Holographic image intelligent interaction system and method based on AI

Abstract

The invention relates to the technical field of virtual reality interaction, in particular to an AI-based holographic image intelligent interaction system and method, and the system comprises a layer structure fusion module, a pixel boundary scheduling module, a user posture recognition module, a dynamic behavior judgment module and an interaction layer response module. According to the method, the pixel matrix of the multi-resolution image layer and the gradient direction and frequency information of the overlapping area are extracted, the image layer overlapping sequence is defined, the spatial logic and visual hierarchy of image layer fusion are optimized, the transparency is adjusted in combination with the boundary brightness difference and the pixel density, the image transition naturalness is improved, and the tearing and aliasing feeling is relieved; based on symmetric axis construction and offset sequence analysis of skeleton key points, fine tracking and action recognition of posture changes are achieved, an image response matrix is generated through the corresponding relation between postures and transparency changes, the content of an image layer is made to have dynamic linkage, the interaction response efficiency and immersion are enhanced, and a collaborative mechanism of image processing and behavior feedback is constructed.

Description

Technical Field

[0001] This invention relates to the field of virtual reality interaction technology, and in particular to an AI-based holographic intelligent interaction system and method. Background Technology

[0002] The field of virtual reality (VR) interaction technology encompasses techniques that allow users to interact with virtual objects in a natural way within a computer-generated 3D virtual environment. The core components of this technology include virtual scene construction, sensory integration, interactive behavior recognition, and feedback response. VR interaction typically integrates technologies such as visual display, motion capture, speech recognition, and haptic feedback to create an immersive human-computer interaction environment. In terms of overall system composition, VR interaction systems not only involve graphics and image generation and rendering technologies but also encompass the integrated application of user behavior recognition, environmental perception, and interactive decision-making mechanisms, finding wide application in scenarios such as education, healthcare, military simulation, and industrial design.

[0003] Among them, the AI-based holographic intelligent interaction system refers to a system that integrates artificial intelligence and holographic image generation technology to construct a three-dimensional visual interactive platform capable of intelligent semantic understanding and behavior recognition. This system primarily addresses users' interactive needs in three-dimensional virtual scenes, including natural language expression, gestures, and facial expressions. It employs semantic recognition methods based on natural language parsing, image recognition strategies based on deep learning, and holographic light field imaging technology to achieve multimodal recognition and image presentation control of user input. The system performs acoustic feature analysis based on user-input voice signals to extract semantic intent; simultaneously, it combines gesture trajectory recognition algorithms from camera images to determine the type of action command; and through a three-dimensional holographic image projection mechanism, it performs image reconstruction and spatial distribution management of the displayed content in the virtual scene. This technical solution utilizes a neural network model to analyze user behavior patterns and generates interactive feedback based on a decision rule base. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an AI-based holographic image intelligent interaction system and method.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An AI-based holographic image intelligent interaction system includes: The layer structure fusion module collects the pixel matrix of multi-resolution layers in the fusion display area, extracts the pixel gradient direction features in the overlapping area, determines the degree of consistency with the direction of the neighboring area, and combines the pixel frequency to determine the layer stacking order and generate a layer arrangement position mapping table.

[0006] The pixel boundary scheduling module extracts the gradient direction and brightness difference of the layer boundary position based on the layer arrangement position mapping table, determines the priority of boundary pixel superposition, calibrates the transparency adjustment order, and establishes a transparency control map of the boundary area.

[0007] The user posture recognition module uses the boundary region transparency control map to extract the frame sequence of skeletal key points within a time period, constructs a symmetry axis with the shoulder and hip as the reference, records the distribution difference of key points on both sides of the symmetry axis, and generates a symmetry axis key point offset change sequence.

[0008] The dynamic behavior determination module analyzes the time trend and amplitude fluctuation of the offset change sequence of the key points of the symmetry axis, identifies abnormal action segments in combination with the stability benchmark, and generates a set of abnormal body posture change segments of the user.

[0009] The interactive layer response module calls the set of abnormal body posture changes of the user, locates the layer range in the transparency control image, extracts the relationship between transparency and key point coordinates, and generates an interactive frame display adjustment record table.

[0010] As a further embodiment of the present invention, the layer arrangement position mapping table includes gradient direction consistency, pixel frequency distribution, and overlay order calibration; the boundary region transparency control map includes boundary gradient features, brightness difference value, pixel density distribution, overlay priority calibration, and transparency adjustment sequence; the symmetry axis key point offset change sequence includes skeletal key point frame sequence, body symmetry axis reference, and key point difference time series; the user abnormal body posture change segment set includes offset amplitude fluctuation range, stability reference segment, and key point trajectory data; and the interactive frame segment display adjustment record table includes layer range positioning data, transparency coordinate relationship matrix, and display adjustment integration record.

[0011] As a further aspect of the present invention, the layer structure fusion module includes: The pixel matrix extraction submodule collects the pixel matrix of multi-resolution layers in the blended display area, filters the pixels in the overlapping area, divides the area according to the layer number, extracts the grayscale information and coordinate position of the pixels in the area, constructs a unified coordinate index structure, and obtains the pixel matrix value of the layer overlay. The gradient direction evaluation submodule extracts the gradient direction of pixels in the overlapping area based on the pixel matrix values ​​of the layer, obtains the direction data of adjacent pixels, determines the consistency between the current pixel direction and the neighborhood direction, and obtains the direction consistency matching degree by combining the matching statistics. The layer sorting and mapping submodule counts the pixel frequency of the layers in the overlapping area based on the directional consistency matching degree, combines the directional matching and frequency results for joint sorting, calibrates the arrangement order of the layers in the blended display area, and generates a layer arrangement position mapping table.

[0012] As a further aspect of the present invention, the pixel boundary scheduling module includes: The gradient extraction submodule extracts the horizontal and vertical brightness values ​​of the pixel positions at the layer boundary based on the layer order information in the layer arrangement position mapping table. Based on the brightness changes and pixel coordinate positions of adjacent areas, it analyzes the boundary direction features and brightness differences to generate boundary gradient direction feature values. The density determination submodule calls the boundary gradient direction feature value to obtain the pixel distribution of a local area in the layer. Based on the direction feature distribution and the degree of pixel aggregation, it determines the boundary density state within the area and generates a local pixel density distribution coefficient. The transparency sorting submodule determines the priority level and overlay rules of boundary pixels based on the local pixel density distribution coefficient and the boundary gradient direction feature value, combined with the layer arrangement order and pixel distribution state, and generates a boundary region transparency control map.

[0013] As a further aspect of the present invention, the specific calculation formula for the spatial distribution dispersion parameter is as follows: ; Where F represents the spatial dispersion parameter for computation. This represents the intensity value of the pixel in the i-th row and j-th column of the local region. This represents the absolute value of the horizontal coordinate difference between the pixel and the center point of the region. This represents the absolute value of the vertical coordinate difference between the pixel and the center point of the region. The gradient direction angle of the pixel is represented by , k represents the horizontal feature compensation coefficient, m represents the vertical aggregation correction coefficient, N represents the total number of pixels in the local region, n represents the number of rows in the region, and b represents the number of columns in the region.

[0014] As a further aspect of the present invention, the user posture recognition module includes: The transparency region identification submodule obtains the transparency control map of the boundary region, extracts continuously changing regions based on the gray value distribution in the map, identifies the time frame number range corresponding to the transparency partition, filters the frame segments with significant transparency, and generates transparent frame interval segments. The skeleton sequence extraction submodule calls the user skeleton key point information in the corresponding frame according to the transparent frame interval, extracts the key point coordinate set of each frame, locates the shoulder and hip positions, constructs the axis of symmetry direction, divides the key point position distribution according to the direction, and generates the axis of symmetry reference coordinate sequence. The symmetry offset calculation submodule calls the symmetry axis reference coordinate sequence, identifies the distribution differences of key points on both sides of the symmetry axis, calculates the fusion geometric correlation, sorts out the changing trend of differences in continuous frames, extracts the difference fluctuation features of key positions, and generates a dynamic change sequence of symmetry axis key point offset.

[0015] As a further aspect of the present invention, the specific calculation formula for calculating the fusion geometric correlation degree is as follows: ; in, Represents the geometric correlation correction coefficient. Representing the Left distance measurement at key points, Representing the Distance measurement to the right of each key point. This represents the radian value between the left measuring point and the baseline of the axis of symmetry. This represents the radian value between the right measuring point and the baseline of the axis of symmetry. Representing the Dynamic weights of key points Represents the spatial distribution dispersion (take the arithmetic root of the sum of squares of the weights of all key points). This represents the number of key points on one side.

[0016] As a further aspect of the present invention, the dynamic behavior determination module includes: The offset sequence detection submodule obtains the time and position information recorded in the offset change sequence of the symmetry axis key points, identifies the changes in the coordinates of key points in adjacent frames, sorts out the position change amplitude of key points in the sequence, generates the position change trend in time order, and obtains the inter-frame offset trend value of key points. The amplitude fluctuation analysis submodule extracts the offset amplitude sequence over a time period based on the inter-frame offset trend value of the key points, identifies the changes in offset amplitude between consecutive frames, determines the correspondence between the fluctuation range and the offset amplitude stability benchmark value, screens out the time periods of abnormal fluctuations, and obtains the time interval of abnormal offset fluctuations. The abnormal segment extraction submodule retrieves the spatial trajectory data of key points within the corresponding time period based on the abnormal offset fluctuation time interval, integrates the continuous position information of key points within the abnormal segment, constructs a position trajectory sequence of multiple key points, and obtains a set of abnormal body posture change segments of the user.

[0017] As a further aspect of the present invention, the interactive layer response module includes: The frame segment calibration submodule obtains the marked frame segment range in the set of abnormal body posture change segments of the user, extracts the corresponding sequence frame image number, counts the number of image frames and constructs a time interval list, divides the non-continuous segments in the frame segment, establishes a number index corresponding structure, and obtains the abnormal frame segment number index group. The region extraction submodule calls the frame sequence number in the abnormal frame segment number index group to locate the layer region index, obtain the position coordinate value of the boundary region in the corresponding frame, determine the contour closure feature, extract the transparency parameter value in the region, calculate the combination interval of change frequency and amplitude, and obtain the layer transparency fluctuation interval group. The relationship mapping submodule matches the set of key point spatial coordinates based on the layer transparency fluctuation range group, extracts the overlap and offset distance between the points and the layer boundary, constructs the correspondence sequence between key points and transparency, judges the consistency of direction and the difference of fluctuation, establishes a response matrix, and obtains the interactive frame display adjustment record table.

[0018] The AI-based intelligent interaction method for holographic images includes the following steps: S1: Collect the pixel matrix of the different resolution layers in the blended display area, detect the gradient direction of the pixels in the overlapping area of ​​the layers, combine the directional distribution of neighboring pixels and the frequency of pixel occurrence, determine the directional consistency level between the layers, and generate a layer arrangement position mapping table. S2: Based on the layer order information in the layer arrangement position mapping table, extract the gradient direction and brightness difference features of the layer boundary region, combine the pixel density distribution in the local region, determine the superposition priority level of the boundary pixels in the layer fusion process, and establish a boundary region transparency control map. S3: Based on the transparency region in the boundary region transparency control map, extract the user's skeletal key point frame sequence within the corresponding time period, construct the axis of symmetry with the shoulder and hip key points, record the difference in the number of key points on both sides of the axis of symmetry in the frame, and generate the axis of symmetry key point offset change sequence. S4: Based on the time change trend in the key point offset change sequence of the symmetry axis, analyze the fluctuation range of the key point offset amplitude between frames, combine the stability benchmark to delineate the frame segment interval of abnormal change, extract the key point position trajectory within the segment, and generate a set of abnormal body posture change segments of the user. S5: Call the frame range in the set of abnormal body posture changes of the user, locate the layer area in the transparency control map of the boundary area, extract the transparency change record of the corresponding area and the key point coordinate relationship, integrate the area change information, and generate an interactive frame display adjustment record table.

[0019] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by extracting the pixel matrix and gradient direction and frequency information of overlapping areas of multi-resolution layers, the layer stacking order is clarified, the spatial logic and visual hierarchy of layer fusion are optimized, and the transparency is adjusted by combining boundary brightness differences and pixel density to improve the naturalness of image transitions and reduce tearing and aliasing. Based on the symmetry axis construction and offset sequence analysis of skeletal key points, fine tracking of body posture changes and action recognition are achieved. An image response matrix is ​​generated through the correspondence between body posture and transparency changes, enabling the layer content to have dynamic linkage, enhancing interactive response efficiency and immersion, and constructing a collaborative mechanism for image processing and behavioral feedback. Attached Figure Description

[0020] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a system block diagram of the present invention; Figure 3 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0022] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0023] Please see Figure 1 The AI-based holographic intelligent interaction system includes: The layer structure fusion module collects the pixel matrix of multi-resolution layers in the holographic image fusion display area, extracts the gradient direction features of pixels in the overlapping area of ​​layers, determines the consistency between the gradient direction features and the direction between neighboring pixels, and combines the pixel frequency information to determine the layer stacking order and generate a layer arrangement position mapping table. The pixel boundary scheduling module extracts the gradient direction features and brightness difference values ​​of the layer boundary positions based on the layer order information in the layer arrangement position mapping table. Combined with the pixel density information in the local area, it determines the superposition priority of boundary pixels in the fusion process, marks the transparency adjustment order of each boundary point, and establishes a transparency control map of the boundary area. The user posture recognition module extracts the user's skeletal key point frame sequence within the time period of the region based on the transparency distribution area in the boundary region transparency control map, constructs a posture symmetry axis with the shoulder and hip as the reference, performs difference recording on the distribution of key points on both sides of the symmetry axis in each frame, forms a time series, and generates a symmetry axis key point offset change sequence. The dynamic behavior determination module analyzes the fluctuation range of key point offset amplitude between frames based on the time change trend recorded in the key point offset change sequence of the symmetry axis, combines the stability judgment benchmark to delineate abnormal action segments, extracts the key point position trajectory within the segment, and generates a set of abnormal body posture change segments of the user. The interactive layer response module calls the marked frame range in the set of abnormal body posture changes of the user, locates the layer range in the transparency control map of the corresponding boundary area, extracts the transparency change data of the corresponding position and the relationship between the key point coordinate position, constructs the layer response matrix, integrates the change information of the corresponding layer area, and generates an interactive frame display adjustment record table.

[0024] The layer arrangement and position mapping table includes gradient direction consistency, pixel frequency distribution, and overlay order calibration. The boundary region transparency control map includes boundary gradient features, brightness difference values, pixel density distribution, overlay priority calibration, and transparency adjustment sequence. The symmetry axis keypoint offset change sequence includes skeletal keypoint frame sequence, body symmetry axis benchmark, and keypoint difference time series. The user abnormal body posture change segment set includes offset amplitude fluctuation range, stability benchmark segment, and keypoint trajectory data. The interactive frame segment display adjustment record table includes layer range positioning data, transparency coordinate relationship matrix, and display adjustment integration record.

[0025] Please see Figure 2 The layer structure blending module includes: The pixel matrix extraction submodule collects the pixel matrix of multi-resolution layers in the blended display area, filters the pixels in the overlapping area, divides the area according to the layer number, extracts the grayscale information and coordinate position of the pixels in the area, constructs a unified coordinate index structure, and obtains the pixel matrix value of the layer overlay. First, the spatial range of the blending area needs to be defined. This can be done using the system's default display area size, such as 1920×1080 pixels, or it can be limited according to the region of interest defined in the scene. Layers may contain different resolutions, for example, layer A is 800×600 pixels and layer B is 1024×768 pixels. Before mapping the coordinates of each layer, they need to be unified to the base size of the blending area and a resampling operation is performed. This is usually done using bilinear interpolation to insert the layer pixels into a standard size matrix. After unifying the size, an index relationship is built based on the layer numbers. For example, layer A is numbered 1 and layer B is numbered 2. The system binds each pixel to its layer number and establishes a coordinate mapping model by calculating the relative positional relationship between the layer and the blending coordinate system, recording the position of each pixel in the original layer and its corresponding position in the blending area. When filtering overlapping pixels, the system extracts a set of pixels with overlapping areas in the blending coordinate system from all layers based on the established mapping relationship. For example, if the overlapping area is 500 to 700 pixels horizontally and 300 to 500 pixels vertically, the pixel coordinates under all layer numbers within this range will be extracted and aligned. Subsequently, the grayscale values ​​under these coordinates are extracted. If an 8-bit grayscale image is used, the grayscale value of each pixel is in the range of 0 to 255. For example, when the coordinates are 520×350, the grayscale value of a pixel in layer 1 is 128, and that in layer 2 is 143. These grayscale values, along with their respective layer numbers and coordinates, are recorded uniformly. The generated pixel matrix data structure can be represented as a triplet of layer number, pixel position, and corresponding grayscale value, thereby realizing the construction of a joint grayscale matrix for the overlapping areas between layers.

[0026] The gradient direction evaluation submodule extracts the gradient direction of pixels in the overlapping area based on the pixel matrix values ​​of the layer overlay, obtains the direction data of adjacent pixels, judges the consistency between the current pixel direction and the neighborhood direction, and obtains the direction consistency matching degree by combining the matching results. First, the gradient direction of each pixel needs to be calculated. This process depends on the grayscale change direction of its surrounding pixels. This is done by obtaining the horizontal and vertical grayscale differences at each pixel location. For example, if the left and right neighbors of the current pixel have grayscale values ​​of 150 and 100 respectively, and the top and bottom neighbors have grayscale values ​​of 130 and 90 respectively, then the horizontal change is 50 and the vertical change is 40. The gradient direction is the angle formed by these horizontal and vertical changes. This angle is recorded as the current pixel's direction information. Then, within a 3×3 area centered on the current pixel, the directions of the other 8 pixels are obtained, and the difference between the current pixel's direction and its surrounding directions is compared. If the difference between the direction angle of a certain pixel and the current pixel's direction angle is less than a set angle threshold, such as 20 degrees, the direction is considered consistent. The number of neighboring pixels that meet this consistency condition is counted. If there are 5 pixels, the current pixel's direction consistency is 5 units. This value is then divided by the total number of neighboring pixels (8 units) to obtain the direction consistency matching degree, for example, 62.5%. This matching degree ultimately constitutes the direction consistency matrix of the entire overlapping area, containing the direction consistency ratio of each pixel, which is used for subsequent layer sorting calculations.

[0027] The layer sorting and mapping submodule counts the pixel frequency of layers in the overlapping area based on the direction consistency matching degree, combines the direction matching and frequency results for joint sorting, marks the arrangement order of layers in the blended display area, and generates a layer arrangement position mapping table. First, the number of pixels in each layer whose directional consistency matches a preset threshold is counted. For example, if the threshold is set to 60%, then each pixel with a directional consistency exceeding 60% is considered a valid match. In the overlapping area, the number of pixels in each layer that meet this condition is counted, and then divided by the total number of pixels in the overlapping area to obtain the directional matching frequency. Assuming that layer 1 has 1000 pixels in the overlapping area, and 750 pixels match the directional consistency, the matching frequency is 75%. Then, the average directional consistency of all pixels in the overlapping area of ​​each layer is combined, for example, 73%. A weighted calculation method is used to merge the two indicators for scoring. If directional consistency is weighted at 70% and matching frequency at 30%, then layer 1's weighted score is 0.7 multiplied by 73% plus 0.3 multiplied by 75%, resulting in 73.6%. This scoring process is performed on each layer, ultimately forming a weighted score list. The layers are then sorted according to their scores, and the sorted results are used to generate a layer mapping table, representing the order and priority of each layer in the merged display area.

[0028] Please see Figure 2 The pixel boundary scheduling module includes: The gradient extraction submodule extracts the horizontal and vertical brightness values ​​of the pixel positions at the layer boundary based on the layer order information in the layer arrangement position mapping table. Based on the brightness changes and pixel coordinate positions of adjacent areas, it analyzes the boundary direction features and brightness differences to generate boundary gradient direction feature values. First, a structured list containing layer numbers and their hierarchical relationships within the overall image is established. This list records the relative order of all layers and is used for priority determination in subsequent boundary processing. During the extraction of layer boundary pixel positions, grayscale processing is typically used to extract the image's brightness values, which can be achieved by weighted summation of the red, green, and blue channels. Taking an image size of 1024×768 as an example, the brightness value changes at the layer boundaries are obtained by scanning in both horizontal and vertical directions. The extraction of boundary brightness changes can be achieved by using the difference in brightness values ​​between pixels on both sides. A 3×3 neighborhood sliding window is used to calculate the brightness difference between the center pixel and its horizontally and vertically adjacent pixels, thus obtaining the horizontal and vertical trends. The direction angle is determined by the magnitude of the brightness change values. For example, if the horizontal brightness difference is significantly higher than the vertical difference, the boundary is considered a horizontal feature; conversely, if the difference is lower, it is considered a vertical feature. This type of judgment is categorized by setting a threshold for the difference ratio. In practice, the direction can be divided into 8 intervals, each covering 22.5°, corresponding to typical main directions in 360° space. By recording the direction of each boundary pixel and the corresponding brightness change, a layer boundary direction feature set is constructed. Each item in this set corresponds to the location, direction category, and brightness difference of a boundary pixel. Taking the edge of a region in a layer as an example, if its brightness decreases from 180 to 80, corresponding to a direction of 45°, then the direction is recorded as the second interval, and the brightness change is 100. After being labeled with layer order information, it is stored in the feature list to provide a reference for subsequent judgment.

[0029] The density determination submodule calls the boundary gradient direction feature value to obtain the pixel distribution of a local area in the layer. Based on the direction feature distribution and the degree of pixel aggregation, it determines the boundary density state within the area and generates a local pixel density distribution coefficient. The specific formula for calculating the spatial distribution dispersion parameter is as follows: ; Where F represents the spatial dispersion parameter for computation. This represents the intensity value of the pixel in the i-th row and j-th column of the local region. This represents the absolute value of the horizontal coordinate difference between the pixel and the center point of the region. This represents the absolute value of the vertical coordinate difference between the pixel and the center point of the region. The gradient direction angle of the pixel (in radians) is represented by , k represents the horizontal feature compensation coefficient (0.5≤k≤1.2), m represents the vertical aggregation correction coefficient (0.3≤m≤0.9), N represents the total number of pixels in the local region, n represents the number of rows in the region, and b represents the number of columns in the region. Taking a 5×5 pixel area as an example, the center point coordinates are (3,3), the total number of pixels in the area is N=25, the number of rows is n=5, and the number of columns is b=5; Pixel intensity value ρ 22 The value obtained through grayscale histogram detection is 128 (8-bit grayscale range 0-255), Δx 22 =|2-3|=1,Δy 22 =|2-3|=1, gradient direction angle θ 22 The horizontal gradient Gx = 15 and the vertical gradient Gy = 30 are calculated using the Sobel operator. θ = 15. 22 =arctan(30 / 15) = 1.107 radians; The horizontal feature compensation coefficient k = 0.8 (based on the ratio of the horizontal gradient magnitude to the total gradient magnitude, 0.8 = 15 / (15 + 30 × 0.5)), and the vertical clustering correction coefficient m = 0.6 (based on the ratio of the vertical gradient magnitude to the total gradient magnitude, 0.6 = 30 × 0.5 / (15 + 30 × 0.5)). Numerator: Δx 22 ^2+Δy 22 ^2=1 2 +1 2 =2; Denominator term: ; Fractional term: 2 / 1.944≈1.029; Absolute value of strength difference: |128-1.029|=126.971; Single pixel contribution value: 126.971; Summing across the entire region: Assuming the average contribution of other pixels is 120, the total cumulative value ≈ 120 × 25 = 3000 Dispersion parameter F: 3000 / 25 = 120; The results show that the spatial distribution dispersion parameter F=120 reflects the degree of deviation between the current region pixel intensity and the spatial location weighted distribution. When the F value exceeds the threshold of 100, it is judged as a boundary dense state. The dispersion parameter is negatively correlated with the local pixel density distribution coefficient. For every 20 units increase in the F value, the density coefficient decreases by 0.1.

[0030] The transparency sorting submodule determines the priority level and overlay rules of boundary pixels based on the local pixel density distribution coefficient and the boundary gradient direction feature value, combined with the layer arrangement order and pixel distribution state, and generates a transparency control map of the boundary area. The transparency ranking submodule determines the priority level of boundary pixels in each layer based on the extracted local pixel density distribution coefficients and boundary gradient direction feature values, according to the layer arrangement order. During processing, the layer arrangement order is first converted into layer weights, with the lowest layer weight set to 1 and the highest layer weight determined by the total number of layers (e.g., the topmost layer in a 5-layer array has a weight of 5). Next, at each boundary pixel location, the brightness change and density value corresponding to that pixel are obtained by combining its layer, orientation, and region density information. These two values ​​are multiplied to obtain the boundary salience. This value is then multiplied by the layer weight to obtain the transparency priority level. If multiple layers contain boundary pixels at the same pixel location, their respective transparency priority levels are compared, and the one with the highest priority is selected as the primary display content at that location. For example, if layer 3 has a vertical orientation, a brightness difference of 70, a region density of 180, and a corresponding layer weight of 3, then the salience is 70 multiplied by 180, which equals 12600, and the transparency level is 3 multiplied by 12600, which equals 37800. If the salience of adjacent layer 4 at this location is 14000 and its layer weight is 4, then its transparency level is 56000, which is higher. Therefore, the content of layer 4 is used at this location. Finally, the transparency levels at all pixel locations are normalized so that the transparency values ​​are distributed within the range of 0 to 1. Based on this result, the transparency parameters during layer blending are adjusted to form the final transparency control map. This map is used to represent the display priority and blending ratio of different layer contents at various locations.

[0031] Please see Figure 2 The user posture recognition module includes: The transparency region identification submodule obtains the transparency control map of the boundary region, extracts continuously changing regions based on the gray value distribution in the map, identifies the time frame number range corresponding to the transparency partition, filters the frame segments with significant transparency, and generates transparent frame interval segments. In the transparency region recognition submodule, the image processing tool is first called to perform grayscale conversion on the transparency control map to obtain the grayscale distribution of the image pixels. At each pixel, the gradient change of each local region in the image is calculated by comparing the grayscale difference of adjacent pixels. By setting the recognition threshold value of grayscale change fluctuation to 20, the boundary change segment with continuous grayscale difference can be identified. The distribution range of these regions in the horizontal axis of the image is recorded, and the pixel index on the horizontal axis is converted into a time frame number. For example, if the image width is 1280 pixels, then the 500th column represents the 500th frame. On this basis, the time frame segment corresponding to the continuous boundary change interval is extracted, the overall grayscale value of each frame is further extracted, and the grayscale change amplitude between adjacent frames is analyzed. If the grayscale difference of a frame compared with the previous frame exceeds 10%, it is marked as a frame with significant grayscale change. These frames are merged into a continuous time period to construct a transparent frame interval segment with significant transparency.

[0032] The skeleton sequence extraction submodule calls the user skeleton key point information in the corresponding frame according to the transparent frame interval, extracts the key point coordinate set of each frame, locates the shoulder and hip positions, constructs the axis of symmetry direction, divides the key point position distribution according to the direction, and generates the axis of symmetry reference coordinate sequence. The skeleton sequence extraction submodule needs to call the pose estimation model frame by frame according to the transparent frame intervals to extract the two-dimensional coordinate information of human key points in the image, mainly including key positions such as the left shoulder, right shoulder, left hip, and right hip. First, the left and right shoulders are connected and their midpoints are taken. Then, the left and right hips are connected and their midpoints are taken, forming a vector direction from the midpoint of the shoulder to the midpoint of the hip, which serves as a reference for the axis of symmetry. The spatial relative position of all key points in each frame with respect to this axis of symmetry is calculated. The vertical distance between the key point and the axis of symmetry is determined by judging the cross product of the direction vectors. Located on the left or right side of the axis of symmetry, the distribution of skeletal key points is classified. Then, the frame sequence is traversed sequentially to record the number and position changes of key points on both sides of the axis of symmetry in each frame, and a reference coordinate sequence of the axis of symmetry is constructed for subsequent offset analysis. For example, in a frame, the coordinates of the left shoulder are (220, 150), the right shoulder is (380, 150), the left hip is (230, 300), and the right hip is (370, 300). Then the midpoints of the shoulder and hip are (300, 150) and (300, 300), respectively, and the corresponding direction axis is vertically downward.

[0033] The symmetry offset calculation submodule calls the symmetry axis reference coordinate sequence, identifies the distribution differences of key points on both sides of the symmetry axis, calculates and fuses the geometric correlation, sorts out the changing trend of differences in continuous frames, extracts the difference fluctuation features of key positions, and generates a dynamic change sequence of key point offsets on the symmetry axis. The specific formula for calculating the fusion geometric correlation degree is as follows: ; in, Represents the geometric correlation correction coefficient. Representing the Left distance measurement of each key point (Euclidean distance from the key point to the axis of symmetry). Representing the Distance measurement to the right of each key point. This represents the radian value between the left measuring point and the baseline of the axis of symmetry. This represents the radian value between the right measuring point and the baseline of the axis of symmetry. Representing the The dynamic weights of each key point (calculated jointly by the radius of curvature and motion acceleration of the key point) Represents the spatial distribution dispersion (take the arithmetic root of the sum of squares of the weights of all key points). Represents the number of key points on one side; Euclidean distance: , ; Coordinate data was collected through a motion capture system, and the coordinates of the symmetry axis reference point are: The key point of the first group is the left side sitting position. Right coordinates Calculated , ; Radius value: , ; Substituting the coordinates, we get , ; Dynamic weights: radius of curvature The acceleration of motion is calculated using the three-point method. Calculated by the displacement difference between adjacent frames First key point: radius of curvature (Calculation of three-point coordinates), acceleration (Displacement difference 0.13m / 0.1s) 2 ),have to ; Spatial distribution dispersion: ,when When the group of key points has weights of 4.06, 3.92, and 3.75 respectively, the calculated results are as follows: ; Geometric difference terms: ; After squaring, we get Multiply by weight ; Overall offset item: ; Cubic root synthesis: ; Parameter values ​​are based on The radius of curvature was calculated using the three-point method: the radius of the circle was calculated by taking the coordinates of three adjacent frames, and the measured data were in the range of 5-10 mm. Acceleration calculations are based on a 30fps sampling rate, with displacement difference accuracy reaching 0.01m, and measured data in the range of 0.5-2.0m / s². The weighting coefficients of 0.4 and 0.6 are assigned according to the importance of the features, and the contribution ratio of curvature and acceleration is determined by principal component analysis. Cube root operations reduce the impact of dimensional differences, and geometric difference terms and overall offset terms are combined after their dimensions are unified to length units.

[0034] Numerical result correlation Calculation results This is a dimensionless geometric correlation correction coefficient. A value greater than 1 indicates that the key point distribution exhibits asymmetric characteristics. This coefficient will serve as the amplitude modulation factor of the dynamically changing sequence and, together with the D value of subsequent frames, will form a time series for trend analysis.

[0035] Please see Figure 2 The dynamic behavior determination module includes: The offset sequence detection submodule obtains the time and position information recorded in the offset change sequence of key points on the axis of symmetry, identifies the changes in the coordinates of key points in adjacent frames, sorts out the position change amplitude of key points in the sequence, generates the position change trend in time order, and obtains the inter-frame offset trend value of key points. First, the coordinate values ​​of key points on both sides of the axis of symmetry are collected, for example, the pixel positions of the left and right shoulders are recorded respectively, and the midpoint between the two points is calculated as the reference point of the axis of symmetry. Then, the coordinates of the same key points are repeatedly extracted in adjacent frames, the reference point position is calculated again, and the coordinate difference with the reference point position of the previous frame is calculated. The difference and time interval are recorded to form time position offset information. For example, in a video of 30 frames per second, from frame 10 to frame 11, the reference point of the axis of symmetry is detected to move 3 pixels horizontally and 2 pixels vertically, and the time is recorded as 33 milliseconds. A set of offset records is generated. After continuous acquisition, an offset sequence arranged by time is formed. By taking the square root of the sum of the squares of the coordinate differences in each set of records, the movement amplitude of the key point between each frame is obtained, and a complete sequence of inter-frame position change amplitude is compiled. For example, the offset amplitudes between frames 10 and 20 are 2.2, 2.5, 2.8, 3.0, 2.6, 2.9, 3.1, 3.4, 3.3, and 3.0 respectively. This sequence can be used to further generate trend trajectories to form a sequence of offset change trend values ​​between consecutive frames.

[0036] The amplitude fluctuation analysis submodule extracts the offset amplitude sequence of a time period based on the inter-frame offset trend value of key points, identifies the changes in offset amplitude between consecutive frames, determines the correspondence between the fluctuation range and the offset amplitude stability benchmark value, screens out the time period of abnormal fluctuation, and obtains the time interval of abnormal offset fluctuation. The amplitude fluctuation analysis submodule, based on the offset trend value sequence generated in the previous stage, extracts the offset amplitude set using a fixed time interval and sets the sliding window size, such as 10 frames per time period. It calculates the difference between the maximum and minimum offset amplitude within each segment as the fluctuation amplitude for that segment. For example, if the offset amplitude value is 2.1 to 2.9 from frame 30 to frame 39, the fluctuation amplitude is 0.8 pixels. The fluctuation amplitudes of all time periods are summarized and compared with a preset stability benchmark value. If the fluctuation amplitude of a certain segment exceeds the benchmark value, that time period is marked as abnormal. The stability benchmark value can be obtained through statistical analysis of different video data under normal conditions. For example, analysis under static conditions shows that offset fluctuations are mostly concentrated between 0.3 and 0.7, and the average value of 0.5 is taken as the reference benchmark. If the fluctuation amplitude of a certain time period is 0.9, exceeding the benchmark value, it is marked as an abnormal fluctuation segment, and the start and end time period numbers are recorded. For example, if the fluctuation value is 1.1 from frame 40 to frame 49, it is marked as an abnormal time period.

[0037] The abnormal segment extraction submodule retrieves the spatial trajectory data of key points within the corresponding time period based on the abnormal time interval of offset fluctuation, integrates the continuous position information of key points in the abnormal segment, constructs a position trajectory sequence of multiple key points, and obtains a set of abnormal body posture change segments of the user. The abnormal segment extraction submodule retrieves the coordinate trajectory data of key points within the marked abnormal offset fluctuation time periods. It continuously arranges the coordinate values ​​of key points in each frame to form a continuous trajectory data set. For example, from frame 60 to frame 80, the coordinates of the left knee key point are (340, 720), (342, 724), (345, 726)..., and the coordinates of the right hip point are (500, 820), (503, 825), (506, 828)... These key point trajectory data sets are synchronously combined frame by frame to generate a multi-key point trajectory set, forming a spatially continuous trajectory sequence. This sequence reflects the movement of various body parts within the time period. Areas with prominent trajectory changes within abnormal time periods, such as large changes in trajectory curvature or frequent changes in movement direction, can be classified as abnormal body posture segments. This process is repeated for all marked time periods, ultimately forming a set of abnormal body posture change segments. Each segment in the set contains elements such as the corresponding time interval, key point trajectory group, and displacement change characteristics.

[0038] Please see Figure 2 The interactive layer response module includes: The frame segment calibration submodule obtains the marked frame segment range from the set of abnormal body posture change segments of the user, extracts the corresponding sequence frame image number, counts the number of image frames and constructs a time interval list, divides the non-continuous segments in the frame segment, establishes the corresponding structure of number index, and obtains the abnormal frame segment number index group. First, image sequences of each frame are extracted from the video source. Then, using a pose recognition model, the positional changes of skeletal keypoints are analyzed frame by frame. By continuously comparing the changes in joint angles and movement speeds in adjacent frames, frames with significant changes are identified as anomalies. At a sampling rate of 30 frames per second, if any joint angle change exceeds 15 degrees for three consecutive frames, or the displacement speed of a keypoint exceeds 0.2 meters per second, that frame is designated as the starting frame of an anomaly. The ending frame of an anomaly is the frame in which the angle or speed returns to the normal range in subsequent frames. The image number corresponding to each anomaly segment is its start and end frame numbers in the sequence, and the number of images is the difference between the start and end numbers plus 1. Converting the image numbers to a timestamp list requires calculation based on the frame rate. For example, if the starting frame number is 105, the corresponding time is 3.5 seconds; if the ending frame number is 135, the corresponding time is 4.5 seconds. If the interval between frame numbers exceeds two frames, for example, if numbers 105, 106, and 110 appear, then an intermediate frame is considered missing, and the segment is divided into two discontinuous segments. Finally, all abnormal segments are recorded using a dictionary structure, with each group of numbers forming an index. For example, the first group has start and end numbers from 105 to 120, and the second group has numbers from 123 to 135. After establishing the corresponding structure, all frame segment numbers are summarized to form an abnormal index set.

[0039] The region extraction submodule calls the frame sequence number in the abnormal frame segment number index group to locate the layer region index, obtains the position coordinate value of the boundary region in the corresponding frame, determines the contour closure feature, extracts the transparency parameter value in the region, calculates the combination interval of change frequency and amplitude, and obtains the layer transparency fluctuation interval group. The region extraction submodule retrieves image data from each frame of the image sequence based on the frame segment number and quickly locates the target region using the region indexes already marked in the layer structure. After each frame is loaded, a segmentation model is used to identify the layer regions, converting the boundary positions of the regions into the coordinates of the four corner points of a rectangular border. By comparing the connectivity of the first and last pixels of the region boundary and the contour area, it is determined whether the region is closed. For example, if the distance between the starting and ending pixels of the boundary is less than 3 pixels and the region area is non-zero, it is considered a closed contour. The transparency parameter is taken from the image's alpha channel, and the transparency values ​​of all pixels within the contour region are averaged to obtain the average transparency of the layer region in each frame. A sliding window is set in the time series, for example, every 3 frames is a group. The transparency changes between two consecutive groups are compared. If the difference is greater than 10%, a fluctuation event is recorded, and the maximum and minimum transparency values ​​are calculated as the fluctuation amplitude. The frequency of all fluctuation events is counted throughout the entire frame segment, and the transparency change range is categorized. For example, if the transparency fluctuates from 45% to 72% in a certain segment, with 4 fluctuations, it is recorded as the transparency fluctuation interval group for that segment, which is convenient for subsequent processing.

[0040] The relation mapping submodule matches the set of key point spatial coordinates based on the layer transparency fluctuation range group, extracts the overlap and offset distance between the points and the layer boundary, constructs the correspondence sequence between key points and transparency, judges the consistency of direction and the difference of fluctuation, establishes the response matrix, and obtains the interactive frame display adjustment record table. The relationship mapping submodule uses transparency fluctuation ranges and keypoint coordinate information for matching. In each frame, a pose recognition model is used to obtain several keypoint coordinates, which are stored by number, for example, P1 to Pn. Each keypoint coordinate is compared with the layer boundary to calculate whether it falls within the contour area. The judgment criterion is the ratio of the number of pixels overlapping within the layer to the total number of pixels in the area; if the overlap ratio is 17.5%, partial overlap is considered to exist. The offset distance is calculated by taking the square root of the sum of the squares of the horizontal and vertical distances between the keypoint position and the layer center point, i.e., the straight-line distance between the point and the center point. When constructing the correspondence between keypoints and transparency, the pairing information of keypoint position and transparency is recorded for each frame. The consistency between the keypoint direction and the transparency change direction is analyzed by judging whether the movement trend of the keypoint is consistent with the transparency rise and fall trend. If the two change directions are highly similar, the direction similarity can be qualitatively defined as strong consistency. Simultaneously, the amplitude of transparency value change and the amplitude of keypoint position movement are statistically analyzed in the frame sequence. The degree of fluctuation difference is obtained by comparing the standard deviation of the two changes; a larger standard deviation indicates more drastic fluctuation. Finally, the displacement data, transparency pairing data, orientation consistency, and variation differences of each key point in all frames are summarized and recorded in an image adjustment record table. Each record contains fields such as frame number, key point number, transparency value, transparency change value, key point displacement change value, orientation similarity, and fluctuation difference.

[0041] Please see Figure 3 The AI-based holographic intelligent interaction method includes the following steps: S1: Collect the pixel matrix of the different resolution layers in the blended display area, detect the gradient direction of the pixels in the overlapping area of ​​the layers, combine the directional distribution of neighboring pixels and the frequency of pixel occurrence, determine the directional consistency level between the layers, and generate a layer arrangement position mapping table. S2: Based on the layer order information in the layer arrangement position mapping table, extract the gradient direction and brightness difference features of the layer boundary region, combine the pixel density distribution in the local region, determine the superposition priority of boundary pixels in the layer fusion process, and establish a boundary region transparency control map. S3: Based on the transparency area in the boundary area transparency control map, extract the user's skeletal key point frame sequence within the corresponding time period, construct the axis of symmetry with the shoulder and hip key points, record the difference in the number of key points on both sides of the axis of symmetry in the frame, and generate the axis of symmetry key point offset change sequence. S4: Based on the temporal variation trend of the key point offset change sequence of the symmetry axis, analyze the fluctuation range of the key point offset amplitude between frames, combine the stability benchmark to delineate the frame segment interval of abnormal changes, extract the key point position trajectory within the segment, and generate a set of abnormal body posture change segments of the user. S5: Call the frame range in the set of abnormal body posture changes of the user, locate the layer area in the transparency control map of the boundary area, extract the transparency change record of the corresponding area and the key point coordinate relationship, integrate the area change information, and generate an interactive frame display adjustment record table.

[0042] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An AI-based holographic image intelligent interaction system, characterized in that: The system comprises: The layer structure fusion module collects the pixel matrix of the multi-resolution layers in the fusion display area, extracts the gradient direction features of the pixels in the overlapping area, judges the consistency degree with the neighborhood direction, combines the pixel frequency to mark the layer superposition order, and generates a layer arrangement position mapping table; The pixel boundary scheduling module extracts the gradient direction and brightness difference of the layer boundary position based on the layer arrangement position mapping table, judges the boundary pixel superposition priority, marks the transparency adjustment order, and establishes a boundary area transparency regulation map; The user body posture recognition module extracts the frame sequence of the skeleton key points in a time period based on the boundary area transparency regulation map, constructs a symmetry axis with the shoulder and hip as the reference, records the distribution difference of the key points on both sides of the symmetry axis, and generates a key point offset change sequence of the symmetry axis; The dynamic behavior judgment module analyzes the time trend and amplitude fluctuation of the key point offset change sequence of the symmetry axis, identifies the abnormal motion section combined with the stability reference, and generates a user abnormal body posture change segment set; The interactive layer response module calls the user abnormal body posture change segment set, locates the layer range in the transparency regulation map, extracts the transparency and key point coordinate relationship, and generates an interactive frame segment imaging adjustment record table. 2.The AI-based holographic image intelligent interaction system according to claim 1, characterized in that: The layer arrangement position mapping table comprises gradient direction consistency degree, pixel frequency distribution, and superposition order marking, the boundary area transparency regulation map comprises boundary gradient features, brightness difference value, pixel density distribution, superposition priority marking, and transparency adjustment sequence, the key point offset change sequence of the symmetry axis comprises frame sequence of the skeleton key points, body posture symmetry axis reference, and key point difference time sequence, the user abnormal body posture change segment set comprises offset amplitude fluctuation range, stability reference section, and key point trajectory data, and the interactive frame segment imaging adjustment record table comprises layer range positioning data, transparency coordinate relationship matrix, and imaging adjustment integration record. 3.The AI-based holographic image intelligent interaction system according to claim 1, characterized in that: The layer structure fusion module comprises: The pixel matrix extraction submodule collects the pixel matrix of the multi-resolution layers in the fusion display area, filters the pixels in the overlapping area, divides the area according to the layer number, extracts the gray information and coordinate position of the pixels in the area, constructs a unified coordinate index structure, and obtains the layer superposition pixel matrix value; The gradient direction evaluation submodule extracts the gradient direction of the pixels in the overlapping area based on the layer superposition pixel matrix value, obtains the direction data of the adjacent pixels, judges the consistency degree of the current pixel direction and the neighborhood direction, and obtains the direction consistency matching degree combined with the matching condition statistical result; The layer sorting mapping submodule counts the pixel frequency of the layers in the overlapping area according to the direction consistency matching degree, combines the direction matching and frequency results for joint sorting, marks the arrangement order of the layers in the fusion display area, and generates a layer arrangement position mapping table. 4.The AI-based holographic image intelligent interaction system according to claim 3, characterized in that: The pixel boundary scheduling module comprises: The gradient extraction submodule extracts the horizontal and vertical brightness values of the layer boundary pixel position based on the layer order information in the layer arrangement position mapping table, analyzes the boundary direction features and brightness difference according to the brightness change and pixel coordinate position of the adjacent area, and generates the boundary gradient direction feature value; The density determination sub-module calls the boundary gradient direction feature value, acquires pixel distribution of a local region in the layer, determines the boundary density state in the region according to the direction feature distribution and the pixel aggregation degree, and generates a local pixel density distribution coefficient; The transparency ranking sub-module determines the priority level and the superposition rule of the boundary pixel according to the local pixel density distribution coefficient and the boundary gradient direction feature value, combines the layer arrangement order and the pixel distribution state, and generates a boundary region transparency control graph. 5.The AI-based holographic image intelligent interaction system according to claim 4, characterized in that: The specific calculation formula of the calculation of the spatial distribution dispersion parameter is: ； wherein F represents a spatial distribution dispersion parameter, represents the intensity value of the pixel in the i-th row and j-th column of the local region, represents the absolute value of the horizontal coordinate difference between the pixel and the center point of the region, represents the absolute value of the vertical coordinate difference between the pixel and the center point of the region, represents the gradient direction angle of the pixel, k represents a horizontal direction feature compensation coefficient, m represents a vertical direction aggregation correction coefficient, N represents the total number of pixels in the local region, n represents the number of rows of the region, and b represents the number of columns of the region. 6.The AI-based holographic image intelligent interaction system according to claim 4, characterized in that: The user body state recognition module comprises: The transparency region recognition sub-module acquires the boundary region transparency control graph, extracts a continuously changing region according to the gray value distribution in the graph, identifies the time frame number range corresponding to the transparency partition, filters the frame segment with significant transparency, and generates a transparent frame interval segment; The skeleton sequence extraction sub-module calls the user skeleton key point information in the corresponding frame according to the transparent frame interval segment, extracts the key point coordinate set of each frame, locates the shoulder and hip positions, constructs a symmetric axis direction, divides the key point position distribution according to the direction, and generates a symmetric axis reference coordinate sequence; The symmetry offset calculation sub-module calls the symmetric axis reference coordinate sequence, identifies the distribution difference of the key points on both sides of the symmetric axis, calculates the fusion geometric correlation degree, arranges the change trend of the difference in the continuous frames, extracts the difference fluctuation feature of the key position, and generates a symmetric axis key point offset dynamic change sequence. 7.The AI-based holographic image intelligent interaction system according to claim 6, characterized in that: The specific calculation formula of the calculation of the fusion geometric correlation degree is: ； wherein, represents a geometric correlation correction coefficient, represents a first key point left range, represents a first key point right range, represents an arc value of the left measuring point to the symmetry axis reference line, represents an arc value of the right measuring point to the symmetry axis reference line, represents a dynamic weight of the first key point, represents a spatial distribution dispersion (taking the arithmetic root of the sum of squares of all key point weights), represents a number of unilateral key points. 8.The AI-based holographic image intelligent interaction system according to claim 6, characterized in that: The dynamic behavior determination module comprises: The offset sequence detection sub-module acquires the time position information recorded in the symmetric axis key point offset change sequence, identifies the change of the key point coordinates in the adjacent frames, arranges the position change amplitude of the key points in the sequence, generates the position change trend in time sequence, and obtains the key point inter-frame offset trend value; The amplitude fluctuation analysis sub-module extracts the offset amplitude sequence of the time period based on the key point inter-frame offset trend value, identifies the offset amplitude change between the continuous frames, judges the corresponding relationship between the fluctuation range and the offset amplitude stability reference value, filters the time period with abnormal fluctuation, and obtains the offset fluctuation abnormal time interval; The abnormal section extraction sub-module calls the spatial trajectory data of the key points in the corresponding time period according to the offset fluctuation abnormal time interval, integrates the continuous position information of the key points in the abnormal section, constructs the position trajectory sequence of multiple key points, and obtains a user abnormal body state change segment set. 9.The AI-based holographic image intelligent interaction system according to claim 8, characterized in that: The interactive layer response module comprises: The frame segment calibration sub-module acquires the frame segment range marked in the user abnormal body state change segment set, extracts the corresponding sequence frame image number, counts the image frame number and constructs a time interval list, divides the non-continuous section in the frame segment, establishes a number index corresponding structure, and obtains an abnormal frame segment number index group; The region extraction sub-module calls the frame sequence number in the abnormal frame segment number index group, locates the layer region index, acquires the boundary region position coordinate value in the corresponding frame, determines the contour closure feature, extracts the transparency parameter value in the region, calculates the change frequency and amplitude combination interval, and obtains a layer transparency fluctuation interval group; The relationship mapping submodule matches the set of key point spatial coordinate positions according to the set of layer transparency fluctuation intervals, extracts the coincidence degree and offset distance between the point position and the layer boundary, constructs the corresponding sequence of key points and transparency, judges the direction consistency and fluctuation difference, establishes a response matrix, and obtains an interactive frame segment imaging adjustment record table.

10. An AI-based holographic image intelligent interaction method, characterized in that, The AI-based holographic image intelligent interaction system according to any one of claims 1-9, comprising the following steps: S1: Collecting the pixel matrix of the difference resolution layers in the fusion display area, detecting the gradient direction of the pixels in the layer overlap area, combining the direction distribution of the neighborhood pixels and the pixel frequency information, determining the direction consistency level between the layers, and generating a layer arrangement position mapping table; S2: Based on the layer order information in the layer arrangement position mapping table, extracting the gradient direction and brightness difference characteristics of the layer boundary area, combining the pixel density distribution in the local area, determining the superposition priority level of the boundary pixels in the layer fusion process, and establishing a boundary area transparency control map; S3: According to the transparency area in the boundary area transparency control map, extracting the frame sequence of the user's skeletal key points in the corresponding time period, constructing a symmetry axis with the shoulder and hip key points, recording the difference value of the distribution number of the key points on both sides of the symmetry axis in the frame, and generating a symmetry axis key point offset change sequence; S4: Based on the time change trend in the symmetry axis key point offset change sequence, analyzing the fluctuation range of the key point offset amplitude between frames, combining the stability reference to determine the frame segment interval of abnormal changes, extracting the key point position trajectory in the section, and generating a user abnormal body posture change segment set; S5: Calling the frame segment range in the user abnormal body posture change segment set, positioning the layer area in the boundary area transparency control map, extracting the transparency change record and key point coordinate relationship of the corresponding area, integrating the area change information, and generating an interactive frame segment imaging adjustment record table.

Images