Target positioning method based on visual space-time operator

Abstract

The invention discloses a target positioning method based on a visual space-time operator, relates to the technical field of target positioning, is used for solving the problems that the target positioning accuracy is reduced and the error recognition rate is increased, and is used for distinguishing a target from a background through a human-computer interaction page, collecting application information of a recognition method and determining the target recognition credibility. Target identification is used for determining the number of current targets; the targets are marked; gesture prompts are generated; the position of an arm is locked for motion tracking; whether the multi-target recognition mode is executed or not is determined in combination with the page proportion, visual spatio-temporal information is fully utilized, positioning accuracy and system stability are improved, the method is suitable for complex human-computer interaction scenes, multi-target positioning precision and response speed are improved, and the error recognition rate is effectively reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target positioning, more particularly, to a target positioning method based on a visual space-time operator. BACKGROUND

[0002] With the rapid development of artificial intelligence and computer vision technology, visual-based target positioning technology is widely used in intelligent monitoring, human-computer interaction, virtual reality and other fields. Traditional target positioning methods mostly rely on feature extraction and target detection algorithms of static images. However, in dynamic scenes, especially in multi-person interaction environments, a single image recognition technology is difficult to accurately distinguish targets from background subjects, resulting in a decrease in target positioning accuracy.

[0003] The prior art has the following disadvantages: Currently, target recognition is often based on analysis of a single frame of image, lacking comprehensive utilization of the space-time features of target dynamic behavior, making it difficult to effectively capture the gesture motion trajectory and motion intensity of the target, and further affecting the accurate judgment of the number of target subjects and their motion states, making it difficult to comprehensively and quantitatively evaluate the target action, resulting in a decrease in target positioning accuracy and an increase in misrecognition rate. Therefore, a target positioning method based on a visual space-time operator is proposed.

[0004] The above information disclosed in the background section is only intended to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute the prior art known to those of ordinary skill in the art. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a target positioning method based on a visual space-time operator, which realizes comprehensive calculation of target gesture motion speed, trajectory and motion intensity by using space-time feature extraction and analysis algorithms, dynamically sorts and assigns weights to the target, and intelligently switches the multi-target subject recognition mode based on the target weight and the page proportion to solve the problems raised in the above background technology.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme, a target positioning method based on a visual space-time operator, comprising the following steps: Step S1: distinguishing targets and background subjects according to a human-computer interaction page, collecting recognition method application information, determining the recognition trust degree of the target recognition method, and determining whether to use the target recognition method to recognize the targets and background subjects according to the recognition trust degree; Step S2: using the target recognition method to determine the number of targets in the current page, marking the targets and generating gesture prompt characters, locking the target arm position for motion tracking, and collecting gesture motion speed to calculate gesture speed specification features; Step S3: Collect gesture motion length to calculate gesture trajectory specification features, analyze gesture motion intensity in combination with gesture speed specification features to obtain gesture motion intensity coefficient and determine target subject quantity; Step S4: Sort target subjects according to gesture motion intensity coefficient and assign target weight, obtain page proportion of each target subject according to human-computer interaction page, and determine whether to execute multi-target subject recognition mode in combination with target weight.

[0007] In a preferred embodiment, in step S1, image data currently displayed in the human-computer interaction page is obtained, pixel-level region division processing is performed on the image data, and the target region and the background subject region in the image are preliminarily separated; The recognition method application information of the target recognition method to be applied in the historical application scene is collected. The recognition method application information includes bounding box information and effective recognition target information, the bounding box information is a rectangular detection region range, and the effective recognition target information is a region pixel set recognized and defined.

[0008] In a preferred embodiment, in step S1, the intersection area ratio of the two rectangular detection region ranges in the adjacent frames is taken as the detection box overlap rate; The spatial coverage rate of the random frame image is calculated based on the effective recognition target information, and the ratio of the area of the region pixel set accurately recognized and defined to the total pixel area of the page is defined as the spatial coverage rate. The difference between the spatial coverage rate and the preset spatial reference coverage rate is divided by the spatial reference coverage rate to obtain the spatial coverage offset rate.

[0009] In a preferred embodiment, in step S1, after the spatial coverage offset rate is standardized, the detection box overlap rate and the spatial coverage offset rate are integrated to evaluate the recognition trustworthiness by using a weighted average formula; When the recognition trustworthiness is greater than or equal to the preset trustworthiness discrimination threshold, the target recognition method is executed to continue to recognize the target and the background subject; When the recognition trustworthiness is less than the preset trustworthiness discrimination threshold, it is determined that the target recognition method is not stable enough, and the calling of the target recognition method is stopped.

[0010] In a preferred embodiment, in step S2, the input image of the current human-computer interaction page is accurately recognized based on the target recognition method to determine the quantity of targets in the image; The target in the detection box is labeled and a gesture prompt word is generated, motion tracking is achieved by locking the arm position, and the arm position is a key point coordinate obtained by a pose estimation algorithm. In adjacent time frames, the spatial displacement of the arm position is calculated by the displacement of the arm position key point coordinates between adjacent frames; The spatial displacement of the arm position is divided by the inter-frame time interval to obtain the instantaneous velocity as the gesture motion velocity; The variance of the gesture motion velocity in the preset time window is calculated as the gesture velocity specification feature.

[0011] In a preferred embodiment, in step S3, a preset acquisition window is divided into multiple acquisition time points, and the spatial coordinates of the arm end at each acquisition time point are identified by a pose estimation algorithm. The gesture motion length between adjacent acquisition time points is calculated by the Euclidean distance method; The length rate is obtained by dividing each gesture motion length by the corresponding time interval. The length rate stability index of the gesture motion is calculated based on the average value and the standard deviation of each length rate; The burst degree of the gesture motion is obtained by subtracting the minimum value from the maximum value of the length rate and dividing the result by the maximum value of the length rate; The gesture trajectory specification feature is calculated by the geometric mean method after the length rate stability index of the gesture motion is standardized and combined with the burst degree.

[0012] In a preferred embodiment, in step S3, the ratio of the gesture trajectory specification feature to the preset trajectory specification threshold value is subtracted from 1 to obtain the trajectory margin; The ratio of the gesture velocity specification feature to the preset velocity specification threshold value is subtracted from 1 to obtain the velocity margin; The sum of the velocity margin and the trajectory margin is taken as the gesture motion intensity; The gesture motion intensity coefficient is obtained based on the gesture motion intensity by an exponential function.

[0013] In a preferred embodiment, in step S4, the target subjects are sorted in descending order according to the gesture motion intensity coefficient and combined into a target subject set. The sorting index is assigned in sequence according to the sorting position of the target subject in the target subject set. The reciprocal of the sorting index is taken as the target weight of the corresponding target subject; The page proportion of each target subject can be obtained by a target recognition method to obtain the pixel area occupied by the target subject in the human-computer interaction page. The ratio of the pixel area to the total pixel area of the human-computer interaction page is taken as the page proportion of the target subject; The product of the page proportion of the target subject and the target weight is taken as the comprehensive weight index.

[0014] In a preferred embodiment, in step S4, the target subject corresponding to the maximum of the comprehensive weight index is taken as the first subject, the comprehensive weight indexes of all target subjects except the first subject are taken as an arithmetic mean, and the comprehensive weight index of the first subject is subtracted from the arithmetic mean to obtain a weight difference value; The weight difference value is compared with a preset difference threshold to determine whether to execute a multi-target subject identification mode: If the weight difference value is lower than the preset difference threshold, it is determined to execute the multi-target subject identification mode; On the contrary, the single-target subject identification mode is executed for the first subject.

[0015] Technical effects and advantages of the present application: The present application distinguishes the target and the background subject through the human-computer interaction page, collects and identifies the application information, determines the identification confidence of the target identification method, uses the target identification method to identify and determine the target quantity of the current page, marks the target and generates a gesture prompt character, locks the target arm position to track the movement, collects the gesture movement speed to calculate the gesture speed specification feature, collects the gesture movement length to calculate the gesture trajectory specification feature, analyzes the gesture movement intensity to obtain the gesture movement intensity coefficient and determine the target subject in combination with the gesture speed specification feature, assigns the target weight according to the gesture movement intensity coefficient after sorting the target subject, and determines whether to execute the multi-target subject identification mode according to the page proportion of each target subject obtained from the human-computer interaction page in combination with the target weight. The present application fully utilizes the visual space-time information, improves the accuracy of target positioning and the stability of the system, is suitable for complex human-computer interaction scenes, improves the precision and real-time response capability of multi-target positioning, and effectively reduces the misidentification rate. BRIEF DESCRIPTION OF DRAWINGS

[0016] Fig. 1 The present application is a flowchart for the implementation of a target positioning method based on a visual space-time operator.

[0017] Fig. 2 The present application is a flowchart for the implementation of a target positioning method based on a visual space-time operator. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0019] The application distinguishes the target and the background subject through the human-computer interaction page, collects and identifies the application information, determines the identification confidence of the target identification method, uses the target identification method to identify and determine the target quantity of the current page, marks the target and generates a gesture prompt character, locks the target arm position for motion tracking, collects the gesture motion speed to calculate the gesture speed specification feature, collects the gesture motion length to calculate the gesture trajectory specification feature, analyzes the gesture motion intensity to obtain the gesture motion intensity coefficient and determine the target subject in combination with the gesture speed specification feature, assigns the target weight after sorting the target subject according to the gesture motion intensity coefficient, obtains the page proportion of each target subject according to the human-computer interaction page, and determines whether to execute the multi-target subject identification mode in combination with the target weight. The method fully utilizes the visual space-time information, improves the accuracy of target positioning and the stability of the system, and is suitable for complex human-computer interaction scenes.

[0020] Embodiment 1, a target positioning method based on a visual space-time operator, as shown in Figs. 1-2 The method comprises the following steps: Step S1: distinguishing the target and the background subject according to the human-computer interaction page, collecting and identifying the application information, determining the identification confidence of the target identification method, and determining whether to use the target identification method to identify the target and the background subject according to the identification confidence; Step S2: using the target identification method to identify and determine the target quantity of the current page, marking the target and generating a gesture prompt character, locking the target arm position for motion tracking, and collecting the gesture motion speed to calculate the gesture speed specification feature; Step S3: collecting the gesture motion length to calculate the gesture trajectory specification feature, analyzing the gesture motion intensity to obtain the gesture motion intensity coefficient and determine the target subject quantity in combination with the gesture speed specification feature; Step S4: assigning the target weight after sorting the target subject according to the gesture motion intensity coefficient, obtaining the page proportion of each target subject according to the human-computer interaction page, and determining whether to execute the multi-target subject identification mode in combination with the target weight.

[0021] The specific implementation is as follows: The target refers to the user individual in the human-computer interaction scene. In step S1, the image data currently displayed in the human-computer interaction page is obtained, the image data is a digital image frame collected at a predetermined sampling frequency, contains the pixels of all visible elements in the page, the image data is subjected to pixel-level region division processing, the target region and the background subject region in the image are preliminarily separated through an image segmentation algorithm, the target region is a pixel set determined to contain the target outer contour, the target body structure and its key parts; and the background subject region is all pixel sets except the target region.

[0022] It should be noted that the human-computer interaction page refers to a graphical interface for realizing information exchange and operation interaction between a user and a computer system, which displays state information, interaction controls and visual elements of the system in a two-dimensional space; the image segmentation algorithm refers to a calculation method for dividing an image into a plurality of regions or targets with consistency or similarity based on pixel information of the input image.

[0023] Collect the recognition method application information of the target recognition method to be applied in the historical application scene, the recognition method application information includes detection box information and effective recognition target information, the detection box information is the range of the rectangular detection region, and the effective recognition target information is the pixel set of the region accurately recognized and defined.

[0024] The detection box overlap rate is calculated based on the detection box information, the detection box overlap rate is used to measure the position and shape consistency of the rectangular detection region range in adjacent frames, and is defined as the ratio of the intersection area to the union area of the two rectangular detection region ranges in adjacent frames, and the calculation formula is: ; Among them, is the detection box overlap rate, is the rectangular detection region range of the current frame, is the rectangular detection region range of the previous frame, represents the pixel area of the rectangular detection region range. The value range of the detection box overlap rate is , the closer the value is to 1, the smaller the change of the detection box position is, and the higher the stability of the recognition result is; the closer the value is to 0, the more serious the drift of the detection box between adjacent frames is, and the lower the recognition stability is.

[0025] The spatial coverage rate of the random frame image is calculated based on the effective recognition target information, and the ratio of the area of the region pixel set accurately recognized and defined to the total pixel area of the page is defined as the spatial coverage rate; A preset spatial reference coverage rate is obtained by collecting a large number of human-computer interaction page image frames, extracting the target region pixel area based on the image segmentation algorithm, and calculating a spatial coverage rate sequence, and using the median of the spatial coverage rate as the spatial reference coverage rate; The difference between the spatial coverage rate and the spatial reference coverage rate is divided by the spatial reference coverage rate to obtain the spatial coverage offset rate; The value range of the spatial coverage offset rate is , the larger the value is, the greater the difference between the target coverage area of the current frame and the reference value is, and the lower the stability of the recognition result is.

[0026] After the spatial coverage offset rate is standardized, the detection box overlap rate and the spatial coverage offset rate are used to evaluate the recognition trustworthiness by using a weighted average formula, and the specific calculation formula is: ; wherein, is the recognition trust degree, is the detection frame overlap rate, is the spatial coverage offset rate, and is the weight coefficient, a multivariate linear regression analysis method is used, the detection frame overlap rate and the spatial coverage offset rate in the historical sample are used as independent variables, and the recognition accuracy is used as the dependent variable, and the least square method is used for fitting determination.

[0027] The higher the recognition trust degree is, the better the target recognition method performs in position consistency and spatial proportion stability; It should be noted that the standardization processing refers to a process of mapping different physical quantities or original data of different dimensions to a uniform dimension, a uniform numerical range or a uniform statistical distribution through a specific mathematical transformation. The standardization processing methods include but are not limited to standard linear transformation based on interval scaling, Z-Score standardization method based on statistics or normalization method based on nonlinear mapping function. Herein, the application method of the standardization processing is not described in detail.

[0028] Based on the statistical analysis of the historical recognition method application information, a trust degree discrimination threshold is preset; Specifically, based on the historical recognition result data, the corresponding recognition trust degree value is calculated, and combined with the recognition accuracy label manually labeled, the ROC curve is drawn, and the value that makes the difference between true positive rate and false positive rate maximum is selected as the trust degree discrimination threshold; When the recognition trust degree is greater than or equal to the trust degree discrimination threshold, the target recognition method continues to identify the target and the background subject; When the recognition trust degree is less than the trust degree discrimination threshold, it is determined that the target recognition method is not stable enough, and the calling of the target recognition method is stopped.

[0029] In step S2, the input image of the current human-computer interaction page is accurately identified based on the target recognition method to preliminarily determine the number of targets in the image.

[0030] Through frame-by-frame analysis of the input image by the target recognition method, the target recognition method identifies the target in the image based on computer vision technology, outputs a set containing several target detection frames, and obtains the total number of identified targets in the current page as the target number of the current page by counting all the detection frames.

[0031] The target in each detection frame is labeled, and the label content includes a unique target identifier. After labeling, a gesture prompt word is generated on the human-computer interaction page, the gesture prompt word is a predefined text or symbol, and is used to guide the user to perform a specified gesture operation.

[0032] After the gesture prompt character is generated, motion tracking is achieved by locking the arm position in the detection frame area of the recognized target, the arm position being a key point coordinate obtained by a pose estimation algorithm, the pose estimation algorithm outputting a key point coordinate set in a two-dimensional page coordinate system based on a human key point detection technology.

[0033] It should be noted that the pose estimation algorithm refers to a calculation method for detecting and positioning the coordinate position of a human key point in a two-dimensional or three-dimensional space through an input image or video frame.

[0034] The gesture motion speed is calculated based on the collected arm position, the key point coordinate sequence of the arm position in the continuous image frames in the adjacent time frames is collected, the spatial displacement of the arm position is obtained by calculating the displacement of the key point coordinates of the arm position between adjacent frames, the instantaneous speed is obtained by dividing the spatial displacement of the arm position by the inter-frame time interval as the gesture motion speed, and the specific calculation formula is: ; Among them, is the gesture motion speed, and are the key point coordinates of the same arm position in the continuous two frames, is the inter-frame time interval.

[0035] The variance of the gesture motion speed in the preset time window is calculated as the gesture speed specification feature, and the specific calculation formula is: ; Among them, is the gesture speed specification feature, is the gesture motion speed of the i-th image, is the mean value of the gesture motion speed in the time window, is the total number of image frames in the time window.

[0036] The gesture speed specification feature reflects the stability of the gesture motion, and the smaller the value, the smaller the gesture motion speed fluctuation, that is, the more stable and continuous the gesture action; on the contrary, the larger the value, the larger the gesture motion speed fluctuation, and the more likely to affect the accuracy of gesture recognition and interactive experience.

[0037] In step S3, a preset collection window is divided into a plurality of collection time points, the spatial coordinates of the arm end at each collection time point are identified by the pose estimation algorithm, and the spatial coordinates of adjacent collection time points are calculated by the Euclidean distance method to calculate the gesture motion length of adjacent collection time points; The length rate is obtained by dividing each gesture motion length by the corresponding time interval, the average value and the standard deviation are calculated based on each length rate, and the ratio of the standard deviation to the average value is taken as the length rate stability index of the gesture motion; The smaller the length rate stability index of the gesture motion is, the more stable the rhythm of the gesture motion is; the greater the length rate stability index of the gesture motion is, the stronger the fluctuation of the gesture motion is.

[0038] The burst degree of the gesture motion is obtained by subtracting the minimum value of the length rate from the maximum value of the length rate and dividing the result by the maximum value of the length rate; The higher the burst degree of the gesture motion is, the more the high-intensity action and low-intensity action in the collection window exist; the lower the burst degree of the gesture motion is, the more stable the intensity change of the gesture motion is.

[0039] The gesture trajectory specification feature is calculated by geometric average method after the length rate stability index of the gesture motion is standardized and the burst degree of the gesture motion is standardized, and the calculation formula is: wherein, the burst degree is, the standardized value of the length rate stability index is, the gesture trajectory specification feature is.

[0040] The closer the gesture trajectory specification feature is to 1, the more stable the gesture motion is; the closer the gesture trajectory specification feature is to 0, the more fluctuant the gesture motion is.

[0041] It should be noted that the preset collection window is used to limit the statistical and determination time range of the gesture motion length, the collection window is a continuous time interval, which is divided into a plurality of collection time points, the collection time points can be set at equal intervals, for example, the time interval of the collection time points is consistent with the image frame, and a collection time point is determined after receiving each frame of video; the Euclidean distance method is a spatial distance measurement method, which is used to calculate the shortest straight line distance between two points in the same coordinate system, in the embodiment, the Euclidean distance method is used to calculate the gesture motion length of the arm end between adjacent collection time points.

[0042] The gesture motion intensity is compared and analyzed by comparing the preset trajectory specification threshold and the speed specification threshold with the gesture trajectory specification feature and the gesture speed specification feature respectively; The trajectory margin is obtained by subtracting 1 from the ratio of the gesture trajectory specification feature to the trajectory specification threshold; The speed margin is obtained by subtracting 1 from the ratio of the gesture speed specification feature to the speed specification threshold; When the trajectory margin and the speed margin are greater than 0, it indicates that the corresponding target gesture motion intensity is high; when the trajectory margin and the speed margin are less than 0, it indicates that the corresponding target motion intensity is low; The sum of the speed margin and the trajectory margin is taken as the gesture motion intensity; If the gesture motion intensity is greater than 0, the gesture motion intensity coefficient is obtained by an exponential function: wherein, the gesture motion intensity is, a gesture motion intensity coefficient; If the gesture motion intensity is less than or equal to 0, the gesture motion intensity coefficient is set to 0. A target corresponding to the gesture motion intensity coefficient greater than 0 is marked as a target subject, and the number of target subjects is counted.

[0043] It should be noted that the preset trajectory specification threshold is used to measure whether the gesture motion trajectory reaches the baseline value of the trajectory, for example, in a test environment of human-computer interaction, a plurality of standard gesture motions are completed by a training personnel, gesture trajectory specification characteristic values of each gesture motion are calculated, and the minimum value of the gesture trajectory specification characteristic values is taken as the trajectory specification threshold; the preset speed specification threshold is used to measure whether the gesture motion speed reaches the baseline value of the speed, for example, in the same test environment, gesture speed specification characteristic values when the standard target gesture is completed are collected, and the minimum value of the gesture speed specification characteristic values is taken as the speed specification threshold.

[0044] In step S4, the target subjects are sorted in descending order according to the gesture motion intensity coefficient and combined into a target subject set, the target subjects are assigned with sorting indexes in sequence according to the sorting positions of the target subjects in the target subject set, and the reciprocals of the sorting indexes are taken as target weights of the corresponding target subjects. The page proportion of each target subject can be obtained by a target recognition method, and after the target subject is recognized, the pixel area occupied by the target subject in the current human-computer interaction page is counted, and the ratio of the pixel area to the total pixel area of the current human-computer interaction page is taken as the page proportion of the target subject.

[0045] The product of the page proportion of the target subject and the target weight is taken as a comprehensive weight index. The target subject corresponding to the maximum value of the comprehensive weight index is taken as a first subject, the arithmetic mean of the comprehensive weight indexes of all target subjects except the first subject is calculated, and the difference between the comprehensive weight index of the first subject and the arithmetic mean is taken as a weight difference value. The weight difference value is compared with a preset difference threshold to determine whether to execute a multi-target subject recognition mode. If the weight difference value is lower than the preset difference threshold, it is determined to execute the multi-target subject recognition mode; otherwise, a single-target subject recognition mode is executed for the first subject.

[0046] The single-target subject recognition mode refers to a working mode in which, in the current human-computer interaction page, the first subject with the highest comprehensive weight index is subjected to subsequent target recognition and interaction processing, and the multi-target subject recognition mode refers to a working mode in which a plurality of target subjects are subjected to recognition and interaction processing in the current human-computer interaction page.

[0047] It needs to be explained that the preset difference threshold refers to a critical value for measuring the difference between the first subject and other target subjects in the multi-target subject identification and determination process, which can be determined through historical interaction data analysis, for example, the distribution of the difference between the first subject and other target subjects in the comprehensive weight index is calculated in the sample data set of the historical interaction data, and the high quantile point is taken as the difference threshold.

[0048] Finally, it also needs to be explained that in this text, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between them.

[0049] Moreover, the term "comprising" "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article or equipment including the element.

[0050] In this text, the singular form "one" "a" and "said / that" can also include the plural form, unless the context clearly indicates otherwise. It should also be understood that the term "including / containing" or "having" and the like specifies the existence of the stated features, integers, steps, operations, components, parts or combinations thereof, but does not exclude the possibility of the presence or addition of one or more other features, integers, steps, operations, components, parts or combinations thereof. In this specification, the term "and / or" includes any and all combinations of the related listed items.

[0051] The various embodiments in the specification are described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the various embodiments can be combined as needed, and the same and similar parts refer to each other.

[0052] The above description of the disclosed embodiments enables those skilled in the art to implement or use the various modifications of the embodiments of the present application, and it will be obvious to those skilled in the art that the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A target positioning method based on a visual space-time operator, characterized in that: The method comprises the following steps: Step S1: According to the target and background subject of the human-computer interaction page, the application information of the recognition method is collected, the recognition trust degree of the target recognition method is determined, and it is determined whether to use the target recognition method to recognize the target and the background subject according to the recognition trust degree; Step S2: The target quantity of the current page is determined by using the target recognition method to recognize, the target is marked and a gesture prompt character is generated, the target arm position is locked for motion tracking, the gesture motion speed is calculated to obtain the gesture speed specification feature; Step S3: The gesture motion length is collected to calculate the gesture trajectory specification feature, the gesture motion intensity coefficient is obtained by analyzing the gesture motion intensity in combination with the gesture speed specification feature, and the target subject quantity is determined; Step S4: The target subject is sorted according to the gesture motion intensity coefficient and is given a target weight, the page proportion of each target subject is obtained according to the human-computer interaction page, and it is determined whether to execute the multi-target subject recognition mode in combination with the target weight.

2. The target positioning method based on the visual space-time operator according to claim 1, wherein: in step S1, image data currently displayed in the human-computer interaction page is obtained, and pixel-level region division processing is performed on the image data to preliminarily separate the target region and the background subject region in the image; The application information of the target recognition method in the historical application scene is collected; The recognition method application information includes detection box information and effective recognition target information, the detection box information is a rectangular detection region range, and the effective recognition target information is a region pixel set recognized and defined.

3. The target positioning method based on the visual space-time operator according to claim 2, wherein: In step S1, the intersection area ratio of the two rectangular detection region ranges in adjacent frames is taken as the detection box overlap rate; The spatial coverage rate of the random frame image is calculated based on the effective recognition target information, and the ratio of the area of the region pixel set accurately recognized and defined to the total pixel area of the page is defined as the spatial coverage rate; The spatial coverage offset rate is obtained by dividing the difference between the spatial coverage rate and the preset spatial reference coverage rate by the spatial reference coverage rate.

4. The target positioning method based on the visual space-time operator according to claim 3, wherein: In step S1, after the spatial coverage offset rate is standardized, the detection box overlap rate and the spatial coverage offset rate are evaluated by using a weighted average formula to evaluate the recognition trust degree; When the recognition trust degree is greater than or equal to the preset trust degree discrimination threshold, the target recognition method is executed to continue to recognize the target and the background subject; When the recognition trust degree is less than the preset trust degree discrimination threshold, it is determined that the target recognition method is not stable, and the target recognition method is stopped.

5. The target positioning method based on the visual space-time operator according to claim 1, wherein: In step S2, the input image of the current human-computer interaction page is accurately recognized based on the target recognition method to determine the number of targets in the image; ​ Mark the target in the detection frame and generate gesture prompt words, realize motion tracking by locking the arm position, and the arm position is the key point coordinate obtained by the pose estimation algorithm; In adjacent time frames, the spatial displacement of the arm position is obtained by calculating the displacement of the key point coordinates of the arm position between adjacent frames; Divide the spatial displacement of the arm position by the inter-frame time interval to obtain the instantaneous speed as the gesture motion speed; Calculate the variance of the gesture motion speed in the preset time window as the gesture speed specification feature.

6. The target positioning method based on a visual space-time operator according to claim 1, characterized in that: In step S3, a preset acquisition window is divided into multiple acquisition time points, and the spatial coordinates of the arm end at each acquisition time point are identified by a pose estimation algorithm, and the gesture motion length between adjacent acquisition time points is calculated by the Euclidean distance method; Divide each gesture motion length by the corresponding time interval to obtain the length rate, and calculate the length rate stability index of the gesture motion based on the average value and standard deviation of each length rate; The difference between the maximum value and the minimum value of the length rate is divided by the maximum value of the length rate to obtain the burst degree of the gesture motion; After the length rate stability index of the gesture motion is standardized, the burst degree is calculated by the geometric mean method to obtain the gesture trajectory specification feature.

7. The target positioning method based on a visual space-time operator according to claim 6, characterized in that: In step S3, the difference between the ratio of the gesture trajectory specification feature to the preset trajectory specification threshold value and 1 is obtained as the trajectory margin; The difference between the ratio of the gesture speed specification feature to the preset speed specification threshold value and 1 is obtained as the speed margin; The sum of the speed margin and the trajectory margin is taken as the gesture motion intensity; Based on the gesture motion intensity, the gesture motion intensity coefficient is obtained by an exponential function.

8. The target positioning method based on a visual space-time operator according to claim 7, characterized in that: In step S4, the target subjects are sorted in descending order according to the gesture motion intensity coefficient and combined into a target subject set, and the sorting indexes are assigned in turn according to the sorting positions of the target subjects in the target subject set, and the reciprocal of the sorting index is taken as the target weight of the corresponding target subject; The page proportion of each target subject can be obtained by a target recognition method to obtain the pixel area occupied by the target subject in the human-computer interaction page, and the ratio of the pixel area to the total pixel area of the human-computer interaction page is taken as the page proportion of the target subject; The product of the page proportion of the target subject and the target weight is taken as the comprehensive weight index.

9. The target positioning method based on a visual space-time operator according to claim 8, characterized in that: In step S4, the target subject corresponding to the maximum value of the comprehensive weight index is taken as the first subject, the arithmetic mean of the comprehensive weight indexes of all target subjects except the first subject is taken, and the difference between the comprehensive weight index of the first subject and the arithmetic mean is obtained as the weight difference value; The weight difference value is compared with the preset difference threshold value to determine whether to execute the multi-target subject recognition mode: If the weight difference value is lower than the preset difference threshold value, it is determined to execute the multi-target subject recognition mode; Conversely, the single-target subject identification mode is executed on the first subject.

Images