A new media material content collection system based on image data processing

Abstract

This invention discloses a new media content acquisition system based on image data processing, belonging to the field of content acquisition technology. It includes a dynamic feature extraction module, a cross-frame feature bridging module, a dynamic re-identification mapping module, an illumination adaptive normalization module, and a multi-subject temporal fusion module. The dynamic feature extraction module acquires a continuous sequence of image frames, performs spatial and temporal feature extraction on each frame, and binds the extracted spatial and temporal features according to frame order to generate a dynamic feature sequence with a time index. This invention maintains feature continuity before and after occlusion through dynamic feature sequences and cross-frame bridging, achieving stable identity association and improving target tracking accuracy and acquisition stability. Furthermore, it achieves illumination adaptation and semantic consistency through dynamic re-identification and temporal fusion, ensuring the accuracy and reliability of content acquisition in multi-subject environments.

Description

Technical Field

[0001] This invention relates to the field of content acquisition technology, specifically to a new media content acquisition system based on image data processing. Background Technology

[0002] The new media content acquisition system based on image data processing is a comprehensive system that utilizes image recognition technology to collect, classify, and intelligently filter multi-source visual information in real time. The system acquires raw image data through high-resolution imaging equipment and relies on image recognition algorithms to extract features and perform semantic analysis on the image content, identifying key visual information such as people, scenes, objects, actions, and emotional elements. Based on this, the system uses a deep learning model to intelligently judge the content quality, composition style, and application scenario of the acquired materials, achieving automated acquisition and accurate archiving of image materials required for new media creation. This system not only improves the efficiency and accuracy of image material acquisition but also provides high-value foundational data support for subsequent multimedia editing, content generation, and visual dissemination.

[0003] The existing technology has the following shortcomings: During rapid scene transitions, image recognition models often experience interruptions in frame association when performing target tracking and identity mapping tasks due to the reappearance of the captured object after a brief occlusion. This interruption prevents the model from effectively matching feature vectors before and after occlusion, leading to misclassification of visually similar but actually different objects as the same subject, resulting in incorrect identity mapping. This problem is particularly pronounced in environments with dynamic lighting, complex backgrounds, and multi-subject interactions. When the model lacks temporal constraints and feature re-identification mechanisms, the probability of misjudgment increases significantly, further causing mismatched tags, disordered object trajectories, and semantic shifts in the collected material. Consequently, the content exhibits systematic distortions such as overlapping characters and misplaced scenes during post-production editing, content generation, and multimodal analysis, severely impacting the authenticity and semantic consistency of the collected data.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a new media content acquisition system based on image data processing to solve the problems mentioned in the background art. To achieve the above objectives, the present invention provides the following technical solution: a new media material content acquisition system based on image data processing, including a dynamic feature extraction module, a cross-frame feature bridging module, a dynamic re-identification mapping module, an illumination adaptive normalization module, and a multi-subject temporal fusion module; Dynamic feature extraction module: acquires a continuous frame image sequence, performs spatial and temporal feature extraction on each frame in the image sequence, binds the extracted spatial and temporal features according to the frame order, and generates a dynamic feature sequence with a time index, providing continuous input for subsequent feature association; Cross-frame feature bridging module: Based on dynamic feature sequences, a cross-frame feature bridging layer is constructed. The time index is used to match the features of adjacent frames frame by frame, identify feature breakpoints before and after occlusion, and generate feature association paths across occlusion intervals to establish a continuous channel for feature aggregation. Dynamic Re-identification Mapping Module: Based on the feature association path output by the cross-frame feature bridging layer, a dynamic re-identification mapping layer is constructed on the time axis. Feature points in the feature association path whose similarity meets the set threshold are aggregated to form an identity anchor set, and an identity index structure is generated accordingly to provide a basic index for illumination adaptation. Illumination Adaptive Normalization Module: Based on the identity index structure, dynamic illumination partitioning is performed. Within each illumination partition, illumination normalization is performed according to the feature weights of the identity anchor point set, so that the identity index maintains feature stability under illumination changes and provides consistent feature input for subsequent temporal fusion. Multi-subject temporal fusion module: Based on the feature input after illumination normalization, a multi-subject temporal fusion layer is established. The feature trajectory of the identity anchor point set in each time period is fused to realize continuous identity tracking in multi-subject interaction scenarios. The identity recovery and semantic extension of the sampled object after occlusion are completed, thereby realizing the accurate collection and association of new media material content.

[0006] Preferably, the step of acquiring a sequence of consecutive frame images and generating a dynamic feature sequence with a time index includes:

[0007] Acquire a continuous sequence of images, perform preprocessing on each frame of the image sequence, acquire images at fixed time intervals while keeping the optical focal length and imaging parameters consistent, and perform brightness equalization, color normalization, geometric distortion correction and edge sharpness enhancement on the images.

[0008] Spatial feature extraction is performed on each preprocessed frame image to extract a set of spatial features including appearance contour, color distribution, compositional relationship, regional hierarchy and geometric positional relationship, and the feature scale and expression standard are kept consistent across frames;

[0009] By combining spatial features, temporal features between consecutive frames are extracted to obtain dynamic change information including displacement trends, morphological transformations, motion trajectories, and illumination changes, so that each frame of the image has both static and dynamic feature representations.

[0010] Spatial and temporal features are fused and bound together, and a time index is established according to the frame order to generate a dynamic feature sequence with time identifiers, providing continuous input for subsequent feature bridging, cross-frame association, and identity re-identification.

[0011] Preferably, the steps for constructing a cross-frame feature bridging layer based on dynamic feature sequences include:

[0012] After acquiring the dynamic feature sequence with time index, the feature correspondence between consecutive frames is initialized and matched. The frame sequence identifier of the time index is used to compare the features of adjacent frames, and the spatial and temporal features are combined to ensure the temporal continuity of the comparison.

[0013] After completing the inter-frame matching, the intervals where the feature association is interrupted in the time index sequence are detected. The positions where the change in spatial features of adjacent frames exceeds a preset threshold and the time interval does not exceed a specific frame distance are identified as occlusion breakpoints, and the time indexes of the breakpoint start frame and the recovery frame are recorded.

[0014] Based on the identified occlusion breakpoints, a cross-frame feature bridging path is constructed on the time axis, starting with the last frame feature before occlusion and ending with the first frame feature after occlusion, to ensure that the temporal extension order is consistent with the frame order and to maintain a smooth connection of spatial distribution.

[0015] The generated cross-frame feature association path is subjected to continuity verification. The time indices at both ends of the bridging path are aligned with the dynamic feature sequence. Spatial smoothing correction is performed on the feature points covered by the path to form a feature association channel that is temporally continuous, spatially consistent and semantically coherent.

[0016] Preferably, in the process of constructing the cross-frame feature bridging path, the connection between the last frame feature before occlusion and the first frame feature after occlusion is extended according to the time index order in the dynamic feature sequence, and the spatial position change of the sampled object during the occlusion is inferred by combining the distribution trend of spatial features before and after occlusion, so that the generated cross-frame feature association path remains continuous on the time axis and smoothly connected in the spatial dimension.

[0017] Preferably, the step of constructing a dynamic re-identification mapping layer based on the feature association path output by the cross-frame feature bridging layer includes:

[0018] After obtaining the feature association path output by the cross-frame feature bridging layer, the feature points of each time segment in the path are temporally aligned, sorted according to the time index, and similarity evaluation is performed on the feature points of adjacent time segments to identify continuous feature segments that conform to the identity continuity rule.

[0019] Based on the temporal alignment results, feature points in the feature association path whose similarity meets the set threshold are aggregated to form a set of identity anchor points that are continuous on the time axis, and the consistency of the time index is maintained to ensure the uniqueness of the identity.

[0020] After obtaining the set of identity anchor points, the set of identity anchor points is structured and summarized. An identity index structure is generated based on the time span, spatial distribution and feature stability, and a mapping relationship between time and space is established.

[0021] The generated identity index structure is subjected to consistency verification and illumination adaptation preparation. The differences in feature distribution within different illumination intervals are compared, and illumination offset information is recorded to provide basic index support for subsequent illumination normalization and identity fusion.

[0022] Preferably, during the generation of the identity index structure, the arrangement of the identity anchor set on the time axis is based on the time index as the primary key, and combined with the positioning range of spatial features for dual-dimensional indexing, so that the features of the same sampled object in different lighting intervals remain temporally continuous and spatially consistent, thereby improving the stability and traceability of the identity index structure under lighting change conditions.

[0023] Preferably, the steps for performing dynamic illumination partitioning based on the identity index structure include:

[0024] After generating the identity index structure, the illumination changes in the collection sequence are detected according to the time index, the boundary points of illumination intensity changes are identified and the illumination partitions are divided, and the identity anchor point set of the corresponding time period is assigned to the corresponding partition to establish an illumination consistency interval.

[0025] After determining the illumination partitions, feature weights are assigned to the set of identity anchor points within each illumination partition. The weight values ​​are determined based on the stability of each anchor point under the current illumination conditions to form a feature expression framework dominated by steady-state illumination.

[0026] Based on feature weights, illumination normalization is performed on the set of identity anchor points within the illumination partition, adjusting the feature intensity and brightness response range of each anchor point, and maintaining a smooth transition of features between adjacent partitions.

[0027] After completing illumination normalization, consistency verification is performed on the global identity index structure, all illumination partition features are reintegrated into a complete time series, and the weights of adjacent partition boundary features are adjusted to ensure the continuity and uniformity of feature expression under illumination changes.

[0028] Preferably, during the illumination normalization process, a weighted smoothing adjustment method is adopted for the transition region of adjacent illumination partitions, so that the identity anchor point features at the partition boundary maintain a gradual consistency in brightness and color distribution, thereby ensuring the feature continuity and semantic expression stability on the time axis, and thus improving the recognition accuracy and feature robustness of the identity index structure under drastic illumination changes.

[0029] Preferably, the steps for establishing a multi-subject temporal fusion layer based on illumination-normalized feature inputs include:

[0030] Based on the feature input after illumination normalization, the temporal distribution of different identity anchor point sets is organized and aligned. The time index structure is used as the main line to arrange each identity anchor point set in chronological order, and the missing time index segments are filled in to ensure the continuity of the trajectory.

[0031] Based on the time alignment results, the feature trajectories of each identity anchor point set are fused spatially and temporally. The features of adjacent frames are fused into continuous trajectories according to the time index order, and the independence and integrity of the trajectories are maintained through feature weight compensation when there is spatial overlap.

[0032] After completing trajectory fusion, the identity relationships between different subjects are identified and distinguished. Based on the unique identity identifier, the identity intersection points on the time axis are detected, and the occlusion extension and interaction events are distinguished according to the illumination normalization characteristics and temporal continuity.

[0033] After completing the fusion of multi-subject trajectories and identity recognition, the fusion results are semantically extended. A semantic chain is established based on the temporal information of the identity anchor set, so that the collected objects maintain the continuity and consistency of identity expression and semantic orientation in the time dimension.

[0034] Preferably, the semantic extension processing further includes: when establishing the semantic chain, based on the time sequence information of the feature input after illumination normalization and the identity anchor set, dynamically completing the feature change trend of different subjects in the interaction process, so that the semantic chain maintains logical coherence and semantic consistency under illumination changes and occlusion recovery conditions, thereby ensuring the continuous stability of identity expression and behavioral semantics in multi-subject interaction scenarios.

[0035] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0036] This invention introduces dynamic feature sequences and cross-frame feature bridging mechanisms during image acquisition, enabling features between frames before and after occlusion to maintain continuous correlation on the time axis. This effectively avoids identity interruption and mismatch issues when an object reappears after a brief disappearance, achieving smooth connection of features in the temporal dimension and persistent mapping of identity information. Consequently, it significantly improves the accuracy of target tracking and the stability of material acquisition under conditions of rapid scene switching.

[0037] This invention establishes a dynamic re-identification mapping layer and a multi-subject temporal fusion layer, enabling identity anchors to maintain semantic consistency and trajectory integrity in environments with varying lighting and multi-subject interactions. It achieves continuous fusion of feature normalization and identity indexing adapted to lighting, allowing the acquisition system to maintain consistency of material tags and integrity of semantic structure in complex dynamic scenes, thereby significantly improving the accuracy and reliability of image material acquisition and content association. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0039] Figure 1 This is a schematic diagram of a new media material content acquisition system based on image data processing according to the present invention. Detailed Implementation

[0040] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0041] This invention provides, for example Figure 1 The system shown is a new media content acquisition system based on image data processing, including a dynamic feature extraction module, a cross-frame feature bridging module, a dynamic re-identification mapping module, an illumination adaptive normalization module, and a multi-subject temporal fusion module. Dynamic feature extraction module: acquires a continuous frame image sequence, performs spatial and temporal feature extraction on each frame in the image sequence, binds the extracted spatial and temporal features according to the frame order, and generates a dynamic feature sequence with a time index, providing continuous input for subsequent feature association; The specific implementation method of this module is as follows:

[0042] A continuous sequence of images is acquired, and preprocessing operations are performed on each frame. To ensure the stability of spatial information and the continuity of the temporal sequence, the acquisition of consecutive frames should be carried out at fixed time intervals, and the optical focal length and imaging parameters should be kept consistent during the acquisition process so that each frame has an alignable geometric reference. After the image sequence is acquired, basic processing is performed on the original images, including brightness equalization, color normalization, geometric distortion correction, and edge sharpness enhancement, to ensure that each input frame is comparable and correlated at the visual information level. The image sequence processed in the above way forms a stable input source that can be used for feature extraction, providing a reliable data foundation for the subsequent simultaneous extraction of spatial and temporal features.

[0043] Spatial feature extraction is performed on each preprocessed image frame. In this process, a single frame is used as the processing unit, and a structured analysis is performed on the visual elements contained in the image to extract the main spatial feature information. These spatial features include the appearance outline, color distribution, compositional relationships, regional hierarchy, and geometrical positional relationships of the target object in the image. By collaboratively extracting local and global features of the image, a multi-dimensional feature set representing the spatial content of the image can be formed. This spatial feature set can accurately reflect the compositional features, positional distribution, and visual weight relationships of the captured object in the two-dimensional image, establishing a spatial foundation for the continuity of the time series. It is worth noting that, during the spatial feature extraction process, to ensure the feature expression has cross-frame scalability, the spatial feature extraction of each frame is performed using the same feature expression standard and scale specification, ensuring the alignment and stability of spatial features between different frames.

[0044] By combining the spatial features of each frame, temporal feature information is further extracted to construct a temporal continuity representation between frames. Temporal feature extraction takes the continuous frame image sequence as the overall analysis object, focusing on depicting the dynamic changes between adjacent frames, including the displacement trend, morphological transformation, motion trajectory, and scene illumination changes of the captured object. In the process of extracting temporal features, the spatial feature change trends of adjacent frames need to be quantified and described, forming a temporal difference expression between continuous frames. This ensures that each frame not only has its own static spatial description but also carries dynamic attributes reflecting the laws of temporal evolution. Through this process, each frame of the image sequence is endowed with dual spatial and temporal feature information, making the image content no longer an isolated static image but a dynamic information node with coherence, directionality, and evolutionary logic. This embedding of temporal features enables subsequent feature association not only based on spatial similarity but also relying on temporal evolution relationships, achieving continuous recognition under occlusion and recurrence conditions.

[0045] Finally, the spatial and temporal features of each frame are fused and bound, and time indices are sequentially built according to frame order to generate a dynamic feature sequence with time markers. In this process, the combination of spatial and temporal features is based on the ordered integration of frame order. The comprehensive features of each frame use the frame number or timestamp as a unique index unit, maintaining a strict temporal order within the sequence. Through the binding of time indices, feature data between different frames achieves traceability and locationability in the temporal dimension, enabling accurate location and retrieval of image features from any time segment. Simultaneously, the establishment of the dynamic feature sequence unifies spatial variations and temporal extensions between consecutive frames into a single temporal structure, providing a complete and continuous input channel for subsequent feature bridging, cross-frame association, and identity re-identification. In practical applications, the structured organization of this dynamic feature sequence maintains a stable feature transfer chain during subsequent occlusion recovery, identity mapping, and semantic extension processes, avoiding identity misjudgment and semantic drift caused by information breaks between frames.

[0046] This implementation constructs a dynamic feature sequence with a time index by acquiring consecutive frame images, accurately extracting spatial features, dynamically describing temporal features, and orderly binding the two, enabling synchronous structured expression of image data in both temporal and spatial dimensions. This implementation not only ensures the continuity and consistency of feature information but also provides a high-quality input foundation for subsequent feature association, identity re-identification, and illumination adaptation, thereby significantly improving feature stability and recognition accuracy under complex scenes, dynamic lighting, and multi-subject interaction conditions during new media material acquisition.

[0047] Cross-frame feature bridging module: Based on dynamic feature sequences, a cross-frame feature bridging layer is constructed. The time index is used to match the features of adjacent frames frame by frame, identify feature breakpoints before and after occlusion, and generate feature association paths across occlusion intervals to establish a continuous channel for feature aggregation.

[0048] The specific implementation method of this module is as follows:

[0049] After acquiring the dynamic feature sequence with time index, the feature correspondence between consecutive frames is initialized and matched. The core of this stage is to identify potential corresponding targets in adjacent frames and compare the features of consecutive frames one-to-one using the frame order identifier of the time index. During the matching process, it is necessary to ensure that the feature information of each frame maintains the same index order as the time axis, so that the feature comparison is based on temporal continuity. The feature comparison between each pair of adjacent frames is performed in conjunction with the spatial and temporal features in the aforementioned dynamic feature sequence, so that the comparison process is based not only on the consistency of appearance features, but also on the continuity of temporal changes. Through this temporally constrained feature correspondence, the logical continuity of features between adjacent frames can be maintained under visual changes, pose adjustments, or scene transitions, providing basic matching information for subsequent identification of occlusion breakpoints.

[0050] After initial inter-frame matching, potential feature breakpoints in the time series are identified and located. In actual acquisition scenarios, the captured object may experience partial, complete, or temporary occlusion during a certain time segment, preventing direct matching of features from the previous frame. In such cases, interrupted feature association regions should be detected in the time index sequence—specifically, intervals where time is continuous but feature association is missing—to identify potential breakpoints. Feature breakpoint identification requires dual confirmation of inter-frame spatial consistency and temporal continuity: when the spatial feature variation between adjacent frames exceeds a preset threshold and the time interval does not exceed a specific frame distance, the location can be identified as a suspected occlusion breakpoint. After identifying the breakpoint, the time index positions of its starting and resuming frames should be recorded on the timeline to provide a temporal localization basis for subsequent bridging association. Precise identification of breakpoints effectively delineates feature intervals requiring cross-frame bridging, avoiding redundant feature extensions in non-occluded areas, thus maintaining the clarity of the time series structure and the specificity of feature association.

[0051] Based on the identification of feature breakpoints, bridging processing is performed on the features on both sides of the occlusion interval to restore the continuous expression of the sampled object before and after occlusion. This process requires establishing a feature association path across the occlusion interval on the time axis. Specifically, a bridging channel connecting the last frame feature before occlusion should be constructed along the time axis, starting from the last frame feature before occlusion and ending with the first frame feature after occlusion. When establishing the bridging channel, the time index information of adjacent frames in the dynamic feature sequence should be fully utilized to ensure that the temporal extension order of the path is consistent with the actual acquisition frame order. At the same time, to ensure the continuity of the bridging path in spatial expression, it is necessary to refer to the spatial feature distribution before and after occlusion to infer the possible spatial position change trend of the object during occlusion, so that the bridging path is not only temporally coherent but also spatially smooth. Through joint constraints between the time axis and spatial distribution, the cross-frame feature bridging layer can form a logically consistent, temporally correct, and spatially stable feature association path within the occlusion interval, so that the sampled object can seamlessly connect with the features before occlusion when it reappears, avoiding identity confusion or misjudgment caused by feature breakage.

[0052] After generating cross-frame feature association paths, the bridging results undergo continuity verification and feature aggregation channel construction. The focus at this stage is to ensure that the paths output by the cross-frame feature bridging layer are consistent and traceable in both temporal and spatial dimensions. To this end, the time indices at both ends of the bridging path should be aligned with the original time indices in the dynamic feature sequence to ensure that the temporal signature of the bridging path remains consistent with the overall sequence. Simultaneously, in the spatial dimension, the feature points covered by the bridging path should be continuously smoothed to maintain consistent visual representation with adjacent frames in terms of spatial location, pose contour, and illumination distribution. This continuity verification process further filters out feature drift and positional deviations that may be caused by occlusion, making the final feature association path a truly temporally continuous, spatially smooth, and semantically consistent feature extension channel. After verification, the output of the cross-frame feature bridging layer is the corrected cross-frame feature association path. This path achieves continuous connection of features before and after occlusion in the time dimension, maintains the consistency of visual expression in the spatial dimension, and maintains the coherence of object identity in the semantic dimension, thus providing a stable input foundation for subsequent dynamic re-identification and identity index construction.

[0053] This implementation method achieves cross-frame bridging and extension of feature information within occlusion intervals by performing frame-by-frame matching, breakpoint identification, bridging path construction, and continuity verification on dynamic feature sequences with time indexes. This process not only ensures the complete transmission of image features along the time axis but also maintains the consistency of object appearance features at the spatial level, enabling the captured object to maintain the continuity of the feature chain even under complex conditions such as occlusion, reappearance, or changes in lighting. Through this feature bridging method under dual constraints of temporal and spatial constraints, the problem of feature breakage and identity mismatch that easily occurs in traditional image recognition methods in rapid scene switching and complex dynamic environments is effectively eliminated. This significantly improves the stability and semantic consistency of object recognition during the acquisition of new media materials, laying a high-precision foundation for subsequent identity mapping and multi-subject interactive fusion.

[0054] Dynamic Re-identification Mapping Module: Based on the feature association path output by the cross-frame feature bridging layer, a dynamic re-identification mapping layer is constructed on the time axis. Feature points in the feature association path whose similarity meets the set threshold are aggregated to form an identity anchor set, and an identity index structure is generated accordingly to provide a basic index for illumination adaptation.

[0055] The specific implementation method of this module is as follows:

[0056] After obtaining the feature association paths output by the cross-frame feature bridging layer, temporal alignment and similarity evaluation are performed on the feature points of each time segment within the path. Since the cross-frame feature association paths encompass continuous features before and after occlusion, each path contains feature representations across multiple time segments. To ensure that features from different time segments are comparable in the temporal dimension, the feature paths are first sorted according to their time indices, ensuring that feature points within the path strictly follow a temporal ascending relationship. Subsequently, similarity evaluation is performed on feature points of adjacent time segments to determine their consistency at the identity level. This similarity evaluation considers not only the spatial distribution and appearance similarity of features but also the continuous change trend of features in the temporal dimension, thereby avoiding misjudgments caused by short-term illumination changes or pose adjustments. Through this temporal alignment and similarity analysis process, continuous feature segments conforming to the identity continuity pattern can be identified within the feature association paths, laying the foundation for subsequent feature aggregation.

[0057] After completing temporal alignment and similarity assessment, feature points that meet a set similarity threshold in the feature association path are aggregated to form a preliminary identity anchor set. In this step, the similarity threshold is used as the aggregation standard, and feature points with high consistency on the time axis are grouped into the same identity trajectory. This ensures that the composition of the identity anchor set not only reflects the stability of spatial features but also embodies the continuity of the temporal dimension. Each identity anchor represents the feature aggregation center within a specific time segment, and its formation process is a concise expression of the feature consistency of the sampled object within that time period. To ensure the integrity of the anchor set in the temporal dimension, the continuity of the time index must be maintained during the aggregation process. That is, the time series of feature points in the same identity anchor set must not be broken, thus ensuring that the anchor set has both temporal continuity and identity uniqueness. Through this temporally ordered feature aggregation method, the feature trajectories of different objects in the same scene can be effectively distinguished, avoiding identity confusion caused by local appearance similarities.

[0058] After obtaining the identity anchor set, each identity anchor set is structurally summarized to generate an identity index structure. This identity index structure is a feature mapping system that arranges multiple identity anchor sets in an ordered manner on a time axis, with the time index as the primary key. Specifically, the arrangement relationship of each identity anchor set on the time axis should be sorted out according to the time span, spatial distribution, and feature stability of each set, so that the same sampled object forms a complete identity trajectory in different time periods. To ensure the scalability and relevance of the index structure, the identity index structure needs to retain the correspondence between time hierarchy and spatial features during construction. That is, each identity index unit not only records the time sequence position of the feature anchor but also includes its spatial positioning range and appearance description. Through this time-space dual-dimensional index organization, the structure can be directly referenced in subsequent illumination adaptation and identity fusion to achieve efficient identity retrieval and associated positioning. At the same time, the establishment of the identity index structure also provides a foundation for multi-subject management in complex scenarios, enabling the system to distinguish and track multiple independent identity sequences within the same time period, thereby improving the overall recognition accuracy and temporal consistency.

[0059] After the identity index structure is generated, consistency verification and illumination adaptation preparation are performed. Since the identity anchor point set originates from cross-frame feature association paths, and image acquisition at different time periods often involves changes in illumination conditions, it is necessary to ensure that the identity index structure maintains feature stability under varying illumination conditions. Therefore, illumination consistency analysis is performed on each anchor point set in the identity index structure. By comparing the feature distribution of the same captured object in different illumination intervals, feature shifts caused by illumination differences are identified, and this shift information is recorded in the index structure. Subsequently, this shift data can provide a reference for the subsequent illumination normalization process, enabling the identity index to quickly match and adjust feature weights during the illumination adaptation stage, ensuring the continuity and accuracy of the recognition results. Through this verification and preparation process, the identity index structure not only possesses temporal and spatial consistency but also adaptive capabilities to environmental changes, thus forming a robust identity index foundation layer.

[0060] This implementation achieves a complete establishment of a dynamic re-identification mapping layer by temporal alignment of cross-frame feature association paths, feature aggregation, identity anchor formation, and index structure construction. This process fully utilizes the dual constraints of temporal index and spatial features, ensuring that the captured object maintains continuous expression of identity features even under complex conditions such as occlusion, lighting changes, and pose adjustments. The generation of the identity anchor set not only enhances the aggregation of features along the time axis but also achieves feature traceability and scalability through the establishment of the identity index structure. Through this structured dynamic re-identification method, stable mapping of cross-frame identity information is achieved during the acquisition of new media content, providing a high-precision index foundation for subsequent lighting normalization processing and multi-subject temporal fusion, thereby effectively improving the reliability and semantic consistency of identity recognition during the content acquisition process.

[0061] Illumination Adaptive Normalization Module: Based on the identity index structure, dynamic illumination partitioning is performed. Within each illumination partition, illumination normalization is performed according to the feature weights of the identity anchor point set, so that the identity index maintains feature stability under illumination changes and provides consistent feature input for subsequent temporal fusion.

[0062] The specific implementation method of this module is as follows:

[0063] After generating the identity index structure, dynamic partitioning and identification are performed on the illumination changes in the acquisition sequence. Since new media material acquisition often occurs in environments with constantly changing illumination conditions, such as changes in the angle of natural light, switching of indoor light sources, and interference from reflective surfaces, it is necessary to partition the illumination changes along the time axis. The determination of illumination partitions is based on the time index structure. By detecting the brightness change trend within continuous time periods, the boundary points where significant changes in illumination intensity occur are identified. Each partition represents an acquisition interval with relatively consistent illumination conditions within a certain time range. After the illumination partitions are determined, the set of identity anchor points corresponding to the time period is assigned to that partition, ensuring that each illumination partition corresponds to a set of identity anchor points formed under that illumination condition. This joint time and illumination partitioning method ensures that subsequent feature normalization processing can specifically eliminate the impact of illumination differences on feature expression, maintaining a consistent baseline state of illumination within each partition.

[0064] After the illumination partitions are determined, feature weights are assigned to the set of identity anchor points within each partition. Since the contribution of various visual features (such as color, brightness, and shadow contours) of the sampled object to the overall feature representation varies under different illumination conditions, it is necessary to assign corresponding feature weights to each identity anchor point based on its stability under the current partition's illumination conditions. This weight reflects the reliability of the anchor point's feature in the illumination representation, i.e., whether the representation under that illumination condition is sufficiently stable and representative. During the weight assignment process, the consistency of the anchor point's performance in adjacent partitions should be comprehensively considered to prevent illumination anomalies in a single interval from excessively affecting the overall feature stability. Through this feature weight allocation within partitions, feature points with more stable illumination can obtain a higher weight ratio in subsequent normalization processing, while feature points with higher illumination sensitivity will have their participation intensity reduced accordingly. This forms an illumination-steady expression framework in the overall feature aggregation, laying the foundation for accurate execution of illumination normalization.

[0065] After assigning feature weights, illumination normalization is performed on the set of identity anchor points within each illumination zone. The core of illumination normalization is to eliminate the disruption of feature expression consistency caused by differences in illumination conditions, ensuring that the feature representation of the same sampled object tends to be uniform under different illuminations. In this process, the feature intensity and brightness response range of each anchor point are adjusted based on the feature weights within the illumination zone, making the feature distribution of all anchor points within that zone more consistent. Specifically, while maintaining the continuity of the temporal index, the feature expressions of adjacent zones should be smoothly connected to prevent abrupt feature changes at illumination transitions. During normalization, the transition regions between illumination zones must also be considered to ensure smooth feature transitions and semantic consistency. For example, when illumination changes from strong to weak or from warm to cool colors, the feature normalization process should use weight balancing to gradually change the feature expression, maintaining visual consistency over time. Through this unified correction method within and outside illumination zones, a continuous and illumination-adaptive feature expression chain can be formed on the time axis, enabling the identity index to maintain stable recognition capabilities even under drastic illumination changes.

[0066] After normalizing each illumination partition, the global identity index structure undergoes consistency verification to ensure that the features of all illumination partitions have a unified representation benchmark in the overall time series. The main task at this stage is to detect the feature consistency and temporal coherence between different partitions after illumination normalization. During verification, the feature trajectories of all illumination partitions should be reintegrated into a complete time series, using the time index as the main thread, and checking for feature gaps or representational shifts at the boundaries of adjacent partitions. If feature discontinuities are detected, the feature weights of adjacent partitions need to be readjusted to ensure a smooth transition of boundary features in both spatial and temporal dimensions. This global consistency verification ensures that the identity index structure maintains a stable feature extension relationship across illumination intervals, preventing identity misjudgments or semantic mismatches due to illumination changes. Furthermore, based on the consistency verification, illumination adaptation index records can be generated to provide an illumination reference benchmark for subsequent temporal fusion stages, allowing the subsequent fusion process to directly reference the unified feature input after illumination normalization, thereby achieving dynamic feature integration in a multi-agent environment.

[0067] This implementation achieves dynamic adaptation and feature stability of the identity index structure under complex lighting conditions through illumination partitioning recognition, feature weight allocation, illumination normalization processing, and global consistency verification. This process not only effectively eliminates feature shifts and recognition errors caused by illumination changes but also maintains the continuity and smoothness of feature expression over time. This combination of partitioning and dynamic normalization enables the new media material acquisition method to maintain high-precision identity preservation and semantic stability even under complex lighting environments. It provides consistent input conditions for subsequent temporal fusion and multi-subject interactive recognition, thereby significantly improving the authenticity, consistency, and application quality of the acquired materials in visual communication.

[0068] Multi-subject temporal fusion module: Based on the feature input after illumination normalization, a multi-subject temporal fusion layer is established, which merges the feature trajectories of identity anchor points in each time period to realize continuous identity tracking in multi-subject interaction scenarios, complete the identity restoration and semantic extension after the object being collected is occluded, and thus realize the accurate collection and association of new media material content;

[0069] The specific implementation method of this module is as follows:

[0070] Based on the feature input after illumination normalization, the temporal distribution of different identity anchor point sets is organized and aligned. Since the feature input after illumination normalization has achieved uniformity in brightness and contrast, the key at this stage is to uniformly align the feature trajectories of each identity anchor point set on the time axis, so that the trajectories of different subjects are described using the same time index system. To this end, the time index structure should be the main thread, arranging each identity anchor point set according to its order of appearance in the time dimension, and filling in time nodes in the missing time index segments through interpolation or mapping to ensure that each identity trajectory has a continuous time reference. This time dimension alignment not only ensures the temporal consistency between the trajectories of multiple subjects, but also provides a comparable temporal benchmark for subsequent feature fusion. It is worth noting that in multi-subject coexistence scenarios, a dual correspondence between the time index and the identity index should be maintained simultaneously, so that the features of different subjects within the same time segment will not be confused, thus providing clear identity boundaries for trajectory fusion.

[0071] After time alignment, the feature trajectories of each identity anchor set are fused spatially and temporally. The key to this step is to temporally integrate the identity feature trajectories from different time periods, creating a continuous dynamic feature chain for each sampled object on the timeline. To achieve this, the identity anchor set is traversed frame by frame on the timeline, and features from adjacent frames are fused according to the time index order, thus forming a trajectory line with temporal continuity. During trajectory fusion, potential spatial overlap and occlusion between subjects must also be considered. When different subjects overlap spatially within the same time segment, the trajectory of the subject with stronger feature continuity should be prioritized, and a feature weight allocation mechanism should be used to partially compensate for potentially occluded subject features to ensure the independence and integrity of each subject's identity trajectory during interaction. Through this dual temporal and spatial fusion method, the feature trajectories of multiple subjects in the same scene can be continuously extended within a unified time frame, thereby achieving continuous expression of interactive actions and stable mapping of identity tracking in subsequent processing stages.

[0072] After completing the initial multi-subject trajectory fusion, the identity relationships between different subjects are identified and distinguished to prevent identity overlap or misassociation in interactive scenarios. In multi-subject interaction scenarios, different objects may experience feature confusion due to similar postures, overlapping actions, or spatial proximity. Therefore, it is necessary to distinguish the relationships between the trajectories of different identities during the fusion stage. To this end, based on the unique identity identifier defined in the aforementioned identity index structure, the feature trajectories of different subjects should be separated according to the correspondence of identity anchor points, and possible identity intersections should be detected on the time axis. When multiple identity trajectories are found to have overlapping features or spatial proximity within the same time period, it should be determined whether the overlapping area belongs to the extended trajectory of the same subject or is an interactive transient of different subjects, based on the feature consistency index after illumination normalization and the temporal continuity constraint. If it is determined to be an occlusion extension of the same subject, the trajectory continuity is preserved; if it is determined to be an interactive event of different subjects, the boundary is clearly defined in the identity index and the trajectory independence is maintained. This identity relationship identification process can effectively avoid identity mismatch caused by subject contact or similar posture, ensuring that each trajectory in the multi-subject fusion process has a clear identity attribution, thereby maintaining data consistency and reliability in the subsequent semantic processing stage.

[0073] After multi-subject trajectory fusion and identity relationship discrimination are completed, the fusion results undergo semantic extension processing to ensure the continuity and consistency of the identity expression and semantic orientation of the sampled objects in the temporal dimension. The goal of semantic extension is to ensure that the fused trajectory not only reflects the spatial movement relationship and temporal continuity of the sampled objects, but also reflects their behavioral logic and scene context at the semantic level. To this end, based on the fused trajectory set, the feature change trends of each subject in different time periods should be semantically correlated to form a semantic chain across time segments. During the semantic extension process, the feature input after illumination normalization should be fully utilized to ensure that the semantic chain can maintain recognition consistency under illumination changes. At the same time, the temporal information of the aforementioned identity anchor point set should be combined to dynamically complete the semantic chain, ensuring that the behavioral trajectory of each subject presents a continuous logical evolution at the semantic level. Through this semantic layer extension processing, not only can the identity of the sampled objects be seamlessly connected after occlusion restoration, but the overall semantic consistency of the scene can also be maintained during multi-subject interaction, thus making the entire acquisition process coherent and realistic in both semantic understanding and visual expression.

[0074] This implementation method constructs a multi-subject temporal fusion layer through time alignment, trajectory fusion, identity recognition, and semantic extension. Relying on feature inputs after illumination normalization, this process achieves dynamic identity fusion and semantic preservation under complex lighting and multi-subject interaction conditions. This ensures that the captured objects maintain continuous identity trajectories and consistent semantic expression even under complex situations such as occlusion, lighting changes, and overlapping actions. This temporal fusion mechanism significantly improves the accuracy and semantic coherence of new media material collection, providing high-quality data support with complete structure and semantic consistency for subsequent content generation, behavioral analysis, and visual dissemination.

[0075] This invention introduces dynamic feature sequences and cross-frame feature bridging mechanisms during image acquisition, enabling features between frames before and after occlusion to maintain continuous correlation on the time axis. This effectively avoids identity interruption and mismatch issues when an object reappears after a brief disappearance, achieving smooth connection of features in the temporal dimension and persistent mapping of identity information. Consequently, it significantly improves the accuracy of target tracking and the stability of material acquisition under conditions of rapid scene switching.

[0076] This invention establishes a dynamic re-identification mapping layer and a multi-subject temporal fusion layer, enabling identity anchors to maintain semantic consistency and trajectory integrity in environments with varying lighting and multi-subject interactions. It achieves continuous fusion of feature normalization and identity indexing adapted to lighting, allowing the acquisition system to maintain consistency of material tags and integrity of semantic structure in complex dynamic scenes, thereby significantly improving the accuracy and reliability of image material acquisition and content association.

[0077] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A new media material content acquisition system based on image data processing, characterized in that, It includes a dynamic feature extraction module, a cross-frame feature bridging module, a dynamic re-identification mapping module, an illumination adaptive normalization module, and a multi-subject temporal fusion module; Dynamic feature extraction module: acquires a continuous frame image sequence, performs spatial and temporal feature extraction on each frame in the image sequence, binds the extracted spatial and temporal features according to the frame order, and generates a dynamic feature sequence with a time index; Cross-frame feature bridging module: Based on dynamic feature sequences, a cross-frame feature bridging layer is constructed. The time index is used to match the features of adjacent frames frame by frame, identify feature breakpoints before and after occlusion, and generate feature association paths that cross the occlusion interval. Dynamic Re-identification Mapping Module: Based on the feature association path output by the cross-frame feature bridging layer, a dynamic re-identification mapping layer is constructed on the time axis. Feature points in the feature association path whose similarity meets the set threshold are aggregated to form an identity anchor set, and an identity index structure is generated accordingly. Illumination Adaptive Normalization Module: Based on the identity index structure, dynamic illumination partitioning is performed. Within each illumination partition, illumination normalization is performed according to the feature weights of the identity anchor point set, so that the identity index maintains feature stability under illumination changes. Multi-subject temporal fusion module: Based on the feature input after illumination normalization, a multi-subject temporal fusion layer is established, which fuses the feature trajectories of identity anchor points in each time period to complete the identity restoration and semantic extension of the sampled object after occlusion.

2. The new media material content acquisition system based on image data processing according to claim 1, characterized in that, The steps for acquiring a sequence of consecutive frame images and generating a dynamic feature sequence with a time index include: Acquire a continuous sequence of images, perform preprocessing on each frame of the image sequence, acquire images at fixed time intervals while keeping the optical focal length and imaging parameters consistent, and perform brightness equalization, color normalization, geometric distortion correction and edge sharpness enhancement on the images. Spatial feature extraction is performed on each preprocessed frame image to extract a set of spatial features including appearance contour, color distribution, compositional relationship, regional hierarchy and geometric positional relationship, and the feature scale and expression standard are kept consistent across frames; By combining spatial features, temporal features between consecutive frames are extracted to obtain dynamic change information including displacement trends, morphological transformations, motion trajectories, and illumination changes, so that each frame of the image has both static and dynamic feature representations. Spatial and temporal features are fused and bound together, and a time index is established according to frame order to generate a dynamic feature sequence with time identifiers.

3. The new media material content acquisition system based on image data processing according to claim 1, characterized in that, The steps for constructing a cross-frame feature bridging layer based on dynamic feature sequences include: After acquiring the dynamic feature sequence with time index, the feature correspondence between consecutive frames is initialized and matched. The frame sequence identifier of the time index is used to compare the features of adjacent frames, and the spatial and temporal features are combined to ensure the temporal continuity of the comparison. After completing the inter-frame matching, the intervals where the feature association is interrupted in the time index sequence are detected. The positions where the change in spatial features of adjacent frames exceeds a preset threshold and the time interval does not exceed a specific frame distance are identified as occlusion breakpoints, and the time indexes of the breakpoint start frame and the recovery frame are recorded. Based on the identified occlusion breakpoints, a cross-frame feature bridging path is constructed on the time axis, starting with the features of the last frame before occlusion and ending with the features of the first frame after occlusion. The generated cross-frame feature association path is subjected to continuity verification. The time indices at both ends of the bridging path are aligned with the dynamic feature sequence. Spatial smoothing correction is performed on the feature points covered by the path to form a feature association channel that is temporally continuous, spatially consistent and semantically coherent.

4. A new media material content acquisition system based on image data processing according to claim 3, characterized in that, In the process of constructing the cross-frame feature bridging path, the connection between the last frame feature before occlusion and the first frame feature after occlusion is extended according to the time index order in the dynamic feature sequence, and the spatial position change of the sampled object during the occlusion period is inferred by combining the distribution trend of spatial features before and after occlusion.

5. A new media material content acquisition system based on image data processing according to claim 3, characterized in that, The steps for constructing a dynamic re-identification mapping layer based on the feature association paths output by the cross-frame feature bridging layer include: After obtaining the feature association path output by the cross-frame feature bridging layer, the feature points of each time segment in the path are time-series aligned, sorted according to the time index, and similarity evaluation is performed on the feature points of adjacent time segments to identify continuous feature segments that conform to the identity continuity rule. Based on the temporal alignment results, feature points in the feature association path whose similarity meets a set threshold are aggregated to form a set of identity anchor points that are continuous on the time axis, while maintaining the consistency of the time index. After obtaining the set of identity anchor points, the set of identity anchor points is structured and summarized. An identity index structure is generated based on the time span, spatial distribution and feature stability, and a mapping relationship between time and space is established. The generated identity index structure is subjected to consistency verification and illumination adaptation preparation. The differences in feature distribution within different illumination intervals are compared, and illumination offset information is recorded to provide basic index support for illumination normalization and identity fusion.

6. A new media material content acquisition system based on image data processing according to claim 5, characterized in that, In the process of generating the identity index structure, the arrangement of the identity anchor point set on the time axis is based on the time index as the primary key, and combined with the positioning range of spatial features for dual-dimensional indexing, so that the features of the same sampled object in different lighting intervals remain temporally continuous and spatially consistent.

7. A new media material content acquisition system based on image data processing according to claim 5, characterized in that, The steps for performing dynamic lighting partitioning based on the identity index structure include: After generating the identity index structure, the illumination changes in the collection sequence are detected according to the time index, the boundary points of illumination intensity changes are identified and the illumination partitions are divided, and the identity anchor point set of the corresponding time period is assigned to the corresponding partition to establish an illumination consistency interval. After determining the illumination partitions, feature weights are assigned to the set of identity anchor points within each illumination partition. The weight values ​​are determined based on the stability of each anchor point under the current illumination conditions, forming a feature expression framework dominated by illumination steady state. Based on feature weights, illumination normalization is performed on the set of identity anchor points within the illumination partition, adjusting the feature intensity and brightness response range of each anchor point, and maintaining a smooth transition of features between adjacent partitions. After completing illumination normalization, consistency verification is performed on the global identity index structure, all illumination partition features are reintegrated into a complete time series, and the weights of adjacent partition boundary features are adjusted.

8. A new media material content acquisition system based on image data processing according to claim 7, characterized in that, During the illumination normalization process, a weighted smoothing adjustment method is used for the transition area of ​​adjacent illumination zones to ensure that the identity anchor point features at the zone boundary maintain a consistent gradient in brightness and color distribution.

9. A new media material content acquisition system based on image data processing according to claim 7, characterized in that, The steps for establishing a multi-subject temporal fusion layer based on illumination-normalized feature inputs include: Based on the feature input after illumination normalization, the temporal distribution of different identity anchor sets is organized and aligned. The identity anchor sets are arranged in chronological order with the time index structure as the main line, and the missing time index segments are filled in. Based on the time alignment results, the feature trajectories of each identity anchor point set are fused spatially and temporally. The features of adjacent frames are fused into continuous trajectories according to the time index order, and the independence and integrity of the trajectories are maintained through feature weight compensation when there is spatial overlap. After completing trajectory fusion, the identity relationships between different subjects are identified and distinguished. Based on the unique identity identifier, the identity intersection points on the time axis are detected, and the occlusion extension and interaction events are distinguished according to the illumination normalization characteristics and temporal continuity. After completing the fusion of multi-subject trajectories and identity recognition, the fusion results are semantically extended. A semantic chain is established based on the temporal information of the identity anchor set, so that the collected objects maintain the continuity and consistency of identity expression and semantic orientation in the time dimension.

10. A new media material content acquisition system based on image data processing according to claim 9, characterized in that, Semantic extension processing further includes: when establishing semantic chains, based on the time sequence information of the feature input after illumination normalization and the identity anchor set, dynamically completing the feature change trend of different subjects in the interaction process, so that the semantic chains maintain logical coherence and semantic consistency under illumination changes and occlusion recovery conditions.

Images