A window intelligent control method and system based on indoor activities

Abstract

The application provides a window intelligent control method and system based on indoor activities, and relates to the technical field of intelligent window control. The method comprises the following steps: real-time monitoring of confidence indicators and manual intervention signals; when the confidence indicators are lower than a preset threshold or a manual intervention signal is received, the visual feature information and spatial position information of the newly added shielding entity are extracted, and the functional attributes of the shielding entity are determined in combination with a preset semantic mapping relationship; and a window adjustment instruction is generated based on the environmental requirements implied by the functional attributes to control the window execution mechanism to execute the window control strategy. The method aims to solve the problems of the intelligent window control system adopting a mismatched control strategy and the system's perception barrier being aggravated due to the introduction of physical partitions, can automatically perceive the physical partitions set by the user, and adjust the window control strategy according to the functional attributes, thereby achieving the effects of improving the control accuracy and meeting the environmental requirements in a composite activity scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent window control, in particular to a window intelligent control method and system based on indoor activities. BACKGROUND

[0002] In modern intelligent building environments, window intelligent control systems have become indispensable key technologies for improving residential comfort, energy efficiency, and security. Such systems automatically adjust window states by identifying indoor activity types, and perform well in single-activity scenarios. However, with the popularity of open and multi-functional spaces, the emergence of composite activity scenarios poses fundamental challenges to existing systems.

[0003] In composite activity scenarios, multiple different activities overlap in space and time, making it difficult for systems to accurately identify the dominant activity. For example, in an open office environment, a video conference and light exercise may occur simultaneously, and the system cannot effectively distinguish the priority of these activities. This identification ambiguity forces the system to adopt a general strategy, often failing to meet the environmental needs of specific activities. When the system mistakenly identifies a video conference as a general activity and opens the window, external noise and air flow can severely interfere with the conference quality.

[0004] Temporary measures taken by residents to correct system errors, such as setting up physical barriers such as screens or plants, while temporarily solving the problem, inadvertently create new technical difficulties. These temporary barriers block the field of view of indoor activity detectors, resulting in incomplete environmental information obtained by the system. Information loss further exacerbates the difficulty of activity recognition, forming a vicious cycle: system errors due to incorrect identification lead to inappropriate operations, and user correction behaviors weaken the system's perception ability, ultimately making the system completely lose the possibility of self-correction.

[0005] Therefore, in open or multi-functional indoor spaces, when intelligent window control systems apply mismatched default security strategies due to their activity recognition models failing to accurately distinguish parallel composite indoor activities, resulting in external interference affecting the normal conduct of high-priority activities, how to enable the system to effectively cope with temporary physical barriers that residents introduce to alleviate the above interference, which inadvertently partially block the field of view of indoor activity detectors, so that in the case of incomplete sensor information or blind areas, the system can still accurately infer the true composite activity type and its priority, and adaptively adjust the window control strategy to meet the differentiated safety, comfort, and privacy needs of different residents in the same space, avoiding the system falling into a vicious cycle of further exacerbating perception barriers due to user correction behaviors, has become a technical problem that needs to be solved. SUMMARY

[0006] The purpose of the present application is to provide a window intelligent control method and system based on indoor activities, aiming to solve the problem that in open or multi-functional indoor spaces, the intelligent window control system cannot accurately distinguish between composite indoor activities due to the activity recognition model, leading to mismatched control strategies, and the physical partition introduced by the user to correct the system error further exacerbates the system's perception barrier. The present application can automatically perceive the physical partition set by the user and adjust the window control strategy according to its functional attributes, thereby improving the control accuracy and meeting the environmental requirements in a composite activity scenario.

[0007] In a first aspect, the present application provides a window intelligent control method based on indoor activities, for a window intelligent control system, which at least includes a visual acquisition unit and a window execution mechanism, comprising the following steps:

[0008] S1. Real-time monitoring of the confidence index of the indoor activity recognition result and the manual intervention signal of the window execution mechanism;

[0009] S2. When the confidence index is lower than the preset threshold or the manual intervention signal is received, controlling the visual acquisition unit to acquire indoor environment images, and detecting whether a new shielding entity appears in the indoor activity area based on the indoor environment images;

[0010] S3. When a new shielding entity appears in the indoor activity area, extracting the visual feature information and spatial position information of the shielding entity, and determining the functional attribute of the shielding entity according to the visual feature information and spatial position information, combined with the preset semantic mapping relationship;

[0011] S4. When a shielding entity with a specific functional attribute is identified, the identification event is taken as a negative feedback signal to the current window control strategy, and the window control strategy is corrected based on the negative feedback signal, and a window adjustment instruction is generated based on the environmental requirements implied by the functional attribute, to control the window execution mechanism to execute the corrected window control strategy through the window adjustment instruction.

[0012] The window intelligent control method based on indoor activities provided by the present application can effectively deal with the physical partition introduced by the user to correct the system error. By identifying the functional attribute of the shielding entity, the identification event is taken as a negative feedback signal, and then the window control strategy is corrected, so that the window control strategy can still be adaptively adjusted in the case of incomplete sensor information, avoiding the vicious cycle of further exacerbating the perception barrier caused by the user's corrective behavior.

[0013] In a second aspect, the present application provides a window intelligent control system, comprising a visual acquisition unit and a window execution mechanism, further comprising:

[0014] a monitoring module, configured to monitor a confidence index of the indoor activity recognition result and a manual intervention signal of the window actuator in real time;

[0015] a control module, configured to control the visual acquisition unit to acquire an indoor environment image when the confidence index is lower than a preset threshold or the manual intervention signal is received, and detect whether a new shielding entity appears in an indoor activity area based on the indoor environment image;

[0016] a determination module, configured to extract visual feature information and spatial position information of the shielding entity when the new shielding entity appears in the indoor activity area, and determine a functional attribute of the shielding entity according to the visual feature information and the spatial position information in combination with a preset semantic mapping relationship;

[0017] a generation module, configured to take the shielding entity with the specific functional attribute as a negative feedback signal of a current window control strategy when the shielding entity with the specific functional attribute is recognized, correct the window control strategy based on the negative feedback signal, and generate a window adjustment instruction based on an environmental requirement implied by the functional attribute, so as to control the window actuator to execute the corrected window control strategy through the window adjustment instruction.

[0018] As can be seen from the above, the window intelligent control method based on indoor activity provided by the present application can automatically perceive the physical partition set by the user and adjust the window control strategy according to the functional attribute of the shielding entity, so as to improve the control precision and meet the environmental requirement in the complex activity scene.

[0019] Other features and advantages of the present application will be described in the following description, and become apparent from the description, or be learned from the practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the structure particularly pointed out in the written description and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flow chart of the window intelligent control method based on indoor activity provided by the embodiment of the present application.

[0021] Figure 2 A structural schematic diagram of the window intelligent control system provided by the embodiment of the present application.

[0022] Label explanation:

[0023] 100, monitoring module; 200, control module; 300, determination module; 400, generation module. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0025] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0026] With reference to the accompanying drawings of Figure 1 , the present application provides a window intelligent control method based on indoor activities, which is used for a window intelligent control system, and the window intelligent control system at least includes a visual acquisition unit and a window execution mechanism, and comprises the following steps:

[0027] S1. Real-time monitoring of a confidence index for an indoor activity recognition result, and a manual intervention signal of the window execution mechanism;

[0028] S2. When the confidence index is lower than a preset threshold or the manual intervention signal is received, controlling the visual acquisition unit to acquire an indoor environment image, and detecting whether a new shielding entity appears in an indoor activity area based on the indoor environment image;

[0029] S3. When the new shielding entity appears in the indoor activity area, extracting visual feature information and spatial position information of the shielding entity, and determining a functional attribute of the shielding entity according to the visual feature information and the spatial position information, in combination with a preset semantic mapping relationship; the semantic mapping relationship includes a corresponding logic between an entity category, a spatial topological relationship and a functional attribute;

[0030] S4. When the shielding entity with the specific functional attribute is recognized, taking the recognition event as a negative feedback signal for a current window control strategy, and correcting the window control strategy based on the negative feedback signal, and generating a window adjustment instruction based on an environmental demand implied by the functional attribute, so as to control the window execution mechanism to execute the corrected window control strategy through the window adjustment instruction.

[0031] For ease of understanding, some key terms in this embodiment are explained as follows:

[0032] Window Intelligent Control System: This system is an automated system that integrates various sensors, actuators, and control logic, aiming to intelligently adjust the window state based on indoor environment and user needs. Its core function is to autonomously decide and execute window opening, closing, ventilation, lighting, and privacy adjustment operations, to optimize living experience, improve energy efficiency, and ensure safety, by sensing indoor activities, environmental parameters, and user feedback.

[0033] Visual Acquisition Unit: As an important component of the window intelligent control system, the visual acquisition unit is mainly responsible for obtaining image or video data of the indoor environment. This unit usually contains one or more image sensors (such as wide-angle cameras) and can cooperate with embedded processors for preliminary image data processing. Its role is to provide visual information input for the system to perform indoor activity recognition, environmental monitoring, and occlusion entity detection tasks.

[0034] Window Actuator: The window actuator is a component in the window intelligent control system responsible for physically executing window actions. It is usually composed of a motor, transmission device (such as a DC motor with gear transmission), and corresponding control circuit, capable of receiving instructions from the control system and accurately controlling the opening angle, closing, locking, and other states of the window. In addition, the actuator may also integrate angle encoders or force sensors to feedback the real-time state of the window or detect manual intervention.

[0035] Confidence Index: The confidence index is a quantitative value that measures the system's certainty of indoor activity recognition results, usually represented as a score between 0 and 1. Higher confidence indicates that the system has stronger confidence in the recognition result, while lower confidence suggests that the recognition result may be ambiguous or uncertain. This index is a key parameter for evaluating the performance of activity recognition programs and can be used as a condition to trigger further environmental detection.

[0036] Manual Intervention Signal: The manual intervention signal is a signal generated by the user through direct operation of the window actuator (such as manually pushing or pulling the window, pressing the physical control button). This signal reflects the user's dissatisfaction with the current window control strategy or specific needs, and is an important way for the system to obtain implicit user feedback. The system can adjust its automated control strategy in a timely manner to better meet user intentions by detecting such signals.

[0037] Indoor Activity Area: The indoor activity area refers to the specific range of indoor space where users perform major activities (such as resting, working, meeting, and exercising). Defining this area helps the system focus on the space most relevant to user behavior, thereby improving the accuracy and efficiency of activity recognition and environmental monitoring.

[0038] Blocking entity: Blocking entity refers to objects that appear within the indoor activity area, are not fixed furniture, and may hinder sensor information collection. These entities are usually temporarily introduced by users to meet specific needs (e.g., privacy, sound insulation), such as screens, tall potted plants, mobile bookshelves, etc. Detecting and identifying these blocking entities is crucial for understanding user intent and correcting window control strategies.

[0039] Semantic mapping relationship: Semantic mapping relationship is a pre-set logical correspondence that links the physical characteristics of blocking entities (e.g., entity category, spatial topological relationship) with their implied functional attributes (e.g., privacy line-of-sight blocking, acoustic environment isolation). Through this mapping, the system can rise from physical entity recognition to understanding of user deep needs, thereby generating window adjustment instructions that better meet user intent.

[0040] Window control strategy: Window control strategy is a series of window adjustment rules pre-set by the system according to indoor activity types, environmental parameters, and user needs. These strategies define specific actions such as opening angle, ventilation mode, and locking state that windows should take in different scenarios. The system corrects these strategies to adapt to changing environments and user needs.

[0041] Window adjustment instruction: Window adjustment instruction is a specific operation command generated by the system according to the corrected window control strategy and sent to the window execution mechanism. These instructions guide the execution mechanism to perform window opening, closing, angle adjustment, etc., to achieve the expected environmental regulation effect.

[0042] The present application proposes an indoor activity-based window intelligent control method, which is used in a window intelligent control system. The window intelligent control system at least includes a visual acquisition unit and a window execution mechanism.

[0043] In step S1, the confidence indicator for the indoor activity recognition result is monitored in real time, as well as the manual intervention signal of the window actuator. The purpose of this step is to continuously assess the accuracy of the system in recognizing indoor activities, and to timely capture the direct feedback of the user on the window control. For example, the system can be configured with an activity recognition module that continuously analyzes data from various sensors (such as visual sensors, acoustic sensors) and outputs the recognition result of the current indoor activity and its confidence score. When the confidence score is below a pre-set threshold (e.g., 0.6), it indicates that the system is not confident enough in the current activity recognition, and there may be a misjudgment. At the same time, the system monitors whether the window is manually operated by the user through sensors (e.g., force sensors or angle encoders) on the window actuator. If it is detected that the user manually opens or closes the window, a manual intervention signal is generated. These monitoring mechanisms ensure that the system can timely discover its own limitations in recognition or the user's dissatisfaction with the current strategy.

[0044] In step S2, when the confidence indicator is below the pre-set threshold or the manual intervention signal is received, the control of the visual acquisition unit to obtain indoor environment images, and based on the indoor environment images to detect whether there are new shielding entities in the indoor activity area. The purpose of this step is to further explore the environmental changes through visual information when the system recognition is ambiguous or the user actively corrects. For example, when the confidence is below the threshold or the manual intervention signal is received in step S1, the system will immediately activate the visual acquisition unit (e.g., a wide-angle camera) to obtain real-time images of the current indoor environment. Subsequently, the image processing unit can analyze these images to detect whether there are new, non-fixed furniture objects in the indoor activity area. One implementation is that the system can pre-store a baseline environment model in a non-shielding state, which contains the spatial layout information of the indoor fixed furniture. When the new indoor environment image is obtained, the system will compare the image with the baseline environment model to extract the difference area. Then, object recognition is performed on the difference area, and if the recognition result is a non-human and non-fixed furniture object, it is determined that there is a new shielding entity.

[0045] In step S3, when a new occlusion entity appears in the indoor activity area, the visual feature information and spatial position information of the occlusion entity are extracted, and the functional attribute of the occlusion entity is determined according to the visual feature information and the spatial position information, in combination with a preset semantic mapping relationship. The semantic mapping relationship includes the corresponding logic between the entity category, the spatial topological relationship, and the functional attribute. The purpose of this step is to deeply understand the potential intention of the user introducing the occlusion entity. For example, once a new occlusion entity is detected in step S2, the system will further analyze the visual features (such as texture, color distribution, shape contour) and spatial position information (such as its specific coordinates in the indoor activity area, the relative position with the window execution mechanism) of the entity. The semantic mapping module inside the system stores a preset semantic mapping relationship, which associates common entity categories (such as screens, curtains, sound-absorbing panels, mobile walls) and their topological relationships in space (such as located between the indoor activity area and the window execution mechanism, surrounding the indoor activity area) with specific functional attributes (such as privacy line-of-sight blocking, acoustic environment isolation). By matching the extracted visual features and spatial position information of the occlusion entity with the semantic mapping relationship, the system can infer the functional attribute implied by the occlusion entity. For example, if a rectangular object with a folding texture and a height of about 1.8 meters is detected, and its spatial position information shows that it is located between the indoor activity area and the window execution mechanism, the system will determine its functional attribute as "privacy line-of-sight blocking" according to the semantic mapping relationship.

[0046] In step S4, when an occlusion entity with a specific functional attribute is identified, the identification event is taken as a negative feedback signal to the current window control policy, and the window control policy is revised based on the negative feedback signal, and a window adjustment instruction is generated based on the environmental requirement implied by the functional attribute, to control the window actuator to execute the revised window control policy. This step aims to convert the user's introduction of the occlusion entity's behavior into a basis for system learning and optimization. For example, when the system identifies a new occlusion entity and determines its functional attribute (e.g., "privacy view blocking") in step S3, the system considers it as a negative feedback to the current window control policy. The system records the environmental context information (such as external noise level, indoor light intensity, temperature, humidity) at the time of the negative feedback and the specific window policy (such as window opening angle, ventilation mode) executed before. These data are stored in the policy feedback database. When similar scenarios to the recorded negative feedback environmental context and activity mode are encountered in the future, the system will preferentially try to use a window control policy opposite to or more conservative than the policy that led to the negative feedback. For example, if the system previously opened the window due to the judgment of the need for ventilation, resulting in the introduction of external noise interference, causing the user to erect a screen (functional attribute: privacy view blocking), then when similar high-priority activities and environmental conditions are detected again in the future, the system may prefer to keep the window closed to ensure acoustic isolation and privacy. At the same time, the system generates a corresponding window adjustment instruction (e.g., closing the window, lowering the curtain) based on the environmental requirement implied by the functional attribute (e.g., privacy view blocking requirement), and executes the revised window control policy through the window adjustment instruction to control the window actuator. The system also continuously monitors whether the user manually intervenes again or introduces new occlusion entities, forming a "behavior-policy reverse learning cycle" to continuously optimize the window control policy.

[0047] The above technical solutions are further described in more detail through a more specific example as follows:

[0048] Suppose in an open office space, user A is conducting an important video conference, while in another area of the same space, user B is conducting light physical exercise. A traditional window intelligent control system may have a low confidence index for the activity recognition result due to the presence of two activities at the same time, or the system may mistakenly identify the entire scene as a low-priority "light exercise" and therefore open the window to increase ventilation. However, the opening of the window introduces external noise, which interferes with user A's video conference. User A manually closes the window to ensure the stability and focus of the conference environment, and places a movable screen near his conference area.

[0049] At this time, the window intelligent control method proposed in this application begins to play a role. In step S1, the system monitors in real time that the confidence index of the indoor activity recognition result is lower than the preset threshold (for example, the system's recognition confidence of "video conference" is only 0.4, which is lower than the threshold of 0.6), and receives a manual intervention signal of user A manually closing the window. These signals trigger the system to further detect the environment.

[0050] Then, in step S2, the system controls the vision acquisition unit to obtain indoor environment images. Based on these images, the system detects whether a new shielding entity appears in the indoor activity area. Specifically, the system calls a pre-stored baseline environment model in the unshielded state, which contains the spatial layout information of the fixed office desks and chairs in the office space. The system differentiates and compares the real-time indoor environment images obtained with the baseline environment model, successfully extracts the difference area where user A placed the screen. Then, the system performs object recognition on the difference area, and the recognition result shows an object that is not a human body and not a fixed furniture, thereby determining that a new shielding entity (i.e. the screen) appears in the indoor activity area.

[0051] In step S3, the system further processes the new shielding entity. The system extracts the visual feature information (such as its folding texture, color, rectangular contour) and spatial position information (such as the screen is located between the conference area of user A and the window actuator) of the screen. The system determines the functional attribute of the screen in combination with the preset semantic mapping relationship. The semantic mapping relationship contains the corresponding logic that "when the object category is a screen and the spatial topological relationship represents that the shielding entity is located between the indoor activity area and the window actuator, the functional attribute of the shielding entity is determined to be a private line-of-sight blocking". Therefore, the system successfully determines that the functional attribute of the screen is "private line-of-sight blocking".

[0052] Finally, in step S4, the system identifies the event of the screen with the "privacy sight blocking" functional attribute as a negative feedback signal to the current window control strategy. Based on this negative feedback signal, the system corrects the window control strategy. For example, the system records the entire process of the user A manually closing the window and setting the screen due to the previous opening of the window caused by the "light exercise" identification and the external noise interference. In the corrected strategy, the system will give priority to the needs of "privacy sight blocking" and "acoustic environment isolation". Based on the environmental needs implied by the "privacy sight blocking" functional attribute, the system generates window adjustment instructions, such as instructing the window actuator to keep the window closed and possibly further instructing the smart curtain to lower to enhance privacy and sound insulation. At the same time, in order to compensate for the ventilation needs, the system may try to ventilate indirectly through the windows of other unaffected areas or activate the smart fresh air system for silent ventilation. Through the window adjustment instructions, the window actuator executes the corrected window control strategy, ensuring the stability and concentration of user A's video conference.

[0053] As can be seen from the above example, the method proposed in the present application effectively solves the problems of incorrect control operation caused by ambiguous activity identification and sensor information acquisition obstacles caused by temporary physical barriers introduced by users to correct errors in traditional intelligent window system in complex indoor activity scenarios. When facing user A's video conference and user B's physical exercise, the traditional system may incorrectly open the window due to the inability to accurately identify high-priority activities, introducing noise interference. When user A manually closes the window and sets the screen, the traditional system not only fails to understand the user's intention, but may further reduce the accuracy of activity identification due to the screen obstruction, falling into a vicious cycle.

[0054] In contrast, the method of the present application can timely discover the limitations of system identification or user dissatisfaction by monitoring the confidence index and manual intervention signal in real time. When user A manually closes the window and sets the screen, the system can identify it as a new shielding entity and further accurately infer the "privacy sight blocking" functional attribute of the screen through visual features and spatial location information combined with semantic mapping relationship, thereby understanding the real needs of user A. This "semantic understanding" of user behavior is a key innovation point of the present application, which converts the user's passive corrective behavior into a feedback signal for system active learning. Based on this negative feedback, the system can correct its window control strategy from being driven by activity identification alone to being driven by user intention, generating window adjustment instructions that better meet actual needs. For example, the system no longer blindly opens the window for ventilation, but prioritizes the privacy and quiet environment required for video conferencing. This closed-loop "behavior-strategy reverse learning cycle" enables the system to continuously learn and optimize from the user's actual "corrective behavior", breaking the vicious cycle of perception and action of traditional systems, achieving truly adaptive intelligent control and significantly improving user experience.

[0055] In some embodiments, in step S2, the step of detecting whether a new occluding entity appears in the indoor activity area based on the indoor environment image comprises:

[0056] S21. calling a pre-stored baseline environment model in an unoccluded state; the baseline environment model contains spatial layout information of fixed furniture in the indoor environment;

[0057] S22. differentiating the indoor environment image from the baseline environment model, and extracting a difference region;

[0058] S23. performing object recognition on the difference region, and when the recognition result is an object other than a human body and fixed furniture, determining that a new occluding entity appears in the indoor activity area.

[0059] The baseline environment model aims to provide a stable reference framework for distinguishing fixed elements in the indoor environment from newly appearing entities that may have an occluding effect. This model can be established by a comprehensive scan or modeling of the indoor space when the system is first deployed. For example, a LiDAR system can be used to perform a three-dimensional point cloud scan of the indoor space, generating a three-dimensional point cloud model containing the precise spatial positions and geometric shapes of all fixed furniture (such as walls, doors, window frames, fixed cabinets, etc.). Alternatively, a series of images can be collected by a visual acquisition unit in an unoccluded state, and a sparse or dense three-dimensional map of the indoor environment can be constructed using SLAM technology, which contains the texture and geometric information of the fixed furniture. In addition, a pre-designed indoor space CAD model can also be used as a baseline, which records the layout of fixed structures and furniture in detail.

[0060] The purpose of differential comparison is to identify regions in the current indoor environment image that are inconsistent with the baseline environment model, which usually indicate newly appearing objects. One implementation is to perform image registration and alignment between the indoor environment image obtained by the current visual acquisition unit and the baseline environment model, and then perform pixel-level difference calculation, such as using image subtraction or SSIM to quantify the difference and highlight the changed regions. Another way is to use background subtraction algorithms such as GMM or motion detection methods based on inter-frame difference, treating the baseline environment model as "background" and new objects in the real-time indoor environment image as "foreground", thereby extracting the foreground region as the difference region. Scene reconstruction can also be performed through a deep learning model, and the reconstruction result is compared with the baseline environment model in three-dimensional geometry to identify newly added geometric bodies.

[0061] The purpose of object recognition in the difference area is to accurately classify newly appearing objects to ensure that only those objects that may act as occlusion entities and affect system perception are identified. This can be achieved by deploying advanced object detection and recognition algorithms, such as the YOLOv7 or Mask R-CNN algorithms based on deep learning. These algorithms can analyze the pixels within the difference area and identify the object categories contained therein, such as "person", "chair", "table", "screen", "potted plant", etc. The system will filter the identification results, excluding "human body" and "fixed furniture" categories (such as tables and chairs that are already in the baseline model), and only when the identification result is a non-human and non-fixed furniture object, it will be determined as a newly added occlusion entity. In addition, auxiliary judgment can also be combined with geometric features, for example, if the identified object has a specific aspect ratio (e.g., aspect ratio greater than 2:1) and height (e.g., height exceeding 1.5 meters), it further enhances the possibility of being an occlusion entity.

[0062] The present solution solves the challenge of accurately identifying newly added occlusion entities in complex indoor environments by introducing a baseline environment model and performing differential comparison. When the system detects that the confidence index of indoor activity recognition results is below a preset threshold or receives a manual intervention signal, it will trigger the acquisition of indoor environment images. At this time, the system first calls the pre-stored baseline environment model in the unoccluded state, which accurately records the spatial layout information of indoor fixed furniture, providing a stable reference for subsequent detection. Then, the system performs fine differential comparison between the currently acquired indoor environment images and the baseline environment model. This comparison can effectively filter out the background elements that do not change in the room, thereby accurately extracting all areas in the image that do not match the baseline model, i.e., the difference area. These difference areas represent newly appearing or changing objects in the indoor environment. To further confirm whether these difference areas are true occlusion entities, the system will perform object recognition on these areas. Through advanced object recognition algorithms, the system can distinguish whether the objects in the difference area are human bodies, fixed furniture, or other types of objects. Only when the identification result clearly indicates that the object is neither a human body nor indoor fixed furniture, the system will finally determine that a newly added occlusion entity has appeared in the indoor activity area. This hierarchical screening mechanism ensures the accuracy and robustness of detection, avoiding the misjudgment of human activity or existing fixed furniture as a newly added occlusion. In this way, the present solution can provide a reliable and high-precision occlusion entity detection capability for the above-mentioned window intelligent control method, enabling the system to accurately identify temporary physical barriers introduced by users to respond to system misjudgments, thereby providing a solid data foundation for subsequent correction of window control strategies, avoiding strategy misjudgments due to inaccurate detection, and thus improving the effectiveness of the overall control strategy and user experience.

[0063] In some embodiments, in step S3, determining the functional attribute of the occluding entity based on the visual feature information and the spatial position information, in combination with a pre-set semantic mapping relationship, comprises:

[0064] S31. Identifying the object category of the occluding entity based on the visual feature information, and determining the spatial topological relationship of the occluding entity relative to the window execution mechanism and the indoor activity area based on the spatial position information;

[0065] S32. When the object category is a screen or curtain, and the spatial topological relationship indicates that the occluding entity is located between the indoor activity area and the window execution mechanism, determining the functional attribute of the occluding entity as a private line-of-sight blocking;

[0066] S33. When the object category is an acoustic panel or a movable wall, and the spatial topological relationship indicates that the occluding entity encloses the indoor activity area, determining the functional attribute of the occluding entity as an acoustic environment isolation.

[0067] Wherein, identifying the object category of the occluding entity based on the visual feature information aims to ensure classification based on the actual appearance of the object, avoiding misjudgment due to similar appearance. This can be achieved, for example, by analyzing the image obtained from the visual acquisition unit through a deep learning model, extracting the texture, color distribution, shape contour, etc. of the occluding entity as the visual feature vector, and matching it with the pre-trained object category model, so as to identify its object category. Alternatively, traditional image processing techniques such as edge detection, shape matching, color histogram analysis, etc. can also be used to extract the geometric features of the occluding entity from the image, and then compared with the pre-set feature template to determine its object category.

[0068] Determining the spatial topological relationship of the occluding entity relative to the window execution mechanism and the indoor activity area based on the spatial position information aims to provide the location context of the object relative to the window and the activity area, laying a solid foundation for accurate judgment of the functional attribute. This can be achieved, for example, by obtaining the three-dimensional coordinate data of the occluding entity, the window execution mechanism and the indoor activity area through three-dimensional reconstruction technology, and then calculating the relative position, distance and direction between these entities to determine whether the occluding entity is located between the indoor activity area and the window execution mechanism, or whether it encloses the indoor activity area. Alternatively, indoor positioning technology can also be used to obtain two-dimensional or three-dimensional position information of the occluding entity, the window execution mechanism and the indoor activity area, and combined with the pre-set indoor layout map, the spatial relationship between the occluding entity and the window execution mechanism and the indoor activity area can be determined through geometric analysis, such as whether it is within a certain pre-set sector, or whether it forms a closed area.

[0069] When the object category is a screen or curtain, and the spatial topology relationship represents that the shielding entity is located between the indoor activity area and the window actuator, the function attribute of the shielding entity is determined as a private line-of-sight blocking, which is aimed at solving the problem of identifying privacy needs in a specific position and type, and avoiding false classification due to ambiguous position or confused type. Specifically, the system receives the identification result that the object category is "screen" or "curtain", and at the same time receives the judgment that the spatial topology relationship indicates that the entity is located "between the indoor activity area and the window actuator". At this time, the system logic judgment module performs a logical "and" operation on the two conditions, and if the result is true, the function attribute of the shielding entity is explicitly marked as "private line-of-sight blocking".

[0070] When the object category is a sound-absorbing panel or a movable wall, and the spatial topology relationship represents that the shielding entity surrounds the indoor activity area, the function attribute of the shielding entity is determined as acoustic environment isolation, which is aimed at effectively identifying the acoustic isolation needs under the surrounding structure, and ensuring the accurate determination of the function attribute in the acoustic sensitive scene. Specifically, the system receives the identification result that the object category is "sound-absorbing panel" or "movable wall", and at the same time receives the judgment that the spatial topology relationship indicates that the entity "surrounds the indoor activity area". At this time, the system logic judgment module performs a logical "and" operation on the two conditions, and if the result is true, the function attribute of the shielding entity is explicitly marked as "acoustic environment isolation".

[0071] The scheme of the present application works in the following way: first, the window intelligent control system acquires indoor environment images through the visual acquisition unit, and detects whether a new shielding entity appears in the indoor activity area based on the images. Next, for the shielding entity, the system will perform step S31, that is, according to its visual feature information, the specific object category is identified, for example, screen, curtain, sound-absorbing board, movable wall, etc., and at the same time, according to its spatial position information, the spatial topological relationship of the entity relative to the indoor activity area and the window actuator is accurately determined, for example, whether it is located between the two or whether it surrounds the activity area. Then, the system will use the preset semantic mapping relationship combined with the specific rules defined in S32 and S33. For example, if the object category is identified as "screen" or "curtain", and the spatial topological relationship shows that it is located "between the indoor activity area and the window actuator", the system will logically deduce that the functional attribute of the shielding entity is "privacy line blocking". Similarly, if the object category is identified as "sound-absorbing board" or "movable wall", and the spatial topological relationship shows that it "surrounds the indoor activity area", the system will deduce that its functional attribute is "acoustic environment isolation". This judgment mechanism based on explicit rules solves the problem that in a complex activity scene, only relying on object category and spatial position information cannot accurately distinguish the functional attribute. By accurately matching the specific object category with the specific spatial topological relationship, the system can more accurately understand the user's deep intention of introducing the shielding entity, thereby avoiding the ambiguity of functional attribute recognition and providing a solid foundation for subsequent correction of window control strategy. For example, when the system identifies the functional attribute of "privacy line blocking", it will tend to adjust the window to enhance privacy; when it identifies "acoustic environment isolation", it will tend to adjust the window to reduce noise. This refined functional attribute recognition enables the window intelligent control system to more accurately respond to the user's actual needs, avoiding the situation where the window action does not match the user's needs due to misjudgment.

[0072] In some embodiments, in step S31, the step of identifying the object category of the shielding entity according to the visual feature information comprises:

[0073] S31A1. Obtain multi-view images or sequence images of the shielding entity;

[0074] S31A2. Extract visual feature information of the shielding entity from the multi-view images or sequence images, and obtain fused visual feature information by fusing the visual feature information;

[0075] S31A3. Identify the object category of the shielding entity based on the fused visual feature information.

[0076] Among them, the multi-view image or sequence image of the shielding entity is obtained, aiming to capture visual information of the shielding entity from different angles or time sequences to provide more abundant original data and avoid the limitations of feature missing or perspective deviation under a single perspective. This step can be realized in various ways. For example, multiple visual acquisition units can be deployed at different positions in the indoor space, enabling them to capture images of the shielding entity from multiple perspectives simultaneously. Alternatively, a single visual acquisition unit can be used to obtain a series of sequence images by moving or rotating it over a period of time or continuously capturing images as the shielding entity moves.

[0077] The visual feature information of the shielding entity is extracted from the multi-view image or sequence image, and the fused visual feature information is obtained by fusing the visual feature information, aiming to integrate multi-angle or dynamic change feature data to generate more complete and robust feature representation and reduce feature incompleteness caused by partial shielding or perspective change. This step can be realized by image processing technology and machine learning algorithm. For example, a deep learning model such as convolutional neural network (CNN) can be used to extract high-dimensional feature vectors from each frame of multi-view image or sequence image. Subsequently, these feature vectors from different angles or time points can be integrated through various fusion strategies, such as simple feature splicing, weighted average, or more complex attention mechanism fusion network, to generate a unified, more rich semantic information fusion feature vector.

[0078] Based on the fused visual feature information, the object class of the shielding entity is identified, aiming to use the integrated robust features for accurate classification to ensure the reliability and accuracy of object class identification. This step is usually realized by a classifier. For example, the fused visual feature information can be input into a pre-trained classifier, which can be a support vector machine (SVM), random forest, or a multi-layer perceptron (MLP) neural network model. The classifier is trained to map the fused features to predefined object classes, such as "screen", "curtain", "tall potted plant", "moving bookshelf", etc.

[0079] The scheme of the present application overcomes the limitation that a single-view image may not fully capture the visual features of the occluding entity by acquiring multi-view images or sequence images of the occluding entity. By extracting and fusing visual feature information from these multi-source images, the system can obtain a more comprehensive and robust representation of the occluding entity's features. Based on this fused visual feature information, the system can more accurately identify the object class of the occluding entity. This improved object class recognition accuracy directly enhances the reliability of determining the functional attribute of the occluding entity based on visual feature information and spatial location information. For example, when the system needs to identify whether the occluding entity is a "screen" or a "tall potted plant", the multi-view or sequence images provide more rich texture, shape and structure information, enabling the classifier to make a more confident decision. This accurate determination of the functional attribute further ensures the effectiveness and relevance of subsequent window control strategy modifications, enabling the window intelligent control system to more accurately respond to the environmental needs implied by the user's introduction of temporary physical partitions, thereby avoiding window misoperation caused by inaccurate object class recognition and improving the system's intelligence level and user experience.

[0080] In some embodiments, in step S31, determining the spatial topological relationship of the occluding entity relative to the window actuator and the indoor activity area based on the spatial location information comprises:

[0081] S31B1. Obtain the spatial location information of the newly added occluding entity, the indoor activity area, and all window actuators;

[0082] S31B2. Based on the spatial location information of the newly added occluding entity, the indoor activity area, and all window actuators, for each window actuator, determine whether the newly added occluding entity is located within the preset spatial sector between the window actuator and the indoor activity area to identify the relevant window actuator;

[0083] S31B3. When multiple relevant window actuators are identified, select the relevant window actuator closest to the newly added occluding entity as the target window actuator based on the distance between the newly added occluding entity and the multiple relevant window actuators;

[0084] S31B4. Determine the spatial topological relationship of the newly added occluding entity relative to the target window actuator and the indoor activity area based on the spatial location information of the newly added occluding entity, the indoor activity area, and the spatial location information of the target window actuator.

[0085] The above scheme aims to provide accurate geometric data basis for subsequent spatial relationship analysis. Among them, obtaining the spatial position information of the newly added shielding entity, the indoor activity area and all window actuators can be through three-dimensional reconstruction of the indoor environment by a visual acquisition unit (such as a camera equipped with a depth sensor or a laser radar scanner) to directly obtain the three-dimensional coordinate data of each entity in the preset coordinate system. In addition, the spatial position of each entity can also be calculated by capturing images from different angles using multiple visual acquisition units, using multi-view geometry or triangulation principle. Another way is to combine the pre-stored indoor environment layout map and object recognition technology, to identify the known reference objects and newly added entities in the image, and to estimate their approximate spatial position by using perspective principle.

[0086] On the basis of obtaining the above spatial position information, the system will judge whether the newly added shielding entity is located in the preset spatial sector between the window actuator and the indoor activity area for each window actuator, in order to identify the relevant window actuator. The purpose of this step is to effectively screen out the window actuators that may have direct spatial relationship with the newly added shielding entity in a complex environment with multiple window actuators, so as to focus the subsequent analysis on the most relevant window. The preset spatial sector can be defined as a conical or sectorial spatial region with the window actuator as the vertex and the indoor activity area as the base. Whether the shielding entity is located in the sector can be realized by calculating whether the geometric center point or the bounding box of the shielding entity intersects with the sector. For example, it can be calculated whether the line connecting the shielding entity and the window actuator passes through the boundary of the indoor activity area, or whether the shielding entity is on the virtual straight line connecting the window actuator and the indoor activity area or within a preset distance range near it.

[0087] When multiple relevant window actuators are identified, the system will select the relevant window actuator closest to the newly added shielding entity as the target window actuator according to the distance between the newly added shielding entity and the multiple relevant window actuators. This step is used to further accurately determine the window most likely to be affected by the newly added shielding entity among multiple potential relevant window actuators. By calculating the Euclidean distance between the geometric center point (or the center of its bounding box) of the newly added shielding entity and the geometric center point (or key reference point) of each relevant window actuator, the system can select the window actuator with the smallest distance. This selection mechanism ensures that the subsequent spatial topological relationship analysis is based on the most direct and most significant physical association, avoiding false judgments due to too far distance.

[0088] Finally, the system determines the spatial topological relationship of the newly added shielding entity relative to the target window execution mechanism and the indoor activity area according to the newly added shielding entity, the spatial position information of the indoor activity area, and the spatial position information of the target window execution mechanism. This step aims to clarify the precise relative position relationship between the newly added shielding entity and the target window execution mechanism and the indoor activity area, and to provide key spatial context information for subsequent functional attribute semantic mapping. The spatial topological relationship can include qualitative descriptions such as "between the target window execution mechanism and the indoor activity area", "surrounding the indoor activity area", "close to one side of the target window execution mechanism", etc. In specific determination, the relative coordinates and geometric shapes of the three can be used to determine through spatial geometric calculation. For example, if the shielding entity is in the projection area of the connecting segment of the target window execution mechanism and the indoor activity area, it can be determined that it is between the two.

[0089] In a window intelligent control system, when the system needs to revise the window control strategy according to the changes in the indoor environment, it is crucial to accurately identify the functional attributes of the newly added shielding entity. However, in a multi-window scenario, how to accurately determine the spatial topological relationship between the shielding entity and a specific window is a key factor affecting the accuracy of functional attribute judgment. The scheme of the present application solves this problem through systematic steps. First, the system comprehensively acquires the spatial position information of the newly added shielding entity, the indoor activity area, and all window actuators. These information constitutes a complete geometric description of the current indoor environment, laying a foundation for subsequent spatial analysis. Then, for each window actuator, the system determines whether the newly added shielding entity is located within the preset spatial sector between the window actuator and the indoor activity area. This judgment process effectively filters out those windows that may have a direct spatial relationship with the shielding entity, thereby avoiding unnecessary analysis of irrelevant windows and improving processing efficiency and accuracy. When there are multiple window actuators identified as "relevant", the system further selects the closest one as the "target window actuator" based on the distance between the newly added shielding entity and these relevant window actuators. This selection mechanism ensures that the subsequent spatial topological relationship analysis is based on the most direct and significant physical association, eliminating the ambiguity that may occur in a multi-window scenario, making the judgment of the shielding entity's influence range more focused and accurate. Finally, based on the spatial position information of the newly added shielding entity, the indoor activity area, and the determined target window actuator, the system can accurately determine the spatial topological relationship of the newly added shielding entity relative to the target window actuator and the indoor activity area, such as being located between them or surrounding the activity area. This accurate spatial topological relationship, as a key input, can be combined with the preset semantic mapping relationship (such as the corresponding logic between entity categories, spatial topological relationships, and functional attributes) to accurately infer the functional attributes of the newly added shielding entity. Through the above series of steps, the scheme can overcome the challenge of inaccurate spatial relationship judgment in a multi-window environment, providing a reliable spatial context for subsequent functional attribute determination. This enables the system to more accurately identify the user's intention to introduce temporary physical barriers, such as for privacy visual obstruction or acoustic environment isolation, and then generate window adjustment instructions that better meet the user's expectations based on the environmental needs implied by these functional attributes, effectively revising the window control strategy. This accurate spatial relationship recognition capability, combined with the overall window intelligent control method and the scheme of semantic mapping relationship and functional attribute determination, enhances the system's understanding and response ability to user needs in complex indoor activity scenarios, avoiding the problem of window actions not meeting user needs due to misjudgment.

[0090] Reference is made to the accompanying Figure 2The application provides a window intelligent control system, comprising a visual acquisition unit and a window execution mechanism, and further comprising:

[0091] a monitoring module 100 for monitoring a confidence index of an indoor activity recognition result and a manual intervention signal of the window execution mechanism in real time;

[0092] a control module 200 for controlling the visual acquisition unit to acquire an indoor environment image when the confidence index is lower than a preset threshold or the manual intervention signal is received, and detecting whether a new shielding entity appears in an indoor activity area based on the indoor environment image;

[0093] a determination module 300 for extracting visual feature information and spatial position information of the shielding entity when the new shielding entity appears in the indoor activity area, and determining a functional attribute of the shielding entity according to the visual feature information and the spatial position information in combination with a preset semantic mapping relationship;

[0094] a generation module 400 for taking the recognition event of the shielding entity with the specific functional attribute as a negative feedback signal of a current window control strategy when the shielding entity with the specific functional attribute is recognized, correcting the window control strategy based on the negative feedback signal, and generating a window adjustment instruction based on an environmental demand implied by the functional attribute to control the window execution mechanism to execute the corrected window control strategy through the window adjustment instruction.

[0095] In some embodiments, the control module 200 performs the following when detecting whether a new shielding entity appears in the indoor activity area based on the indoor environment image:

[0096] S21. A reference environment model in an unshielded state is called, and the reference environment model contains spatial layout information of indoor fixed furniture;

[0097] S22. The indoor environment image is compared with the reference environment model to extract a difference area;

[0098] S23. Object recognition is performed on the difference area, and when the recognition result is an object other than a human body and fixed furniture, it is determined that a new shielding entity appears in the indoor activity area.

[0099] In some embodiments, the semantic mapping relationship comprises a corresponding logic between an entity category, a spatial topological relationship and a functional attribute;

[0100] The determination module 300 performs the following when determining the functional attribute of the shielding entity according to the visual feature information and the spatial position information in combination with the preset semantic mapping relationship:

[0101] S31. Identify the object category of the occlusion entity according to the visual feature information, and determine the spatial topological relationship of the occlusion entity relative to the window execution mechanism and the indoor activity area according to the spatial position information;

[0102] S32. When the object category is a screen or curtain, and the spatial topological relationship indicates that the occlusion entity is located between the indoor activity area and the window execution mechanism, determine that the functional attribute of the occlusion entity is privacy view blocking.

[0103] S33. When the object category is an acoustic panel or a movable wall, and the spatial topological relationship indicates that the occlusion entity surrounds the indoor activity area, determine that the functional attribute of the occlusion entity is acoustic environment isolation.

[0104] In some embodiments, the determining module 300, when used to identify the object category of the occlusion entity according to the visual feature information, performs:

[0105] S31A1. Obtain a multi-view image or a sequence image of the occlusion entity;

[0106] S31A2. Extract visual feature information of the occlusion entity from the multi-view image or the sequence image, and obtain fused visual feature information by fusing the visual feature information;

[0107] S31A3. Identify the object category of the occlusion entity based on the fused visual feature information.

[0108] In some embodiments, the determining module 300, when used to determine the spatial topological relationship of the occlusion entity relative to the window execution mechanism and the indoor activity area according to the spatial position information, performs:

[0109] S31B1. Obtain spatial position information of the newly added occlusion entity, the indoor activity area, and all window execution mechanisms;

[0110] S31B2. According to the spatial position information of the newly added occlusion entity, the indoor activity area, and all window execution mechanisms, for each window execution mechanism, determine whether the newly added occlusion entity is located in a preset spatial sector between the window execution mechanism and the indoor activity area, to identify the relevant window execution mechanism;

[0111] S31B3. When multiple relevant window execution mechanisms are identified, select the relevant window execution mechanism closest to the newly added occlusion entity as the target window execution mechanism according to the distance between the newly added occlusion entity and the multiple relevant window execution mechanisms;

[0112] S31B4. Determine the spatial topological relationship of the newly added occlusion entity relative to the target window execution mechanism and the indoor activity area according to the spatial position information of the newly added occlusion entity, the indoor activity area, and the target window execution mechanism.

[0113] In this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.

[0114] The above description is merely illustrative of the application and not intended to be limiting. Other modifications of the application will occur to those skilled in the art upon reading the description. Therefore, the scope of the application should be determined not with reference to the description, but should instead be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.

Claims

1. A window intelligent control method based on indoor activities, used in a window intelligent control system, the window intelligent control system including at least a vision acquisition unit and a window actuator, characterized in that, Includes the following steps: S1. Real-time monitoring of confidence indicators for indoor activity recognition results, as well as manual intervention signals for window actuators; S2. When the confidence index is lower than the preset threshold or a manual intervention signal is received, the visual acquisition unit is controlled to acquire indoor environmental images, and the indoor environmental images are used to detect whether any new obstructing entities appear in the indoor activity area. S3. When a new occluding entity appears in the indoor activity area, extract the visual feature information and spatial location information of the occluding entity, and determine the functional attributes of the occluding entity based on the visual feature information and spatial location information, combined with the preset semantic mapping relationship. S4. When an obstructing entity with specific functional attributes is identified, the identification event is used as a negative feedback signal to the current window control strategy. The window control strategy is corrected based on the negative feedback signal, and a window adjustment command is generated based on the environmental requirements implied by the functional attributes. The window actuator is then controlled to execute the corrected window control strategy through the window adjustment command.

2. The intelligent window control method based on indoor activities according to claim 1, characterized in that, Step S2, the step of detecting whether a new occluding entity appears within the indoor activity area based on the indoor environment image, includes: S21. Call the pre-stored baseline environment model in an unobstructed state; the baseline environment model contains spatial layout information of the fixed indoor furniture; S22. Compare the indoor environment image with the benchmark environment model to extract the difference areas; S23. Perform object recognition on the difference area. When the recognition result is an object that is neither a human body nor a fixed piece of furniture, it is determined that a new occluding entity has appeared in the indoor activity area.

3. The intelligent window control method based on indoor activities according to claim 1, characterized in that, Semantic mapping relationships include the correspondence logic between entity categories, spatial topological relationships, and functional attributes.

4. The intelligent window control method based on indoor activities according to claim 3, characterized in that, Step S3, which involves determining the functional attributes of the occluded entity based on visual feature information and spatial location information, combined with a preset semantic mapping relationship, includes: S31. Based on visual feature information, identify the object category of the occluding entity, and based on spatial location information, determine the spatial topological relationship of the occluding entity relative to the window actuator and the indoor activity area; S32. When the object category is a screen or curtain, and the spatial topology indicates that the occluding entity is located between the indoor activity area and the window actuator, the functional attribute of the occluding entity is determined to be privacy visual obstruction. S33. When the object category is a sound-absorbing panel or a movable wall, and the spatial topology represents the shading entity surrounding the indoor activity area, the functional attribute of the shading entity is determined to be acoustic environmental isolation.

5. The intelligent window control method based on indoor activities according to claim 4, characterized in that, In step S31, the step of identifying the object category of the occluded entity based on visual feature information includes: S31A1. Acquire multi-view images or image sequences of occluded entities; S31A2. Extract visual feature information of occluded entities from multi-view images or image sequences, and obtain fused visual feature information by fusing the visual feature information; S31A3. Based on the fused visual feature information, identify the object category of the occluded entity.

6. The intelligent window control method based on indoor activities according to claim 4, characterized in that, Step S31, which involves determining the spatial topological relationship between the obstructing entity and the window actuator and the indoor activity area based on the spatial location information, includes: S31B1. Obtain spatial location information of newly added obstructing entities, indoor activity areas, and all window actuators; S31B2. Based on the newly added obstructing entity, the indoor activity area, and the spatial location information of all window actuators, for each window actuator, determine whether the newly added obstructing entity is located within a preset spatial sector between the window actuator and the indoor activity area, so as to identify the relevant window actuator. S31B3. When multiple related window actuators are identified, the related window actuator closest to the newly added occluding entity is selected as the target window actuator based on the distance between the newly added occluding entity and the multiple related window actuators. S31B4. Based on the spatial location information of the newly added occlusion entity, the indoor activity area, and the target window actuator, determine the spatial topological relationship between the newly added occlusion entity and the target window actuator and the indoor activity area.

7. A window intelligent control system, comprising a vision acquisition unit and a window actuator, characterized in that, Also includes: The monitoring module is used to monitor the confidence index of indoor activity recognition results in real time, as well as the manual intervention signals of the window actuator. The control module is used to control the visual acquisition unit to acquire indoor environmental images when the confidence index is lower than a preset threshold or when a manual intervention signal is received, and to detect whether any new obstructing entities appear in the indoor activity area based on the indoor environmental images. The determination module is used to extract the visual feature information and spatial location information of the occluding entity when a new occluding entity appears in the indoor activity area, and determine the functional attributes of the occluding entity based on the visual feature information and spatial location information, combined with the preset semantic mapping relationship. The generation module is used to take the recognition event as a negative feedback signal to the current window control strategy when an occluding entity with specific functional attributes is identified, and to correct the window control strategy based on the negative feedback signal. It also generates window adjustment instructions based on the environmental requirements implied by the functional attributes, so as to control the window actuator to execute the corrected window control strategy through the window adjustment instructions.

8. The window intelligent control system according to claim 7, characterized in that, The control module executes the following when detecting whether a new obstructing entity has appeared within the indoor activity area: S21. Call the pre-stored baseline environment model in an unobstructed state; the baseline environment model contains spatial layout information of the fixed indoor furniture; S22. Compare the indoor environment image with the benchmark environment model to extract the difference areas; S23. Perform object recognition on the difference area. When the recognition result is an object that is neither a human body nor a fixed piece of furniture, it is determined that a new occluding entity has appeared in the indoor activity area.

9. The window intelligent control system according to claim 7, characterized in that, Semantic mapping relationships include the correspondence logic between entity categories, spatial topological relationships, and functional attributes; The determination module is executed when determining the functional attributes of occluded entities based on visual feature information and spatial location information, combined with a preset semantic mapping relationship: S31. Based on visual feature information, identify the object category of the occluding entity, and based on spatial location information, determine the spatial topological relationship of the occluding entity relative to the window actuator and the indoor activity area; S32. When the object category is a screen or curtain, and the spatial topology indicates that the occluding entity is located between the indoor activity area and the window actuator, the functional attribute of the occluding entity is determined to be privacy visual obstruction. S33. When the object category is a sound-absorbing panel or a movable wall, and the spatial topology represents the shading entity surrounding the indoor activity area, the functional attribute of the shading entity is determined to be acoustic environmental isolation.

10. The window intelligent control system according to claim 9, characterized in that, The module determines the object category when identifying occluded entities: S31A1. Acquire multi-view images or image sequences of occluded entities; S31A2. Extract visual feature information of occluded entities from multi-view images or image sequences, and obtain fused visual feature information by fusing the visual feature information; S31A3. Based on the fused visual feature information, identify the object category of the occluded entity.

Images