Display device non-aware control method and system based on visual sector entry and exit events

Abstract

The application relates to the technical field of human-computer interaction, in particular to a display device non-sensing control method and system based on visual sector entry and exit events; based on a deployed remote control base station, position information of a wearable device worn by a user is acquired; a coordinate system is constructed with the remote control base station as an origin and the normal of a display device screen as an X axis; in the coordinate system, a viewing area of the display device is identified and determined as a visual sector; based on the wearable device position information and the visual sector, identification analysis is carried out to determine a first motion mode and a second motion mode; for the first motion mode, an entering state of the wearable device position information is judged; for the second motion mode, a state characteristic vector containing a space coordinate and a suitable viewing index is constructed; based on the state characteristic vector, identification is carried out, and an in-sector behavior state is obtained through analysis; and the application carries out non-sensing control of the display device based on the identified entering state and the in-sector behavior state.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction technology, specifically to a method and system for seamless control of display devices based on visual sector entry and exit events. Background Technology

[0002] With the deep integration of computer technology and smart hardware, human-computer interaction is evolving from traditional physical contact input to contactless, natural spatial interaction. In smart home and multimedia interaction scenarios, seamless control technology based on user spatial status is gradually becoming a research hotspot. This type of technology aims to automatically trigger the wake-up, suspension, or multimedia state switching of computing devices by sensing the user's presence and behavior, thereby improving the smoothness and intelligence of the interaction.

[0003] Current contactless interaction solutions mainly rely on two technological approaches: one is image recognition technology based on computer vision, which uses cameras to capture the user's skeleton or face to determine their presence; the other is proximity detection technology based on infrared or simple wireless signal strength. However, the above-mentioned existing technologies have significant limitations in practical applications.

[0004] First, while vision-based solutions offer high recognition accuracy, they raise serious privacy concerns in private spaces like homes and are highly susceptible to ambient light and occlusion, leading to input data interruptions. Second, solutions based on infrared or simple signal strength are essentially threshold-triggered mechanisms, lacking a deep understanding of user behavior semantics. These technologies struggle to construct accurate three-dimensional interaction spaces and are prone to boundary jitter due to signal multipath effects. More critically, existing technologies cannot effectively distinguish between non-interactive body adjustments and substantive interactive intentions. For example, when a user merely adjusts their posture causing a slight signal fluctuation, it is often misinterpreted as the user leaving, triggering device lock or pause, severely disrupting the continuity of the user experience.

[0005] Therefore, without relying on image acquisition, how to accurately define the interaction area by processing spatial perception data and intelligently identify the user's deep behavioral intentions has become a difficult problem that urgently needs to be solved in this field.

[0006] To address this, a method and system for seamless control of display devices based on visual sector entry and exit events are proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for seamless control of display devices based on visual sector entry and exit events, which enables seamless control of display devices by recognizing the upcoming entry state and the behavior state within the sector.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for seamless control of display devices based on visual sector in / out events includes:

[0010] Based on the deployed remote control base station, the location information of the wearable device worn by the user is obtained; the location information of the wearable device includes real-time angle information and real-time distance information;

[0011] A coordinate system is constructed with the remote control base station as the origin and the normal of the display device screen as the X-axis; in the coordinate system, the viewing area of ​​the display device is identified and determined as the visible sector;

[0012] Based on the location information of the wearable device and the visual sector, an identification and analysis are performed to determine the first motion mode and the second motion mode;

[0013] For the first motion mode, determine the upcoming state of the wearable device's location information;

[0014] For the second motion mode, a state feature vector containing spatial coordinates and a suitable viewing index is constructed; based on the state feature vector, the in-sector behavior state in the second motion mode is identified and analyzed; the in-sector behavior state includes the maintaining viewing state, the dynamic adjustment state, and the imminent exit state.

[0015] Based on the identified upcoming state and the behavior state within the sector, the display device is controlled seamlessly.

[0016] The process of obtaining location information for wearable devices includes:

[0017] The wireless spatial sensing unit built into the remote control base station receives the wireless signal sent by the wearable device. The signal arrival angle algorithm is used to analyze the phase difference of the wireless signal and calculate the horizontal azimuth angle and vertical pitch angle of the wearable device relative to the remote control base station. The real-time angle information is then generated by combining these calculations.

[0018] Using wireless signal ranging technology, the physical transmission characteristics of wireless signals are measured, the straight-line distance between the wearable device and the remote control base station is calculated, and the real-time distance information is generated.

[0019] Real-time angle information and real-time distance information are synchronized in time and transformed in coordinates to synthesize wearable device location information.

[0020] The specific process for identifying the visible sectors of a display device includes:

[0021] Obtain the installation height data of the remote control base station relative to the physical ground plane, and establish a spatial mapping of the ground plane in a coordinate system with the remote control base station as the origin;

[0022] On a horizontal projection plane parallel to the ground plane, with the normal of the display device screen as the central axis, a horizontally symmetrical sector coverage area is defined according to the preset horizontal opening angle parameters.

[0023] An effective distance ring is defined along the radial dimension centered on the display device screen, based on the preset minimum monitoring radius and maximum viewing distance;

[0024] In the vertical dimension perpendicular to the ground plane, a suitable viewing height range based on human posture and line of sight is set, and combined with the installation height data, it is converted into a vertical effective coordinate range in the coordinate system;

[0025] The intersection area of ​​the horizontal sector coverage area, the effective distance ring, and the vertical effective coordinate range in three-dimensional space is defined as the visible sector, forming a three-dimensional monitoring space limited by ground height and viewing angle.

[0026] The process of identifying and judging the first motion pattern and the second motion pattern includes:

[0027] A hysteresis dual-threshold strategy for determining the visible sector space is preset, including an inner threshold group and an outer threshold group; wherein the spatial range of the outer threshold group is larger than that of the inner threshold group.

[0028] The location of the wearable device at the moment the remote control base station is turned on is obtained. If the location of the wearable device is outside the outer threshold group, it is determined to be the first motion mode; if the location of the wearable device is inside the outer threshold group, it is determined to be the second motion mode.

[0029] When the first motion mode is determined, the inner ring threshold group is used for judgment; only when the wearable device location information falls within the range defined by the inner ring threshold group is the state determined to be reversed to the second motion mode.

[0030] When the second motion mode is determined, the outer ring threshold group is used for judgment; the prediction of flipping to the first motion mode is triggered only when the position information of the wearable device exceeds the range defined by the outer ring threshold group.

[0031] In the first motion mode, the process of identifying the state to be entered specifically includes:

[0032] Establish a sliding time window, extract the position coordinate sequence of the wearable device within the sliding time window, and construct the first motion trajectory vector; calculate the tangential velocity component and radial velocity component of the first motion trajectory vector relative to the center point of the visible sector.

[0033] The first motion trajectory vector and velocity component are input into a pre-trained trajectory prediction model; the trajectory prediction model is constructed based on a long short-term memory network and is used to output a predicted position sequence within a preset time period in the future; when the endpoint of the predicted position sequence falls within the visible sector and the velocity component shows a decreasing trend that conforms to the deceleration characteristics of sitting down, it is determined that the state is about to be entered.

[0034] In the second motion mode, the process of analyzing the behavioral state within the sector specifically includes:

[0035] Simultaneously collect three-axis acceleration data, three-axis angular velocity data, and spatial position coordinates of wearable devices. Calculate the suitable viewing index based on the wearable device's position information, and perform time alignment and normalization processing to obtain a state feature vector.

[0036] The state feature vector is input into a preset behavior recognition neural network model; the behavior recognition neural network model uses a convolutional neural network to extract spatial displacement features, posture frequency features, energy decay features, and the rate of change of the suitable viewing index, and combines an attention mechanism to perform weighted analysis of temporal position changes; the output includes classification results for maintaining the viewing state, dynamically adjusting the state, and about to exit the state.

[0037] The process of obtaining the suitable viewing index specifically includes:

[0038] A weighted scoring model is used to calculate and fuse location suitability, posture suitability, and lighting suitability in real time to obtain a suitable viewing index.

[0039] The location suitability is calculated as follows: based on the Gaussian mixture model, the historical location data of the user while maintaining a viewing state is clustered to generate a probability distribution cluster of suitable furniture locations. The probability value of belonging to the probability distribution cluster is calculated based on the current location coordinates to obtain the location suitability.

[0040] The calculation of posture suitability involves: decomposing the position information into vertical height, combining the user's height data with the inertial sensor data of the wearable device to determine the user's posture; and identifying the posture suitability based on the time-series data of the user's posture.

[0041] The calculation of the light suitability is as follows: the ambient light intensity is detected by the ambient light sensor integrated in the remote control base station, and the light suitability is obtained by combining and analyzing the temporal changes of the ambient light intensity with the pre-identified suitable light intensity.

[0042] Based on the identified state, the seamless control of the display device includes: when it is determined that it is about to enter a state, triggering the display device to perform pre-wake-up and volume increase operations; when it is determined that it is in a viewing state, maintaining the current playback parameters of the display device; when it is determined that it is in a dynamic adjustment state, blocking false triggering commands caused by position jitter; and when it is determined that it is about to exit a state, triggering the display device to perform pause and mute operations.

[0043] A seamless control system for display devices based on visual sector in / out events includes:

[0044] The location acquisition module acquires the location information of the wearable device worn by the user based on the deployed remote control base station; the location information of the wearable device includes real-time angle information and real-time distance information.

[0045] The sector determination module constructs a coordinate system with the remote control base station as the origin and the normal of the display device screen as the X-axis; in the coordinate system, it identifies the viewing area of ​​the display device and determines it as the visible sector.

[0046] The pattern recognition module performs identification and analysis based on the wearable device's location information and the visible sector to determine the first motion mode and the second motion mode;

[0047] For the first motion mode, determine the upcoming state of the wearable device's location information;

[0048] For the second motion mode, a state feature vector containing spatial coordinates and a suitable viewing index is constructed; based on the state feature vector, the in-sector behavior state in the second motion mode is identified and analyzed; the in-sector behavior state includes the maintaining viewing state, the dynamic adjustment state, and the imminent exit state.

[0049] The contactless control module performs contactless control of the display device based on the recognized upcoming state and the behavior state within the sector.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] 1. This invention utilizes the wireless spatial sensing unit of a remote control base station, and integrates the signal arrival angle algorithm with multi-channel phase ranging technology to achieve sub-meter level seamless positioning at the physical level, completely avoiding the risk of privacy leakage in the home environment. Furthermore, it introduces the concept of a stereoscopic viewing sector limited by ground height and viewing angle, and combines it with a hysteresis dual-threshold strategy to construct an asymmetric spatial judgment logic, effectively filtering invalid signals generated by users at non-viewing heights, and also using the buffer zone of physical space to absorb random noise generated by the multipath effect of wireless signals; significantly improving the anti-interference capability and boundary judgment stability in complex home environments.

[0052] 2. This invention constructs a state feature vector containing spatial coordinates and a suitable viewing index; by introducing a convolutional neural network and an attention mechanism, it can deeply analyze the energy decay characteristics of user actions, accurately identify closed actions with fast decay and convergence characteristics and continuous actions with no decay characteristics, and intelligently block false triggering commands caused by position jitter when it is determined to be a dynamic adjustment state, maintaining the playback parameters unchanged; greatly ensuring the continuity and immersion of the user's viewing experience, and avoiding the mechanical judgment in the traditional threshold triggering scheme.

[0053] 3. This invention introduces a trajectory prediction model based on a long short-term memory network, which can capture the user's movement trend as they approach the visible sector in real time. By recognizing the biological behavioral characteristics of deceleration upon sitting down, it effectively improves the accuracy of recognition. In addition, a Gaussian mixture model is used to cluster the user's historical location data to generate personalized furniture location probability distribution clusters. Combined with posture and lighting suitability calculations, the control logic can adapt to different users' heights, sitting habits, and home lighting environments, ensuring that it can learn users' habits over time and improve the accuracy of recognition. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the sensorless control method for display devices based on visual sector entry and exit events according to the present invention.

[0055] Figure 2 This is a data logic diagram of the seamless control method for display devices based on visible sector entry and exit events according to the present invention.

[0056] Figure 3 This is a schematic diagram of the structure of the seamless control system for display devices based on visual sector entry and exit events according to the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Example 1:

[0059] This invention proposes a method for contactless control of display devices based on visual sector entry and exit events. The flow of the method is as follows: Figure 1 As shown, the data logic of the method is as follows: Figure 2 As shown, it includes:

[0060] Based on the deployed remote control base station, the location information of the wearable device worn by the user is obtained; the location information of the wearable device includes real-time angle information and real-time distance information;

[0061] A coordinate system is constructed with the remote control base station as the origin and the normal of the display device screen as the X-axis; in the coordinate system, the viewing area of ​​the display device is identified and determined as the visible sector;

[0062] Based on the location information of the wearable device and the visual sector, an identification and analysis are performed to determine the first motion mode and the second motion mode;

[0063] For the first motion mode, determine the upcoming state of the wearable device's location information;

[0064] For the second motion mode, a state feature vector containing spatial coordinates and a suitable viewing index is constructed; based on the state feature vector, the in-sector behavior state in the second motion mode is identified and analyzed; the in-sector behavior state includes the maintaining viewing state, the dynamic adjustment state, and the imminent exit state.

[0065] Based on the identified upcoming state and the behavior state within the sector, the display device is controlled seamlessly.

[0066] The process of obtaining location information for wearable devices includes:

[0067] The wireless spatial sensing unit built into the remote control base station receives the wireless signal sent by the wearable device. The signal arrival angle algorithm is used to analyze the phase difference of the wireless signal and calculate the horizontal azimuth angle and vertical pitch angle of the wearable device relative to the remote control base station. The real-time angle information is then generated by combining these calculations.

[0068] Using wireless signal ranging technology, the physical transmission characteristics of wireless signals are measured, the straight-line distance between the wearable device and the remote control base station is calculated, and the real-time distance information is generated.

[0069] Real-time angle information and real-time distance information are synchronized in time and transformed in coordinates to synthesize wearable device location information.

[0070] The remote control base station integrates a wireless spatial sensing unit, which is specifically a low-power Bluetooth array antenna, such as a rectangular or circular array; when the wearable device broadcasts a positioning signal containing a specific header, the base station's antenna array will capture the signal.

[0071] Since the physical distance between different elements in an antenna array is fixed, there are slight differences in the arrival time of the signal at different elements, which manifest as differences in carrier phase. Using a signal arrival angle algorithm, in-phase orthogonal data (IQ data) of the same source signals received by different antenna elements are collected, and the phase difference is calculated. Based on the correspondence between the phase difference and the geometric arrangement of the antenna array, the horizontal azimuth and vertical elevation angles of the incident signal are calculated. These two angles determine the orientation vector of the device in space.

[0072] In this embodiment, the signal arrival angle algorithm and its specific implementation process are as follows:

[0073] Antenna Array Physical Arrangement: The wireless spatial sensing unit built into the remote control base station adopts a uniform antenna array arranged in a 4x4 square array, totaling 16 antenna elements. The physical spacing between adjacent antennas in both the horizontal and vertical directions is strictly set to half the wavelength of the received signal. This specific spacing design can effectively avoid positional overlap or ambiguity during spatial angle calculation, ensuring the uniqueness of angle measurements.

[0074] Phase extraction and data preprocessing: The spatial sensing unit acquires the raw data stream of the wireless signal through high-speed sampling, including in-phase and quadrature components. The system extracts the instantaneous phase deviation by calculating the phase difference between the same source signals received by different antenna elements. To cope with multipath reflection interference in the home environment, the system adopts a multi-band joint processing strategy: it performs rapid frequency hopping measurements within a preset wireless frequency band to acquire multiple sets of phase data at different frequencies, and uses a linear fitting method to eliminate spurious signals caused by reflection, thereby extracting phase information that reflects the true straight-line propagation path.

[0075] Spatial angle calculation process: Angle estimation is performed using a multi-signal classification algorithm based on subspace decomposition. Specific steps include: integrating signal data from multiple sampling times to construct a correlation matrix reflecting the spatial characteristics of the signal; mathematically decomposing this matrix into a signal space representing the true signal and a noise space representing interference and noise; constructing a spatial search function to find the function's maximum point in a two-dimensional space of horizontal azimuth and vertical elevation angles; the location where the function energy is most concentrated corresponds to the wearable device's true direction relative to the base station, thus calculating the accurate horizontal and elevation angles.

[0076] Complex Environment Optimization Solution: To address the multipath reflection problem caused by indoor walls or furniture, spatial smoothing calibration technology is introduced; by dividing the large antenna array into multiple overlapping sub-regions and averaging them, the correlation interference between signals is reduced; this ensures that the system can maintain high-precision angle resolution even in harsh environments with indirect line-of-sight, achieving sub-meter level stable positioning.

[0077] Simultaneously, wireless signal ranging protocols, such as Bluetooth channel sensing technology, are used. Two-way handshake communication is established between the base station and the wearable device to measure the phase rotation slope or round-trip time of flight of the signal on a specific channel. To improve accuracy, rapid frequency hopping measurements are performed on different physical channels. By analyzing the phase response curves under multiple frequency bands, interference caused by multipath reflections is eliminated, and the straight-line distance is calculated. Compared to traditional ranging based on signal strength, this method more accurately reflects the true physical distance.

[0078] Since angle and distance measurements may not be completed in exactly the same microsecond, high-precision timestamps are applied to the acquired angle and distance data streams, and interpolation algorithms are used to align them to the same moment. Finally, using the transformation logic from spherical to Cartesian coordinates, the horizontal azimuth, vertical elevation, and straight-line distance are converted into three-dimensional Cartesian coordinates with the remote control base station as the origin, forming the final position information. Furthermore, combined with the display device size information, the offset vector of the screen center relative to the remote control base station is established, and this offset compensation is introduced when calculating the visible sector.

[0079] This invention overcomes the shortcomings of single-technology paths by integrating signal angle of arrival (SAA) algorithms and channel sounding ranging technology. Traditional Bluetooth signal strength ranging exhibits significant fluctuations, failing to meet precise control requirements. This solution, however, achieves sub-meter positioning accuracy, precisely distinguishing whether a user is sitting on or standing behind a sofa. Secondly, time synchronization and coordinate transformation mechanisms ensure real-time data consistency, capturing the user's true trajectory during rapid movement and avoiding positional blurring or jumps caused by data asynchrony. Finally, this solution is entirely based on radio electromagnetic wave characteristics, eliminating the need for camera image acquisition and completely avoiding the risk of user privacy leaks at the physical level. Furthermore, compared to ultra-wideband solutions, its Bluetooth ecosystem-based implementation cost is lower, making it more suitable for widespread adoption in home consumer electronics.

[0080] The specific process for identifying the visible sectors of a display device includes:

[0081] Obtain the installation height data of the remote control base station relative to the physical ground plane, and establish a spatial mapping of the ground plane in a coordinate system with the remote control base station as the origin;

[0082] On a horizontal projection plane parallel to the ground plane, with the normal of the display device screen as the central axis, a horizontally symmetrical sector coverage area is defined according to the preset horizontal opening angle parameters.

[0083] An effective distance ring is defined along the radial dimension centered on the display device screen, based on the preset minimum monitoring radius and maximum viewing distance;

[0084] In the vertical dimension perpendicular to the ground plane, a suitable viewing height range based on human posture and line of sight is set, and combined with the installation height data, it is converted into a vertical effective coordinate range in the coordinate system;

[0085] The intersection area of ​​the horizontal sector coverage area, the effective distance ring, and the vertical effective coordinate range in three-dimensional space is defined as the visible sector, forming a three-dimensional monitoring space limited by ground height and viewing angle.

[0086] Furthermore, the process of constructing the coordinate system includes:

[0087] An attitude reference unit is pre-installed in the display device. The attitude reference unit has a built-in nine-axis inertial measurement sensor, which includes a three-axis gyroscope, a three-axis accelerometer, a three-axis magnetometer, and a wireless communication module.

[0088] The attitude reference unit collects the attitude quaternion of the display device in the geophysical coordinate system in real time, calculates the normal vector perpendicular to the screen plane of the display device, and sends it to the remote control base station via wireless signal.

[0089] The remote control base station receives the normal vector and first establishes a physical coordinate system with the center of the remote control base station antenna array as the origin. Then, it calculates the spatial rotation matrix between the physical coordinate system and the normal vector, including combining the gravitational acceleration vector detected by the attitude reference unit with the normal vector, and calculating the horizontal axis using the cross product. Using this rotation matrix, all measurement data in the physical coordinate system is mapped and transformed to the virtual monitoring coordinate system in real time. In the virtual monitoring coordinate system, the X-axis is forcibly aligned to the normal direction of the display device screen, the Y-axis is parallel to the screen plane, and the Z-axis is perpendicular to the ground.

[0090] When the screen normal of the display device is not perpendicular to the direction of gravity, such as when the TV is tilted and suspended, the screen normal is locked as the X-axis first, and the Z-axis is projected and corrected to ensure that the calculated Y-axis always remains horizontal, thereby ensuring the accuracy of the height dimension in the subsequent determination of the visible sector.

[0091] Furthermore, the remote control base station may be placed on a TV cabinet (low position) or hung on top of the TV (high position). First, the installation height of the base station needs to be obtained. This can be done by the user inputting it during initial setup, or by having the user wear a ring and touch the ground for calibration. After obtaining the installation height, the plane equation of the ground is determined in a coordinate system with the base station as the origin. For example, if the base station is 1.5 meters above the ground, the ground plane is at -1.5 meters on the Z-axis, establishing a reference point for the physical world.

[0092] On a horizontal plane parallel to the ground, with the normal direction of the TV screen, i.e., the vertical outward direction, as the central axis; according to preset parameters, such as 45 degrees to the left and right, a fan-shaped angle range is defined; this range represents the optimal viewing angle of the TV screen, excluding areas on the sides of the screen where the picture cannot be seen clearly.

[0093] Distance limits are defined radiating outwards from the display device screen. A minimum monitoring radius is set to exclude areas where the user is close to the TV operating device or wiping the screen; a maximum viewing distance is set, and these two radii form a ring-shaped band. The minimum monitoring radius and maximum viewing distance are determined based on historical location data.

[0094] When watching TV, users typically sit on a sofa or recline, with their hands usually between 0.5 and 1.2 meters above the ground. Based on this ergonomic data, combined with the obtained base station installation height, the height above the ground is converted into a Z-axis coordinate range in a coordinate system. For example, if the base station is 1.5 meters high, and the suitable height is 1 meter above the ground, then the corresponding Z-axis value in the coordinate system is -0.5 meters. The hand height can be detected using a wearable device similar to a ring.

[0095] Finally, the intersection of the horizontal sector, distance ring, and vertical height range is taken to form a three-dimensional space; only coordinates falling within this space are considered to be within the visible sector.

[0096] This invention introduces constraints on the ground plane and a suitable height range, constructing a three-dimensional space limited by ground height, which can effectively filter out a large number of invalid signals. For example, if a user stands on a ladder to change a light bulb, although the angle and distance may meet the requirements, the height will not match the viewing characteristics and will be automatically excluded; this greatly reduces the false trigger rate and ensures that only actions of actually watching TV in the visible area will trigger a response, improving the product's intelligent experience.

[0097] Furthermore, it also includes a boundary self-evolution calibration step based on reinforcement learning, specifically including: defining the spatial parameters of the visible sector as the state space, defining the user's misjudgment correction command as the reward signal; constructing an online reinforcement learning agent that generates a negative reward signal when it receives a resume playback command issued by the user shortly after the user has decided to exit; updating the spatial parameters of the visible sector using a policy gradient algorithm, dynamically expanding the effective radius or adjusting the angle of the misjudgment location, so that the physical coverage of the visible sector can adaptively grow and reshape according to the user's actual activity habits.

[0098] This embodiment utilizes reinforcement learning to achieve self-growth of the visible sector, using the sector's geometric parameters (radius, opening angle) as the agent's state. When a pause operation is performed, if the user presses the play button via remote control within a short period, this behavior is defined as a negative reward (penalty signal). The reinforcement learning algorithm uses this signal to trace back the coordinates of the position that triggered the exit and fine-tunes the sector boundary parameters, such as re-drawing the coordinate point into the sector. As the number of uses increases, the sector boundary gradually evolves from a regular geometric shape into an irregular polygon that fits the user's actual home activity area.

[0099] This embodiment addresses the pain point that rigid rules cannot adapt to complex home layouts. It can continuously learn and evolve through user error correction, ultimately forming a customized monitoring area, which greatly reduces the false trigger rate caused by special apartment layouts or non-standard furniture placement.

[0100] The process of identifying and judging the first motion pattern and the second motion pattern includes:

[0101] A hysteresis dual-threshold strategy for determining the visible sector space is preset, including an inner threshold group and an outer threshold group; wherein the spatial range of the outer threshold group is larger than that of the inner threshold group.

[0102] The location of the wearable device at the moment the remote control base station is turned on is obtained. If the location of the wearable device is outside the outer threshold group, it is determined to be the first motion mode; if the location of the wearable device is inside the outer threshold group, it is determined to be the second motion mode.

[0103] When the first motion mode is determined, the inner ring threshold group is used for judgment; only when the wearable device location information falls within the range defined by the inner ring threshold group is the state determined to be reversed to the second motion mode.

[0104] When the second motion mode is determined, the outer ring threshold group is used for judgment; the prediction of flipping to the first motion mode is triggered only when the position information of the wearable device exceeds the range defined by the outer ring threshold group.

[0105] Two sets of spatial parameters are maintained. The first set is the inner threshold set, which has a smaller range; the second set is the outer threshold set, which has a larger range. The area between the two sets of thresholds is the hysteresis band.

[0106] When the device is first started or connected, the current position is obtained; if the position is directly outside the outer ring, it is initialized to the first motion mode; if it is inside the inner ring, it is initialized to the second motion mode.

[0107] When in the first motion mode, i.e., when determined to be outside the visible sector, only the inner circle threshold group is used as the judgment standard. The user must completely enter the smaller inner circle range before the state will be flipped to the second motion mode. If the user is just wandering around the boundary, although physically close, the entry event will not be triggered.

[0108] When in the second motion mode, i.e., when determined to be within the visible sector, the outer threshold group is used as the judgment standard. The user must completely move out of the larger outer range before the flip prediction is triggered. If the user leans back or moves slightly while watching TV, causing the position to slightly exceed the inner circle, the system still considers the user present and will not cut off the signal.

[0109] This invention solves the signal jitter problem under critical conditions, ensuring control stability. In wireless positioning, signal noise is unavoidable, causing calculated coordinates to fluctuate within a small range. Without a hysteresis strategy, when a user sits near the boundary, coordinate jumps lead to high-frequency switching between inside and outside the sector, causing frequent screen flickering or intermittent sound, severely impacting the user experience. This solution, by setting a buffer zone in physical space and utilizing asymmetric logic of high entry threshold and low exit threshold, absorbs random signal noise and unconscious minor user displacements, improving anti-interference capabilities.

[0110] In the first motion mode, the process of identifying the state to be entered specifically includes:

[0111] Establish a sliding time window, extract the position coordinate sequence of the wearable device within the sliding time window, and construct the first motion trajectory vector; calculate the tangential velocity component and radial velocity component of the first motion trajectory vector relative to the center point of the visible sector.

[0112] The first motion trajectory vector and velocity component are input into a pre-trained trajectory prediction model; the trajectory prediction model is constructed based on a long short-term memory network and is used to output a predicted position sequence within a preset time period in the future; when the endpoint of the predicted position sequence falls within the visible sector and the velocity component shows a decreasing trend that conforms to the deceleration characteristics of sitting down, it is determined that the state is about to be entered.

[0113] In this embodiment, the trajectory prediction model based on the Long Short-Term Memory network and its training and recognition process are as follows:

[0114] A sliding window with a length of 50 time steps is established, with a sampling interval of 100 milliseconds, meaning each input sequence represents continuous motion features within the past 5 seconds, resulting in the first motion trajectory vector. To accurately describe the motion trend, two key velocity components of this trajectory vector relative to the center point of the visible sector are calculated:

[0115] Tangential velocity component: represents the user's velocity as they rotate or move laterally around the base station; radial velocity component: represents the user's velocity as they approach or move away from the base station; finally, the first motion trajectory vector, tangential velocity component, and radial velocity component are jointly identified. Before being input into the model, all coordinate and velocity data are scaled to between 0 and 1 using a min-max normalization method to accelerate network convergence.

[0116] Network structure and parameter configuration: The trajectory prediction model consists of an input layer, two hidden layers, and a fully connected output layer.

[0117] Hidden layer: It adopts a double-layer long short-term memory unit stacked structure, with each layer containing 128 hidden neurons, which capture the long-term dependencies of motion trajectories through its internal forgetting gate mechanism.

[0118] Activation function: The hidden layer uses the hyperbolic tangent activation function, while the output layer uses the linear activation function.

[0119] Output dimension: The model output layer generates a sequence of predicted coordinates for future time periods, directly mapping the user's possible landing points.

[0120] Model training process and strategies:

[0121] Sample construction: The training data comes from human walking trajectory samples in real home environments. The sample labels are divided into categories such as "sitting down", "passing by", and "leaving" through manual annotation.

[0122] Loss function: The mean squared error is used as the target loss function for model training, which aims to minimize the Euclidean distance between the predicted trajectory points and the actual observed trajectory points.

[0123] Hyperparameter settings: The adaptive moment estimation algorithm is used for weight optimization during the training process. The initial learning rate is set to 0.001, the batch size is set to 64, and an early stopping mechanism is introduced during the training process to prevent overfitting.

[0124] Behavioral semantic recognition logic: The endpoint of the future position sequence output by the model is compared with the inner circle range of the visible sector. If the predicted endpoint is located within the inner circle and the radial velocity component in the trajectory vector shows a smooth decreasing trend, then the biological characteristics of deceleration upon sitting are met. At this point, it is determined that the user has a clear intention to enter, thereby triggering the pre-wake operation of the display device.

[0125] The determination requires two conditions: first, the predicted endpoint coordinates must fall within the inner circle of the visible sector; second, the velocity component must exhibit specific deceleration characteristics. This is based on behavioral principles: when a user wants to sit down to watch TV, they will inevitably slow down as they approach the sofa area to prepare to sit down; while if the user is just passing by, they will usually maintain a constant speed or accelerate as they pass through. When both conditions are met simultaneously, it is determined that the user is about to enter the state.

[0126] Traditional sensor control is often trigger-based, meaning the TV only wakes up after the user is seated and the signal is stable. This results in a disjointed experience, as the user has to wait for the TV to start or the sound to gradually increase after sitting down. This invention introduces LSTM trajectory prediction, which can detect the user's intention to enter the TV in advance from changes in the user's walking path and speed. In particular, by recognizing the key biometric behavior of slowing down upon sitting down, it effectively distinguishes between passing by and walking to watch TV. This allows the wake-up command to be sent in advance before the user actually sits down, so that the TV is already in optimal working condition when the user's gaze falls on the screen.

[0127] In the second motion mode, the process of analyzing the behavioral state within the sector specifically includes:

[0128] Simultaneously collect three-axis acceleration data, three-axis angular velocity data, and spatial position coordinates of wearable devices. Calculate the suitable viewing index based on the wearable device's position information, and perform time alignment and normalization processing to obtain a state feature vector.

[0129] The state feature vector is input into a preset behavior recognition neural network model; the behavior recognition neural network model uses a convolutional neural network to extract spatial displacement features, posture frequency features, energy decay features, and the rate of change of the suitable viewing index, and combines an attention mechanism to perform weighted analysis of temporal position changes; the output includes classification results for maintaining the viewing state, dynamically adjusting the state, and about to exit the state.

[0130] Three types of data are collected simultaneously: inertial sensor data uploaded by the wearable device, including acceleration and angular velocity; spatial location coordinates calculated by the base station; and a calculated suitable viewing index. Due to the high sampling rate of inertial data and the low sampling rate of location data, time alignment is first performed using an interpolation algorithm, and the data is normalized to a unified numerical range to construct a state feature vector matrix containing multi-dimensional information.

[0131] The matrix is ​​then input into a pre-built behavior recognition neural network, which contains convolutional neural network layers to extract features over time. For example, it extracts patterns of spatial displacement (translation or oscillation), posture frequency features (high-frequency jitter or low-frequency tilt), and the rate of change of the viewing comfort index. Next, the data flows through an attention mechanism layer that automatically assigns higher weight to key time points in the data; for example, the dramatic acceleration change at the moment a user stands up is given special attention, while the subsequent stable standing is relatively ignored.

[0132] In this embodiment, the identification of data features includes:

[0133] Spatial displacement characteristics: Based on the position coordinates of continuous time steps, the mean and variance of the movement distance, radial velocity and acceleration in three-dimensional space are calculated to characterize the displacement stability of the user in the visible sector.

[0134] Posture frequency characteristics: The frequency domain analysis of the triaxial angular velocity data is performed using Fast Fourier Transform to extract the main frequency components and energy distribution of the signal, thereby identifying whether the user is in a static viewing state or in a high-frequency limb swaying state.

[0135] Energy decay characteristics: The synthetic modulus of the wearable device's inertial data is calculated as the instantaneous energy, and the time-domain morphology after the energy peak is analyzed using an exponential decay fitting method. If the energy rapidly returns to the baseline value shortly after the burst, exhibiting a fast decay characteristic, it is characterized as attitude adjustment; if the energy remains high without decay, it is characterized as continuous displacement behavior.

[0136] Suitable viewing index change rate: Calculate the first derivative of the suitable viewing index within adjacent time windows to capture the dynamic correlation trend between user behavior and environmental suitability.

[0137] Furthermore, the energy decay feature is specifically used to characterize the temporal convergence trend of the user's action intensity, so as to accurately distinguish between brief posture adjustments and continuous departure behaviors. First, the synthetic modulus or variance of the data from the wearable device's inertial sensors (accelerometers and angular velocity meters) is calculated as the instantaneous motion energy; the convolutional neural network layer extracts features by analyzing the temporal morphology after the energy peak: if the energy is detected to decay rapidly and return to the static reference value in a short time after the burst, showing a fast decay convergence feature, it indicates that the action is a closed self-terminating action, such as adjusting sitting posture or stretching, and is classified as a dynamic adjustment state; if the energy is detected to maintain a high level of oscillation or show a divergent trend, showing no decay or slow decay features, it indicates that the action is a continuous displacement action, such as getting up and walking, and is classified as an impending exit state; thus, through the temporal morphological analysis of the energy dimension, the interference of instantaneous large-scale movements on the state determination is effectively shielded.

[0138] The fully connected layers of the model output the probability distribution of the classification results, and the current state is determined based on the maximum probability.

[0139] Maintain a viewing posture: move slightly in position, keep the posture stable.

[0140] Dynamic adjustment status: The posture data fluctuates drastically (such as stretching or changing sitting posture), but the position coordinates still mainly remain within the sector, and the suitable viewing index does not continue to decline.

[0141] Imminent Exit Status: The position moves outward, the suitable viewing index continues to change, and it is accompanied by a posture characteristic of standing up.

[0142] Existing technologies often mistakenly pause playback as a departure simply by detecting significant user movement or positional shift. However, in real life, users frequently adjust their posture, drink water, or shift their position on the sofa while watching TV, which do not signify the end of viewing. This invention utilizes the feature extraction capabilities and attention mechanisms of CNNs to capture key actions, combined with multimodal data on position and posture, to distinguish the nature of actions at a microscopic level. It filters out all interfering commands caused by posture adjustments, triggering a pause only when a clear intention to leave is confirmed; this greatly ensures the continuity and immersion of the viewing experience.

[0143] The process of obtaining the suitable viewing index specifically includes:

[0144] A weighted scoring model is used to calculate and fuse location suitability, posture suitability, and lighting suitability in real time to obtain a suitable viewing index.

[0145] The location suitability is calculated as follows: based on the Gaussian mixture model, the historical location data of the user while maintaining a viewing state is clustered to generate a probability distribution cluster of suitable furniture locations. The probability value of belonging to the probability distribution cluster is calculated based on the current location coordinates to obtain the location suitability.

[0146] The calculation of posture suitability involves: decomposing the position information into vertical height, combining the user's height data with the inertial sensor data of the wearable device to determine the user's posture; and identifying the posture suitability based on the time-series data of the user's posture.

[0147] The calculation of the light suitability is as follows: the ambient light intensity is detected by the ambient light sensor integrated in the remote control base station, and the light suitability is obtained by combining and analyzing the temporal changes of the ambient light intensity with the pre-identified suitable light intensity.

[0148] A Gaussian mixture model runs in the background, recording the user's current 3D coordinates whenever the user remains in a viewing state for a certain period. The Gaussian mixture model uses this historical data for cluster training, automatically learning the probability distribution clusters of frequently used locations in the home, such as sofas and armchairs—essentially a heatmap. In real-time calculations, the degree to which the current coordinates fall into these high-probability clusters is calculated; the closer to the cluster center, the higher the location suitability score; this allows the model to identify the user's location preferences.

[0149] In this embodiment, the calculation process of the location suitability and the operating mechanism of the Gaussian mixture model are as follows:

[0150] Construction and parameter setting of Gaussian mixture model: A Gaussian mixture model is used to model the spatial distribution of users; the number of Gaussian distributions (i.e., the number of cluster centers) preset in the model is usually set to two to five, and the specific value is initialized according to the number of commonly used furniture such as sofas and armchairs in the home environment. Each Gaussian distribution represents a potential suitable viewing area, and its central position and coverage in three-dimensional space are defined by the mean vector and covariance matrix.

[0151] Historical location data screening and preprocessing: Only location data from users who maintained a viewing state for more than five minutes were collected as training samples. Before being stored in the training library, a moving average filter was used to remove outliers caused by the multipath effect of wireless signals, ensuring that the input samples could truly reflect the user's spatial preferences during stable viewing.

[0152] Clustering learning and expectation-maximization process: The expectation-maximization algorithm is used to iteratively optimize the model parameters.

[0153] Initialization: The center point of the initial Gaussian distribution is selected through random sampling;

[0154] Iterative process: In each iteration, the posterior probability of the current sample belonging to each Gaussian distribution is first calculated; then, the mean, covariance and weight of each distribution are re-estimated based on these probability values.

[0155] Convergence criterion: When the growth rate of the model likelihood function is less than one ten-thousandth, or when the number of iterations reaches the preset upper limit of one hundred, the operation is stopped and the model parameters are locked.

[0156] Real-time calculation of location suitability: After obtaining the user's current location coordinates, they are input into a pre-trained Gaussian mixture model to calculate the weighted sum of the probability density functions of that coordinate point under all Gaussian distributions. This value represents the probability that the current location belongs to a suitable viewing area. Finally, this probability value is mapped to a location suitability score between zero and one, with a higher score for closer proximity to the cluster center.

[0157] Dynamic update and adaptive mechanism: To adapt to changes in user behavior patterns, such as furniture movement or changes in viewing habits, an online update mechanism is introduced. When a user is detected to frequently and consistently start viewing in a new location, a background update procedure is triggered. This process uses new samples to fine-tune the model parameters, or, if necessary, adds new Gaussian distribution clusters, ensuring that the location suitability evaluation logic can self-evolve over time.

[0158] Real-time calculation of posture suitability: Extract the Z-axis data from the three-dimensional coordinate system and combine it with the user's preset height information to estimate the current line of sight height; at the same time, combine the inertial sensor to determine the ring's tilt angle. If the height indicates that the user is in a sitting position and the ring's tilt angle is stable, the score is higher; if the height indicates that the user is standing, the score is lower.

[0159] Environmental fusion for suitable lighting: The ambient light sensor on the base station collects light intensity in real time and compares the current light intensity with the pre-identified suitable light intensity. If the ambient light dims or no strong light shines directly on the base station, the lighting suitability score is higher.

[0160] The three scores are weighted and identified according to preset weights, and the final suitable viewing index is output.

[0161] In this embodiment, the specific implementation process of the weighted scoring model and index fusion of the suitability for viewing index is as follows:

[0162] Sub-indicator numerical normalization: First, the value range of the three sub-indicators is unified to ensure that all input parameters are within the range of zero to one.

[0163] Location suitability: The probability density values ​​of the current coordinate point under each Gaussian distribution cluster are mapped to values ​​between zero and one through a normalization function, reflecting the degree of location preference.

[0164] Posture suitability: The posture categories identified in real time are mapped to prior probability values. For example, based on the average duration of a user in a certain posture in historical data, normalization and quantification are performed to assign a suitability value to different postures.

[0165] Light suitability: The deviation between ambient light intensity and preset suitable light intensity is mapped to a score of zero to one using a Gaussian kernel function.

[0166] The fusion formula of the weighted scoring model is as follows: The total score is calculated using a linear weighted summation model. The fusion formula is defined as follows: The suitable viewing index is equal to the weighted sum of the three suitability scores and their corresponding weights. In the initial state, the weight coefficients of the three scores are all set to one-third, focusing on the physical consistency of position and posture.

[0167] Furthermore, the calculation of the posture suitability employs a deep learning-based posture intent mapping strategy, specifically including:

[0168] A pre-trained deep neural network model for posture recognition is constructed. The model takes the time-series position coordinates of the wearable device and inertial sensor data as input and outputs the user's specific posture category. The posture category includes at least standard sitting posture, semi-reclining posture, standing posture, and walking posture. A posture viewing probability database based on the user's historical behavior data is established to calculate the prior probability of the user effectively viewing in each posture category. The posture category recognized in real time is mapped to the prior probability value, and a long short-term memory network is used to perform a temporal comprehensive analysis of the probability values ​​of continuous time steps to output the posture suitability that represents the user's depth of immersion, thereby distinguishing between static sitting posture with high viewing probability and temporary standing posture with low viewing probability.

[0169] Specifically, a lightweight deep neural network is constructed and deployed on a base station. The input data consists of time-series data from wearable devices over the past 3 seconds, including three-axis acceleration, three-axis angular velocity, and the Z-axis height calculated by the base station. The model is pre-trained and can output a current-moment posture classification label for the user; for example, Class0 - standard sitting posture, Class1 - semi-reclining posture, Class2 - standing posture, and Class3 - walking posture. A user profile database is maintained, recording the historical frequency of the user's TV viewing in each posture. For example, statistics show that the user spends 90% of their viewing time in a semi-reclining posture and only 5% in a standard sitting posture; based on this, the real-time identified semi-reclining posture is assigned a very high prior probability, such as 0.95. Finally, a sliding time window is used to smooth the continuous probability values, outputting the final posture suitability score to ensure that only stable postures that conform to the user's habits are considered valid viewing.

[0170] This invention addresses the problem of traditional fixed threshold methods failing to adapt to the diverse sitting postures of different individuals by incorporating deep learning and personalized historical data. It not only identifies physical posture but also combines this with semantic understanding based on the user's personal habits, significantly improving the accuracy of depth perception and avoiding misjudgments caused by non-standard user postures.

[0171] Furthermore, the calculation of the light suitability adopts a light environment scene perception model, specifically including: collecting time-series light intensity data output by an ambient light sensor, and extracting light feature vectors including the mean light intensity, fluctuation variance, and spectral energy; inputting the light feature vectors into a light environment classifier to identify the current lighting scene category, which includes immersive viewing mode, natural lighting mode, and strong light interference mode; based on a preset scene weight table, assigning a high baseline suitability score to the immersive viewing mode and a low baseline suitability score to the strong light interference mode, and dynamically calculating the final light suitability by combining the current screen brightness feedback of the display device.

[0172] This embodiment assists in decision-making by identifying lighting scenarios; the light sensor in the base station collects light intensity data at a frequency of 10Hz. A lighting environment classifier, such as a random forest or MLP, is constructed. Input features include the mean and variance of light intensity (to determine fluctuations), and frequency domain features (to determine whether it is flickering from artificial light sources or natural light). The classifier outputs the current scene label: if it is identified as an immersive viewing mode (low light intensity, low fluctuations), the baseline score for lighting suitability is set to 1.0; if it is identified as a strong light interference mode (high light intensity, high fluctuations, such as direct sunlight), it is set to 0.4 and its weight is reduced. In addition, the average brightness of the TV screen is considered; if the ambient light fluctuates synchronously with the screen brightness, it further confirms that the user is focused on watching the movie.

[0173] This invention possesses environmental perception capabilities, no longer mechanically relying on light intensity values, but understanding scene semantics. In particular, it can recognize users' actions of actively turning off lights to create an atmosphere, or recognize harsh environments with sunlight interference, thereby making control decisions that conform to human intuition under various complex lighting conditions.

[0174] This invention introduces a context-aware dimension to seamless control, making the control logic more human-centered and personalized. Traditional control logic is black-and-white; being within a designated area is considered presence. However, in reality, a user might be cleaning in front of the TV, within the designated area, but not necessarily wanting the TV to continue playing. This solution uses a suitable viewing index, integrating three dimensions: viewing habits, body posture, and ambient lighting. It can identify situations significantly different from the usual viewing pattern, thus providing a lower index to avoid accidental playback triggers. Furthermore, the introduction of a Gaussian mixture model gives the system self-learning capabilities, eliminating the need for manual sofa placement and allowing it to analyze and identify family layout habits over time.

[0175] Based on the identified state, the seamless control of the display device includes: when it is determined that it is about to enter a state, triggering the display device to perform pre-wake-up and volume increase operations; when it is determined that it is in a viewing state, maintaining the current playback parameters of the display device; when it is determined that it is in a dynamic adjustment state, blocking false triggering commands caused by position jitter; and when it is determined that it is about to exit a state, triggering the display device to perform pause and mute operations.

[0176] Example 2:

[0177] This invention also proposes a seamless control system for display devices based on visual sector entry and exit events. Building upon Embodiment 1, it further discloses the microscopic execution flow of the seamless control method for display devices based on visual sector entry and exit events. The structure of the system is as follows: Figure 3 As shown, it includes:

[0178] The location acquisition module acquires the location information of the wearable device worn by the user based on the deployed remote control base station; the location information of the wearable device includes real-time angle information and real-time distance information.

[0179] The sector determination module constructs a coordinate system with the remote control base station as the origin and the normal of the display device screen as the X-axis; in the coordinate system, it identifies the viewing area of ​​the display device and determines it as the visible sector.

[0180] First, the remote control base station is physically connected to the data port on the back of the display device via a universal serial bus interface. After the remote control base station is powered on, it establishes a handshake communication with the operating system of the display device through the wired interface and reads the operating status parameters of the display device in real time, including the screen backlight brightness value and the current volume value.

[0181] The remote control base station uses its built-in nine-axis inertial sensor to collect its own gravitational acceleration vector, thereby determining the vertical direction (i.e., the Z-axis) in physical space. Simultaneously, it receives the screen normal vector sent by the display device. The remote control base station constructs a virtual monitoring coordinate system with its own antenna array center as the origin, the screen normal as the X-axis, the direction perpendicular to the X-axis and parallel to the ground as the Y-axis, and the direction of gravity reversal as the Z-axis. In this coordinate system, the inner threshold group is set as a semicircle with a radius of 2 meters, the outer threshold group as a semicircle with a radius of 5 meters, and the height range of the visible sector is set to 0.5 to 2 meters above the ground.

[0182] The pattern recognition module performs identification and analysis based on the wearable device's location information and the visible sector to determine the first motion mode and the second motion mode;

[0183] For the first motion mode, determine the upcoming state of the wearable device's location information;

[0184] For the second motion mode, a state feature vector containing spatial coordinates and a suitable viewing index is constructed; based on the state feature vector, the in-sector behavior state in the second motion mode is identified and analyzed; the in-sector behavior state includes the maintaining viewing state, the dynamic adjustment state, and the imminent exit state.

[0185] When a user appears within the monitoring range of the remote control base station, the four-by-four rectangular array antenna inside the base station receives the Bluetooth positioning signal emitted by the smart ring; the base station uses a multi-signal classification algorithm to calculate the horizontal and vertical angles of the signal arrival, and combines multi-band phase ranging technology to calculate the straight-line distance.

[0186] The system detects that the user is outside the outer threshold group and determines it to be in the first motion mode. At this time, the system activates a trajectory prediction model based on a long short-term memory network. The specific structure of this model is as follows: the input layer accepts a six-dimensional feature vector, which includes the current three-dimensional spatial coordinates and three-axis velocity components; the hidden layer consists of two layers of stacked long short-term memory units, each containing 128 neurons, with a time step set to 50 frames, corresponding to five seconds of historical data; the output layer is a fully connected layer that outputs the predicted coordinate sequence for the next two seconds.

[0187] As the user walks briskly towards the sofa from four meters away, the system extracts trajectory data from the past five seconds and inputs it into the model. The model predicts that the user's future landing point coordinates are within the inner threshold group. Simultaneously, the system calculates the rate of change of the velocity component and finds that the user's radial velocity decreases uniformly from 1.5 meters per second to 0.2 meters per second, consistent with the biological characteristics of deceleration upon sitting down. Based on the landing point being within the sector and accompanied by deceleration characteristics, it is determined that the user is about to enter a sitting state. The remote control base station then sends a pre-wake command to the display device via the universal serial bus interface. The display device lights up the screen and begins playing buffered content at a low volume of 10%.

[0188] Once the user has fully entered the inner threshold group, the system determines that the state has reversed to the second motion mode. At this point, the system begins to perform fine-grained identification of the behavior state within the sector in a loop.

[0189] Step 1: Quantitative calculation of the suitable viewing index;

[0190] The system collects data in real time and performs the following dimensionless calculations:

[0191] Location suitability calculation: A pre-stored Gaussian mixture model is used to calculate the Euclidean distance between the current user's coordinates and the cluster center. This distance is then mapped to a score from 0 to 1, with higher scores for closer distances.

[0192] Posture suitability calculation: A pre-trained deep neural network model for posture recognition is constructed. The model takes the time-series position coordinates of the wearable device and inertial sensor data as input and outputs the user's specific posture category. The posture category includes at least standard sitting posture, semi-reclining posture, standing posture, and walking posture. A posture viewing probability database based on the user's historical behavior data is established to calculate the prior probability of the user effectively viewing in each posture category. The posture category recognized in real time is mapped to the prior probability value. In addition, a long short-term memory network is used to perform temporal comprehensive analysis on the probability values ​​of continuous time steps to output the posture suitability that represents the user's depth of immersion.

[0193] Light suitability calculation: The base station's light sensor detects that the ambient light intensity is relatively low, and at the same time, the data interface reads that the display device's screen brightness is in a high-brightness state, which is determined to be an immersive viewing scenario.

[0194] Index fusion: The total score is calculated using a linear weighting method.

[0195] Step 2: Behavior classification based on convolutional neural networks;

[0196] During the viewing process, the user stretched once; the system collected the triaxial acceleration, triaxial angular velocity, position coordinates and suitable viewing index sequence during this process, constructed a state feature vector, and input it into the behavior recognition neural network model.

[0197] The model structure is as follows: The first layer is a temporal convolutional layer containing 64 convolutional kernels with a kernel size of 5, used to extract short-term action features; the second layer is a channel attention mechanism layer, used to assign high weights to acceleration peak moments; the third layer is a global average pooling layer; and the last layer is a three-class fully connected layer, used for viewing, dynamic adjustment, and exiting.

[0198] During a stretch, the acceleration energy characteristic bursts forth, but the attention mechanism layer detects this burst and rapidly decays back to the baseline value within 0.8 seconds. Furthermore, since the user did not stand up, although the position coordinates fluctuated, they never exceeded the outer threshold group, and the position component in the suitable viewing index remained high. After comprehensive analysis, the model outputs the classification result of the dynamically adjusted state with a confidence level of 95%.

[0199] The contactless control module performs contactless control of the display device based on the recognized upcoming state and the behavior state within the sector.

[0200] Based on the dynamic adjustment status determination, the non-interactive control module executes a shielding strategy; although the sensor detects violent hand shaking, the system determines that this is a non-interactive body adjustment, thus blocking all possible erroneous operation commands that could trigger pause, maintaining the current playback progress and volume parameters of the display device unchanged, and ensuring the continuity of the user experience.

[0201] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for seamless control of display devices based on visual sector entry and exit events, characterized in that, include: Based on the deployed remote control base stations, the location information of the wearable device worn by the user is obtained; The wearable device location information includes real-time angle information and real-time distance information; A coordinate system is constructed with the remote control base station as the origin and the normal of the display device screen as the X-axis; in the coordinate system, the viewing area of ​​the display device is identified and determined as the visible sector; Based on the location information of the wearable device and the visual sector, an identification and analysis are performed to determine the first motion mode and the second motion mode; For the first motion mode, determine the upcoming state of the wearable device's location information; For the second motion mode, a state feature vector containing spatial coordinates and a suitable viewing index is constructed; based on the state feature vector, the behavior state within the sector under the second motion mode is identified and analyzed. The behavioral states within the sector include the "stay watching" state, the "dynamic adjustment" state, and the "about to exit" state. The process of obtaining the suitable viewing index specifically includes: A weighted scoring model is used to calculate and integrate location suitability, posture suitability, and lighting suitability in real time to obtain a suitable viewing index. The location suitability is calculated as follows: based on the Gaussian mixture model, the historical location data of the user while maintaining a viewing state is clustered to generate a probability distribution cluster of suitable furniture locations. The probability value of belonging to the probability distribution cluster is calculated based on the current location coordinates to obtain the location suitability. The calculation of posture suitability involves: decomposing the position information into vertical height, combining the user's height data with the inertial sensor data of the wearable device to determine the user's posture; and identifying the posture suitability based on the time-series data of the user's posture. The calculation of the light suitability is as follows: the ambient light intensity is detected by the ambient light sensor integrated in the remote control base station, and the light suitability is obtained by combining and analyzing the temporal changes of the ambient light intensity with the pre-identified suitable light intensity. Based on the identified upcoming state and the behavior state within the sector, the display device is controlled seamlessly.

2. The method for seamless control of a display device based on visual sector entry and exit events according to claim 1, characterized in that: The process of obtaining location information for wearable devices includes: The wireless spatial sensing unit built into the remote control base station receives the wireless signal sent by the wearable device. The signal arrival angle algorithm is used to analyze the phase difference of the wireless signal and calculate the horizontal azimuth angle and vertical pitch angle of the wearable device relative to the remote control base station. The real-time angle information is then generated by combining these calculations. Using wireless signal ranging technology, the physical transmission characteristics of wireless signals are measured, the straight-line distance between the wearable device and the remote control base station is calculated, and the real-time distance information is generated. Real-time angle information and real-time distance information are synchronized in time and transformed in coordinates to synthesize wearable device location information.

3. The method for seamless control of a display device based on visual sector entry and exit events according to claim 1, characterized in that: The specific process for identifying the visible sectors of a display device includes: Obtain the installation height data of the remote control base station relative to the physical ground plane, and establish a spatial mapping of the ground plane in a coordinate system with the remote control base station as the origin; On a horizontal projection plane parallel to the ground plane, with the normal of the display device screen as the central axis, a horizontally symmetrical sector coverage area is defined according to the preset horizontal opening angle parameters. An effective distance ring is defined along the radial dimension centered on the display device screen, based on the preset minimum monitoring radius and maximum viewing distance; In the vertical dimension perpendicular to the ground plane, a suitable viewing height range based on human posture and line of sight is set, and combined with the installation height data, it is converted into a vertical effective coordinate range in the coordinate system; The intersection area of ​​the horizontal sector coverage area, the effective distance ring, and the vertical effective coordinate range in three-dimensional space is defined as the visible sector, forming a three-dimensional monitoring space limited by ground height and viewing angle.

4. The method for seamless control of a display device based on visual sector entry / exit events according to claim 1, characterized in that: The process of identifying and judging the first motion pattern and the second motion pattern includes: A hysteresis dual-threshold strategy for determining the visible sector space is preset, including an inner threshold group and an outer threshold group; wherein the spatial range of the outer threshold group is larger than that of the inner threshold group. The location of the wearable device at the moment the remote control base station is turned on is obtained. If the location of the wearable device is outside the outer threshold group, it is determined to be the first motion mode; if the location of the wearable device is inside the outer threshold group, it is determined to be the second motion mode. When the first motion mode is determined, the inner ring threshold group is used for judgment; only when the wearable device location information falls within the range defined by the inner ring threshold group is the state determined to be reversed to the second motion mode. When the second motion mode is determined, the outer ring threshold group is used for judgment; the prediction of flipping to the first motion mode is triggered only when the position information of the wearable device exceeds the range defined by the outer ring threshold group.

5. The method for seamless control of a display device based on visual sector entry / exit events according to claim 4, characterized in that: In the first motion mode, the process of identifying the state to be entered specifically includes: Establish a sliding time window, extract the position coordinate sequence of the wearable device within the sliding time window, and construct the first motion trajectory vector; calculate the tangential velocity component and radial velocity component of the first motion trajectory vector relative to the center point of the visible sector. The first motion trajectory vector and velocity component are input into a pre-trained trajectory prediction model; the trajectory prediction model is constructed based on a long short-term memory network and is used to output a predicted position sequence within a preset time period in the future; when the endpoint of the predicted position sequence falls within the visible sector and the velocity component shows a decreasing trend that conforms to the deceleration characteristics of sitting down, it is determined that the state is about to be entered.

6. The method for seamless control of a display device based on visual sector entry / exit events according to claim 5, characterized in that: In the second motion mode, the process of analyzing the behavioral state within the sector specifically includes: Simultaneously collect three-axis acceleration data, three-axis angular velocity data, and spatial position coordinates of wearable devices. Calculate the suitable viewing index based on the wearable device's position information, and perform time alignment and normalization processing to obtain a state feature vector. The state feature vector is input into a preset behavior recognition neural network model; the behavior recognition neural network model uses a convolutional neural network to extract spatial displacement features, posture frequency features, energy decay features, and the rate of change of the suitable viewing index, and combines an attention mechanism to perform weighted analysis of temporal position changes; the output includes classification results for maintaining the viewing state, dynamically adjusting the state, and about to exit the state.

7. The method for seamless control of a display device based on visual sector entry / exit events according to claim 6, characterized in that, Based on the identified state, the seamless control of the display device includes: when it is determined that it is about to enter a state, triggering the display device to perform pre-wake-up and volume increase operations; when it is determined that it is in a viewing state, maintaining the current playback parameters of the display device; when it is determined that it is in a dynamic adjustment state, blocking false triggering commands caused by position jitter; and when it is determined that it is about to exit a state, triggering the display device to perform pause and mute operations.

8. A seamless control system for display devices based on visual sector entry and exit events, characterized in that, include: The location acquisition module acquires the location information of the wearable device worn by the user, based on the deployed remote control base station. The wearable device location information includes real-time angle information and real-time distance information; The sector determination module constructs a coordinate system with the remote control base station as the origin and the normal of the display device screen as the X-axis; in the coordinate system, it identifies the viewing area of ​​the display device and determines it as the visible sector. The pattern recognition module performs identification and analysis based on the wearable device's location information and the visible sector to determine the first motion mode and the second motion mode; For the first motion mode, determine the upcoming state of the wearable device's location information; For the second motion mode, a state feature vector containing spatial coordinates and a suitable viewing index is constructed; based on the state feature vector, the behavior state within the sector under the second motion mode is identified and analyzed. The behavioral states within the sector include the "stay watching" state, the "dynamic adjustment" state, and the "about to exit" state. The process of obtaining the suitable viewing index specifically includes: A weighted scoring model is used to calculate and integrate location suitability, posture suitability, and lighting suitability in real time to obtain a suitable viewing index. The location suitability is calculated as follows: based on the Gaussian mixture model, the historical location data of the user while maintaining a viewing state is clustered to generate a probability distribution cluster of suitable furniture locations. The probability value of belonging to the probability distribution cluster is calculated based on the current location coordinates to obtain the location suitability. The calculation of posture suitability involves: decomposing the position information into vertical height, combining the user's height data with the inertial sensor data of the wearable device to determine the user's posture; and identifying the posture suitability based on the time-series data of the user's posture. The calculation of the light suitability is as follows: the ambient light intensity is detected by the ambient light sensor integrated in the remote control base station, and the light suitability is obtained by combining and analyzing the temporal changes of the ambient light intensity with the pre-identified suitable light intensity. The contactless control module performs contactless control of the display device based on the recognized upcoming state and the behavior state within the sector.

Images