An event encoding, dynamic vision generation and timeline adaptive synchronization method and system based on multi-source time series data

Abstract

The application discloses a kind of event encoding, dynamic visual generation and time axis self-adaptive synchronization method and system based on multi-source time series data, it is related to the technical field of cleaning device, including obtaining at least two kinds of heterogeneous time series data, unified time reference and quality control;Extract trend, peak and valley, mutation, period and confidence feature, form event encoding sequence with time anchor point;Event encoding is converted into visual control vector constrained by continuity, drive dynamic image or animation sequence generation;Establish the double time reference of sensing time axis and media rendering time axis, calculate drift error according to event anchor point and correct frame length, transition time and rendering parameter;Compare target event trajectory with actual rendering characteristic trajectory, close-loop correction visual control vector.The present application can reduce the cumulative drift and mutation jitter when heterogeneous time series data is converted into dynamic visual content, improve time consistency, continuity and real-time processing stability.

Description

Technical Field

[0001] This invention relates to the technical field of cleaning devices, and specifically to a method and system for event encoding, dynamic visual generation, and adaptive synchronization of time axis based on multi-source time-series data. Background Technology

[0002] Wearable devices, motion sensors, interactive terminals, and industrial equipment can continuously output time-series data such as heart rate, activity load, speed, acceleration, posture, interaction frequency, and environmental parameters. To make large amounts of continuous data easier to understand, existing systems typically plot it as curves, dashboards, or directly map the values ​​to media parameters such as pitch, color, graphic size, and animation speed according to preset rules.

[0003] However, the sampling frequency, timestamp accuracy, and communication latency of different data sources are not consistent. If the sampling points of each data source are directly mapped to the media timeline point by point, event boundary misalignment, cumulative time drift, and sudden jumps will occur; especially in the generation of dynamic images or animations that last for tens of seconds to several minutes, small clock deviations will accumulate into obvious audiovisual or data-visual asynchrony.

[0004] On the other hand, simple one-to-one parameter mapping usually only retains instantaneous values ​​and cannot stably express temporal structures such as trends, peaks and troughs, abrupt changes, and periods. Generative image or animation models may also suffer from problems such as excessive structural changes between adjacent frames, a mismatch between visual complexity and the intensity of input events, and frame drops that cause further time axis shifts when the computational load increases. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for event encoding, dynamic visual generation and adaptive synchronization of time axis based on multi-source time series data, so as to overcome the above-mentioned defects in the prior art.

[0006] A method for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data includes the following steps: S1. Obtain time series data from at least two data sources, assign source identifiers and collection timestamps to each data source, and establish a time mapping from the source time of each data source to the unified time based on a unified reference time to obtain time-aligned multi-source time series data. S2. Perform quality control and adaptive windowing on the above multi-source time series data, extract trend features, peak and valley features, mutation features, periodic features and data confidence in each time window, determine the event boundaries based on the features, and form an event coding sequence sorted by uniform time. Each event code includes at least the event start time, event end time, event intensity, event duration, change polarity and confidence. S3. Based on the current event encoding, the visual control vector corresponding to the previous event, and the preset visual continuity constraints, calculate the visual control vector corresponding to the current event. The visual control vector includes at least three of the following: brightness change, contrast change, image edge density target value, visual object density target value, motion amplitude, frame duration, and transition rate. S4. Generate keyframes and transition frames corresponding to each event based on the visual control vector, and combine them to obtain dynamic visual content. S5. Establish a unified time-corresponding sensing timeline and a media rendering timeline corresponding to dynamic visual content. Using the event boundary in event coding as the event anchor point, calculate the drift error of the same event anchor point on the two timelines. When the drift error exceeds the drift threshold, while keeping the key frame time corresponding to the event anchor point unchanged, allocate the drift correction amount to multiple non-key frames after the event anchor point to correct the frame duration and / or transition time. S6. Extract rendering feature trajectories from the actual generated dynamic visual content, compare the rendering feature trajectories with the target event trajectories obtained by event encoding, calculate the comprehensive error including time synchronization error, rendering feature error and visual continuity error, and perform closed-loop correction on the subsequent visual control vectors based on the comprehensive error.

[0007] Furthermore, in step S1, the data source comes from at least two of the following: wearable device physiological data, motion data, posture or acceleration data, interactive behavior data, device operation data, and environmental sensing data, and the different data sources have different sampling frequencies.

[0008] Furthermore, the time mapping in step S1 employs a linear or piecewise linear clock correction model. ; in, No. For each data source's original timestamp, To standardize timestamps, For the first Clock scale correction parameters for each data source For the first Clock offset parameters for each data source, and Determined based on at least two common events, synchronization markers, and / or periodic calibration packages.

[0009] Furthermore, in step S2, the saliency of events is calculated for each candidate time window:

[0010] and satisfy ;in, The change in mean For peak and valley amplitude, The trend slope This represents the change in the frequency spectrum. For data confidence, For normalization function, to The corresponding feature weights are used; the event boundary is determined when the event significance exceeds the adaptive threshold or the mutation feature exceeds the mutation threshold.

[0011] Furthermore, each event in step S2 is encoded as follows: ; in, and The first The start and end times of each event are unified. For the intensity of the event, For duration, For changing polarity, For periodic parameters, , where is the confidence level.

[0012] Furthermore, step S3 uses a rule mapping model, a machine learning regression model, a neural network model, or a combination thereof to convert the event encoding into a visual control vector according to the following formula:

[0013] in, For mapping functions, and These are the visual control vectors for the current event and the previous event, respectively. The scene constraints include visual continuity constraints, which include at least one of the following: the brightness change of adjacent events does not exceed a first threshold, the motion amplitude change does not exceed a second threshold, and the structural similarity of adjacent keyframes is not lower than a third threshold.

[0014] Furthermore, step S4 uses a parametric renderer and / or a generative image model to generate dynamic images or animations; when using a generative image model, the visual control vector is encoded as a conditional vector, and at least one of the structural features, edge map, depth map, object layout, or low-dimensional latent variables of the previous keyframe is used as the structural constraint of the current keyframe.

[0015] Furthermore, in step S5, the first The drift error of each event anchor point is calculated according to the following formula: ; when When the drift exceeds the drift threshold, the drift correction amount is allocated to the subsequent K non-critical frames, where the allocation weight of the r-th non-critical frame is calculated according to the following formula: ; The single-frame correction amount is calculated according to the following formula: ; in, The moment of the event anchor point on the media rendering timeline. The moment of the event anchor point on the unified sensing timeline. To standardize the mapping of time to media time. This is the preset maximum single-frame correction amount. Furthermore, step S6 extracts at least three items from the actual generated dynamic visual content: average brightness, brightness variance, edge density, object density, average optical flow amplitude, and keyframe structural similarity, and calculates the comprehensive error according to the following formula: ; Wherein, the weights satisfy ; For comprehensive error; subscripts in the formula The corresponding time synchronization error term is the average value of the absolute value of the event anchor point drift error after normalization by the reference drift amount; subscript The corresponding rendering feature error term is the normalized weighted error between the target event trajectory and the actual rendered feature trajectory in dimensions such as brightness, edge density, object density, and optical flow amplitude; subscript The corresponding visual continuity penalty term is the penalty for the difference between adjacent visual control vectors or the structural difference between adjacent keyframes after normalization by the reference scale. Among them, the time synchronization error term is the normalized event anchor time error, the rendering feature error term is the error between the normalized target event trajectory and the actual rendering feature trajectory, and the visual continuity penalty term is the normalized adjacent visual control vector or keyframe structure difference; when the rendering queue latency exceeds the threshold, the event anchor keyframe is kept from being downgraded, and at least one of the following is performed on non-keyframes: reducing resolution, reducing the number of intermediate frames, or reducing the number of non-key visual objects.

[0016] An event encoding, dynamic visual generation, and timeline adaptive synchronization system based on multi-source time-series data, including... The data access module is used to receive time-series data from at least two data sources and attach source identifiers and collection timestamps; The time synchronization and quality control module is used to estimate clock scale correction parameters and clock offset parameters, generate uniform timestamps, mark abnormal data, and perform adaptive windowing. The event encoding module is used to extract temporal features, determine event boundaries, and generate event encoding sequences; the visual control vector generation module is used to map event encodings into visual control vectors that satisfy visual continuity constraints. The dynamic vision generation module is used to generate event keyframes and transition frames based on visual control vectors; The timeline synchronization correction module is used to establish a unified sensor timeline and a media rendering timeline, calculate event anchor point drift error, and allocate drift correction amounts to non-key frames. The feedback evaluation and resource scheduling module is used to extract actual rendering features, calculate comprehensive errors, correct subsequent visual control vectors, and reduce the resource consumption of non-key frames while keeping event anchor keyframes from degrading when resources are insufficient. Each module is communicatively connected and configured to perform the method described in any one of claims 1 to 9.

[0017] The beneficial effects achieved by this invention are as follows: 1. This invention no longer employs the traditional point-to-point mapping method. Instead, it extracts multi-source temporal data into semantic units with clear start and end times, intensity, and polarity through event encoding, fundamentally avoiding frequent jumps caused by noise. Simultaneously, by establishing a dual time axis (sensor axis and media axis) and introducing an event anchor mechanism, when time drift occurs, the correction amount is directionally allocated to non-critical transition frames, strictly protecting the key frame time corresponding to the event anchor point from becoming unchanged. This completely solves the problems of cumulative drift and key event misalignment in long-term dynamic visual generation.

[0018] 2. This invention overcomes the limitations of unidirectional open-loop mapping. Through a closed-loop rendering feature feedback, it extracts features such as brightness, edge density, and optical flow from the actually generated image and compares them with the target trajectory, automatically correcting subsequent visual control vectors so that the visual output continuously follows changes in data semantics. Furthermore, when computational resources are insufficient, the system employs an event anchor-priority protection strategy, reducing only the resolution or number of non-critical frames to ensure the continuity and recognizability of critical events on the main timeline, balancing real-time stability and visual quality under high load. Attached Figure Description

[0019] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a flowchart illustrating the overall process of the method of the present invention.

[0020] Figure 2 This is a schematic diagram of the timing event encoding structure of the present invention.

[0021] Figure 3 This is a schematic diagram illustrating the dual-time-axis synchronization and drift correction of the present invention.

[0022] Figure 4 This is a schematic diagram illustrating the correction of the rendering feature closed loop in this invention.

[0023] Figure 5 This is a block diagram of the system of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of the phrase "embodiment" in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0026] like Figures 1 to 5 As shown, the present invention adopts the following overall technology chain: multi-source time-series data acquisition → time unification and quality control → adaptive event segmentation → event encoding → visual control vector generation → dynamic visual generation → dual time axis drift correction → rendering feature feedback → subsequent parameter closed-loop correction.

[0027] I. Multi-source data access and unified time base Each data packet contains at least a source identifier (source_id), a raw timestamp (t), a sample value (x), and an optional data quality field (q). For the k-th data source, the system estimates clock scale parameters using common events, synchronization markers, or periodic calibration packets. and clock offset parameters And map the original timestamp to a unified time. :

[0028] For data sources with different sampling frequencies, the system chooses to preserve the original event points, interpolate to a uniform sampling grid, or perform window statistics based on the data type. Missing, out-of-limit, anomalous, and low-confidence sampling points are marked with quality flags; low-quality points do not directly trigger event boundaries.

[0029] II. Adaptive Windowing and Event Detection The system uses a combination of basic windows and candidate event windows to process time-series data. For each candidate time window, event saliency is constructed. :

[0030]

[0031] In the formula, This represents a normalization function that maps the corresponding features to the interval between 0 and 1. The change in mean For peak and valley amplitude, The trend slope This represents the change in the frequency spectrum. For data confidence; to The five items are assigned non-negative weights, with a sum of 1. Weights can be set by domain priors or determined through grid search, cross-validation, or supervised training using historical annotation windows. When labeled samples are lacking, equal weights can be used initially, followed by calibration based on false positive and false negative rates. The system can use quantile thresholds based on historical window distribution, exponential moving average thresholds, or trained change point detectors to form adaptive thresholds. When Q exceeds the adaptive threshold, or the mutation score exceeds the mutation threshold, candidate event boundaries are established; candidate boundaries with adjacent distances less than the minimum event interval are merged.

[0032] III. Event Coding Construct event encoding for the i-th event: ; in, and The first The start and end times of each event are unified. For the intensity of the event, For duration, For changing polarity, For periodic parameters, For confidence levels, the contribution weights of each data source can be further recorded for multi-source events. Event coding allows the system to retain the temporal structure of "when it happened, how long it lasted, how strong the change was, the direction of the change, and the confidence level," rather than directly turning a single sampled value into a visual attribute.

[0033] IV. Visual Control Vectors and Continuity Constraints The system generates a visual control vector V based on event encoding. iThe visual control vector may include parameters that can be directly executed by the rendering engine, such as average brightness target, contrast target, edge density target, visual object density, motion amplitude, deformation amplitude, frame duration, and transition rate. An example is:

[0034] in, For rule mapping models, machine learning regression models, neural network models, or combinations thereof, and These are the visual control vectors for the current event and the previous event, respectively. Scene constraints are implemented. To avoid visual jitter, the system sets an upper limit on the rate of change of adjacent visual control vectors; when the model output exceeds the upper limit, amplitude limiting, smoothing, or secondary optimization is employed. Continuity constraints prioritize the protection of structural similarity, continuity of motion direction, and the rate of brightness change.

[0035] V. Dynamic Visual Content Generation For each event, the system generates at least one event keyframe and zero or more transition frames based on the event duration. The generation method can be parametric 2D / 3D rendering or a generative image model controlled by conditional vectors. When using a generative model, V... i Encode it as a conditional vector, and use the structural features, edge map, depth map, object layout or low-dimensional latent variables of the previous keyframe as structural constraints to reduce irrelevant changes in the main structure between adjacent events.

[0036] VI. Dual Time Axis and Event Anchor Point Drift Correction The system also maintains a unified timeline for sensing. and media rendering timeline An event anchor point is generated at each event boundary. For the first For each event anchor point, the following formula is used for calculation:

[0037]

[0038]

[0039]

[0040] In the formula, To standardize the mapping ratio between time and target media time. This represents the drift error of the media time relative to the mapped unified time. The number of subsequent non-keyframes participating in this correction. This is a non-keyframe sequence number. For the first The weights of non-key frames are assigned, with the weights of non-key frames being greater the closer they are to the event anchor point. For the first The correction amount applied during the duration of a non-keyframe or at a transition moment. This is the maximum allowable correction amount per frame. This is the clipping function. If the sum of the corrections for each frame after clipping has not yet reached the limit... The system calculates the remaining correction amount and iteratively distributes it among subsequent non-key frames that have not reached the amplitude limit according to the remaining weight; if necessary, the correction window is expanded, but the target time of the event anchor key frame is not changed.

[0041] VII. Rendering Feature Feedback Closed Loop After dynamic visual content is generated, the system extracts rendering features such as average brightness, brightness variance, edge density, object density, average optical flow amplitude, and keyframe structural similarity from the actual rendered frames, forming the actual rendered feature trajectory R(t). The target event trajectory T(t) is constructed from the event encoding sequence. The overall error can be expressed as:

[0042]

[0043] In the formula, For comprehensive error; subscripts in the formula The corresponding time synchronization error term is the average value of the absolute value of the event anchor point drift error after normalization by the reference drift amount; subscript The corresponding rendering feature error term is the normalized weighted error between the target event trajectory and the actual rendered feature trajectory in dimensions such as brightness, edge density, object density, and optical flow amplitude; subscript The corresponding visual continuity penalty term is the penalty for the difference between adjacent visual control vectors or the structural difference between adjacent keyframes after normalization by the reference scale. , and These are non-negative weights for time synchronization, rendering consistency, and visual continuity, respectively, and the sum of the three is 1. The weights can be configured according to the application's priorities. For example, , , The initial values ​​were set to 0.4, 0.4, and 0.2 respectively, and then calibrated using historical qualified samples or manual evaluation results. The system performs bounded corrections to the visual control vectors of subsequent events based on J, without backtracking and destroying the already locked event anchor points.

[0044] 8. Real-time resource scheduling The system monitors rendering queue latency, processor load, and video memory usage. When resources are insufficient, event anchor frames are defined as non-degradable frames, and strategies such as reducing resolution, decreasing the number of intermediate frames, reducing the number of non-critical objects, or using frame interpolation are prioritized for non-critical transition frames. Quality is restored step by step after resources are restored. This strategy ensures that real-time performance adjustments do not disrupt the event anchor and main event structure.

[0045] IX. System Modules and Their Functions (1) Data access module: receives time-series data with different sampling frequencies and different data formats, adds source identifier, original timestamp and quality field to data packets, and completes interface format conversion.

[0046] (2) Time synchronization and quality control module: Estimate the clock scale correction parameters and clock offset parameters of each data source, map the source time to a unified time, detect missing, out-of-bounds, abnormal mutations and low confidence data, and perform interpolation, quality marking and adaptive windowing.

[0047] (3) Event coding module: Extract trend, peak and valley, mutation, spectrum, period and confidence features from the candidate window, calculate event salience, determine event boundary, and output event coding sequence containing start and end time, intensity, duration, polarity, periodicity and confidence.

[0048] (4) Visual control vector generation module: Through rule mapping, regression model, neural network or combination thereof, the event encoding is converted into parameters such as brightness, contrast, edge density, object density, motion amplitude, frame duration and transition rate, and amplitude limiting, smoothing and continuity constraints are executed.

[0049] (5) Dynamic visual generation module: Generate event keyframes and transition frames based on visual control vectors; when using a generative model, the structural features, edge map, depth map, object layout or latent variables of the previous keyframe are used as structural constraints.

[0050] (6) Time axis synchronization correction module: maintains the sensor unified time axis and media rendering time axis, forms event anchor points with event boundaries, calculates drift error, and corrects the frame duration or transition time of subsequent non-key frames according to the assigned weight and single frame limit value.

[0051] (7) Feedback evaluation and resource scheduling module: extract feature trajectories from actual rendering results and calculate comprehensive error to correct subsequent visual control vectors; protect event anchor keyframes when computing resources are insufficient, and only reduce the resolution, number or object complexity of non-keyframes.

[0052] The above modules can be implemented as software functional units in the same processor, or they can be distributed in local terminals, edge computing nodes, and cloud servers; computer programs can be stored in non-transitory computer-readable storage media and executed by the processor.

[0053] This embodiment illustrates how to convert continuous motion and physiological time-series data into dynamic images / animations without requiring music generation as a necessary step. Data sources include activity load sequences, heart rate sequences, and acceleration sequences output from exercise equipment. The sampling frequencies for these three types of data can be 1 Hz, 1 Hz, and 25 Hz, respectively, and the system converts them to a unified time axis through a unified time mapping.

[0054] The system calculates the trend slope, peak and trough amplitudes, and abrupt change scores using a 5-second base window; for acceleration data, it further calculates the window energy. If the significance of an event in a certain window exceeds a preset quantile threshold for the significance of historical windows in the most recent 120 seconds, a candidate event boundary is formed. Candidate boundaries that are less than 2 seconds apart are merged.

[0055] Example calculation: Within a specific 5-second candidate window, assume the normalized value of the mean change in heart rate. =0.80, normalized peak-to-valley amplitude value =0.90, normalized value of trend slope =0.70, normalized value of spectral variation =0.60, data confidence level =0.90; take =0.30、 =0.25、 =0.20、 =0.10、 =0.15, then:

[0056] Calculated =0.80. If the current adaptive threshold is 0.75, then If the intensity exceeds the threshold, the system classifies the window as a candidate event and establishes an event boundary; when no other intensity conversion model is set, the normalized event intensity can be... The initial value is set to 0.80. The above values ​​are configuration examples used to illustrate the calculation process and do not represent unverified product performance data.

[0057] For each event, the system calculates the normalized event strength A. i For any visual parameter y that is positively correlated with the event intensity, linear interpolation can be performed according to the following formula:

[0058] In the formula, and These are the preset lower and upper limits for the visual parameters, respectively; for parameters such as keyframe interval that are negatively correlated with event intensity, the following can be used:

[0059] A configurable mapping relationship is shown in Table 1, where both the motion amplitude target and the edge density target are normalized to the 0-1 range, and the specific range can be calibrated according to the renderer capabilities and scene style.

[0060] Table 1. Example mapping relationship between event intensity and dynamic visual parameters

[0061] The system also calculates the duration D of the event. i Determine the number of keyframes or the length of the time slice, according to the changing polarity P. i Determine the direction of motion or deformation. Smoothing is performed when the change in the visual control vector exceeds the amplitude limit threshold to avoid frequent image jumps caused by input sensor noise.

[0062] After generating event keyframes and transition frames, the system compares the unified timeline and media timeline based on the event anchor point. If the event anchor point drifts, the correction amount is distributed to subsequent non-keyframes. After rendering, the optical flow amplitude, edge density, and average brightness are extracted, compared with the target event trajectory, and the visual control vector for the next event is modified accordingly.

[0063] The data sources in this embodiment include dwell time in front of exhibits, touch interaction frequency, body movement speed, and the operating status output by the exhibit equipment. The system does not require wearable devices. Each data source is time-aligned to form an event encoding sequence, which drives the dynamic visual content in the exhibition wall, interactive screen, or projection.

[0064] When an interaction is detected to transition from a low-frequency stable state to a high-frequency abrupt change, the intensity and abrupt change parameters in the event encoding increase; the visual control vector increases motion amplitude and visual object density, but continuity constraints limit structural changes in adjacent keyframes. If the device rendering load increases, non-critical transition frames and non-critical objects are reduced, while the event anchor keyframes and their temporal positions are maintained.

[0065] This embodiment shows that the core of the present invention does not depend on a certain type of physiological data, a certain visual style, or a certain generative model, but rather on the combination of multi-source temporal event encoding, constrained visual generation, dual timeline synchronization, and rendering feedback closed loop.

[0066] For already recorded motion data or equipment operation data, the system can perform event detection and encoding offline, and generate a media timeline according to the event sequence. Each event automatically creates a timeline segment, with the event start and end times used to determine the segment range, the event intensity used to determine visual parameter targets, and the event anchor points used to determine keyframes.

[0067] When the actual project frame rate of the media editing software is not divisible by the data time base, the system records fractional frame errors and accumulates them in the drift variable. When a threshold is reached, the errors are distributed across subsequent non-keyframes instead of being adjusted centrally on a single keyframe. This allows for the creation of an editable timeline consistent with the original data event structure.

[0068] Without departing from the core idea of ​​this invention, the event detection algorithm can employ statistical change point detection, Bayesian change point detection, hidden Markov models, recurrent neural networks, Transformer temporal models, etc.; the visual content can be two-dimensional graphics, three-dimensional graphics, particle animations, patterns, digital exhibition images, information visualization animations, or generative image sequences; time synchronization can be completed on local devices, edge computing nodes, or in the cloud. All the above substitutions do not alter the technical chain of "event encoding—visual control—dual timelines—rendering feedback".

[0069] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other occasions without modification, are all within the protection scope of the present invention.

Claims

1. A method for event encoding, dynamic visual generation, and adaptive synchronization of time axis based on multi-source time-series data, characterized in that: Includes the following steps: S1. Obtain time series data from at least two data sources, assign source identifiers and collection timestamps to each data source, and establish a time mapping from the source time of each data source to the unified time based on a unified reference time to obtain time-aligned multi-source time series data. S2. Perform quality control and adaptive windowing on the above multi-source time series data, extract trend features, peak and valley features, mutation features, periodic features and data confidence in each time window, determine the event boundaries based on the features, and form an event coding sequence sorted by uniform time. Each event code includes at least the event start time, event end time, event intensity, event duration, change polarity and confidence. S3. Based on the current event encoding, the visual control vector corresponding to the previous event, and the preset visual continuity constraints, calculate the visual control vector corresponding to the current event. The visual control vector includes at least three of the following: brightness change, contrast change, image edge density target value, visual object density target value, motion amplitude, frame duration, and transition rate. S4. Generate keyframes and transition frames corresponding to each event based on the visual control vector, and combine them to obtain dynamic visual content. S5. Establish a unified time-corresponding sensing timeline and a media rendering timeline corresponding to dynamic visual content. Using the event boundary in event coding as the event anchor point, calculate the drift error of the same event anchor point on the two timelines. When the drift error exceeds the drift threshold, while keeping the key frame time corresponding to the event anchor point unchanged, allocate the drift correction amount to multiple non-key frames after the event anchor point to correct the frame duration and / or transition time. S6. Extract rendering feature trajectories from the actual generated dynamic visual content, compare the rendering feature trajectories with the target event trajectories obtained by event encoding, calculate the comprehensive error including time synchronization error, rendering feature error and visual continuity error, and perform closed-loop correction on the subsequent visual control vectors based on the comprehensive error.

2. The method for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data according to claim 1, characterized in that: In step S1, the data source comes from at least two of the following: wearable device physiological data, motion data, posture or acceleration data, interactive behavior data, device operation data, and environmental sensing data, and the different data sources have different sampling frequencies.

3. The method for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data according to claim 1, characterized in that: The time mapping in step S1 uses a linear or piecewise linear clock correction model. ； in, No. For each data source's original timestamp, To standardize timestamps, For the first Clock scale correction parameters for each data source For the first Clock offset parameters for each data source, and Determined based on at least two common events, synchronization markers, and / or periodic calibration packages.

4. The method for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data according to claim 1, characterized in that: In step S2, the saliency of events is calculated for each candidate time window: and satisfy ;in, The change in mean For peak and valley amplitude, The trend slope This represents the change in the frequency spectrum. For data confidence, For normalization function, to The corresponding feature weights are used; the event boundary is determined when the event significance exceeds the adaptive threshold or the mutation feature exceeds the mutation threshold.

5. The method for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data according to claim 1, characterized in that: Each event in step S2 is encoded as follows: ； in, and The first The start and end times of each event are unified. For the intensity of the event, For duration, For changing polarity, For periodic parameters, , where is the confidence level.

6. The method for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data according to claim 1, characterized in that: Step S3 uses a rule mapping model, machine learning regression model, neural network model, or a combination thereof to convert the event encoding into a visual control vector according to the following formula: in, For mapping functions, and These are the visual control vectors for the current event and the previous event, respectively. The scene constraints include visual continuity constraints, which include at least one of the following: the brightness change of adjacent events does not exceed a first threshold, the motion amplitude change does not exceed a second threshold, and the structural similarity of adjacent keyframes is not lower than a third threshold.

7. The method for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data according to claim 1, characterized in that: Step S4 uses a parametric renderer and / or a generative image model to generate dynamic images or animations; when using a generative image model, the visual control vector is encoded as a conditional vector, and at least one of the structural features, edge map, depth map, object layout or low-dimensional latent variables of the previous keyframe is used as the structural constraint of the current keyframe.

8. The method for event encoding, dynamic visual generation, and adaptive synchronization of time axis based on multi-source time-series data according to claim 1, characterized in that: In step S5, the first The drift error of each event anchor point is calculated according to the following formula: ； when When the drift exceeds the drift threshold, the drift correction amount is allocated to the subsequent K non-critical frames, where the allocation weight of the r-th non-critical frame is calculated according to the following formula: ； The single-frame correction amount is calculated according to the following formula: ； in, The moment of the event anchor point on the media rendering timeline. The moment of the event anchor point on the unified sensing timeline. To standardize the mapping of time to media time. This is the preset maximum single-frame correction amount.

9. The method for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data according to claim 1, characterized in that: Step S6 extracts at least three of the following from the actually generated dynamic visual content: average brightness, brightness variance, edge density, object density, average optical flow amplitude, and keyframe structural similarity, and calculates the comprehensive error according to the following formula: ； Wherein, the weights satisfy ; For comprehensive error; subscripts in the formula The corresponding time synchronization error term is the average value of the absolute value of the event anchor point drift error after normalization by the reference drift amount; subscript The corresponding rendering feature error term is the normalized weighted error between the target event trajectory and the actual rendered feature trajectory in dimensions such as brightness, edge density, object density, and optical flow amplitude; subscript The corresponding visual continuity penalty term is the penalty for the difference between adjacent visual control vectors or the structural difference between adjacent keyframes after normalization by the reference scale. Among them, the time synchronization error term is the normalized event anchor time error, the rendering feature error term is the error between the normalized target event trajectory and the actual rendering feature trajectory, and the visual continuity penalty term is the normalized adjacent visual control vector or keyframe structure difference; when the rendering queue latency exceeds the threshold, the event anchor keyframe is kept from being downgraded, and at least one of the following is performed on non-keyframes: reducing resolution, reducing the number of intermediate frames, or reducing the number of non-key visual objects.

10. A system for event encoding, dynamic visual generation, and adaptive timeline synchronization based on multi-source time-series data, characterized in that: include The data access module is used to receive time-series data from at least two data sources and attach source identifiers and collection timestamps; The time synchronization and quality control module is used to estimate clock scale correction parameters and clock offset parameters, generate uniform timestamps, mark abnormal data, and perform adaptive windowing. The event encoding module is used to extract temporal features, determine event boundaries, and generate event encoding sequences; the visual control vector generation module is used to map event encodings into visual control vectors that satisfy visual continuity constraints. The dynamic vision generation module is used to generate event keyframes and transition frames based on visual control vectors; The timeline synchronization correction module is used to establish a unified sensor timeline and a media rendering timeline, calculate event anchor point drift error, and allocate drift correction amounts to non-key frames. The feedback evaluation and resource scheduling module is used to extract actual rendering features, calculate comprehensive errors, correct subsequent visual control vectors, and reduce the resource consumption of non-key frames while keeping event anchor keyframes from degrading when resources are insufficient. Each module is communicatively connected and configured to perform the method described in any one of claims 1 to 9.

Images