as the horizontal axis. In this curve, each node contains the position coordinates of the reflection point in the three-dimensional space, the angle between the incident light and the reflected light, the phase shift amount and the polarity identifier value. Through the continuous frame of phase polarity data record, a phase reference line with time continuity and geometric dependence is formed, which is called reflection phase polarity baseline, and is used for subsequent construction of topological mapping.

[0054] The double-path topology fingerprint is established by using the reflection phase polarity baseline. In this step, two reflection paths with obvious differences are selected from the candidate reflection area. One is the main reflection path, which has the maximum illumination intensity and reflection stability. The other is the auxiliary path, which has the most severe change in reflection angle or the most frequent surface disturbance. The geometric characteristics of the two paths in the same time period are compared synchronously, mainly including: path length difference, reflection point displacement speed difference, phase polarity change amplitude, and reflection energy difference. Based on the time frame, the above four types of difference data are constructed into a topology mapping node. Each node is composed of a unique timestamp, a pair of spatial coordinates of the main path and the auxiliary path, a pair of phase polarity states, and an energy difference value. All nodes are arranged in time sequence to form a topology map, which is the double-path topology fingerprint. The fingerprint has stable structure and phase coupling continuity, and can be used as a core reference data for identifying the evolution characteristics of reflection behavior.

[0055] The double-path topology fingerprint is constructed as a subsequent detection reference standard. The static threshold and dynamic offset tolerance indicators are introduced into the double-path fingerprint to establish the reference standard. The static threshold is defined as the maximum tolerance range of phase polarity difference under normal reflection behavior. The dynamic offset tolerance is defined as the acceptable offset interval of the reflection point caused by the object's own motion in consecutive frames. By marking these threshold boundaries in the topology fingerprint map, a partition map containing three types of nodes, stable zone, warning zone, and abnormal zone, is formed. In the subsequent detection process, the real-time collected reflection path data can be matched with the reference map at the node level. The number of nodes falling into the abnormal zone in the matching result can be used to determine whether the current reflection is abnormal. In this way, in the scene of high-speed motion of mirror objects, sudden change of water surface refraction angle, or rapid change of viewing angle, the established double-path topology fingerprint and phase polarity baseline can still be used to accurately compare and capture abnormal reflections, thereby laying a solid foundation for subsequent counterfactual reconstruction and light path correction.

[0056] S200, based on the reflection phase polarity baseline, a time-frequency polarity consistency discriminator is run to analyze the double-path topology fingerprint, identify and locate the phase inversion window with abnormal reflection, calculate the frequency band coherence deviation corresponding to the phase inversion window, and generate a misleading risk cone with boundary features;

[0057] In order to accurately identify the phase inversion phenomenon caused by mirror reflection, based on the established reflection phase polarity baseline and double-path topology fingerprint, a polarity consistency discrimination method combining time domain, frequency domain and spatial coupling information is proposed. The implementation process includes the following steps:

[0058] The time alignment and phase polarity comparison of the dual-path topology fingerprint are performed. Using the reflection phase polarity baseline generated in the previous step as the main time axis reference, each pair of nodes in the dual-path topology fingerprint is mapped frame by frame. This mapping includes two dimensions of information: one is the time frame number, and the other is the three-dimensional spatial coordinates of the reflection points of the main path and the auxiliary path in the same frame. During the mapping process, the position of each pair of path reflection points is extracted, and the X, Y, and Z axis coordinates are recorded respectively. At the same time, the incident angle, reflection angle, and surface normal direction of the point are extracted and compared with the phase identifier of the corresponding frame in the reflection phase polarity baseline. According to the reflection phase identifier, it is judged whether the phase of the main path and the auxiliary path in the current frame is consistent. If it is consistent, it is marked as "0", if it is opposite, it is marked as "1", and the results of each frame are arranged into a sequence to form a frame-level polarity difference sequence. At the same time, in order to avoid the influence of error fluctuations on the determination result, the illumination intensity, surface material reflectivity, and specular highlight offset in each frame also need to be recorded, and the stability of the reflection environment of the frame is evaluated accordingly. Finally, the sequence is input into the next frequency feature analysis.

[0059] The local frequency feature of the frame-level polarity difference sequence marked as "1" is developed, and the time window where the phase reversal may exist is identified. In this process, the time period where the polarity difference exists continuously is selected, and 3 frames before and after are selected as context frames, forming an analysis window with a length of 7 frames. For the reflection area in the window, the brightness value trend graph, the specular highlight spot center position drift trajectory, the edge texture gradient direction distribution graph, and the reflection energy normalized value of each frame are extracted. Then, the frame parameters are arranged in time order to form a frequency sequence, and the main frequency and secondary frequency of each parameter are extracted by fast Fourier transform. For the brightness value spectrum, the energy distribution peak position in the frequency domain is extracted; for the highlight drift trajectory spectrum, the main direction change rate is calculated; for the reflection energy spectrum, the main frequency band range of the energy decay rate is extracted. If these main frequencies have mutations in the time period, and the mutation direction is consistent with the phase polarity change direction, it is determined that there is a high risk of phase reversal in this time window. The analysis results recorded by each window include: the main frequency migration direction, the main frequency point offset amplitude, the reflection point trajectory rotation angle, and the illumination stability score, which are used as input for subsequent coherent deviation evaluation.

[0060] Based on the identified high-risk time window, coherent deviation calculation is performed, and combined with the topological structure information, a misleading risk cone is constructed. This step first performs three-dimensional space backtracking on the main path and auxiliary path reflection points in each high-risk time window, extracts the spatial trajectory points of the reflection points in the continuous three frames in this time period, calculates the reflection point motion trajectory curvature, the angle between the trajectories, and the distance deviation between the space paths, and forms the path space deviation index set. Then, the phase polarity difference of each frame of reflection point, the reflection energy jump value and the brightness distribution deviation are used to construct the frequency domain deviation index set. After normalization, the above two types of indicators are aggregated in multiple dimensions by frame to generate a three-dimensional error vector. Taking the center frame of the window as the coordinate origin, the error vector variation trajectory is drawn in the three-dimensional space with time as the Z-axis, frequency offset as the X-axis, and spatial deviation as the Y-axis, and a misleading risk cone with clear boundaries and direction is constructed with the error extreme point as the radius. The shape of the cone reflects the diffusion trend of coherent misleading risk in time, the concentration degree in frequency and the influence range in space. Each point on the boundary of the cone has a unique corresponding time frame number, spatial position coordinate and frequency offset value.

[0061] The misleading risk cone is combined with the topological fingerprint constructed in the previous step to establish a complete polarity anomaly identification reference model. In this process, the extension path of the central axis of each cone on the time axis is overlapped and compared with the main path node sequence in the topological fingerprint, the path segment overlapping with the cone is marked, and the reflection surface position, normal direction and motion vector in space are tracked. According to the position level of the intersection area in the cone, different abnormal risk weights are assigned, for example, the frames located at the top edge of the cone are marked as “slight risk”, and the frames close to the bottom of the central axis are marked as “high risk”. These weighted node information will be used as the target area for the subsequent counterfactual playback chain start, and provide a basis for the selection of alternative light path, reconstruction of reflection mapping relationship and priority sorting of light field intervention.

[0062] S300, according to the misleading risk cone, a counterfactual playback chain is started, a low reflection shadow light path sequence is constructed to replace the abnormal segment in the detection sequence, the mapping relationship between the incident light path and the reflection light path is reconstructed, and the steady-state consistency score is calculated based on the reconstruction result and the deviation information of the misleading risk cone, and the phase correction budget is determined;

[0063] To cope with the phase inversion phenomenon caused by the sudden change of mirror reflection path, the counterfactual playback chain is started by combining the information of the misleading risk cone, the incident-reflection mapping relationship is reconstructed, and the steady-state consistency is evaluated, and finally the quantifiable phase correction budget is determined. The implementation process includes the following steps:

[0064] Based on the time frame range and spatial boundary coordinates marked in the misleading risk cone, the original reflection path information of the abnormal reflection area is extracted. In the previous step, the misleading risk cone has provided three types of parameters for each high-risk area: first, the start and end frame numbers define the analysis window in time; second, the spatial projection boundary describes the pixel area where the reflection anomaly is concentrated in this time period; third, the frequency offset direction and spatial disturbance intensity indicate the severity of the current area deviating from the normal reflection path. On this basis, all reflection paths located within the boundary of the cone are extracted from the original video data, including the three-dimensional coordinates of the incident point and the reflection point, the corresponding incident light and reflection light direction, the reflection surface normal vector, the specular reflectivity, and the illumination energy value. Subsequently, the continuity of the extracted paths in each time frame is evaluated to determine which paths have mutations, breaks, or irregular jumps. These paths are defined as replacement paths and are used as replacement targets for shadow light paths.

[0065] A low-reflection shadow light path sequence is constructed for replacement to simulate ideal reflection paths in a stable state. The specific method is as follows: In the current frame or adjacent frame, find the same type of reflection area that is not affected by abnormal interference, and the screening criteria are: the reflection energy is in the medium-low range, the mirror high light area is less than the set threshold, the surface normal direction is stable and has no sharp corner. After selection, these areas are finely modeled to extract their surface attribute parameters, including material refractive index, micro-texture roughness, mirror reflection intensity, and reflection surface slope angle. Based on these parameters, combined with the direction of the ambient light source and the observation angle of the current frame, the light path trajectory from the incident point to the reflection point and then to the observation point is simulated and generated, and the extension direction, reflection angle, energy attenuation distribution, and phase delay change of each path in space are recorded. With time sequence as the axis, the shadow light paths of continuous multiple frames are arranged and combined into a sequence to form a stable optical path database for subsequent replacement operations.

[0066] The shadow light path replacement operation on the original abnormal reflection path is performed to complete the preliminary reconstruction of the mapping relationship. In this process, in frame units, for each original reflection path judged as abnormal, the most matching path is found from the shadow light path database for one-to-one replacement. The matching criteria include three-dimensional spatial position proximity (i.e., the Euclidean distance between the reflection points), reflection direction angle difference (the angle difference between the incident light and the reflection light must not exceed the set threshold), and phase change rate similarity (comparing the phase drift amount per unit time). After a successful match, the incident point, reflection point, phase parameter, and reflection energy value of the original path are all replaced with the corresponding data in the shadow light path, while the original texture, shape, and boundary of the image frame remain unchanged to ensure visual continuity. After the replacement is completed, the incident-reflection mapping structure of each frame in the entire sequence is reconstructed to form a complete alternative light path time sequence.

[0067] The steady-state consistency of the reconstructed light path sequence is evaluated and compared with the initial deviation information of the misguidance risk cone to evaluate the replacement effect. The specific evaluation process includes the following three quantitative operations: first, the continuity score between the phase polarity of the replacement path in each frame and the adjacent frame is calculated to determine whether there is mutation or inversion; second, the spatial propagation direction change rate of each light path and the smoothness of the reflection point trajectory are extracted to evaluate the stability of the path on the time axis; third, the frequency response curve of the reconstructed path is compared with the frequency offset characteristics of the misguidance risk cone to determine whether the main frequency band is restored and the energy concentration is recovered. Through the weighted average of the above three indicators, a steady-state consistency score curve is generated, and the minimum value, average value and change rate of the curve are used as a reference to reflect the stability and rationality of the replacement path. The scoring result not only quantifies the success of the current replacement operation, but also reveals the time and spatial location of the potential abnormal area.

[0068] According to the steady-state consistency score result, a phase correction budget is formulated to provide a quantitative basis for subsequent intervention operations. The budget parameters include three items: first, the intervention time range, defined as the continuous frame segment in the score curve where the score is below the threshold, which is the core execution window for subsequent intervention operations; second, the light path correction region, defined as the reflection path projection region in the frame with the lowest score, whose spatial range is expressed in the form of specific pixel boundaries in the original image; third, the phase offset correction amount, defined as the phase drift difference between the current shadow light path and the original path, which will be used as the phase control parameter for subsequent inverse phase pattern injection operations. The final output phase correction budget includes the three parameter sets of time coordinates, spatial positions and phase offsets, providing accurate and executable technical support for the active control stage.

[0069] S400, set double-mirror time scale according to phase correction budget, perform time delay rearrangement and difference correction on the reconstructed cross-frame video sequence, align the time axis of the video sequence, output the corresponding intervention time window, and determine the light path control target point set corresponding to the intervention time window;

[0070] To ensure that the CG animation video frame sequence after the shadow light path replacement and phase reconstruction maintains consistency in the time and spatial dimensions, according to the phase correction budget output in the previous stage, set the double-mirror time scale, perform time delay rearrangement and difference correction on the cross-frame sequence, generate a strictly aligned time axis structure, and output the light path control target point set. The process includes the following steps:

[0071] Based on the time window parameter in the phase correction budget, a double-mirror time scale structure of the time axis is constructed. In the previous step, the time range in which the phase offset occurs is determined, including the starting frame number, the ending frame number, and the center frame number. Taking the center frame as the time symmetry axis, the time symmetry interval is set forward and backward respectively, forming a double-mirror time section with a symmetric structure. For example, if the center frame is the 50th frame and the correction budget sets the window to 20 frames, the front mirror section contains the 40th to 49th frames, and the rear mirror section contains the 51st to 60th frames. All frames in the double-mirror section are numbered and remapped, and the light path data related to the reflection in each frame is recorded, including the three-dimensional coordinates of the reflection point, the incident light and reflected light vector direction, the phase polarity state, the reflection energy intensity, the mirror high light profile change amount, and the relative change rate of the observation angle. Subsequently, based on the time symmetry relationship of the mirror frame, the offset point on the reflection path behavior timeline is spatially mapped with the reflection point in the symmetric frame, and it is determined whether there is an asymmetric phase behavior on the time axis, such as inconsistent phase drift direction, sharp reflection energy difference, or reflection angle mutation phenomenon.

[0072] According to the time inconsistency region found in the mirror time scale, the time delay rearrangement operation is performed on the reconstructed cross-frame video sequence, and the phase space trajectory difference correction is implemented. The specific operation is as follows: for each pair of front and back symmetric frames, the time difference value of the reflection path is calculated. If the timestamp offset between the front and back frames exceeds the set threshold (such as more than 10% of the average sampling interval), it needs to be time-retracted to the center frame, so that the front and back two frames are symmetrically distributed with the same time interval. Subsequently, the spatial trajectory of the reflection point in each frame is drawn as a three-dimensional trajectory line in time sequence, and if it is found that the trajectory line has a mutation or a break, a transition point is inserted between the two adjacent points to repair the trajectory continuity. The generation method of the transition point is as follows: taking the vector average of the two adjacent points as the center point in space, and setting the phase polarity and reflection direction of the point according to the intermediate value of the phase change between the two points. The above operation is performed on the continuous time sequence to obtain a set of three-dimensional reflection trajectories after continuous interpolation correction. In addition, the reflection angle curve with time needs to be smoothed, and the weighted sliding mean method is used to balance the angle fluctuation caused by the insertion of shadow light path, so that the overall reflection behavior has physical continuity and time sequence consistency. Through the above time rearrangement and trajectory difference correction, a time-space coupled unified reflection behavior model is established, which effectively smooths the discontinuity effect caused by path replacement.

[0073] After completing the time domain alignment of the reconstruction sequence, according to the corrected sequence structure, the intervention time window is extracted, and the light path control target point set is identified. First, the most stable continuous frame interval on the time axis is confirmed, and the stability criteria are: the reflection point trajectory offset between adjacent three frames does not exceed the set threshold (such as 0.5 pixels), the reflection energy difference is less than the preset light jitter tolerance (such as 5%), and the phase polarity has no mutation. The selected frame segment is used as the optical intervention time window. In this window, spatial clustering analysis is performed on all reflection paths, and the reflection points in three-dimensional space are clustered according to density and stability to identify the reflection path area with the strongest response and the highest stability. The specific clustering criteria are: high concentration of reflection points, strong consistency of reflection direction, and continuous phase change trend. After clustering, the representative coordinate points of each reflection area are extracted, and their pixel coordinates in the image frame, incident direction, reflection direction, reflection angle, and phase polarity state are recorded to form a target point set with time sequence attributes. The set is exported as a data table structure, where each item contains five parameters: time frame number, pixel position, incident direction vector, reflection direction vector, and phase polarity value. The target point set not only provides the spatial positioning basis for light path intervention operation, but also clearly defines the timing of intervention execution, ensuring that subsequent injection operations can accurately reach the most sensitive reflection path position in space and time, significantly improving the targeting and intervention efficiency of light field regulation.

[0074] S500, synthesizing a time-varying virtual impedance light field based on the light path control target point set, inhibiting the reflection backflow signal according to the topological barrier energy landscape, generating an inverse phase micro-pattern spectrum in the intervention time window, and outputting rhythm parameters for execution;

[0075] To ensure effective control of the backflow behavior of abnormal reflection paths in CG animation videos within the specified intervention time window, a time-varying virtual impedance light field needs to be constructed based on the light path control target point set, and the reflection signal propagation mode needs to be adjusted relying on the topological barrier energy landscape model, so as to generate an inverse phase micro-pattern spectrum opposite to the phase inversion direction, and output rhythm parameters with time structure. This technical path includes the following steps:

[0076] According to the generated light path control target point set, a time-varying virtual impedance light field is synthesized in three-dimensional space. Each target point in the target point set contains the following physical parameters: time frame number, two-dimensional pixel position, three-dimensional space coordinates, incident light direction, reflected light direction, reflection angle, reflection point normal vector, reflection energy value and phase polarity state. The synthesis process starts from the reflection area of the target point, establishes a local coordinate system with the reflection direction of each target point as the axis, and defines the impedance distribution function in the coordinate system. The setting of the impedance value is based on the reflection energy value and the phase shift trend of the target point. The greater the reflection energy and the more severe the phase shift, the lower the corresponding impedance value, so as to enhance the modulation ability of the area to the light signal. In each frame of time, the impedance distribution formed by all target points is dynamically encoded into a set of spatial distribution maps, and a high-frequency light modulation matrix is used to drive the laser source to form a weak interference field with variable phase control ability in the target point projection area. This interference field does not directly reflect the visible image content, but changes the propagation path and phase response of the original reflected signal in a very weak fluctuation interference manner, forming a virtual impedance light field that is dynamically changing in time and space domains. In order to maintain the timing consistency, the update frequency of the light field is set to be completely consistent with the video playback frame rate, for example, 30 frames per second, corresponding to impedance distribution redraw every 33.3 milliseconds.

[0077] A topological barrier energy landscape is established to control the reflection signal backflow path and implement energy suppression. The construction basis of the energy landscape is the reflection energy fluctuation spectrum of each target point area in multiple frames of time. The specific operation is as follows: in the intervention time window, five consecutive frames of images are extracted, the reflection energy values of all target points in each frame are recorded, and the energy fluctuation sequence is formed in time sequence. Then, the spatial distance and reflection angle difference between adjacent target points are calculated to construct the topological connection diagram of the reflection path. The reflection energy peak value of the target point is taken as the energy source point, and the energy contour line diagram is drawn according to the energy gradient direction to form the energy flow spectrum in the reflection path. In the spectrum, the path area where the energy quickly gathers is identified as a high-risk backflow path. In these paths, phase interference points are further inserted, and artificial reverse offset phase change values are set, for example, taking phase inversion 180 degrees as the standard, to generate phase gap areas. The phase gap can be realized by controlling the laser grating array to inject reverse phase interference signals in the corresponding area. The specific injection method is to apply a weak phase disturbance wave opposite to the main reflection signal to each target point in the area in its reflection direction, and keep the interference phase stable. On this basis, the energy flow diagram and the phase gap jointly constitute a topological barrier structure, so that the light path that can form backflow interference is suppressed from the energy and phase dimensions at the same time, thereby blocking its propagation link in the time continuous frame.

[0078] Based on the topological barrier map and the phase blocking area, the inverse phase micro-pattern spectrum is generated, and the rhythm parameters are extracted for synchronous execution. First, in the marked reflection reflux blocking area, according to the time sequence trajectory of the reflux path, the spatial position of the target point in each frame, the reflection direction and the main phase polarity are extracted, and the phase disturbance data set is established. Then, according to the spatial arrangement order and the time advancement order of the reflection points in the data set, a set of inverse phase pattern sequences is designed, each pattern in the sequence contains the following parameters: target point number, injection time (identified by frame number), inverse phase value (identified by ±π phase inversion degree), injection direction vector, injection energy level (used to control the interference intensity). These inverse phase patterns constitute a complete interference sequence, with time consistency and spatial continuity. The arrangement of the patterns in space must conform to the energy distribution structure in the topological barrier energy landscape, and preferentially cover the energy accumulation points and phase shift strong points to form an accurate suppression coverage. To ensure that the pattern and the video frame play time are aligned, corresponding rhythm parameters need to be generated, each set of rhythm parameters including: corresponding pattern injection time frame number, injection duration, interference frequency (number of injections per unit time) and injection intensity level. All rhythm parameters are arranged to form a time sequence table, which is used as a reference for operation driving in the subsequent light field injection stage.

[0079] S600, according to the rhythm parameters, time reversal light field traction is implemented in the intervention time window, inverse phase micro-patterns are injected into the light path, combined with the shadow shading array and the programmable liquid surface micro-ripple array to trigger frequency band migration, to perform loop breaking processing and closed loop dynamic regulation of coherent misdirection path, to realize detection and correction of reflection phase inversion in CG animation video and real signal identification;

[0080] To realize the identification and correction of phase inversion behavior caused by reflection path abnormalities in CG animation video, the rhythm parameters and light path target points obtained in the previous sequence are used to perform time reversal light field traction operation in the intervention time window, and combined with physical structure intervention means, the reflection path is synchronously broken and closed loop regulated, the whole process includes the following steps:

[0081] According to the rhythm parameters, the time reversal pulling operation is performed on the light path within the intervention time window. By precisely injecting the inverse phase micro-pattern, the reflection path is driven back to the initial state. In the previous stage, each rhythm parameter corresponds to a set of light path control target points, including specific injection time, pattern inverse phase value, injection energy level, spatial direction vector, and other physical quantities. In this step, the pattern injection time defined in the rhythm parameters is synchronized with the frame time, and the interference phase of the light source is configured according to the inverse phase value. Each pattern is composed of high-frequency beams with a phase shift of 180 degrees, which are projected to the target point area at a precisely controlled angle, causing the incident-reflection geometry in the reflection path to be forced to reverse, forming a directional destructive interference region. The core mechanism of light field pulling is to use phase reversal interference to suppress the phase inversion interval in the reflection path, thereby preventing the backflow path from continuously evolving. In order to maintain the consistency of the light field and the video timeline, the pattern injection needs to be updated every rhythm cycle, and the update frequency should be consistent with the video playback frame rate, for example, 30 frames per second, which means that the light field pulling is completed every 33.3 milliseconds. This pulling behavior forms a three-dimensional light region in the physical space and a continuous wave field in the time domain, constituting a sustained interference effect on the abnormal reflection path.

[0082] The light field pulling region is operated in cooperation with the shadow light shielding array to shield non-target path light and prevent the inverse phase pattern propagation path from being disturbed by irrelevant reflections. In order to accurately control the spatial distribution of light energy and avoid the interference of stray light generated by highly reflective surfaces with the pulling operation, a set of electrically controlled variable light transmission arrays is used to construct the shadow light shielding structure. This light shielding structure is placed around the pattern injection area to control the passage of light in a specific direction. Before each frame injection, the direction vectors of all light path control target points in the frame are pre-read, and their projection angles are compared and analyzed with other non-target reflection directions. If the angle difference is less than the set threshold (e.g. 3°), it is judged that there may be overlapping interference. Then, the control unit corresponding to the direction in the light shielding array is set to be opaque, forming a local light field isolation region. This step ensures that the inverse phase micro-pattern only propagates on the target reflection path and is not disturbed by multiple light sources and multiple reflective surfaces in the dynamic scene. The light shielding array uses real-time update control logic, with a running frequency synchronized with the frame refresh rate, and adjusts the geometry of the light shielding region adaptively according to the target direction changes. In this way, the pattern pulling region at each time point is kept optically pure, enhancing the interference accuracy.

[0083] The programmable liquid micro-rhythm array is used to cause frequency band migration on the pattern traction path, so that the reflected signal is separated from the original coherent path in the frequency domain, and the interference is broken. The liquid micro-rhythm array is composed of a group of deformable liquid units with dynamically controllable surface tension. Through precise voltage control, each liquid unit surface forms a micro-rhythm structure that oscillates at a predetermined frequency. When the incident light passes through the array area, the liquid disturbance will cause a slight shift in the refraction angle of the light path, and cause energy transfer of the frequency component. In this embodiment, the array is placed above the phase shift high-risk area of the reflection path, and the corresponding liquid unit is activated at the specified time point of the rhythm parameter, so that the surface wave frequency matches the rhythm of the traction pattern. For example, when the pattern rhythm is 30 injections per second, the liquid micro-rhythm frequency is set to 30 Hz, realizing the synchronous control of spatial position and time rhythm. Since frequency band migration can cause the main component of the original frequency to shift to the adjacent non-main frequency area, the coherence on the original path is destroyed, and the reflected signal no longer has sustained interference capability, achieving the effect of breaking the coherent chain. In addition, the array can also adjust the surface tension strength frame by frame, and can also adjust the migration frequency bandwidth, realizing the response suppression of signals of different energy levels.

[0084] After realizing reverse traction, optical shielding and frequency chain breaking, a full-frame closed-loop control structure is constructed to complete the closed control of the coherent path in spatial and temporal dimensions, realizing the identification and correction of the reflection phase inversion content in CG animation video. Under this structure, the phase inversion backflow signal generated by abnormal reflection behavior is suppressed in time by the inverse phase pattern, the propagation path is isolated in space by the light shielding array, and the frequency response is migrated in the frequency spectrum domain by the liquid disturbance. The three form a three-dimensional physical level interruption mechanism of reflection interference. On this basis, the system can establish a dynamic closed-loop tracking channel to check the consistency of the reflection path phase state of each frame with the previous and next frames. The checking content includes the reflection path continuity, phase polarity continuity and energy backflow closure. When non-physical continuous features such as path mutation, inversion and frequency transition are detected, the reflection content in this area is marked as non-real signal, and the real-time repair process or labeling output is triggered.

[0085] The present application can capture the geometric change of the reflection path in the dynamic scene by constructing a mirror reflection-motion coupling observation layer, form a reflection phase polarity baseline and a double-path topology fingerprint, and provide a high-fidelity reference for subsequent identification; further, through time-frequency polarity consistency identification, the phase inversion window and frequency band coherence deviation information are accurately extracted, and the false optical features that are easy to be misjudged are effectively identified; by introducing the counterfactual playback chain and the stable state consistency scoring mechanism, the real light path mapping relationship is reconstructed and the correction demand is quantized, so that the detection process has high recovery ability; in addition, based on the phase correction budget, the time axis alignment and the light path control target point set generation are performed, a controllable intervention window is constructed, and the precise positioning of the intervention operation is ensured; on this basis, by synthesizing the time-varying virtual impedance light field and injecting the inverse phase micro-pattern, and combining the shadow shading array and the liquid surface micro-ripple array, multi-dimensional interference and regulation from physical reflection behavior to frequency band signal are realized, and finally the breaking and closed-loop control of the coherent misleading path are completed, the misjudgment problem caused by the high similarity false reflection signal in the prior art is effectively avoided, and the recognition accuracy, anti-interference ability and engineering practicability of the real and non-real content in the CG synthesized video are improved as a whole.

[0086] The foregoing merely describes certain exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present application. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present application.

Claims

1. A CG animation video detection method based on large model multi-feature data processing, characterized in that, The method comprises the following steps: S100, constructing a mirror reflection-motion coupling observation layer, collecting geometric characteristic parameters of a reflected light path in a dynamic scene, generating a reflection phase polarity baseline, and establishing a double-path topology fingerprint based on the reflection phase polarity baseline as a detection reference standard; S200, running a time-frequency polarity consistency discriminator, analyzing the double-path topology fingerprint, identifying a phase inversion window, calculating a coherent deviation of a corresponding frequency band, and generating a misleading risk cone with boundary characteristics; S300, starting a counterfactual playback chain according to the misleading risk cone, constructing a low-reflection shadow light path sequence to replace an abnormal segment, reconstructing a mapping relationship between an incident light path and a reflected light path, calculating a steady-state consistency score, and determining a phase correction budget; S400, setting a double-mirror time scale according to the phase correction budget, performing time delay rearrangement and difference correction on the reconstructed cross-frame video sequence, outputting an intervention time window, and determining a light path control target point set; S500, synthesizing a time-varying virtual impedance light field based on the light path control target point set, using a topological barrier energy landscape to suppress a reflected backflow signal, generating an inverse phase micro-pattern spectrum in the intervention time window, and outputting an execution rhythm parameter; S600, implementing time reversal light field traction in the intervention time window according to the rhythm parameter, injecting an inverse phase micro-pattern into the light path, and triggering frequency band migration in combination with a shadow light shielding array and a programmable liquid surface micro-rhythm array.

2. The CG animation video detection method based on large model multi-feature data processing according to claim 1, characterized in that, Step S100 comprises: extracting image frames from a video sequence, identifying a reflection region with mirror characteristics, and collecting incident light direction, reflection point spatial displacement parameters and rotation angle; based on the collected reflection path geometric parameters, generating a reflection phase polarity baseline, and calculating a phase polarity state identifier through optical phase offset direction; according to the reflection phase polarity baseline, selecting a main reflection path and an auxiliary reflection path, and constructing a double-path topology fingerprint atlas; set a static threshold and a dynamic offset tolerance index, construct the topology atlas into a partition map, and realize the division of stable area, warning area and abnormal area.

3. The CG animation video detection method based on large model multi-feature data processing according to claim 1, characterized in that, Step S200 comprises: taking the reflection phase polarity baseline as a time axis reference, mapping the double-path topology fingerprint frame by frame, comparing the phase polarity difference, and forming a frame-level polarity difference sequence; extracting the time period with continuous difference in the polarity difference sequence as an analysis window, obtaining four types of physical characteristic parameters including brightness value change trend, high light spot position drift trajectory, edge texture gradient distribution and reflection energy normalization value, and performing frequency domain analysis to identify the time window with phase inversion risk; based on the identified time window, calculate the path trajectory curvature, spatial offset, phase polarity change and energy mutation characteristics, construct a three-dimensional error vector, and generate a misleading risk cone; compare the misleading risk cone with the double-path topology fingerprint, mark the risk level of the cross-overlapping path segment, and use it as the basis for starting the counterfactual playback chain and the priority sorting basis for light path intervention.

4. The CG animation video detection method based on large model multi-feature data processing according to claim 1, characterized in that, Step S300 comprises: extracting the geometric and optical parameters of the original reflection path in the abnormal area according to the time frame, spatial boundary and frequency offset information of the misleading risk cone; construct a low-reflection shadow light path sequence in the undisturbed area and simulate a stable light path; The abnormal path is matched and replaced according to the spatial position, direction angle and phase change, and the path reconstruction is completed; Based on the phase continuity, spatial stability and frequency consistency of the replaced path, a steady-state consistency score is generated; According to the score results, the intervention time range, light path correction area and phase offset correction amount are output to form a phase correction budget.

5. The CG animation video detection method based on large model multi-feature data processing according to claim 1, characterized in that, Step S400 includes: Based on the time window in the phase correction budget, a double-mirror time section with the center frame as the symmetry axis is constructed, and the reflection light path geometry and optical parameters in each frame are extracted; Based on the phase behavior difference between the mirror frames, a time delay rearrangement operation is performed on the cross-frame video sequence, and the reflection path continuity is corrected through trajectory interpolation and reflection angle smoothing; After completing the time axis alignment, the stable frame interval is identified and the aggregation area of the reflection points in space is extracted, and the light path control target point set is generated combining the phase, direction and coordinate information.

6. The CG animation video detection method based on large model multi-feature data processing according to claim 5, characterized in that, Each target point in the light path control target point set includes a time frame number, a pixel position in an image frame, an incident direction vector, a reflection direction vector and a phase polarity value, which is used to accurately mark the time position and direction parameters of subsequent light field intervention.

7. The CG animation video detection method based on large model multi-feature data processing according to claim 1, characterized in that, Step S500 includes: Based on the light path control target point set, a local coordinate system with the reflection direction as the axis is constructed, the impedance value is set and the laser source is driven to form a dynamic interference field, generating a time-varying virtual impedance light field with variable phase control capability; According to the reflection energy change of the target points in the continuous frames, a topological barrier energy landscape is constructed, the backflow path is identified and a reverse phase disturbance wave is injected to form a topological barrier structure with energy suppression and phase gap characteristics; Combined with the spatial distribution and time sequence of the reflection path in the topological barrier structure, an inverse phase micro-pattern spectrum is constructed, the injection time, injection direction, injection intensity and phase inversion value of each pattern are extracted, and rhythm parameters are output synchronously.

8. The CG animation video detection method based on large model multi-feature data processing according to claim 1, characterized in that, Step S600 includes: According to the rhythm parameters, the inverse phase micro-pattern is injected into the light path to control the reflection path to form a directional destructive interference area to realize phase inversion suppression; Combined with the shadow shading array, the non-target path is spatially shielded to enhance the propagation accuracy and interference effect of the inverse phase pattern in the target path; Through the programmable liquid surface micro-array, frequency migration is induced in the phase drift area, so that the abnormal reflection path loses coherence in the frequency domain to form a link disconnection; A full-frame closed-loop control structure is established, and the authenticity of the reflection path is determined according to the consistency verification results in time and space, and identification and correction are completed.

Images