Intelligent interactive projection method and interactive projection device

Abstract

The invention discloses an intelligent interactive projection method and an interactive projection device, relates to the technical field of interactive projection, and solves the technical problems that the recognition accuracy is low, key features of user actions cannot be accurately detected by an algorithm, and interaction is affected. For static action signals, the signals are generated by detecting detailed features of user gestures and multi-dimensional information of specific objects, compared with some simple action detection methods, static actions can be recognized more accurately, and for real-time action signals, the accuracy of real-time action recognition is improved by comprehensively capturing action features, so that the accuracy of real-time action recognition is improved. An advanced computer vision algorithm and a signal processing technology are adopted to efficiently process image data. In the real-time action signal processing process, the real-time requirement is met through rapid preprocessing, feature extraction and segmentation processing of the image, and the interaction fluency is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of interactive projection, and specifically to an intelligent interactive projection method and an interactive projection device. Background Art

[0002] With the development of the intelligent era, the human-computer interaction technology has become a research hotspot. The action recognition technology based on computer vision has emerged as the times require. By capturing the user's actions through a camera and using computer vision algorithms to analyze and recognize the actions, corresponding signals are generated to achieve human-computer interaction. This technology has a wide range of application requirements in many fields such as smart home, intelligent teaching, and intelligent medical rehabilitation, and can provide users with a more convenient, efficient, and natural interaction experience.

[0003] According to the patent application with the publication number CN106774827B, this patent discloses a projection interaction method, a projection interaction device, and an intelligent terminal. The method includes: receiving a gesture image captured by a camera during projection; obtaining a projection image at the same moment as the gesture image and magnifying the projection image to the same size as the gesture avatar; comparing the magnified projection image with the gesture image, and intercepting the different regions in the two images as the target region; performing static gesture recognition on the target region, extracting the gesture in the target region, matching the gesture with a preset gesture template, and obtaining the instruction corresponding to the gesture; using the obtained instruction to control the projection process.

[0004] However, when some existing interactive projection methods are used, traditional action recognition algorithms are difficult to accurately extract and recognize user actions, resulting in a low recognition accuracy rate, making the algorithm unable to accurately detect the key features of user actions, thus resulting in misjudgment. At the same time, when matching the recognized actions with a preset action library, it is easy to have inaccurate matching or no matching, further affecting the overall interaction. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides an intelligent interactive projection method and an interactive projection device, which solve the problems of low recognition accuracy rate, the algorithm being unable to accurately detect the key features of user actions, and affecting the interaction.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: An intelligent interactive projection method, which specifically includes the following steps:

[0007] Capturing the user's actions through a camera, analyzing and recognizing them using computer vision algorithms, and generating static or real-time action signals;

[0008] Analyze the generated static action signal, crop the user action image to obtain a cropped image, and extract features to obtain a feature action. At the same time, compare it with the gesture image to generate a comparison match or non-match signal;

[0009] Analyze the comparison match signal to determine the match result of the gesture image and generate interaction information. Analyze the comparison non-match signal, select the gesture image with the highest similarity as the preselected action, and combine it with the historical interaction record to judge and determine the standard match result to generate interaction information;

[0010] Analyze the real-time action signal, preprocess the user action image to obtain a preprocessed image, extract features to obtain the user action features, segment according to the action nodes to obtain segmented actions, select the preselected pixel points according to the pixel values of the segmented action pixel points, and use them as the standard to generate the user action features. At the same time, combine them to obtain the preprocessed action, and then compare it with the gesture image to generate interaction information or a secondary analysis signal;

[0011] Process the generated secondary analysis signal, match the segmented action with the gesture image to obtain a match result, select the result to be analyzed according to the proportion of the number of segmented actions, and combine the historical interaction record to determine the interaction standard and generate interaction information.

[0012] As a further solution of the present invention, the specific method for generating a static or real-time action signal is:

[0013] Obtain an action image sequence, use computer vision algorithms to identify the image sequence. If the user action is recognized as a gesture in a relatively static state, generate a static action signal. Conversely, if the user action is recognized as a gesture generated during movement, generate a real-time action signal.

[0014] As a further solution of the present invention, the specific method for analyzing the generated static action signal is:

[0015] Obtain the user action image and crop it to obtain a cropped image. At the same time, extract the features of the user action in the cropped image to obtain a feature action, and compare and match the feature action with the gesture image;

[0016] If the feature action exists in the gesture image, generate a comparison match signal. Conversely, if the feature action does not exist in the gesture image, generate a comparison non-match signal, and analyze both respectively.

[0017] As a further solution of the present invention, the specific method for analyzing the comparison match signal and the comparison non-match signal is:

[0018] Analyze the comparison match signal, obtain the corresponding match result in the gesture image, and use the match result as the standard to generate interaction information;

[0019] For the analysis of the non-matching signal of the comparison pair, calculate the similarity between the characteristic action and the gesture image, select the gesture image corresponding to the maximum similarity and record it as the preselected action. Then obtain the historical interaction record and judge the relationship between the preselected action and the historical interaction record;

[0020] If the preselected action exists in the historical interaction record, interact with it as the standard to generate interaction information. Conversely, if the preselected action does not exist in the historical interaction record, generate secondary interaction information.

[0021] As a further solution of the present invention, the specific method for analyzing the real-time action signal is as follows:

[0022] Obtain the user action image corresponding to the real-time action, and perform grayscale conversion, noise reduction, and normalization processing to obtain a preprocessed image. At the same time, extract the features of the preprocessed image and record them as user action features, and segment them according to action nodes to obtain segmented actions, and screen out preselected pixel points according to the pixel values of the pixel points of the segmented actions.

[0023] As a further solution of the present invention, the specific method for screening out the preselected pixel points is as follows:

[0024] Obtain the action image corresponding to the segmented action, and perform single-frame segmentation processing to obtain single-frame pictures. Compare the pixel points and corresponding pixel values of the user action features in the single-frame pictures, compare the obtained pixel values with a preset value, and the specific value of the preset value is set by the operator, and screen out the pixel points whose pixel values meet the preset value and record them as preselected pixel points, and at the same time eliminate the non-conforming pixel points.

[0025] As a further solution of the present invention, the specific method for obtaining the preprocessed action to generate interaction information or secondary analysis signals is as follows:

[0026] Generate preprocessing features based on the preselected pixel points, and by analogy, obtain the preprocessing features of all segmented actions. At the same time, combine the preprocessing features to obtain a preprocessed action;

[0027] Compare and analyze the preprocessed action with the gesture image. If the preprocessed action exists in the gesture image, obtain the corresponding matching image and generate interaction information. Conversely, if the preprocessed action does not exist in the gesture image, generate secondary analysis signals.

[0028] As a further solution of the present invention, the specific method for processing the generated secondary analysis signals is as follows:

[0029] Obtain all segmented actions, match them with the gesture images to get the corresponding matching results. At the same time, calculate the proportion of the number of segmented actions in the matching results, and compare the proportion of the number of matching results with the screening threshold. Screen the matching results with a proportion greater than the screening threshold and record them as the results to be analyzed;

[0030] Obtain the historical interaction records corresponding to the results to be analyzed. At the same time, obtain the number of interactions corresponding to the results to be analyzed, sort them in descending order according to the number of interactions, calculate the similarity between the gesture images and the real-time actions in the results to be analyzed in turn, and select the result to be analyzed with the largest similarity as the interaction standard to generate interaction information.

[0031] An intelligent interactive projection device, which includes a camera, a data processing center and a projection module, and the data processing center includes an image processing module, an action analysis module and an interaction information generation module;

[0032] The camera is used to capture the user action image between the camera and the projection surface after being turned on and send it to the data processing center;

[0033] The data processing center is used to analyze the obtained user action images, identify user actions to generate static or real-time action signals, and transmit both to the action analysis module;

[0034] The action analysis module is used to analyze the obtained static or real-time action signals. For the analysis of static action signals, the cropped image is obtained by cropping the user action image, and the feature action is obtained by feature extraction. At the same time, it is compared with the gesture image to generate a comparison match or comparison mismatch signal. For the analysis of the comparison match signal, the matching result of the gesture image is determined to generate interaction information. For the analysis of the comparison mismatch signal, the gesture image with the largest similarity is screened and recorded as the preselected action, and it is judged in combination with the historical interaction record to determine the standard matching result to generate interaction information and transmit it to the interaction information generation module;

[0035] For the analysis of real-time action signals, the preprocessing image is obtained by preprocessing the user action image, and the user action feature is obtained by feature extraction. At the same time, it is segmented according to the action nodes to obtain segmented actions, the preselected pixel points are screened according to the pixel values of the pixel points of the segmented actions, and the user action feature is generated with them as the standard. At the same time, they are combined to obtain the preprocessing action, and then compared with the gesture image to generate interaction information or a secondary analysis signal;

[0036] For the analysis of the secondary analysis signal, the segmented actions are matched with the gesture images to obtain the matching results. At the same time, the results to be analyzed are screened according to the proportion of the number of segmented actions, and the interaction standard is determined in combination with the historical interaction record to generate interaction information and transmit it to the interaction information generation module;

[0037] The interaction information generation module transmits the acquired interaction information to the projection module;

[0038] The projection module is used to acquire interaction information, generate interaction instructions, and perform projection control according to the instructions.

[0039] The present invention provides an intelligent interaction projection method and an interaction projection device. Compared with the prior art, it has the following beneficial effects:

[0040] In the present invention, when processing static action signals, the user action image is cropped to accurately retain the action area and reduce interference from irrelevant information; at the same time, an advanced feature extraction algorithm is used to extract the characteristic actions that can represent the action, improving the accuracy of feature extraction.

[0041] For static action signals, when comparing with the gesture image library, it can not only judge whether there is a complete match, but also calculate the similarity between the characteristic actions and all gesture images when there is no match, and make a judgment in combination with the historical interaction records, improving the accuracy and flexibility of the match.

[0042] For the processing of real-time action signals, after obtaining the preprocessed action, it is compared and analyzed with the gesture image, considering the similarity of multiple key features such as action contour and posture. At the same time, in the secondary analysis signal processing, by matching the segmented actions with the gesture images and screening the results to be analyzed according to the quantity ratio, and determining the interaction criteria in combination with the historical interaction records, the action classification and matching mechanism is further optimized, improving the matching success rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flowchart of the method steps of the present invention;

[0044] Figure 2 It is a block diagram of the device modules of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0046] Embodiment 1. Please refer to Figure 1 , the present application provides an intelligent interaction projection method, which specifically includes the following steps:

[0047] Step S1,

[0048] The camera is used to capture the user's actions in real time. After obtaining the action image sequence, advanced computer vision algorithms are used to deeply analyze and recognize these image data. Through a series of complex operations such as image processing, feature extraction, and pattern matching, corresponding action signals are generated, which can be specifically divided into static action signals and real-time action signals.

[0049] The generation of static action signals occurs when the algorithm analyzes the user's actions. If the user's actions are recognized as gestures in a relatively static state or the position and posture of a specific object, such signals will be output.

[0050] For example, when the user makes an "OK" gesture in front of the camera, the computer vision algorithm determines that this is an "OK" gesture by detecting features such as the degree of finger bending and the position relationship of the fingertips, thereby generating a corresponding static action signal indicating that the user has made the specific gesture of "OK".

[0051] Another example is that when the camera captures a cup on the table, the algorithm analyzes information such as the position, angle, and contour of the cup in the image to determine that the cup is in an upright position and located in the upper left corner of the table. At this time, a static action signal will also be generated to describe the position and posture of this specific object, the cup.

[0052] Real-time action signals are generated when the algorithm recognizes that the user's actions are in a dynamic movement process. For example, when the user performs a running action in front of the camera, the algorithm will track the position changes of various parts of the user's body (such as the legs and arms) in consecutive images, calculate parameters such as their movement trajectories, speeds, and accelerations, and then generate real-time action signals to represent that the user is in the process of running.

[0053] Another example is that when the user waves a remote control in the air, the algorithm can capture the position changes of the remote control at different times and generate real-time action signals to reflect the movement state of the remote control and its related movement characteristics.

[0054] Step S2

[0055] In the process of recognizing and interacting with the user's actions, first, an in-depth analysis is carried out on the generated static action signals. Through specific signal processing algorithms, the static action signals are converted into corresponding user action images. Next, a cropping operation is performed on the user action images. The purpose is to accurately delete the irrelevant and redundant areas in the images and only retain the action areas directly related to the user's actions, thereby obtaining cropped images.

[0056] After obtaining the cropped image, an advanced feature extraction algorithm is used to extract the features of the user's actions in the cropped image, so as to obtain the characteristic actions that can represent the actions. Subsequently, the characteristic actions are compared and matched one by one with the gesture images in the pre-constructed gesture image library. During the comparison process, if it is found that the current characteristic action exactly matches a certain image in the gesture image library, a comparison matching signal is generated; if no matching gesture image is found, a comparison non-matching signal is generated.

[0057] For example, assume that there are gesture images such as "fist clenching", "thumbs up", and "waving" in the gesture image library. When the characteristic action is exactly the same as the characteristics of the "thumbs up" gesture image, a comparison matching signal will be generated; if the characteristic action does not match any gesture image in the library, a comparison non-matching signal is generated.

[0058] When a comparison matching signal is generated, it is further analyzed. By parsing the signal, the corresponding matching result is obtained from the gesture image library. Based on this matching result, the corresponding interaction information is generated. For example, if the matching result is the "thumbs up" gesture, the interaction information may be to display a prompt of "thumbs up successfully" on the screen, or trigger a certain thumbs up-related function, such as liking a video or liking an article.

[0059] When a comparison non-matching signal is generated, it also needs to be analyzed. At this time, the similarity between the characteristic action and all gesture images in the gesture image library is calculated. During the calculation process, a specific similarity calculation algorithm, such as the cosine similarity algorithm, is used to calculate the similarity between each gesture image and the characteristic action, and they are sorted in descending order of similarity. The gesture image with the largest similarity is selected and denoted as the preselected action. Then, the historical interaction records are obtained, and the preselected action is matched and judged with the actions in the historical interaction records.

[0060] If an action consistent with the preselected action is found in the historical interaction records, the interaction is carried out based on the preselected action, and the corresponding interaction information is generated. For example, if there are multiple "waving" action records in the historical interaction records, and the current preselected action is judged to be "waving" as well, the interaction information may be to activate a function related to waving, such as controlling the intelligent device to turn on or off.

[0061] On the contrary, if the preselected action does not exist in the historical interaction records, secondary interaction information is generated to prompt the user to make the action again, or provide some guiding information to help the user adjust the action to obtain a more accurate matching result.

[0062] Step S3

[0063] After generating the real-time action signal, it is first analyzed in depth. Through signal parsing and conversion technologies, a series of user action images corresponding to the real-time action are obtained. These images completely record the posture changes of the user during the action. For the convenience of subsequent processing, preprocessing operations are performed on the obtained user action images, specifically including grayscale conversion, noise reduction, and normalization.

[0064] Grayscale conversion converts a color image into a grayscale image, reducing the data dimension while retaining the key information in the image. For example, converting a color image of a user's running action into a grayscale image makes subsequent processing focus more on the action contour itself rather than color information. Noise reduction uses algorithms such as median filtering and Gaussian filtering to remove noise interference in the image, ensuring the clarity and accuracy of the image. For example, removing the noise points generated by the camera's photosensitive element noise or transmission interference makes the action details clearer. Normalization maps the image pixel values to a specific range, such as [0, 1] or [-1, 1], to eliminate the influence of differences in brightness and contrast between different images on subsequent analysis, making action images taken at different times and under different lighting conditions comparable. After this series of preprocessing operations, a preprocessed image is obtained.

[0065] Immediately afterwards, feature extraction is performed on the preprocessed image. Through algorithms such as edge detection and contour extraction, user action features are obtained. During this process, the obtained user action features are presented in the form of action contours. For example, for an action image of a user waving, after feature extraction, the contour shape of the arm waving can be clearly outlined. For further refined analysis, the obtained user action features are segmented according to action nodes. Action nodes can be key positions with significant posture changes during the action. For example, when a user does a push-up, the starting position where the arm is straight, the position where it bends to the lowest point, and the ending position where it straightens again, etc. By dividing the action nodes, the continuous action features are segmented into multiple stages, obtaining segmented actions.

[0066] For each segmented action, the corresponding action image is obtained, and single-frame segmentation processing is performed on this action image, splitting the continuous action image sequence into single-frame pictures one by one. This allows for more detailed analysis of each frame of the image. In a single-frame picture, the pixel points of the user action features are obtained, and the pixel values corresponding to each pixel point are obtained one by one.

[0067] For example, in a certain single-frame picture, a pixel point on the action contour of the user's arm may have a pixel value within the range of [0, 255] after grayscale conversion.

[0068] Compare the obtained pixel values with a preset value, which is set by the operator according to actual requirements and scenarios, and the preset value is a range value. For example, the operator sets the pixel value range to [100, 180] according to the expected specific action characteristics.

[0069] The system will filter out the pixel points whose pixel values meet the preset value range, record them as preselected pixel points, and at the same time eliminate the pixel points that do not meet this range. Based on these preselected pixel points, a user action feature is regenerated and recorded as a preprocessing feature. In the same way, each segmented action is processed as described above to obtain its corresponding preprocessing feature. Finally, all the preprocessing features are combined to form a complete preprocessing action.

[0070] After completing the construction of the preprocessing action, compare and analyze it with the gesture image. The gesture image can be a pre-stored standard action image library or a reference action image set in a specific scenario. If there is a match between the preprocessing action and an action in the gesture image, that is, the key features such as the action contour and posture are highly similar, the system obtains the corresponding matching image and generates interaction information according to the preset interaction rules.

[0071] For example, in a smart home control scenario, various gesture action images for controlling household appliances are stored in the gesture image library. When the waving action made by the user after processing matches the waving action for controlling the TV switch in the library, the system obtains this matching image and generates interaction information to turn on or off the TV. Conversely, if no match is found for the preprocessing action in the gesture image, a secondary analysis signal is generated.

[0072] Step S4

[0073] First, extract all relevant segmented actions from the signal. These segmented actions are the results of dividing the user's real-time action characteristics according to action nodes before, and each segmented action represents a key stage in the user's action process.

[0074] Next, for each segmented action, comprehensively match it with the gesture images in the pre-constructed gesture image library. The matching process uses an advanced image recognition algorithm to compare the similarity of the action contour, posture, and other key features of the segmented action with the gesture images. After the matching operation, a series of corresponding matching results are obtained, and each matching result indicates that a certain segmented action has a certain degree of similarity with a specific gesture image in the gesture image library.

[0075] Then, calculate the proportion of the number of segmented actions in each matching result. For example, assume that in a real-time action analysis, a total of 10 segmented actions are obtained, and 3 of them match a specific gesture image successfully. Then the proportion of the number of this matching result is 30% (3÷10×100%). Sort all the matching results in descending order according to the proportion of the number, so that those results with a higher degree of matching with the real-time action segments can be quickly screened out.

[0076] After that, compare the proportion of the number of the sorted matching results with the screening threshold set by the operator according to actual needs. The screening threshold is a key parameter, which determines which matching results have sufficient credibility and are worthy of further analysis. For example, the operator sets the screening threshold to 20% according to experience and the requirements for system accuracy. The system will automatically screen out the matching results with a proportion of the number greater than the screening threshold and record these results as the results to be analyzed. Through this step, those results with a lower degree of matching with the real-time action and likely to be mis-matched can be excluded, thereby narrowing the scope of subsequent analysis and improving the processing efficiency.

[0077] After obtaining the results to be analyzed, the system will obtain the historical interaction records corresponding to each result to be analyzed. The historical interaction records contain the interaction behavior data of the user in similar action scenarios in the past, and these data have important reference value for understanding the intention of the current real-time action. At the same time, the system will also count the number of interactions corresponding to each result to be analyzed, and the number of interactions reflects the frequency of occurrence of this matching result in history. Sort these results to be analyzed in descending order according to the number of interactions, and give priority to those matching results that have appeared more frequently in history, because they are more likely to represent the current true intention of the user.

[0078] Finally, calculate the similarity between the gesture image and the real-time action in the results to be analyzed in turn. The similarity calculation uses a more precise algorithm, comprehensively considering multiple factors such as the time series, spatial position, and posture change of the action. After calculating the similarity of all results to be analyzed, select the result to be analyzed with the maximum similarity as the interaction standard. Based on this interaction standard, the system generates corresponding interaction information according to the preset interaction rules.

[0079] Embodiment 2, please refer to Figure 2 , this application provides an intelligent interactive projection device, including a camera, a data processing center, and a projection module, wherein the data processing center includes an image processing module, an action analysis module, and an interaction information generation module.

[0080] The camera is used to capture the user action image between the camera and the projection surface after being turned on and send it to the data processing center;

[0081] The data processing center is used to analyze the acquired user action images, identify user actions to generate static or real-time action signals, and transmit both to the action analysis module;

[0082] The action analysis module is used to analyze the acquired static or real-time action signals. For the analysis of static action signals, the user action image is cropped to obtain a cropped image, and feature extraction is performed to obtain a feature action. At the same time, it is compared with the gesture image to generate a comparison match or comparison mismatch signal. For the analysis of the comparison match signal, the matching result of the gesture image is determined to generate interaction information. For the analysis of the comparison mismatch signal, the gesture image with the highest similarity is selected and recorded as a preselected action, and it is judged in combination with the historical interaction record to determine the standard matching result and generate interaction information, and then it is transmitted to the interaction information generation module;

[0083] For the analysis of real-time action signals, the user action image is preprocessed to obtain a preprocessed image, and features are extracted to obtain user action features. At the same time, it is segmented according to action nodes to obtain segmented actions. The preselected pixel points are filtered according to the pixel values of the segmented action pixel points, and user action features are generated based on them. At the same time, they are combined to obtain a preprocessed action, and then it is compared with the gesture image to generate interaction information or a secondary analysis signal;

[0084] For the analysis of the secondary analysis signal, the segmented action is matched with the gesture image to obtain a matching result. At the same time, the result to be analyzed is filtered according to the quantity ratio of the segmented actions, and the interaction standard is determined in combination with the historical interaction record to generate interaction information, and then it is transmitted to the interaction information generation module;

[0085] The interaction information generation module transmits the acquired interaction information to the projection module;

[0086] The projection module is used to obtain interaction information and generate interaction instructions, and perform projection control according to the instructions.

[0087] For some data in the above formula, only their numerical values are taken for calculation, and the parameter units are not substituted for calculation. At the same time, the content not described in detail in this specification belongs to the prior art well known to those skilled in the art.

[0088] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent interactive projection method, characterized in that, The method specifically includes the following steps: Capture the user's actions through a camera, analyze and identify them using computer vision algorithms, and generate static or real-time action signals; Analyze the generated static action signals, crop the user action images to obtain cropped images, and perform feature extraction to obtain feature actions. At the same time, compare them with gesture images to generate comparison matching or non-matching signals; Analyze the comparison matching signals to determine the matching results of the gesture images and generate interaction information. Analyze the comparison non-matching signals, select the gesture image with the highest similarity and record it as the preselected action, and combine the historical interaction records to judge and determine the standard matching results to generate interaction information; Analyze the real-time action signals, preprocess the user action images to obtain preprocessed images, extract features to obtain user action features, and segment them according to action nodes to obtain segmented actions. Screen the preselected pixel points according to the pixel values of the segmented action pixel points, and generate user action features based on them. At the same time, combine them to obtain preprocessed actions, and then compare them with gesture images to generate interaction information or secondary analysis signals; Process the generated secondary analysis signals, match the segmented actions with the gesture images to obtain matching results, and screen the results to be analyzed according to the proportion of the number of segmented actions. Combine the historical interaction records to determine the interaction criteria and generate interaction information.

2. The intelligent interactive projection method according to claim 1, wherein The specific method for generating static or real-time action signals is as follows: Obtain the action image sequence, use computer vision algorithms to identify the image sequence. If the user's action is recognized as a gesture in a relatively static state, generate a static action signal; otherwise, if the user's action is recognized as a gesture generated during movement, generate a real-time action signal.

3. An intelligent interactive projection method according to claim 1, characterized in that, The specific method for analyzing the generated static action signals is as follows: Obtain the user action images and crop them to obtain cropped images. At the same time, extract the features of the user actions in the cropped images to obtain feature actions, and compare and match the feature actions with the gesture images; If the feature action exists in the gesture image, generate a comparison matching signal; otherwise, if the feature action does not exist in the gesture image, generate a comparison non-matching signal, and analyze both of them respectively.

4. An intelligent interactive projection method according to claim 3, characterized in that, The specific methods for analyzing the comparison matching signals and comparison non-matching signals are as follows: Analyze the comparison matching signals, obtain the corresponding matching results in the gesture images, and generate interaction information based on the matching results; Analyze the comparison non-matching signals, calculate the similarity between the feature action and the gesture image, select the gesture image with the highest similarity and record it as the preselected action. Then obtain the historical interaction records and judge the relationship between the preselected action and the historical interaction records; If the preselected action exists in the historical interaction records, interact based on the preselected action and generate interaction information; otherwise, if the preselected action does not exist in the historical interaction records, generate secondary interaction information.

5. An intelligent interactive projection method according to claim 1, characterized in that, The specific method for analyzing the real-time action signals is as follows: Obtain the user action image corresponding to the real-time action, and perform grayscale conversion, noise reduction, and normalization processing to obtain a preprocessed image. At the same time, extract the features of the preprocessed image and record them as user action features, and segment them according to action nodes to obtain segmented actions, and screen the preselected pixel points according to the pixel values of the segmented action pixel points.

6. The intelligent interactive projection method according to claim 5, wherein The specific method for screening and obtaining the preselected pixel points is as follows: Obtain the action image corresponding to the segmented action, and perform single-frame segmentation processing to obtain single-frame pictures. For the pixel points and corresponding pixel values of the user action features in the single-frame pictures, compare the obtained pixel values with a preset value, and the specific value of the preset value is set by the operator, and screen the pixel points whose pixel values meet the preset value and record them as preselected pixel points, and at the same time eliminate the pixel points that do not meet the requirements.

7. An intelligent interactive projection method according to claim 1, characterized in that The specific method for obtaining the preprocessed action to generate interaction information or secondary analysis signals is as follows: Generate preprocessing features based on the preselected pixel points, and by analogy, obtain the preprocessing features of all segmented actions. At the same time, combine the preprocessing features to obtain a preprocessed action; Compare and analyze the preprocessed action with the gesture image. If the preprocessed action exists in the gesture image, obtain the corresponding matching image and generate interaction information. On the contrary, if the preprocessed action does not exist in the gesture image, generate a secondary analysis signal.

8. An intelligent interactive projection method according to claim 1, characterized in that, The specific method for processing the generated secondary analysis signal is as follows: Obtain all segmented actions, match them with the gesture image to obtain the corresponding matching results, and at the same time calculate the proportion of the number of segmented actions in the matching results, and compare the proportion of the number of the matching results with the screening threshold, and screen the matching results with the proportion of the number greater than the screening threshold and record them as the results to be analyzed; Obtain the historical interaction records corresponding to the results to be analyzed, and at the same time obtain the number of interactions corresponding to the results to be analyzed, and sort them from largest to smallest according to the number of interactions. Calculate the similarity between the gesture image and the real-time action in the results to be analyzed in turn, and select the result to be analyzed with the largest similarity as the interaction standard to generate interaction information.

9. An intelligent interactive projection device for performing the intelligent interactive projection method according to any one of claims 1-8, characterized in that, The device includes a camera, a data processing center, and a projection module, and the data processing center includes an image processing module, an action analysis module, and an interaction information generation module; The camera is used to capture the user action image between the camera and the projection surface after being turned on and send it to the data processing center; The data processing center is used to analyze the acquired user action image, identify the user action to generate static or real-time action signals, and transmit both to the action analysis module; The action analysis module is used to analyze the acquired static or real-time action signals. For the analysis of static action signals, crop the user action image to obtain a cropped image, and perform feature extraction to obtain a feature action. At the same time, compare it with the gesture image to generate a comparison match or comparison mismatch signal. For the analysis of the comparison match signal, determine the matching result of the gesture image to generate interaction information. For the analysis of the comparison mismatch signal, screen the gesture image with the largest similarity and record it as a preselected action, and judge in combination with the historical interaction record to determine the standard matching result to generate interaction information, and transmit it to the interaction information generation module; Analyze real-time action signals, preprocess the user action image to obtain a preprocessed image, extract features to obtain user action features, segment according to action nodes to obtain segmented actions, screen preselected pixel points based on the pixel values of the segmented action pixel points, generate user action features based on them, and combine them to obtain preprocessed actions. Then, compare with the gesture image to generate interaction information or secondary analysis signals; Analyze the secondary analysis signals, match the segmented actions with the gesture image to obtain a matching result, screen the results to be analyzed according to the proportion of the number of segmented actions, determine the interaction criteria in combination with historical interaction records, generate interaction information, and transmit it to the interaction information generation module; Interaction information generation module, which transmits the obtained interaction information to the projection module; Projection module, used to obtain interaction information and generate interaction instructions, and perform projection control according to the instructions.

Images