An artificial intelligence interaction system

Abstract

The application discloses an artificial intelligence interaction system and relates to the technical field of artificial intelligence. The system collects training images of sports personnel performing sports actions in a stadium through a client terminal, and a characteristic marking unit performs directional characteristic marking. A viscous interaction unit generates a virtual characteristic verification sequence based on each training image in a training characteristic image set in a viscous cycle. An image set verification module verifies the training images in the training characteristic image set based on the virtual characteristic verification sequence of the viscous cycle. On the one hand, the system avoids the deviation of the training direction of the action recognition model caused by the embedding of new images or the insertion of other images similar to the directional characteristics of the training images in the transmission process of the training characteristic image set, thereby causing the loss of model training resources. On the other hand, the system can quickly and accurately distinguish the lost training images embedded by the third party and lost in the transmission process, and retransmits the lost training images.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically to an artificial intelligence interactive system. Background Technology

[0002] Artificial intelligence, abbreviated as AI, is a new technical science that studies and develops theories, methods, technologies, and application systems to simulate, extend, and expand human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. An existing AI interactive system is applied to sports activities performed by athletes in a training field. It collects images of the sports activities performed by athletes in the training field, performs image recognition, and labels them with features. The labeled images are then uploaded to the cloud, and the powerful computing capabilities of the cloud are used to train a sports activity recognition model to improve the accuracy of the model. This method requires that the training image set be absolutely accurate. If new images are inserted or parts of the training image set are tampered with during transmission, the training direction of the activity recognition model will deviate, resulting in a waste of cloud computing power. One current method for verifying training graphs is based on the MD5 module, which (Message-DigestAlgorithm 5) is used to calculate message digests. MD5 uses a complex algorithm to calculate a 128-bit hash value from a piece of information. The generation process of MD5 is quite complex; it evolved from MD2 and MD4. The probability of two different messages yielding the same MD5 value (often called a "collision") is extremely low, about 1 / 2 to the power of 128. Therefore, MD5 can be used as a unique value, similar to a fingerprint. The training image set is obtained by performing MD5 value calculation. The cloud verifies the training image set by verifying the MD5 value. However, this verification method cannot verify the training image set based on a single image. Furthermore, it cannot accurately distinguish between images that are missing or tampered with during transmission. To address the above problems, the present invention proposes a solution; Summary of the Invention

[0003] The purpose of this invention is to provide an artificial intelligence interaction system, which aims to solve the problems of existing technologies that obtain a unique MD5 value for a training image set by performing MD5 value calculations on the training image set, and then verifying it by checking the MD5 value in the cloud. However, this verification method cannot verify the image set based on a single image in the training image set, and it cannot accurately distinguish between images that are missing or tampered with during transmission. The objective of this invention can be achieved through the following technical solutions: An artificial intelligence interactive system, comprising: The client terminal includes several client modules. Each client module is used to acquire training images of a sport in the training field and to annotate them with directional features. Each client module corresponds to a sport that can be performed in the training field. Each client module includes an image acquisition unit, a feature annotation unit, and a sticky interaction unit. The image acquisition unit acquires training images for the corresponding sports, and the feature annotation unit performs directional feature annotation on the training images for the corresponding sports and generates a directional feature-annotated image for the training images based on these annotations. The sticky interaction unit generates a training feature map set for a sticky period based on the directional feature annotation images of all training images within that sticky period, and generates a verification and traceability document for the training feature map set corresponding to the sport within that sticky period according to certain sticky generation rules, as follows: S11: Obtain the total number of pointing feature-labeled images of the training images transmitted by the feature labeling unit to the sticky interaction unit within a sticky period, and label it as p1; S12: First, select the pointing feature annotation image of the second training image received by the sticky interaction unit within the sticky period as the sticky association image of the sticky period, and label the pointing feature annotation images of the previous and next images of the sticky association image received by the sticky interaction unit within the sticky period as the first and second auxiliary images of the sticky association image of the sticky period. S13: According to the first filtering rule, filter the directional features in the sticky association image and the first auxiliary image to obtain the sorted sticky association features D1, D2, ..., Da in the sticky association image and the first auxiliary image, as follows: S131: Obtain the directional features that are labeled in both the sticky association image and the first auxiliary image, and label them as A1, A2, ..., Aa, where a≥1; S132: Obtain all pixels constituting the directional feature A1 in the sticky association image and represent the position of the pixels using number pairs, labeled as B1(Bc1, Bd1), B2(Bc2, Bd2), ..., Bb(Bcb, Bdb). When representing the position of the pixels using number pairs, the position before the comma in the parentheses indicates the column number of the pixel in the sticky association image, and the position after the comma indicates the row number of the pixel in the sticky association image. S133: Obtain the left edge base point value Bdmin, right edge base point value Bdmax, upper edge base point value Bcmin, and lower edge base point value Bcmax that constitute the directional feature A1 in the sticky correlation image, wherein Bdmax and Bdmin are the maximum and minimum values ​​among Bd1, Bd2, ..., Bdb, and Bcmax and Bcmin are the maximum and minimum values ​​among Bc1, Bc2, ..., Bcb, respectively; Using formula Calculate and obtain the edge feature base point value C1 of the directional feature A1; S134: Calculate and obtain the edge feature base point values ​​C1, C2, ..., Aa of the directional features A1, A2, ..., Aa in sequence according to S132 to S133, and re-label the directional features corresponding to the edge feature base point values ​​as sticky association features in descending order of edge feature base point values, and label them as D1, D2, ..., Da; S14: Utilize the formula Calculate the vertical simulation basis F1 of the sticky correlation feature D1 in the acquired sticky correlation image and the first auxiliary image, where E1 is the edge feature base point value of the sticky correlation feature D1, and a1 and a2 are preset first and second vertical simulation factors; S15: Calculate and obtain the vertical pseudobase values ​​F1, F2, ..., Fa of the sticky correlation features D1, D2, ..., Da in the sticky correlation image and the first auxiliary image according to S14; S16: According to the second screening and calibration rules, screen and calibrate the first viscous gradient pixel, the second viscous gradient pixel, ..., the j-th viscous gradient pixel that constitutes the viscous correlation feature D1 in the viscous correlation image; S17: According to S16, filter and label the first viscous gradient pixel, the second viscous gradient pixel, ..., the j-th viscous gradient pixel in the first auxiliary image that constitutes the viscous correlation feature D1, where j>1; S18: Compare h and J. If h > j: S181: Obtain the total number of first viscous gradient pixels constituting viscous correlation feature D1 in the viscous correlation image and the first auxiliary image respectively, and recalibrate them as K1 and L1; Using formula Calculate the vertical fitting value M1 of the first viscous gradient pixel that constitutes the viscous correlation feature D1 in the viscous correlation image and the first auxiliary image; S182: Calculate the vertical fitting values ​​M1, M2, ..., Mj of the first, second, ..., j sticky gradient pixels constituting sticky correlation feature D1 in the sticky correlation image and the first auxiliary image according to S181, and calculate their mean values ​​using the summation and averaging formula. Re-label the mean values ​​of vertical fitting values ​​M1, M2, ..., Mj as the vertical fitting gradient value of sticky correlation feature D1, and label it as N1; S19: Conversely, according to S18, the vertical pseudo-gradient value of the sticky association feature D1 is calculated based on the first, second, ..., h sticky gradient pixels that constitute the sticky association feature D1 in the sticky association image and the first auxiliary image, and is marked as N1; S110: Calculate and obtain the vertical pseudo-gradient values ​​N1, N2, ..., Na of the sticky correlation features D1, D2, ..., Da in the sticky correlation image and the first auxiliary image according to S13 to S19, and aggregate the corresponding vertical pseudo-gradient values ​​with "|" to generate the pseudo-feature verification sequence of the sticky correlation image and the first auxiliary image. S111: Calculate and obtain the pseudo-feature verification sequence of the sticky correlation image and the second auxiliary image according to S12 to S110; S112: Select the pointing feature annotation images of the second, third, ..., p1-1 training images received by the sticky interaction unit within the sticky period in sequence as the sticky association images of the sticky period, and calculate and obtain the pointing feature annotation images of the second, third, ..., p1-1 training images and the pseudo feature verification sequence of the next image according to S11 to S111 respectively. The sticky interaction unit generates a verification traceability document for the training feature map set within the sticky period based on the obtained p1-1 pseudo-feature verification sequences. The model training terminal includes a graph verification module and a model training module. The graph verification module verifies the training feature graph of a sticky period and its corresponding verification traceability document. The model training module stores a motion recognition model that can recognize all sports movements in the training field. The model training module inputs the verified training feature graph of a sticky period into the motion recognition model for training and optimization of the motion recognition model parameters.

[0004] Furthermore, one of the viscosity cycles lasts for a time P1, where P1 is a preset cycle duration.

[0005] Furthermore, the feature annotation unit performs directional feature annotation by using a bounding box method to select the directional features in the training image. It should be noted that all training images acquired by the image acquisition unit have the same image size, which is 2400*1920px.

[0006] Furthermore, the directional features refer to the limbs of an athlete that can be identified by sports movements, including the head, left shoulder, right shoulder, left upper arm, right upper arm, left forearm, right forearm, left hand, right hand, chest, waist, left thigh, right thigh, left calf, right calf, left foot, and right foot.

[0007] Furthermore, the training image for a sport refers to an image of an athlete performing the sport in a training field, and the training image includes the full body of the athlete performing the sport.

[0008] Furthermore, the specific verification rules for the graph set verification module to verify the training feature graph set and its corresponding verification traceability document for a sticky period are as follows: S21: Obtain the first pseudo-feature sequence in the verification traceability document corresponding to the training feature map set within a sticky period, and extract the values ​​in the pseudo-feature sequence according to "|", and obtain the pseudo-feature values ​​V1, V2, ..., Vv corresponding to the pseudo-feature sequence, where v≥1; S22: Obtain the second pointing feature annotation image in the training feature map set within the sticky period and select it as the sticky associated image. Label the first pointing feature annotation image of the sticky associated image within the sticky period as the first subordinate image of the sticky associated image within the sticky period. Label the second pointing feature annotation image of the sticky associated image within the sticky period as the second subordinate image of the sticky associated image within the sticky period. S23: Calculate and obtain the sticky correlation features and their corresponding edge feature base point values ​​of the sticky correlation image and its first subordinate image according to S13, and sort the sticky correlation features in descending order of edge feature base point values. Mark the sorted sticky correlation features as O1, O2, ..., Ov, v≥1. Calculate and obtain the sticky correlation features and their corresponding edge feature base point values ​​of the sticky correlation image and its second subordinate image according to S13, and sort the sticky correlation features in descending order of edge feature base point values. Mark the sorted sticky correlation features as Q1, Q2, ..., Qv. S24: Create the first, second, and third risk weight variables R1, with the initial values ​​of R1, R2, and R3 being 0; S25: Calculate and obtain the vertical pseudogradient values ​​PO1, PO2, ..., POv of the sticky correlation features O1, O2, ..., Ov according to S14 to S110, and calculate and obtain the vertical pseudogradient values ​​PQ1, PQ2, ..., PQv of the sticky correlation features Q1, Q2, ..., Qv; S26: Compare PO1, PQ1, and V1 in sequence: S261: If PO1≠V1, then increment the value of the first risk weight variable R1 by 1; if PQ1≠V1, then increment the value of the second risk weight variable R2 by 1. S262: If PO1≠V1 and if PQ1≠V1, then increment the value of the third risk weight variable R3 by 1. S27: Following S26, compare the vertical pseudogradient values ​​PO1, PO2, ..., POv and the vertical pseudogradient values ​​PQ1, PQ2, ..., PQv with V1, V2, ..., Vv in sequence, and obtain the final values ​​of the first risk weight variables R1, R2 and R3, and label them as PR1, PR2 and PR3 in sequence; S28: If PR1>P4, then the first auxiliary image is determined to be missing, and its corresponding position in the training feature map set of the sticky period is recorded; otherwise, the first auxiliary image is determined to be tampered with, and the first auxiliary image is removed from the training feature map set of the sticky period. If PR2 > P4, then the second auxiliary image is determined to be missing, and its corresponding position in the training feature map set of the sticky period is recorded; otherwise, the second auxiliary image is determined to be tampered with, and the second auxiliary image is removed from the training feature map set of the sticky period. If PR3 > P4, it is determined that the sticky association image is missing, and its corresponding position in the training feature map set of the sticky period is recorded. Otherwise, it is determined that the sticky association image has been tampered with, and the sticky association image is removed from the training feature map set of the sticky period. P4 is a preset threshold. S29: Following S21 to S28, determine the positions of all missing directional feature annotation images in the training feature map set for the viscous period.

[0009] The beneficial effects of this invention are: This invention collects training images of athletes performing various sports movements in a stadium by setting up a client terminal. The feature annotation unit annotates the limbs performing the corresponding sports movements in the corresponding training image set with directional features. The sticky interaction unit generates a simulated feature verification sequence based on each training image with directional feature annotation in the training feature image set within a sticky period, thereby enhancing the correlation between images before and after in the training feature image set. This invention verifies each training image labeled with directional features in the training feature map set by setting up a map set verification module. This module verifies the training image with directional features in the training feature map set within a sticky period using a simulated feature verification sequence. On the one hand, this avoids the training direction of the action recognition model deviating from its original direction and causing loss of model training resources due to the insertion of new or similar images with directional features by third parties during the transmission of the training feature map set. On the other hand, it can quickly and accurately distinguish the training images that have been embedded by third parties or lost during transmission, and retransmit the lost training images, thereby improving the quality of model training. Attached Figure Description

[0010] The invention will now be further described with reference to the accompanying drawings.

[0011] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] like Figure 1 As shown, an artificial intelligence interactive system includes a client terminal and a model training terminal; The client terminal includes several client modules. The client modules are used to collect training images of a sport in the training field and to annotate them with directional features. One client module corresponds to a sport that can be performed in the training field. The training image of a sport refers to an image of an athlete performing the sport in the training field. The training image includes the whole body of the athlete performing the sport. The client module includes an image acquisition unit, a feature annotation unit, and a sticky interaction unit. The image acquisition unit acquires training images of the corresponding sports and transmits them to the feature annotation unit. After receiving the training image of the corresponding sport transmitted by the image acquisition unit, the feature annotation unit performs directional feature annotation on it and generates a directional feature annotation image of the training image based on it. The feature annotation unit then transmits the directional feature annotation image of the training image to the sticky interaction unit. The directional features refer to the limbs of an athlete that can be identified by sports movements, including the head, left shoulder, right shoulder, left upper arm, right upper arm, left forearm, right forearm, left hand, right hand, chest, waist, left thigh, right thigh, left calf, right calf, left foot, and right foot. In this embodiment, the feature annotation unit performs directional feature annotation by using a bounding box method to select the directional features in the training image. It should be noted that the images acquired by the image acquisition unit have the same size. In this embodiment, the image size of the training image is 2400*1920px. The sticky interaction unit receives the pointing feature annotation image of the training image transmitted by the feature annotation unit and stores it. The sticky interaction unit generates a training feature map set for a sticky period based on the pointing feature annotation images of all training images within a sticky period, and generates a verification and traceability document of the training feature map set corresponding to the sport within the sticky period according to certain sticky generation rules. In this embodiment, a sticky period lasts for a time P1, where P1 is a preset period duration. The specific rules for generating verification and traceability documents for the training feature map set corresponding to this sport within a sticky period are as follows: S11: Obtain the total number of pointing feature-labeled images of the training images transmitted by the feature labeling unit to the sticky interaction unit within the sticky period, and mark it as p1; S12: First, select the pointing feature annotation image of the second training image received by the sticky interaction unit within the sticky period as the sticky association image of the sticky period; The pointing feature annotation images of the previous and next images of the viscous associated image received by the viscous interaction unit within the viscous cycle are designated as the first and second auxiliary images of the viscous associated image of the viscous cycle. S13: According to the first filtering rule, filter the directional features in the sticky association image and the first auxiliary image to obtain the sorted sticky association features D1, D2, ..., Da in the sticky association image and the first auxiliary image, as follows: S131: Obtain the directional features that are labeled in both the sticky association image and the first auxiliary image, and label them as A1, A2, ..., Aa, where a≥1; S132: Obtain all pixels constituting the directional feature A1 in the sticky association image and represent the position of the pixels using number pairs, labeled as B1(Bc1, Bd1), B2(Bc2, Bd2), ..., Bb(Bcb, Bdb). When representing the position of the pixels using number pairs, the position before the comma in the parentheses indicates the column number of the pixel in the sticky association image, and the position after the comma indicates the row number of the pixel in the sticky association image. For example, pixel B1 is represented by the ordered pair (Bc1, Bd1). In this case, Bc1 before the comma in the parentheses indicates that pixel B1 is located in column Bc1 in the sticky association image, and Bd1 after the comma indicates that pixel B1 is located in row Bd1 in the sticky association image. S133: Obtain the left edge base point value Bdmin, right edge base point value Bdmax, upper edge base point value Bcmin, and lower edge base point value Bcmax that constitute the directional feature A1 in the sticky correlation image, wherein Bdmax and Bdmin are the maximum and minimum values ​​among Bd1, Bd2, ..., Bdb, and Bcmax and Bcmin are the maximum and minimum values ​​among Bc1, Bc2, ..., Bcb, respectively; Using formula Calculate and obtain the edge feature base point value C1 of the directional feature A1; S134: Calculate and obtain the edge feature base point values ​​C1, C2, ..., Aa of the directional features A1, A2, ..., Aa in sequence according to S132 to S133, and re-label the directional features corresponding to the edge feature base point values ​​as sticky association features in descending order of edge feature base point values, and label them as D1, D2, ..., Da; S14: Utilize the formula Calculate the vertical simulation basis F1 of the sticky correlation feature D1 in the acquired sticky correlation image and the first auxiliary image, where E1 is the edge feature base point value of the sticky correlation feature D1, and a1 and a2 are preset first and second vertical simulation factors; S15: Calculate and obtain the vertical pseudobase values ​​F1, F2, ..., Fa of the sticky correlation features D1, D2, ..., Da in the sticky correlation image and the first auxiliary image according to S14; S16: According to the second screening and calibration rules, screen and calibrate the first viscous gradient pixel, the second viscous gradient pixel, ..., the j-th viscous gradient pixel that constitutes the viscous correlation feature D1 in the viscous correlation image; S161: Obtain all edge pixels of the contour that constitutes the sticky correlation feature D1 in the sticky correlation image, relabel them as first sticky gradient pixels, and label them as G1, G2, ..., Gg, g≥1; S162: First, select the first viscosity gradient pixel G1 as the viscosity gradient point, acquire pixels that are adjacent to the viscosity gradient point and do not belong to the first viscosity gradient pixel among all pixels constituting the viscosity association feature D1 in the viscosity association image, and mark them as second viscosity gradient pixels; S163: According to S162, sequentially select the first viscosity gradient pixels G2, ..., Gg as viscosity gradient points, acquire all second viscosity gradient pixels of the viscosity association feature D1 in the viscosity association image, and acquire the total number thereof; S164: Acquire the total number I1 of pixels remaining after excluding the first viscosity gradient pixels and the second viscosity gradient pixels from all pixels constituting the viscosity association feature D1 in the viscosity association image; S165: If I1>P2, sequentially select each second viscosity gradient pixel as a viscosity gradient point according to S162 to S163, acquire pixels that are adjacent to the viscosity gradient point and do not belong to the first and second viscosity gradient pixels among all pixels constituting the viscosity association feature D1 in the viscosity association image, mark them as third viscosity gradient pixels, and acquire the total number I1 of pixels remaining after excluding the first, second and third viscosity gradient pixels from all pixels constituting the viscosity association feature D1 in the viscosity association image, then compare I1 with P2 in magnitude again; if I1>P2, continue the calculation according to S161 to S164 until there exists a total number I1 of pixels remaining after excluding the first, second, third, ..., h-th viscosity gradient pixels from all pixels constituting the viscosity association feature D1 in the viscosity association image, and at this time, stop the acquisition when I1<P2, where h≥1; Otherwise, no processing is performed, wherein P2 is a preset gradient pixel marking value; S17: Screen and mark the first viscosity gradient pixel, second viscosity gradient pixel, ..., j-th viscosity gradient pixel that constitute the viscosity association feature D1 in the first affiliated image according to S16, where j>1; S18: Compare h and j in magnitude, and if h>j: S181: Acquire the total number of the first viscosity gradient pixels constituting the viscosity association feature D1 in the viscosity association image and the first affiliated image respectively, and re-mark them as K1 and L1; Using the formula Calculate and obtain the vertical拟化 value M1 of the first viscosity gradient pixels constituting the viscosity association feature D1 in the viscosity association image and the first affiliated image; S182: Calculate the vertical fitting values ​​M1, M2, ..., Mj of the first, second, ..., j sticky gradient pixels constituting sticky correlation feature D1 in the sticky correlation image and the first auxiliary image according to S181, and calculate their mean values ​​using the summation and averaging formula. Re-label the mean values ​​of vertical fitting values ​​M1, M2, ..., Mj as the vertical fitting gradient value of sticky correlation feature D1, and label it as N1; S19: Conversely, according to S18, the vertical pseudo-gradient value of the sticky association feature D1 is calculated based on the first, second, ..., h sticky gradient pixels that constitute the sticky association feature D1 in the sticky association image and the first auxiliary image, and is marked as N1; S110: Calculate and obtain the vertical pseudo-gradient values ​​N1, N2, ..., Na of the sticky correlation features D1, D2, ..., Da in the sticky correlation image and the first auxiliary image according to S13 to S19, and aggregate the corresponding vertical pseudo-gradient values ​​with "|" to generate the pseudo-feature verification sequence of the sticky correlation image and the first auxiliary image. S111: Calculate and obtain the pseudo-feature verification sequence of the sticky correlation image and the second auxiliary image according to S12 to S110; S112: Select the pointing feature annotation images of the second, third, ..., p1-1 training images received by the sticky interaction unit within the sticky period in sequence as the sticky association images of the sticky period, and calculate and obtain the pointing feature annotation images of the second, third, ..., p1-1 training images and the pseudo feature verification sequence of the next image according to S11 to S111 respectively. The sticky interaction unit generates a verification traceability document for the training feature map set within the sticky period based on the obtained p1-1 pseudo-feature verification sequences. The sticky interaction unit transmits the training feature map set of the sticky cycle and the corresponding verification traceability document to the model training terminal. The model training terminal includes a graph verification module and a model training module. After receiving a training feature graph set for one sticky period and the corresponding verification traceability document transmitted by the sticky interaction unit, the model training terminal transmits them to the graph verification module. The feature set verification module is used to verify the integrity of the training feature set within the sticky period. After receiving the training feature set and corresponding verification traceability document for one sticky period transmitted by the model training terminal, the feature set verification module verifies them according to certain verification rules, as follows: S21: Obtain the first pseudo-feature sequence in the verification traceability document corresponding to the training feature map set within the sticky period received by the map set verification module, and extract the values ​​in the pseudo-feature sequence according to "|", and obtain the pseudo-feature values ​​V1, V2, ..., Vv corresponding to the pseudo-feature sequence, where v≥1; S22: Obtain the second pointing feature annotation image in the training feature map set within the sticky period and select it as the sticky associated image. Then, label the previous pointing feature annotation image of the sticky associated image within the sticky period as the first subordinate image of the sticky associated image in the sticky period. The next pointing feature annotation image within the viscosity period is designated as the second auxiliary image of the viscosity-related image of the viscosity period. S23: Calculate and obtain the sticky correlation features and their corresponding edge feature base point values ​​of the sticky correlation image and its first subordinate image according to S13, and sort the sticky correlation features in descending order of edge feature base point values. Mark the sorted sticky correlation features as O1, O2, ..., Ov, where v≥1; According to S13, the sticky correlation features and their corresponding edge feature base point values ​​of the sticky correlation image and the second auxiliary image of the sticky correlation image are obtained. The sticky correlation features are sorted in descending order of edge feature base point values. The sorted sticky correlation features are labeled as Q1, Q2, ..., Qv, where v≥1. S24: Create the first, second, and third risk weight variables R1, with the initial values ​​of R1, R2, and R3 being 0; S25: Calculate and obtain the vertical pseudogradient values ​​PO1, PO2, ..., POv of the sticky correlation features O1, O2, ..., Ov according to S14 to S110, and calculate and obtain the vertical pseudogradient values ​​PQ1, PQ2, ..., PQv of the sticky correlation features Q1, Q2, ..., Qv; S26: Compare PO1, PQ1, and V1 in sequence: S261: If PO1≠V1, then increment the value of the first risk weight variable R1 by 1; if PQ1≠V1, then increment the value of the second risk weight variable R2 by 1. S262: If PO1≠V1 and if PQ1≠V1, then increment the value of the third risk weight variable R3 by 1. S27: Following S26, compare the vertical pseudogradient values ​​PO1, PO2, ..., POv and the vertical pseudogradient values ​​PQ1, PQ2, ..., PQv with V1, V2, ..., Vv in sequence, and obtain the final values ​​of the first risk weight variables R1, R2 and R3, and label them as PR1, PR2 and PR3 in sequence; S28: If PR1>P4, then the first auxiliary image is determined to be missing, and its corresponding position in the training feature map set of the sticky period is recorded; otherwise, the first auxiliary image is determined to be tampered with, and the first auxiliary image is removed from the training feature map set of the sticky period. If PR2 > P4, then the second auxiliary image is determined to be missing, and its corresponding position in the training feature map set of the sticky period is recorded; otherwise, the second auxiliary image is determined to be tampered with, and the second auxiliary image is removed from the training feature map set of the sticky period. If PR3 > P4, it is determined that the sticky association image is missing, and its corresponding position in the training feature map set of the sticky period is recorded. Otherwise, it is determined that the sticky association image has been tampered with, and the sticky association image is removed from the training feature map set of the sticky period. P4 is a preset threshold. S29: Following S21 to S28, all pointing feature annotation images in the training feature map set of the sticky period are judged sequentially. The positions of all the pointing feature annotation images that are judged to be missing in the training feature map set of the sticky period are obtained in the training feature map set, and a retransmission instruction is generated based on it and transmitted to the client terminal. After receiving the retransmission instruction transmitted by the map set verification module, the client terminal retransmits all the pointing feature annotation images that are judged to be missing in the sticky period according to the positions of all the pointing feature annotation images that are judged to be missing in the training feature map set carried in it, and the map set verification module performs a second verification. After the verification is passed, the map set verification module transmits the training feature map set of the sticky period to the model training module. The model training module is used to train and optimize the sports action recognition model. The model training module includes several model training units, each of which recognizes a sports action. Each model training unit stores an action recognition model that recognizes the corresponding sports action. After receiving the training feature set within the sticky period transmitted by the image set verification module, the model training module transmits it to the model training unit that recognizes the corresponding sports movements. After receiving the training feature set within the sticky period transmitted by the model training module, the model training unit inputs it into the movement recognition model for training and optimizes the movement recognition model parameters for the corresponding sports movements. In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0014] The above description is merely an example and illustration of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0015] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An artificial intelligence interactive system, characterized in that, include: The client terminal includes several client modules. Each client module is used to acquire training images of a sport in the training field and to annotate them with directional features. Each client module corresponds to a sport that can be performed in the training field. Each client module includes an image acquisition unit, a feature annotation unit, and a sticky interaction unit. The image acquisition unit acquires training images for the corresponding sports, and the feature annotation unit performs directional feature annotation on the training images for the corresponding sports and generates a directional feature-annotated image for the training images based on these annotations. The sticky interaction unit generates a training feature map set for a sticky period based on the directional feature annotation images of all training images within that sticky period, and generates a verification and traceability document for the training feature map set corresponding to the sport within that sticky period according to certain sticky generation rules, as follows: S11: Obtain the total number of pointing feature-labeled images of the training images transmitted by the feature labeling unit to the sticky interaction unit within a sticky period, and label it as p1; S12: First, select the pointing feature annotation image of the second training image received by the sticky interaction unit within the sticky period as the sticky association image of the sticky period, and label the pointing feature annotation images of the previous and next images of the sticky association image received by the sticky interaction unit within the sticky period as the first and second auxiliary images of the sticky association image of the sticky period. S13: According to the first filtering rule, filter the directional features in the sticky association image and the first auxiliary image to obtain the sorted sticky association features D1, D2, ..., Da in the sticky association image and the first auxiliary image, as follows: S131: Obtain the directional features that are labeled in both the sticky association image and the first auxiliary image, and label them as A1, A2, ..., Aa, where a≥1; S132: Obtain all pixels constituting the directional feature A1 in the sticky association image and represent the position of the pixels using number pairs, labeled as B1(Bc1, Bd1), B2(Bc2, Bd2), ..., Bb(Bcb, Bdb). When representing the position of the pixels using number pairs, the position before the comma in the parentheses indicates the column number of the pixel in the sticky association image, and the position after the comma indicates the row number of the pixel in the sticky association image. S133: Obtain the left edge base point value Bdmin, right edge base point value Bdmax, upper edge base point value Bcmin, and lower edge base point value Bcmax that constitute the directional feature A1 in the sticky correlation image, wherein Bdmax and Bdmin are the maximum and minimum values ​​among Bd1, Bd2, ..., Bdb, and Bcmax and Bcmin are the maximum and minimum values ​​among Bc1, Bc2, ..., Bcb, respectively; Using formula Calculate and obtain the edge feature base point value C1 of the directional feature A1; S134: Calculate and obtain the edge feature base point values ​​C1, C2, ..., Aa of the directional features A1, A2, ..., Aa in sequence according to S132 to S133, and re-label the directional features corresponding to the edge feature base point values ​​as sticky association features in descending order of edge feature base point values, and label them as D1, D2, ..., Da; S14: Utilize the formula Calculate the vertical simulation basis F1 of the sticky correlation feature D1 in the acquired sticky correlation image and the first auxiliary image, where E1 is the edge feature base point value of the sticky correlation feature D1, and a1 and a2 are preset first and second vertical simulation factors; S15: Calculate and obtain the vertical pseudobase values ​​F1, F2, ..., Fa of the sticky correlation features D1, D2, ..., Da in the sticky correlation image and the first auxiliary image according to S14; S16: According to the second screening and calibration rules, screen and calibrate the first viscous gradient pixel, the second viscous gradient pixel, ..., the j-th viscous gradient pixel that constitutes the viscous correlation feature D1 in the viscous correlation image; S17: According to S16, filter and label the first viscous gradient pixel, the second viscous gradient pixel, ..., the j-th viscous gradient pixel in the first auxiliary image that constitutes the viscous correlation feature D1, where j>1; S18: Compare h and J. If h > j: S181: Obtain the total number of first viscous gradient pixels constituting viscous correlation feature D1 in the viscous correlation image and the first auxiliary image respectively, and recalibrate them as K1 and L1; Using formula Calculate the vertical fitting value M1 of the first viscous gradient pixel that constitutes the viscous correlation feature D1 in the viscous correlation image and the first auxiliary image; S182: Calculate the vertical fitting values ​​M1, M2, ..., Mj of the first, second, ..., j sticky gradient pixels constituting sticky correlation feature D1 in the sticky correlation image and the first auxiliary image according to S181, and calculate their mean values ​​using the summation and averaging formula. Re-label the mean values ​​of vertical fitting values ​​M1, M2, ..., Mj as the vertical fitting gradient value of sticky correlation feature D1, and label it as N1; S19: Conversely, according to S18, the vertical pseudo-gradient value of the sticky association feature D1 is calculated based on the first, second, ..., h sticky gradient pixels that constitute the sticky association feature D1 in the sticky association image and the first auxiliary image, and is marked as N1; S110: Calculate and obtain the vertical pseudo-gradient values ​​N1, N2, ..., Na of the sticky correlation features D1, D2, ..., Da in the sticky correlation image and the first auxiliary image according to S13 to S19, and aggregate the corresponding vertical pseudo-gradient values ​​with "|" to generate the pseudo-feature verification sequence of the sticky correlation image and the first auxiliary image. S111: Calculate and obtain the pseudo-feature verification sequence of the sticky correlation image and the second auxiliary image according to S12 to S110; S112: Select the pointing feature annotation images of the second, third, ..., p1-1 training images received by the sticky interaction unit within the sticky period in sequence as the sticky association images of the sticky period, and calculate and obtain the pointing feature annotation images of the second, third, ..., p1-1 training images and the pseudo feature verification sequence of the next image according to S11 to S111 respectively. The sticky interaction unit generates a verification traceability document for the training feature map set within the sticky period based on the obtained p1-1 pseudo-feature verification sequences. The model training terminal includes a graph verification module and a model training module. The graph verification module verifies the training feature graph of a sticky period and its corresponding verification traceability document. The model training module stores a motion recognition model that can recognize all sports movements in the training field. The model training module inputs the verified training feature graph of a sticky period into the motion recognition model for training and optimization of the motion recognition model parameters.

2. The artificial intelligence interactive system according to claim 1, characterized in that, One of the viscosity cycles lasts for a time P1, where P1 is a preset cycle duration.

3. The artificial intelligence interactive system according to claim 1, characterized in that, The feature annotation unit performs directional feature annotation by using a bounding box method to select the directional features in the training image. It should be noted that all training images acquired by the image acquisition unit have the same image size, which is 2400*1920px.

4. The artificial intelligence interactive system according to claim 1, characterized in that, The directional features refer to the limbs of an athlete that can be identified by sports movements, including the head, left shoulder, right shoulder, left upper arm, right upper arm, left forearm, right forearm, left hand, right hand, chest, waist, left thigh, right thigh, left calf, right calf, left foot, and right foot.

5. The artificial intelligence interactive system according to claim 1, characterized in that, The training image for a sport refers to an image of an athlete performing the sport in a training field, and the training image includes the full body of the athlete performing the sport.

6. The artificial intelligence interactive system according to claim 1, characterized in that, The specific verification rules for the graph set verification module to verify the training feature graph set and its corresponding verification traceability document for a sticky period are as follows: S21: Obtain the first pseudo-feature sequence in the verification traceability document corresponding to the training feature map set within a sticky period, and extract the values ​​in the pseudo-feature sequence according to "|", to obtain the pseudo-feature values ​​V1, V2, ..., Vv corresponding to the pseudo-feature sequence, where v≥1; S22: Obtain the second pointing feature annotation image in the training feature map set within the sticky period and select it as the sticky associated image. Label the first pointing feature annotation image of the sticky associated image within the sticky period as the first subordinate image of the sticky associated image within the sticky period. Label the second pointing feature annotation image of the sticky associated image within the sticky period as the second subordinate image of the sticky associated image within the sticky period. S23: Calculate and obtain the sticky correlation features and their corresponding edge feature base point values ​​of the sticky correlation image and its first subordinate image according to S13, and sort the sticky correlation features in descending order of edge feature base point values. Mark the sorted sticky correlation features as O1, O2, ..., Ov, v≥1. Calculate and obtain the sticky correlation features and their corresponding edge feature base point values ​​of the sticky correlation image and its second subordinate image according to S13, and sort the sticky correlation features in descending order of edge feature base point values. Mark the sorted sticky correlation features as Q1, Q2, ..., Qv. S24: Create the first, second, and third risk weight variables R1, with the initial values ​​of R1, R2, and R3 being 0; S25: Calculate and obtain the vertical pseudogradient values ​​PO1, PO2, ..., POv of the sticky correlation features O1, O2, ..., Ov according to S14 to S110, and calculate and obtain the vertical pseudogradient values ​​PQ1, PQ2, ..., PQv of the sticky correlation features Q1, Q2, ..., Qv; S26: Compare PO1, PQ1, and V1 in sequence: S261: If PO1≠V1, then increment the value of the first risk weight variable R1 by 1; if PQ1≠V1, then increment the value of the second risk weight variable R2 by 1. S262: If PO1≠V1 and if PQ1≠V1, then increment the value of the third risk weight variable R3 by 1. S27: Following S26, compare the vertical pseudogradient values ​​PO1, PO2, ..., POv and the vertical pseudogradient values ​​PQ1, PQ2, ..., PQv with V1, V2, ..., Vv in sequence, and obtain the final values ​​of the first risk weight variables R1, R2 and R3, and label them as PR1, PR2 and PR3 in sequence; S28: If PR1>P4, then the first auxiliary image is determined to be missing, and its corresponding position in the training feature map set of the sticky period is recorded; otherwise, the first auxiliary image is determined to be tampered with, and the first auxiliary image is removed from the training feature map set of the sticky period. If PR2 > P4, then the second auxiliary image is determined to be missing, and its corresponding position in the training feature map set of the sticky period is recorded; otherwise, the second auxiliary image is determined to be tampered with, and the second auxiliary image is removed from the training feature map set of the sticky period. If PR3 > P4, it is determined that the sticky association image is missing, and its corresponding position in the training feature map set of the sticky period is recorded. Otherwise, it is determined that the sticky association image has been tampered with, and the sticky association image is removed from the training feature map set of the sticky period. P4 is a preset threshold. S29: Following S21 to S28, determine the positions of all missing directional feature annotation images in the training feature map set for the viscous period.

Images