A notebook computer intelligent wake-up method and system based on pattern recognition

Abstract

The application relates to the technical field of pattern recognition, and discloses a notebook computer intelligent wake-up method and system based on pattern recognition, which introduces a brightness difference judgment in a low-power mode, effectively avoids starting high-power image processing when the ambient light changes are not obvious, and thus saves system resources. When the ambient brightness changes reach a preset threshold, the system is switched to an active working mode with high resolution and high frame rate, a second environment image with higher clarity and detail is acquired, and a user's spatial posture sequence is extracted based on the second environment image. In this way, the similarity between the spatial posture sequence and a pre-stored wake-up posture sequence is judged, the recognition difficulty in a complex scene of traditional face recognition is effectively avoided, accurate intelligent wake-up operation can be realized, the success rate of intelligent wake-up and the user experience are significantly improved, and the recognition failure problem caused by damaged image quality and face feature shielding in the prior art is overcome.

Description

Technical Field

[0001] This application relates to the field of pattern recognition technology, and more specifically, to a method and system for intelligent wake-up of a laptop computer based on pattern recognition. Background Technology

[0002] Modern laptops' smart wake-up function typically relies on a built-in camera to capture the user's image for authentication. However, in daily use, camera lenses can become contaminated with grease and dust due to contact, and strong backlighting can severely degrade image quality, resulting in glare, low contrast, and blurriness. When users wear masks, key facial features are obscured, making traditional recognition methods ineffective and causing frequent smart wake-up failures, thus impacting the user experience.

[0003] Specifically, when grease and dust adhere to the surface of a camera lens, light scatters as it passes through the lens, resulting in reduced image contrast and blurred details. In strong backlight conditions, this scattering effect is significantly amplified, producing glare and causing underexposure of the user's face, resulting in a dull, lacking-detail silhouette. At this point, the image data is filled with high levels of noise, extremely low contrast, and large areas of glare, making traditional noise reduction and normalization processes ineffective.

[0004] Furthermore, if a user wears a mask that obscures their mouth and nose, the difficulty of recognition increases significantly. With the facial contours already blurred and lacking detail due to underexposure, the mask completely obscures the key facial feature points that the facial recognition model relies on for location and analysis. Even areas that are not obscured, such as the eyes and eyebrows, are difficult to extract accurately due to the overall low contrast and blurriness of the image. This means that when pattern recognition models receive such highly impaired image input, they are almost unable to calculate any valid and reliable facial landmark vectors, thus failing to construct a feature description for identity comparison.

[0005] Ultimately, under the combined adverse conditions of a camera lens contaminated with everyday dirt, a user's face in a silhouette state with strong backlighting and underexposure, and the user wearing a mask obscuring key facial features, the system's segmentation logic for separating the foreground person from the background environment fails because it cannot find clear edges in the dim outline of low contrast and high glare. Since the user cannot be accurately separated from the background, the subsequent face recognition model cannot obtain a clean and effective recognition area. Without being able to calculate effective facial key point vectors, the system naturally cannot effectively match the currently captured features with pre-stored legitimate user feature data, thus failing to reach the preset confidence threshold. This causes the laptop's smart wake-up function to frequently reject wake-up commands from legitimate users, forcing users to switch to traditional, less efficient authentication methods, significantly degrading the user experience.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] This invention provides a method and system for intelligent wake-up of laptops based on pattern recognition, aiming to solve the technical problem that the image quality of existing laptop intelligent wake-up functions is severely degraded under adverse conditions such as dirty camera lenses, strong backlight, and users wearing masks, leading to recognition failure and affecting user experience.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a smart wake-up method for a laptop computer based on pattern recognition, comprising: S1: acquiring a continuous first environmental image, wherein the continuous first environmental image is an image acquired by a camera in a low-power operating mode; S2: determining whether the brightness difference of the continuous first environmental image exceeds a preset threshold; S3: when the brightness difference exceeds the preset threshold, acquiring a continuous second environmental image, wherein the second environmental image is an image acquired by a camera in an active operating mode, wherein the resolution and frame rate of the image acquired by the camera in the active operating mode are higher than those in the low-power operating mode, and extracting a user's spatial posture sequence based on the continuous second environmental image, wherein the user's spatial posture sequence includes at least one head posture parameter of head position, size, and tilt angle; S4: when the similarity between the spatial posture sequence and a pre-stored wake-up posture sequence reaches a predetermined threshold, performing a wake-up operation.

[0009] In one possible implementation, the step of extracting the spatial pose sequence based on the continuous second environment images includes: S21: performing grayscale processing on the continuous second environment images respectively and identifying and generating edge maps; S22: obtaining the user's head contour from the edge maps; S23: calculating spatial pose data based on the user's head contour, wherein multiple consecutive spatial pose data constitute the user's spatial pose sequence.

[0010] In one possible implementation, in step S22, a closed contour is found from the edge map, and the contour with the largest area and a shape that meets a predetermined shape is selected as the user's head contour, wherein the predetermined shape is an ellipse or a circle.

[0011] In one possible implementation, step S23 specifically includes: S231: extracting multi-source visual motion features from the continuous second environmental image based on the head contour, wherein the multi-source visual motion features include at least two of the following: motion centroid features, direction vector features, and patch center features; S232: weighted fusion of the extracted at least two of the multi-source visual motion features to obtain the user's head pose parameters; S233: smoothing the user's head pose parameters to obtain the user's spatial pose data.

[0012] In one possible implementation, the steps between S232 and S233 further include: S2331: reading data from the inertial measurement unit integrated inside the laptop computer; S2332: using the data from the inertial measurement unit to estimate the jitter offset of the camera between consecutive second environmental image frames in real time; S2333: The user's head posture parameters are reverse-compensated using the jitter offset to eliminate the influence of the camera's own jitter; in S233, the compensated user's head posture parameters are smoothed to obtain the user's spatial posture data.

[0013] In one possible implementation, the data from the inertial measurement unit includes 3-axis acceleration and 3-axis angular velocity. In step S2332, the data from the inertial measurement unit is processed using a sensor fusion algorithm to estimate the angular change and / or linear displacement of the camera around its coordinate axes as the jitter offset. In step S2333, the jitter offset is subtracted from the user's head posture parameters to obtain the compensated user's head posture parameters.

[0014] In one possible implementation, step S231 specifically includes: calculating pixel brightness differences for each pair of consecutive second environmental images to obtain brightness change values; marking pixels whose brightness change values ​​exceed a predetermined threshold as moving pixels; identifying a continuous region composed of all the moving pixels; selecting the region with the largest area and the longest duration of existence; and calculating its geometric center as the motion centroid feature; and / or processing the consecutive second environmental images to obtain pixel brightness gradients; statistically analyzing the histogram of brightness gradient directions to determine the most dominant gradient direction as the direction vector feature; and / or using a preset elliptical or circular template to perform template matching on the consecutive second environmental images; and determining the center of the region with the highest matching score as the patch center feature.

[0015] In one possible implementation, step S4 specifically includes: extracting head pose parameters that are the same as those in the pre-stored wake-up pose sequence from the spatial pose sequence, and calculating the Euclidean distance or cosine similarity between the same head pose parameters as the similarity for comparison.

[0016] In one possible implementation, the method further includes the user initially constructing a wake-up posture sequence for wake-up: capturing and recording a preset head dynamic sequence performed by the user, and extracting the user's head posture parameters from it and storing them as the wake-up posture sequence.

[0017] To achieve the above objectives, the present invention further provides a solution: a pattern recognition-based intelligent wake-up system for laptop computers, comprising: an acquisition module for acquiring continuous first environmental images, wherein the continuous first environmental images are images acquired by a camera in a low-power operating mode; a judgment module for judging whether the brightness difference of the continuous first environmental images exceeds a preset threshold; a processing module for acquiring continuous second environmental images when the brightness difference exceeds the preset threshold, wherein the second environmental images are images acquired by the camera in an active operating mode, wherein the resolution and frame rate of the images acquired by the camera in the active operating mode are higher than those in the low-power operating mode, and extracting a user's spatial posture sequence based on the continuous second environmental images, wherein the user's spatial posture sequence includes at least one head posture parameter of head position, size, and tilt angle; and an execution module for performing a wake-up operation when the similarity between the spatial posture sequence and a pre-stored wake-up posture sequence reaches a predetermined threshold.

[0018] Compared with existing technologies, the present invention provides a method and system for intelligent wake-up of laptop computers based on pattern recognition, which has the following beneficial effects: By introducing brightness difference judgment in low-power mode, high-power image processing is effectively avoided when ambient light changes are not significant, thus saving system resources. When the ambient brightness change reaches a preset threshold, the system switches to a high-resolution and high-frame-rate active working mode to acquire a clearer and more detailed second environmental image, and extracts the user's spatial pose sequence based on this. This head pose-based recognition method effectively avoids the recognition difficulties of traditional facial recognition in complex scenarios such as camera lenses being contaminated with grease and dust, strong backlighting causing image quality degradation, and users wearing masks obscuring key facial features. By judging the similarity between the spatial pose sequence and the pre-stored wake-up pose sequence, this application can achieve accurate intelligent wake-up operation, significantly improving the success rate of intelligent wake-up and user experience, and overcoming the recognition failure problems caused by image quality degradation and facial feature occlusion in the prior art. Attached Figure Description

[0019] Figure 1 This is a schematic flowchart of a smart wake-up method for a laptop computer based on pattern recognition, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the process of extracting spatial pose sequences based on continuous second environmental images according to the present invention; Figure 3 This is a flowchart illustrating the process of calculating spatial posture data based on the user's head contour in this invention. Figure 4 This invention is a schematic diagram illustrating the specific process of eliminating the influence of external mechanical vibrations on head posture parameters. Figure 5 This is a schematic diagram of a smart wake-up system for laptop computers based on pattern recognition, provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] The technical solution of the present invention will be described in detail below with reference to specific embodiments, but these embodiments are not intended to limit the scope of protection of the present invention.

[0023] Please see Figure 1 This is a schematic diagram of a pattern recognition-based intelligent wake-up method for laptops according to the present invention. The pattern recognition-based laptop wake-up method in this embodiment includes: S1: Acquire continuous first environmental images, which are images acquired by the camera in low-power working mode; S2: Determine whether the brightness difference of consecutive first environmental images exceeds a preset threshold; S3: When the brightness difference exceeds the preset threshold, acquire a continuous second environment image. The second environment image is an image acquired by the camera in the active working mode. The resolution and frame rate of the image acquired by the camera in the active working mode are higher than those in the low power working mode. Based on the continuous second environment image, extract the user's spatial posture sequence. The user's spatial posture sequence includes at least one of the head posture parameters, such as head position, size, and tilt angle. S4: When the similarity between the spatial pose sequence and the pre-stored wake-up pose sequence reaches a predetermined threshold, the wake-up operation is executed.

[0024] This application effectively solves the problems of low recognition rate and high wake-up failure rate of traditional methods in complex environments (such as lens contamination, strong backlight, and users wearing masks) by introducing low-power image prediction, high-resolution image acquisition, user space pose sequence extraction, and similarity comparison mechanism with pre-stored wake-up pose sequences, and significantly improves the accuracy of intelligent wake-up and user experience.

[0025] To facilitate a clearer understanding of the technical solutions in this application, some key terms and implementation environments are first explained. In this application, "low-power operating mode" refers to a camera (e.g., a low-resolution, low-frame-rate CMOS sensor, such as the OV7670 or a similar model) acquiring images at a lower frame rate and resolution to save power, suitable for continuous environmental monitoring. For example, it can be set to capture 1 to 2 frames per second, and the image resolution can be limited to 320×240 pixels or even lower. "Active operating mode" refers to a camera acquiring images at a higher frame rate and resolution to obtain more detailed visual information, suitable for scenarios requiring fine pattern recognition. For example, the frame rate can be increased to 15-30 frames per second, and the resolution may be appropriately increased, such as VGA 640×480 pixels. "Brightness difference" refers to the degree of pixel brightness variation between consecutive image frames, reflecting changes in ambient light or significant motion in the image. "Spatial pose sequence" refers to the dynamic trajectory of a user's head over a period of time, including a continuous set of parameters such as head position, size, and tilt angle. "Wake-up posture sequences" are pre-stored specific head dynamic patterns used to trigger wake-up operations. This method is typically deployed inside a laptop, utilizing its integrated camera for image acquisition and performing image processing and pattern recognition algorithms via the built-in processor.

[0026] The image processing unit of a laptop system (e.g., a low-power microcontroller integrated on the laptop motherboard, such as an ARM Cortex-M series processor) receives these low-resolution first ambient images and performs a simple inter-frame difference algorithm for motion detection. Specifically, it compares the pixel brightness differences between the first ambient image of the current frame and the first ambient image of the previous frame. If the difference exceeds a preset threshold (e.g., more than 5% of the pixels have brightness exceeding a certain fixed value), and these changing pixels form a sufficiently large continuous area in space (e.g., greater than 10% of the total pixels in the image), it is determined that there is potential user activity. This method can effectively filter out minor noise in the environment, and even when strong backlighting and lens contamination cause overall image blurring, it can still detect someone approaching or moving through large-area brightness changes, thus avoiding complex and power-consuming analysis of each frame.

[0027] Once initial motion detection detects activity, the system switches to motion stick mode, which is the camera's operating mode, including a simultaneous increase in the frame rate and resolution of the acquired images.

[0028] Specifically, the intelligent wake-up method of this application includes the following detailed description of its main features: In step S1, a series of first environmental images are acquired. These first environmental images are acquired by a camera in low-power operating mode. As one implementation, the camera can be configured to continuously acquire images at a frame rate of 5 frames per second and a resolution of 320x240 pixels. These images are transmitted to a processing unit for preliminary analysis.

[0029] In step S2, it is determined whether the brightness difference between consecutive first environmental images exceeds a preset threshold. For example, the processing unit can calculate the average brightness difference between two consecutive frames of the first environmental images, or calculate the brightness change in a specific area of ​​the image. If the calculated brightness difference value is greater than the preset threshold (e.g., 10 gray levels), it is considered that the environment has changed significantly, and there may be a situation where a user is approaching or moving.

[0030] In step S3, when the brightness difference exceeds a preset threshold, the system acquires consecutive second environmental images. At this time, the camera switches to active working mode to acquire images at a higher resolution and frame rate (e.g., 30 frames per second, 1280x720 pixels). Based on these consecutive second environmental images, the system extracts the user's spatial pose sequence. This spatial pose sequence may include at least one head pose parameter among head position, head size, and head tilt angle. For example, the head position can be determined by analyzing the pixel distribution of the head region in the image, the head size can be estimated by the area of ​​the head region, and the head tilt angle can be determined by the angle between the head's principal axis direction and the vertical direction.

[0031] In step S4, when the similarity between the spatial pose sequence and the pre-stored wake-up pose sequence reaches a predetermined threshold, the system performs a wake-up operation. For example, the system can compare the currently extracted spatial pose sequence with multiple pre-stored wake-up pose sequences. The comparison can be performed by calculating the Euclidean distance or cosine similarity between the sequences. If the similarity reaches a preset threshold (e.g., 0.8), it is considered that the user is performing the expected wake-up action, thereby triggering the laptop to wake up.

[0032] If the similarity between the spatial pose sequence and the stored wake-up pose sequence of a legitimate user reaches a preset threshold, the system will determine that it is a wake-up command issued by a legitimate user. At this time, the laptop's power management module (e.g., ACPI controller) will receive the wake-up signal and immediately wake the device from sleep mode. This includes turning on the screen, loading the operating system, activating all hardware components, and preparing it for user use. If the similarity does not reach the threshold, the system will remain in sleep mode or return to low-power monitoring mode (step S1) to wait for the next potential wake-up attempt. To prevent false wake-ups, a short cooldown period can be set, for example, not responding to wake-up requests for the next 5 seconds after a failed attempt.

[0033] The intelligent wake-up method of this application, by coordinating the aforementioned technical features, effectively solves the problem of low recognition rate in complex environments using traditional methods. First, through steps S1 and S2, the system continuously monitors the environment in a low-power operating mode, switching to a high-power active operating mode only when a significant brightness change is detected. This tiered wake-up mechanism avoids the camera remaining in a high-power state for extended periods, thus extending battery life. When the brightness difference exceeds a preset threshold, indicating potential user activity, the system switches to the active operating mode in step S3 to acquire a second environmental image at higher resolution and frame rate. This high-resolution and high-frame-rate image acquisition provides richer detail information, even under conditions of slight lens contamination or strong backlighting, laying the foundation for subsequent user spatial pose sequence extraction. By extracting the user's spatial pose sequence, this application avoids the strong reliance of traditional face recognition methods on clear facial features. Even when the user is wearing a mask or their facial features are unclear, the user's intent can be identified through dynamic changes in head posture. Finally, in step S4, the extracted spatial pose sequence is compared with a pre-stored wake-up pose sequence. This recognition method based on dynamic posture patterns has higher robustness and anti-interference ability compared to static facial feature recognition. It can effectively cope with adverse conditions such as lens contamination, strong backlight, and users wearing masks, thereby significantly improving the success rate of smart wake-up and user experience.

[0034] The core innovation of this application lies in its hierarchical wake-up mechanism and pattern recognition method based on user spatial pose sequences. Compared with the closest existing technologies, which typically rely directly on high-resolution images for face recognition, the recognition effect is significantly reduced when the ambient light is poor or the user's face is obscured. For example, when the camera lens is dusty or the user is in a strong backlight environment, the image quality acquired by traditional methods will be severely degraded, making it difficult to extract facial features and thus causing recognition failure. This application effectively filters out irrelevant wake-up attempts by introducing brightness difference judgment in a low-power operating mode, and only initiates high-resolution image acquisition when necessary, thereby significantly reducing system power consumption while ensuring recognition effect. More importantly, this application avoids over-reliance on clear facial features by extracting the user's spatial pose sequence and comparing it with a pre-stored wake-up pose sequence. This method can still identify the user's wake-up intention through dynamic changes in head posture even when the user is wearing a mask or their facial features are obscured, greatly improving the robustness and applicability of intelligent wake-up. For example, the user can trigger wake-up through specific head movements or nodding without removing the mask or adjusting the lighting. This innovative approach enables laptops' smart wake-up function to maintain high accuracy and reliability in various complex and adverse usage scenarios, providing users with a more convenient and intelligent interactive experience.

[0035] Specifically, the steps described above for extracting the user's spatial pose sequence based on consecutive second environmental images can be further refined. Please refer to [link / reference]. Figure 2 As shown, the steps for extracting a spatial pose sequence based on consecutive second environment images include: S21: Perform grayscale processing on consecutive second environment images and identify and generate edge maps; S22: Obtain the user's head outline from the edge map; S23: Spatial pose data is calculated based on the user's head contour, and multiple consecutive spatial pose data constitute the user's spatial pose sequence.

[0036] Specifically, in step S21, the consecutive second environmental images are converted to grayscale to simplify the image data and reduce the computational complexity of subsequent processing. Then, edge detection algorithms, such as Canny, Sobel, or Prewitt operators, are applied to identify and generate edge maps from the grayscale images. These edge maps effectively highlight the boundaries and structural information of objects in the image, providing a foundation for subsequent contour extraction.

[0037] Further, in step S22, the user's head contour is obtained from the generated edge map. This process typically involves contour-finding algorithms to identify closed or nearly closed curves in the image. By analyzing the geometric properties of these contours, such as area, perimeter, and shape, the contour that best matches the user's head features can be selected, thereby accurately locating the user's head region.

[0038] In practice, the image can first be converted to grayscale, and then the Canny edge detection algorithm can be applied to identify all edges in the image. The Canny algorithm generates a clear edge map by calculating the gradient intensity and direction of the image and performing non-maximum suppression and double thresholding.

[0039] Next, the system uses a contour search algorithm (e.g., the Suzuki algorithm) to find closed contours in the edge map. In step S22, closed contours are searched from the edge map, and the contour with the largest area and a shape that meets a predetermined shape is selected as the user's head contour. Here, the predetermined shape is an ellipse or a circle. Since the user's head is usually the largest and relatively regular moving object in the image, the system selects the contour with the largest area and a shape that is approximately elliptical or circular as a potential head contour. In this invention, "approximately circular or elliptical" means that, in the view of those skilled in the art, its contour basically conforms to the common understanding of circles or ellipses, allowing for minor deviations due to manufacturing tolerances, design optimizations, and other factors.

[0040] Specifically, finding closed contours from an edge map refers to using image processing techniques, such as connected component analysis or contour tracking algorithms, to identify continuous, end-to-end boundary lines of all pixels in the edge map, thus forming a series of independent closed regions. These closed contours represent the boundary information of different objects in the image.

[0041] The selection process involves choosing the contour with the largest area and a shape that conforms to a predetermined shape as the user's head contour. The goal is to accurately locate and extract the contour that best represents the user's head from among numerous identified closed contours. Specifically, the "largest area" selection criterion is based on the assumption that the human head typically occupies a relatively large area in an image, which helps to exclude small contours caused by noise or small background objects. Meanwhile, the "shape conforms to a predetermined shape" selection criterion utilizes the geometric characteristic that the human head typically appears approximately elliptical or circular in two-dimensional projection, effectively distinguishing the head contour from contours of other irregularly shaped objects. In practical applications, the predetermined shape can be understood as an ellipse or circle. For example, it can be determined whether it conforms to a preset elliptical or circular template by calculating the contour's eccentricity, aspect ratio, or performing shape matching.

[0042] Subsequently, in step S23, spatial pose data is calculated based on the acquired user head contour. Spatial pose data may include parameters such as head position, size, and tilt angle.

[0043] For example, by calculating the minimum bounding rectangle or ellipse of the contour, its major axis, minor axis, and tilt angle can be obtained. Simultaneously, by combining the centroid position of the contour and its vertical offset relative to the bottom of the image (assuming it's the shoulder or body area), the degree of head tilt, rotation, and forward tilt can be roughly estimated. For instance, a large angle between the contour's major axis and the vertical direction may indicate a head tilt; a shift of the centroid upwards in the image may indicate a forward tilt. This method does not rely on facial details but leverages the relatively stable contour information in silhouette mode, providing coarse-grained head shape and pose information even under low contrast, high glare, and mask obscuring conditions.

[0044] For example, the geometric center of the head contour can be calculated as the head position, and the bounding box dimensions of the contour can be used as the head dimensions. For instance, the approximate head dimensions (major axis length L, minor axis length S) can be roughly estimated using the matching template dimensions of C_patch or the bounding box dimensions of the C_motion region. The head tilt angle can then be determined using principal component analysis (PCA) or by fitting the major axis of an ellipse. Multiple consecutive spatial pose data are collected to form a sequence of the user's spatial poses, reflecting the dynamic changes of the user's head over a period of time.

[0045] This application's solution decomposes the complex spatial pose sequence extraction process into a series of clearly defined steps, including grayscale processing, edge map generation, head contour acquisition, and spatial pose data calculation, achieving an effective conversion from raw image data to high-level semantic information (user spatial pose). First, grayscale processing and edge map generation preprocess the image and extract features, removing redundant information and enhancing key structural features. Second, by accurately acquiring the user's head contour from the edge map, the focus can be concentrated on the user's head as the core object. Finally, spatial pose data is calculated based on the head contour and organized into a sequence, making the quantification and analysis of the user's head dynamic behavior possible. This step-by-step processing approach makes the entire pose extraction process clearer and more controllable, and provides structured input for subsequent pattern recognition.

[0046] Through the above technical solution, this application provides a more specific and operable method for extracting user space pose sequences. Compared to merely mentioning the extraction of spatial pose sequences in general terms, this solution, through clearly defined steps S21, S22, and S23, makes the implementation path of the pose extraction process clearer and reduces the ambiguity of the technical solution. This refined processing flow helps improve the accuracy and robustness of pose extraction, ensuring stable acquisition of the user's head pose information under different environmental conditions. This provides a high-quality data foundation for subsequent wake-up pose sequence comparison, thereby improving the reliability of laptop smart wake-up and user experience.

[0047] In some embodiments of this application, when extracting the user's spatial pose sequence based on consecutive second environmental images, it is necessary to obtain the user's head contour from the edge map. However, in practical applications, especially in environments with complex backgrounds, after grayscale processing of consecutive second environmental images and generating edge maps, multiple closed contours may exist in the edge maps. These contours may correspond to the user's head, other objects in the background, or environmental noise. If these contours are not effectively distinguished and filtered, the system may mistakenly identify non-head contours as the user's head, thereby affecting the accuracy of subsequent spatial pose data calculation and reducing the reliability of the intelligent wake-up method.

[0048] This application's solution introduces a mechanism for finding closed contours and filtering by area and shape when acquiring the user's head contour from the edge map. First, by finding closed contours, the system can identify all potential object boundaries. Then, by comparing the area of ​​these closed contours, non-head contours that are too small or too large can be initially eliminated. More importantly, by performing shape matching on the remaining contours to determine if they meet preset elliptical or circular shapes, the system can further and more accurately pinpoint the user's head contour, effectively avoiding misidentification of other misshapen objects as heads. This multi-stage filtering mechanism significantly improves the accuracy and robustness of head contour recognition.

[0049] The above technical solution effectively addresses the problem of accurately identifying a user's head contour when multiple closed contours exist in the edge map against a complex background. By introducing filtering conditions based on maximum area and shape (such as ellipse or circle), the accuracy and robustness of head contour recognition can be significantly improved, and the false recognition rate reduced. This provides a reliable foundation for subsequent accurate calculation of the user's spatial pose data, thereby enhancing the reliability of the entire intelligent wake-up method and the user experience.

[0050] In some preferred embodiments, it is assumed that after acquiring consecutive second environmental images and performing grayscale processing and edge map generation, multiple closed contours are identified in the edge map. For example, there may be a large elliptical contour corresponding to the user's head, while some smaller circular contours correspond to light bulbs in the background, and some irregularly shaped contours correspond to the edges of tables and chairs. In this case, the system first calculates the area of ​​all closed contours and evaluates the degree to which their shapes match preset elliptical or circular templates. Through filtering, the contour with the largest area and the shape closest to an ellipse or circle (e.g., determined by calculating the eccentricity or major-minor axis ratio of the contour) will be ultimately determined as the user's head contour. For example, if the user's head contour area is 5000 pixels, the light bulb contour area is 500 pixels, and the table and chair edge contour area is 2000 pixels and has an irregular shape, then the head contour with the largest area and an elliptical shape will be preferentially selected.

[0051] When the system switches to active working mode, it continuously captures a second environment image at a frequency of, for example, 15-30 times per second. Based on these second environment images, it derives a series of spatial pose data of the user's head, including parameters such as the centroid coordinates (x, y) of the user's head contour, the major axis length L, the minor axis length S, and the tilt angle Theta. The system organizes these continuous pose parameters into a short time series, for example, recording all pose data points within the last 3 seconds.

[0052] To form a "dynamic pose sequence," the system monitors the trends of these parameters over time. For example, it can calculate the displacement vector of the head's centroid between consecutive frames, as well as the rate of change of the tilt angle and the ratio of the major and minor axes.

[0053] A simple sequence representation can be: Sequence=[(x1,y1,L1,S1,Theta1),(x2,y2,L2,S2,Theta2),...,(xn,yn,Ln,Sn,Thetan)] Where n is the length of the sequence (e.g., 30 frames of images in 3 seconds, n=90).

[0054] The system will extract key motion features from these sequences, such as: Average and maximum speed of head movements.

[0055] The magnitude of the change in head tilt angle.

[0056] The number of times the head changes direction.

[0057] The tendency of the head to tilt forward or backward (as indicated by changes in the y-coordinate).

[0058] These features collectively describe the unique movement trajectory and pattern of a user's head during the wake-up process.

[0059] However, in certain application scenarios, this technical solution faces unexpected challenges when performing "dynamic wake-up posture sequence capture." For example, when the system operates in a strong backlight environment, if there is not only a primary strong light source behind the user, but also multiple secondary objects or surfaces with localized dynamic brightness changes in the background, such as an auxiliary monitor playing video, or a polished desktop area that reflects ambient light, a complex physical phenomenon occurs when the camera lens surface has accumulated a film of grease and dust due to daily use. This uneven film causes the incident light to scatter, and the light from these locally dynamically changing light sources or reflections in the background produces additional, transient glare and light spots when passing through the contaminated lens. These glare and light spots are not uniformly distributed, but rather irregularly superimposed on the silhouette area of ​​the user's head, or forming interference near its edges.

[0060] The combined effect results in the system's output edge map no longer representing a single, continuous, and closed head contour when attempting to extract the user's head contour using methods like Canny edge detection. Instead, due to local glare and brightness variations in light spots, the processing method may incorrectly identify these disturbances as valid edges in the image, or cause certain parts of the true head contour to be "washed out" and broken. Consequently, the system frequently identifies multiple small, irregular, and separated contour fragments instead of a complete, coherent head shape. In this fragmented and discontinuous head contour, the system encounters significant stability issues when subsequently attempting to calculate the geometric features of the head contour. For example, if the system identifies multiple potential head fragments in a frame, it might select one as the "primary" head contour based on area size or other heuristics. However, in the next frame, due to changes in dynamic lighting in the background or further lens scattering, this "primary" contour may disappear or be replaced by another newly emerging fragment, forcing the system to switch its recognition target. This switching causes the calculated head centroid coordinates (x, y), major axis length L, minor axis length S, and tilt angle Theta to exhibit drastic, non-physical jumps between consecutive frames. For example, the head centroid may "move" dramatically within a few frames, and the estimated tilt angle may swing rapidly in both positive and negative directions, severely deviating from the user's actual smooth head movement trajectory. These highly unstable posture parameters prevent the system from constructing an accurate and reliable "dynamic wake-up posture sequence" because the data points in the sequence have lost their continuity and physical meaning, failing to reflect the user's true head movement patterns.

[0061] Therefore, in situations where the laptop camera lens has everyday dirt, the user's face is in strong backlight, and there are secondary light sources or reflective surfaces with localized dynamic brightness changes in the background, causing the user's head silhouette to be fragmented and discontinuous, resulting in drastic, non-physical jumps in the estimated geometric features of the head contour between consecutive frames, it is crucial to still stably and accurately capture and construct a dynamic pose sequence reflecting the user's actual head movement pattern. Based on this, this solution further optimizes the specific implementation of calculating the spatial pose data for the user's head contour. The system no longer relies solely on a single edge detection result to obtain the head contour, but instead simultaneously fuses and weights multiple sources of visual motion features to accurately determine the user's head pose parameters. These feature extractions are performed in parallel on the laptop's main processor or a dedicated visual processing unit (e.g., an Intel Movidius Myriad XVPU integrated on the motherboard) to ensure real-time performance.

[0062] For details, please refer to Figure 3 The steps described above for calculating spatial pose data based on the user's head contour specifically include: S231: Extract multi-source visual motion features from consecutive second environment images based on head contours. The multi-source visual motion features include at least two of the following: motion centroid features, orientation vector features, and patch center features. S232: Weighted fusion of at least two extracted multi-source visual motion features to obtain the user's head pose parameters; S233: Smooth the user's head pose parameters to obtain the user's spatial pose data.

[0063] Specifically, in step S231, multi-source visual motion features refer to various types of visual information extracted from the image that can reflect changes in the user's head movement or posture. These features can independently provide characteristics about head posture, and their combination can improve the accuracy and robustness of posture estimation. Among them, the motion centroid feature can be understood as the geometric center of the head motion region, reflecting the overall displacement trend of the head in the image; the direction vector feature can be understood as the dominant direction of the pixel brightness gradient within the head region, reflecting the rotation or tilt of the head; and the patch center feature can be understood as the center of the key head region identified through template matching, reflecting the specific position of the head in the image. The extraction of these features aims to capture key information about head posture from different dimensions.

[0064] Specifically, step S231 can be further specified as follows: calculate the pixel brightness difference for each pair of consecutive second environmental images to obtain the brightness change value, mark the pixels with brightness change values ​​exceeding a predetermined threshold as moving pixels, identify the continuous region composed of all moving pixels, and select the region with the largest area and the longest duration of existence, and calculate its geometric center as the motion centroid feature.

[0065] For example, the system continuously captures image frames and calculates pixel brightness differences between consecutive frames. For each pair of consecutive frames I_t and I_(t-1), its absolute brightness difference map D_t = |I_t - I_(t-1)| is calculated. Then, D_t is binarized, and a brightness change threshold T_diff (e.g., 30 gray levels) is set. Pixels with brightness changes exceeding T_diff are marked as moving pixels. Next, a connected component analysis algorithm (e.g., the findContours function in OpenCV) is used to identify all continuous regions composed of moving pixels. The system selects the moving region with the largest area and the longest duration and calculates its geometric center as the "motion centroid feature".

[0066] For example, for a pixel (u,v), its brightness value in frame t is L(u,v,t). The inter-frame brightness difference D(u,v,t) = |L(u,v,t) - L(u,v,t-1)|. If D(u,v,t) > T_diff, then the pixel is considered a moving pixel.

[0067] The center of motion is C_motion=(Sum(u) / N,Sum(v) / N), where N is the number of moving pixels.

[0068] And / or step S231 can also be specified as: processing the continuous second environment image to obtain the pixel brightness gradient, and statistically analyzing the histogram of the brightness gradient direction to determine the most important gradient direction as the direction vector feature.

[0069] For example, in a pre-defined area of ​​the screen where a user's head might appear (e.g., the central area of ​​the upper half of the screen), the system calculates the dominant directions of the pixel brightness gradients within that area. This is done by processing the image using Sobel or Scharr operators to obtain the horizontal and vertical gradient components Gx and Gy. Then, within the local region, a histogram of gradient directions is plotted to identify the most significant directions. These direction vectors can indicate trends in local brightness changes; for example, if most gradients point in a certain direction, it might mean that the edge of an object is moving or appearing in that direction. The system selects the most dominant gradient directions as "direction vector features."

[0070] For example, the gradient direction Theta(u,v) = atan2(Gy(u,v),Gx(u,v)). Within the region of interest, the gradient directions of all pixels are counted, and the angle intervals that appear most frequently are found.

[0071] And / or step S231 can also be specified as: using a preset elliptical or circular template, performing template matching on consecutive second environmental images, and determining the center of the region with the highest matching score as the patch center feature.

[0072] For example, the system presets several elliptical or circular templates of different sizes and aspect ratios to represent the approximate head shape. On low-resolution grayscale images, the system uses methods such as Normalized Cross-Correlation (NCC) to perform rapid template matching within the image, searching for "moving patches" most similar to these blurred head templates. The matching results provide the center location and matching score of the patch. The system selects the patch center with the highest score and a reasonable location as the "patch center feature." This matching does not pursue high precision, but rather aims to quickly locate the approximate head region in a blurred image.

[0073] In step S232, weighted fusion refers to combining at least two extracted multi-source visual motion features according to preset weights to generate the user's head pose parameters. For example, different weights can be assigned based on the contribution, reliability, or sensitivity to noise of different features to head pose estimation. Through weighted fusion, the advantages of various features can be comprehensively utilized to compensate for the shortcomings of a single feature, thereby obtaining more comprehensive and accurate head pose parameters. The head pose parameters may include the head's position, size, and tilt angle, which together describe the three-dimensional spatial state of the user's head.

[0074] The system will perform a weighted fusion of the three motion features extracted above (motion centroid feature C_motion, direction vector feature V_direction, and patch center feature C_patch) to infer the user's head pose in the current frame. The fusion logic will assign different weights based on the reliability of the features.

[0075] a. Location fusion: The inference of the head center position (x, y) is mainly achieved by fusing C_motion and C_patch. If the matching score of C_patch is high, it indicates good fuzzy shape matching, and the weight of C_patch can be higher. If the motion region corresponding to C_motion is large and continuous, the weight of C_motion is even higher.

[0076] b. Integration of sports trends: The inference of head tilt or rotation trends is primarily achieved by fusing the displacements of V_direction and P_fused between consecutive frames. V_direction provides an instantaneous direction indication, while the continuous changes in P_fused provide a more stable motion trajectory. The system integrates this information; for example, if V_direction indicates a leftward movement and P_fused is also moving to the left, the determination of a leftward rotation is strengthened.

[0077] For example, the head tilt angle Theta_fused can be estimated by analyzing the vertical displacement of P_fused across consecutive frames and the vertical component of V_direction. If V_direction is upward and the y-coordinate of P_fused is decreasing (moving upward), it may indicate a forward head tilt.

[0078] c. Size estimation: The approximate dimensions of the head (major axis length L, minor axis length S) can be roughly estimated using the matching template size of C_patch or the bounding box size of the C_motion region. Since these estimates may be inaccurate, the system employs a smoothing filter (e.g., a Kalman filter or an exponentially weighted moving average) to smooth these dimensional parameters and reduce transient jumps.

[0079] In step S233, smoothing refers to filtering or smoothing the obtained user's head pose parameters to eliminate instantaneous fluctuations caused by image noise, feature extraction errors, or brief atypical movements, thereby obtaining more stable and continuous user spatial pose data. Smoothing can be implemented using moving average, Kalman filtering, or other time series smoothing algorithms, with the aim of improving the stability and usability of the pose data and making it more consistent with the actual head movement trajectory.

[0080] This application's solution effectively addresses the issue of insufficient robustness of single visual features in complex environments by refining the spatial pose data calculation process into the extraction, weighted fusion, and smoothing of head pose parameters from multiple sources of visual motion features. Specifically, by extracting at least two types of multi-source visual motion features, such as motion centroid features, orientation vector features, and patch center features, the motion and pose information of the user's head can be captured from different angles, avoiding pose estimation bias caused by limitations of single features. Subsequently, these multi-source visual motion features are weighted and fused, allowing for optimized combination based on the reliability and importance of each feature, thereby extracting more accurate head pose parameters from information redundancy. Finally, the fused head pose parameters are smoothed to effectively filter out noise and transient jitter, ensuring that the output spatial pose data has good continuity and stability, providing high-quality input for subsequent wake-up pose sequence comparison.

[0081] Through the above technical solutions, this application can significantly improve the accuracy and robustness of user spatial pose data calculation. The introduction of multi-source visual motion features enables the system to effectively capture head pose information even when facing challenges such as changes in lighting, partial occlusion, or complex backgrounds. The weighted fusion mechanism further optimizes the contribution of different features, ensuring the comprehensiveness and accuracy of pose parameters. In addition, the smoothing processing of head pose parameters effectively suppresses noise and instantaneous errors, making the final spatial pose data more stable and reliable. This provides a more accurate and reliable basis for subsequent wake-up judgment, improving the overall performance and user experience of the intelligent wake-up method.

[0082] This improved approach, through the mutual corroboration of multiple features, enables the system to reliably capture the overall movement trajectory of the user's head even when any single visual information is unreliable. In other words, pursuing a single "perfect" visual feature is impractical when visual information is highly impaired. Instead of striving to "see" every detail from the blur, it instead "perceives" multiple blurred motion signals and converges them into a clear motion judgment. This method avoids pursuing pseudo-precision in uncertainty, instead enhancing overall robustness through feature convergence. Thus, even in specific environments where conventional solutions are ineffective, it can still stably and accurately capture and construct a spatial posture sequence reflecting the user's true head movement pattern, achieving intelligent wake-up.

[0083] In some of the above embodiments, the user's head pose parameters can be obtained by extracting and weighted fusion of multi-source visual motion features from consecutive second environmental images. However, in practical applications, the laptop's camera may be affected by external vibrations or user operations, resulting in its own shaking. This shaking introduces additional motion information, leading to inaccurate head pose parameters extracted from the image, thus affecting the accuracy of subsequent spatial pose sequences and the reliability of wake-up operations.

[0084] To address this, this application further proposes an optimization scheme that can eliminate the influence of external mechanical vibrations on head posture parameters. Please refer to [link to relevant documentation]. Figure 4 The step between S232 and S233 includes: S2331: Read data from the inertial measurement unit integrated inside the laptop computer; S2332: Real-time estimation of camera jitter offset between consecutive second environment image frames using data from the inertial measurement unit; S2333: The user's head posture parameters are compensated in reverse using the jitter offset to eliminate the influence of the camera's own jitter. In S233, the compensated user's head pose parameters are smoothed to obtain the user's spatial pose data.

[0085] Specifically, an inertial measurement unit (IMU) is a device capable of measuring an object's three-axis attitude angles (or angular velocities) and acceleration. It is typically integrated inside a laptop computer to sense the device's motion state. In step S2331, the data from the inertial measurement unit may include information such as three-axis acceleration and three-axis angular velocity, which can reflect the camera's minute movements in real time.

[0086] Furthermore, in step S2332, the jitter offset of the camera between consecutive second environmental image frames can be estimated in real time using the data from the inertial measurement unit. This jitter offset refers to the relative displacement or angular change caused by the camera's own jitter between image frames.

[0087] Subsequently, in step S2333, the jitter offset is used to perform reverse compensation on the user's head pose parameters. The purpose of reverse compensation is to remove the motion components caused by the camera's own jitter from the extracted head pose parameters, thereby obtaining parameters that are closer to the user's actual head pose.

[0088] Therefore, in step S233, the user's head posture parameters after jitter offset reverse compensation are smoothed to further eliminate noise and finally obtain more accurate and stable user spatial posture data.

[0089] For example, when a laptop's built-in camera (e.g., a CMOS image sensor such as the OmniVision OV7670 or a more advanced model) captures image frames at an enhanced frame rate (e.g., 15-30 frames per second), the system simultaneously reads real-time data from the laptop's integrated inertial measurement unit (IMU, e.g., a 6-axis MEMS sensor such as the Bosch BMI160 or InvenSense MPU-6050, which are commonly used for hard drive protection or screen orientation detection). This IMU data includes 3-axis acceleration (ax, ay, az) and 3-axis angular velocity (gx, gy, gz). To ensure data time consistency, the timestamps of the camera image frame capture and the IMU data read are precisely synchronized, for example, through hardware interrupts or high-precision timers.

[0090] Precisely synchronized IMU data is crucial for estimating the camera's own micro-shakes. Only by knowing the exact motion state of the camera when capturing each frame can effective compensation be performed.

[0091] The processing unit in the system (e.g., a laptop's main processor or a dedicated sensor fusion microcontroller, such as the STM32L series) uses the synchronously acquired IMU data to estimate in real time the minute displacements and angular changes caused by mechanical vibrations in the camera between consecutive image frames or during a single frame exposure. This is typically achieved through a sensor fusion algorithm, such as a simplified complementary filter or a Kalman filter.

[0092] In practice, gravity compensation can be performed on the IMU's acceleration data first, that is, the gravity component under the current attitude is subtracted from the original acceleration to obtain pure linear acceleration. Then, the angular velocity data is integrated to estimate the angle change, and the linear acceleration data is integrated to estimate the linear displacement.

[0093] For example, for angular velocity data (gx, gy, gz), the small angular changes (Δθx, Δθy, Δθz) of the camera around the X, Y, and Z axes within the frame interval Δt can be approximated as follows: Δθx=gx×Δt Δθy=gy×Δt Δθz=gz×Δt These angular changes directly reflect the camera's own micro-shakes. For linear displacement, since double integration easily introduces drift, in low-amplitude scenarios, it is more practical and robust to focus on the impact of angular jitter on visual pose.

[0094] Accurately estimating the camera's own jitter offset is a prerequisite for effectively compensating for subsequent visual pose parameters.

[0095] The system uses the estimated camera shake offset to compensate for the user's head pose parameters extracted from the image in real time. These pose parameters include the centroid coordinates (Cx, Cy) of the head contour, the major axis length L, the minor axis length S, and the tilt angle Theta.

[0096] In practice, if the estimated camera rotation in the current frame relative to the previous frame is a small angle (Δθx, Δθy, Δθz), these rotation amounts can be applied inversely to the visually estimated head tilt angle Theta. For example, if the camera rotates Δθz around its Z-axis (usually perpendicular to the image plane), then the visually estimated head tilt angle Theta_raw needs to be subtracted from this Δθz to obtain the compensated angle Theta_compensated.

[0097] Theta_compensated=Theta_raw-Δθz For compensation of the centroid coordinates (Cx, Cy), if the camera undergoes linear displacement (Δx, Δy), these displacements can be applied inversely to the centroid coordinates.

[0098] After vibration compensation is completed, in order to further eliminate residual noise and ensure the smoothness of attitude parameters, the system applies a low-pass filter, such as an Exponential Weighted Moving Average (EWMA) filter.

[0099] For example, for the compensated attitude parameter P_compensated, its smoothing value P_smoothed can be calculated as: P_smoothed(t)=α×P_compensated(t)+(1-α)×P_smoothed(t-1) Here, α is a smoothing coefficient (e.g., between 0.1 and 0.3) used to balance the weights of the current value and historical values.

[0100] By directly compensating for pose parameters, rather than through complex image pixel correction, noise introduced by camera micro-shakes can be eliminated efficiently and accurately, making pose parameters closer to the user's actual head movements.

[0101] The system continuously records the head posture parameters (Cx_smoothed, Cy_smoothed, L_smoothed, S_smoothed, Theta_smoothed) after vibration compensation and smoothing in step 3 at a frequency of 15-30 times per second. These parameters are organized into a short time series, for example, recording all posture data points within the most recent 3 seconds.

[0102] Sequence=[(Cx_smoothed1,Cy_smoothed1,L_smoothed1,S_smoothed1,Theta_smoothed1),...,(Cx_smoothedn,Cy_smoothedn,L_smoothedn,S_smoothedn,Theta_smoothedn)] From this sequence, the system extracts key motion features, such as the average speed and maximum speed of head movements, the magnitude of changes in tilt angle, the number of changes in rotation direction, and the tendency to tilt forward or backward. These features can now accurately and stably describe the unique motion trajectory and patterns of the user's head during the wake-up process because they have effectively removed noise introduced by mechanical vibrations.

[0103] After compensation and smoothing, the posture parameters can be used to construct a dynamic posture sequence that truly reflects the user's intent, providing reliable input for subsequent accurate comparison.

[0104] This application's solution effectively addresses the impact of camera shake on the accuracy of head pose parameter extraction by incorporating data from an inertial measurement unit (IMU). Specifically, when the camera shakes, the IMU senses and records the acceleration and angular velocity data generated by this shake in real time. This data is used to accurately estimate the camera's shake offset between consecutive image frames. Subsequently, this shake offset is applied to reverse-correct the head pose parameters extracted through visual features. This reverse compensation mechanism separates the camera's own motion from the user's head motion, thereby eliminating errors introduced by camera shake and resulting in purer and more accurate extracted head pose parameters. Finally, the compensated head pose parameters are smoothed, further improving the stability and reliability of the spatial pose data.

[0105] Through the above technical solution, this application can effectively overcome the interference of camera shake on the accuracy of user head posture parameter extraction in traditional methods, and significantly improve the accuracy and stability of spatial posture sequences. Therefore, when comparing the similarity between spatial posture sequences and pre-stored wake-up posture sequences, it can more accurately reflect the user's true intention, reduce the probability of false or missed wake-ups, and thus improve the reliability of the laptop's smart wake-up method and the user experience.

[0106] In some preferred embodiments, suppose a user places a laptop on a desktop and attempts to wake it up via a head gesture sequence. During this process, the desktop may experience slight vibrations due to minor touches or environmental vibrations, causing minor shaking of the laptop and its integrated camera. Without a shake compensation mechanism, the user's head movement and the camera's shaking would be superimposed in the second environmental image captured by the camera, resulting in inaccurate shake components in the head gesture parameters extracted through visual features.

[0107] Specifically, the solution in this application reads data from the laptop's internal inertial measurement unit (IMU) in real time (e.g., a three-axis accelerometer and a three-axis gyroscope). When the camera shakes, the IMU detects the corresponding changes in acceleration and angular velocity. This data is input into a sensor fusion algorithm, such as Kalman filtering or complementary filtering, to estimate the precise shake offset of the camera between consecutive image frames in real time, including its displacement on the X, Y, and Z axes and its rotation angles around each axis.

[0108] Subsequently, before smoothing the user's head pose parameters (e.g., head position, size, and tilt angle), the estimated jitter offset is subtracted inversely from these head pose parameters. For example, if the camera moves slightly to the left by 1 millimeter, the head position parameters extracted from the image will also shift to the left accordingly. Through inverse compensation, this 1-millimeter offset is subtracted, thereby correcting the user's true head position in space. After this compensation, the user's head pose parameters will more accurately reflect their actual head movement, eliminating interference from camera jitter. Finally, these compensated head pose parameters are smoothed, resulting in more stable and reliable user spatial pose data, thus ensuring the accuracy of smart wake-up.

[0109] In some embodiments described above, to eliminate the impact of camera shake on the extraction of user head posture parameters, a method is proposed to use data from an inertial measurement unit (IMU) integrated within a laptop computer to estimate the camera's shake offset between consecutive second environmental image frames and to perform inverse compensation for the user's head posture parameters. However, in practical applications, the data types of IMUs are diverse, and there may be multiple options for estimating and compensating for shake offset. If efficient and accurate processing methods are not adopted, the accuracy and real-time performance of shake compensation may be affected, thereby impacting the reliability of the final spatial posture data. Therefore, this application further proposes specific processing of the IMU data and clarifies the methods for estimating and compensating for shake offset to improve the accuracy and effectiveness of shake compensation.

[0110] In some embodiments of this application, the data from the inertial measurement unit includes 3-axis acceleration and 3-axis angular velocity. In step S2332, the data from the inertial measurement unit is processed using a sensor fusion algorithm to estimate the angular change and / or linear displacement of the camera around its coordinate axes as the jitter offset. In step S2333, the jitter offset is subtracted from the user's head posture parameters to obtain the compensated user's head posture parameters.

[0111] Specifically, the data from the inertial measurement unit (IMU) can be understood as motion data collected by microelectromechanical systems (MEMS) sensors integrated inside the laptop. Three-axis acceleration refers to the linear acceleration components in the X, Y, and Z orthogonal directions, reflecting the camera's translational motion or the forces acting on it. Three-axis angular velocity refers to the angular velocity components of the camera's rotation around the X, Y, and Z orthogonal axes, reflecting the camera's rotational motion. This data comprehensively describes the camera's instantaneous motion state.

[0112] Sensor fusion algorithms refer to techniques that integrate and process data from different sensors (such as accelerometers and gyroscopes) to obtain more accurate and robust motion state estimates than those from a single sensor. For example, algorithms such as Kalman filtering, extended Kalman filtering (EKF), unscented Kalman filtering (UKF), or complementary filtering can be used. The aim is to effectively filter out noise and accurately estimate the angular changes and / or linear displacements of the camera between consecutive frames of a second environmental image by combining the accuracy of the accelerometer at low frequencies and the accuracy of the gyroscope at high frequencies. Angular changes can refer to the camera's rotation angle in the pitch, yaw, and roll directions, while linear displacements can refer to the camera's translational distance in space. These estimated angular changes and / or linear displacements are used as jitter offsets to quantify the camera's own motion.

[0113] In practical applications, subtracting the jitter offset from the user's head pose parameters can be understood as performing a reverse correction on the head pose parameters. For example, if the camera shifts upwards by a certain distance, causing the head position to shift downwards in the image, then during compensation, the corresponding position component in the head pose parameters is adjusted upwards by the same distance. If the camera rotates, causing a change in the head tilt angle, the true head tilt angle is restored by subtracting the corresponding angle offset. The purpose is to directly counteract the interference of camera jitter on the head pose parameters, thereby obtaining parameters that are closer to the user's true head pose.

[0114] This application's solution explicitly defines the data type of the inertial measurement unit as 3-axis acceleration and 3-axis angular velocity, and utilizes a sensor fusion algorithm to process this data. This enables a more comprehensive and accurate capture of the camera's instantaneous motion state, including its rotation and linear displacement. The sensor fusion algorithm effectively combines the advantages of different sensors, filters out noise, and provides high-precision jitter offset estimation, thus providing a solid data foundation for subsequent compensation operations. By directly subtracting the estimated jitter offset from the user's head pose parameters, this solution can directly and effectively counteract the interference of camera shake on head pose parameters. This achieves precise reverse compensation at the pixel level or pose parameter level, ensuring that the extracted head pose parameters truly reflect the user's actual pose, rather than an apparent pose affected by camera shake.

[0115] Through the above technical solution, this application can utilize richer and more accurate inertial measurement unit data (3-axis acceleration and 3-axis angular velocity), combined with advanced sensor fusion algorithms, to achieve high-precision estimation of camera jitter offset. Compared to solutions relying on a single sensor or simple processing methods, this solution can more accurately identify and quantify the complex motion of the camera, including rotation and linear displacement. Therefore, by directly subtracting the jitter offset for reverse compensation, the accuracy and stability of user head posture parameter extraction can be significantly improved, effectively eliminating errors caused by camera jitter. This provides a more reliable and accurate spatial posture sequence for subsequent pattern recognition and intelligent wake-up, enhancing the robustness and user experience of the entire intelligent wake-up method.

[0116] In some preferred embodiments, the inertial measurement unit integrated within the laptop may include a three-axis accelerometer and a three-axis gyroscope. When the camera experiences slight shaking in low-light conditions or when the user holds the laptop, the inertial measurement unit outputs its three-axis acceleration and three-axis angular velocity data in real time. For example, an extended Kalman filter (EKF) algorithm can be used as the sensor fusion algorithm. The EKF algorithm takes the accelerometer and gyroscope data as input and estimates the camera's attitude change (i.e., the amount of angular change around its coordinate axes) and / or position change (i.e., the amount of linear displacement) between each image frame through prediction and update steps. Suppose that in a certain frame, the EKF estimates that the camera has rotated 2 degrees clockwise around the Z-axis (the axis perpendicular to the screen) and translated 1 millimeter upwards. If the user's head pose parameters extracted from the second environmental image show a head tilt angle of 10 degrees and a 5-pixel upward offset in the head center position, then during compensation, the head tilt angle will be reduced by 2 degrees to obtain a true tilt angle of 8 degrees. Simultaneously, the upward offset of the head center position will be subtracted from the corresponding 1-millimeter translation pixel value, resulting in a more accurate head position. In this way, even with camera shake, it can be ensured that the extracted user head pose parameters are accurately calibrated, thus avoiding misjudgments of the user's wake-up posture due to camera shakiness.

[0117] Specifically, in the above-mentioned intelligent wake-up method for laptops based on pattern recognition, when the similarity between the spatial pose sequence and the pre-stored wake-up pose sequence reaches a predetermined threshold, step S4 of performing the wake-up operation can be further refined.

[0118] Step S4 specifically includes: extracting head pose parameters that are the same as those in the pre-stored wake-up pose sequence from the spatial pose sequence, and calculating the Euclidean distance or cosine similarity between the same head pose parameters as the similarity for comparison.

[0119] The spatial pose sequence is extracted based on continuous second environmental images and includes at least one head pose parameter among the user's head position, size, and tilt angle, reflecting the user's head movement trajectory over a period of time. The pre-stored wake-up pose sequence is a specific head dynamic sequence pre-set and stored by the user to trigger the laptop's wake-up operation. The same head pose parameter refers to head pose data points or feature vectors with corresponding relationships or identical semantics used for comparison between the spatial pose sequence and the pre-stored wake-up pose sequence. For example, if the wake-up pose sequence records the X, Y, Z coordinates and tilt angle of the head at a specific time point, then the corresponding X, Y, Z coordinates and tilt angle at the current time point will also be extracted from the currently acquired spatial pose sequence as the same head pose parameters for comparison.

[0120] Specifically, Euclidean distance is a method for measuring the straight-line distance between two points in multidimensional space; the smaller the value, the higher the similarity. When calculating the Euclidean distance between identical head pose parameters, these parameters can be treated as coordinate points in multidimensional space, and the difference in pose is quantified by calculating the distance between these coordinate points. Cosine similarity measures the similarity between two vectors by calculating the cosine of the angle between them; the closer the value is to 1, the higher the similarity, and the closer it is to -1, the greater the difference. When calculating the cosine similarity between identical head pose parameters, these parameters can be represented as vectors, and the degree of pose similarity is assessed by calculating the cosine of the angle between these vectors. The comparison similarity is the quantified value calculated using the above methods, used for comparison with a predetermined threshold.

[0121] The proposed solution extracts head pose parameters corresponding to a pre-stored wake-up pose sequence from the currently acquired spatial pose sequence and uses mature mathematical methods such as Euclidean distance or cosine similarity for quantitative comparison. This allows for an accurate assessment of the similarity between the user's current pose sequence and the preset wake-up pose sequence. This explicit similarity calculation mechanism enables the system to objectively and stably determine whether the user has performed the expected wake-up action, thereby avoiding false wake-ups or wake-up failures caused by fuzzy judgments.

[0122] Through the above technical solutions, the system can identify the user's wake-up intent in a quantitative and reliable manner. Employing Euclidean distance or cosine similarity as the similarity calculation method ensures the accuracy and consistency of the comparison results, thereby significantly improving the recognition accuracy and user experience of laptop smart wake-up. This helps reduce unnecessary wake-up operations while ensuring timely response when the user performs the correct posture, making the smart wake-up function more stable and intelligent.

[0123] In some embodiments described above, a pattern recognition-based intelligent wake-up method for laptops is proposed. This method performs a wake-up operation by comparing the similarity between a user's spatial posture sequence and a pre-stored wake-up posture sequence. However, in its implementation, the source and customization method of the pre-stored wake-up posture sequence are not clearly defined, which may lead to a lack of system personalization or require users to manually input complex posture data, thereby affecting user experience and system flexibility.

[0124] In response, this application further proposes a method for constructing a wake-up posture sequence. The method also includes the user constructing a wake-up posture sequence for the first time: capturing and recording the user's execution of a preset head dynamic sequence, and extracting the user's head posture parameters from it and storing them as a wake-up posture sequence.

[0125] Specifically, "the user's first construction of a wake-up posture sequence" refers to providing an interactive interface or mode to guide the user through the process of recording the wake-up posture when the system is used for the first time or when the user wishes to customize the wake-up method. This process aims to provide a personalized comparison benchmark for subsequent intelligent wake-up operations.

[0126] The phrase "capturing and recording preset head motion sequences executed by the user" can be understood as follows: after the user enters build mode, the system continuously acquires images of the user's head through the camera and records a series of head movements performed by the user according to preset instructions or freehand actions, such as nodding, shaking the head, and turning left and right. These motion sequences are specific action patterns that the user wishes to use as wake-up commands.

[0127] In practical applications, "extracting the user's head pose parameters" refers to performing image processing and pattern recognition on the recorded head motion sequence to obtain key parameters that constitute the user's head pose sequence, such as head position, size, and tilt angle. These parameters are the basic data for quantifying the user's head movements.

[0128] "Storing as the wake-up posture sequence" means saving the extracted head posture parameter sequence that represents the user's specific wake-up action, so that it can be used as a pre-stored wake-up posture sequence for comparison in the subsequent smart wake-up process.

[0129] When a user first sets up the smart wake-up function, the system enters a registration mode. The user is prompted to perform a preset, unique head dynamic posture sequence, such as: "Please slowly turn your head to the left, then slowly turn your head to the right, and finally tilt your head forward." The system captures and records the user's posture data while performing this sequence and extracts its dynamic features as the "wake-up posture sequence template" for that legitimate user. This template can be a feature vector, for example: Template_Feature=[Average left turn speed, maximum right turn angle, forward tilt amplitude, total duration of attitude change].

[0130] This application's solution effectively addresses the issues of unclear wake-up posture sequence sources and lack of personalization in basic solutions by allowing users to customize and store their own unique wake-up posture sequences. Because users can set wake-up actions according to their own habits and preferences, the system can learn and recognize the user's unique wake-up intent. By capturing preset head dynamic sequences executed by the user, the system can obtain real user behavior data; subsequently, it extracts head posture parameters from this data, transforming complex visual information into quantifiable data sequences; finally, these data sequences are stored as personalized comparison templates for subsequent intelligent wake-up. This process ensures the accuracy and user relevance of the wake-up posture sequences, thereby improving the reliability of wake-up and user satisfaction.

[0131] Through the above technical solution, this application can provide users with a highly personalized intelligent wake-up experience. Users can freely set their own wake-up posture according to their habits and preferences, thereby significantly improving the system's usability and user satisfaction. In addition, since the wake-up posture sequence is constructed by the user, this helps reduce false wake-ups or recognition failures that may be caused by generic posture recognition, further enhancing the system's accuracy and reliability.

[0132] In some preferred embodiments, a specific example is given below. Assume a user is using the laptop's smart wake-up function for the first time. The system prompts the user to enter "Wake-up Posture Entry Mode." In this mode, the system guides the user to perform a preset head movement sequence, such as "slowly turn your head to the left, then slowly turn your head to the right, and finally nod." As the user completes these actions, the laptop's camera continuously captures images of the user's head. The system processes these images in real time, extracting head posture parameters such as the user's head position, size, and tilt angle, and combining them chronologically into a complete spatial posture sequence. Once the user completes the entry, this spatial posture sequence is stored in the laptop's memory as the user's personalized wake-up posture sequence. Subsequently, when the user wishes to wake the laptop, they simply repeat this custom head movement sequence, and the system can recognize and execute the wake-up operation.

[0133] Based on the above-described pattern recognition-based intelligent wake-up method for laptops, this application also discloses a pattern recognition-based intelligent wake-up system for laptops. Please refer to [link to relevant documentation]. Figure 5 As shown, the pattern recognition-based laptop smart wake-up system 100 includes: Acquisition module 11 is used to acquire continuous first environmental images; The judgment module 12 is used to determine whether the brightness difference of the first environmental image exceeds a preset threshold. Processing module 13 is used to acquire a continuous second environment image when the brightness difference exceeds a preset threshold. The frame rate of the second environment image is higher than that of the first environment image. Based on the continuous second environment image, the user's spatial pose sequence is extracted. The user's spatial pose sequence includes at least one of head pose parameters, such as head position, size, and tilt angle. The execution module 14 is used to perform a wake-up operation when the judgment module determines that the similarity between the spatial posture sequence and the pre-stored wake-up posture sequence reaches a predetermined threshold.

[0134] The pattern recognition-based intelligent wake-up system 100 for laptops in this application integrates an acquisition module 11, a judgment module 12, a processing module 13, and an execution module 14 to collaboratively address the problems of low recognition rate and high wake-up failure rate in traditional laptop intelligent wake-up functions under complex environments such as camera lens contamination, strong backlighting, and users wearing masks. The system first utilizes the acquisition module 11 to continuously acquire environmental images in low-power mode, and the judgment module 12 intelligently identifies environmental changes. When potential user activity is detected, the processing module 13 is activated to acquire detailed images at higher resolution and frame rate, and extracts the user's spatial pose sequence from them. Finally, the execution module 14 determines whether to perform a wake-up operation based on the similarity between this pose sequence and a pre-stored wake-up pose sequence. This modular design and hierarchical processing mechanism ensures that the system effectively reduces power consumption while maintaining recognition accuracy, and significantly improves the user's wake-up experience under various adverse conditions.

[0135] To facilitate a clearer understanding of the technical solution presented in this application, some key terms and implementation environments are first explained. In this application, "low-power operating mode" refers to the camera acquiring images at a lower frame rate and resolution to save power, suitable for continuous environmental monitoring. "Active operating mode" refers to the camera acquiring images at a higher frame rate and resolution to obtain more detailed visual information, suitable for scenarios requiring fine pattern recognition. "Brightness difference" refers to the degree of pixel brightness variation between consecutive image frames, reflecting changes in ambient light or significant motion in the image. "Spatial pose sequence" refers to the dynamic trajectory of a user's head over a period of time, including a continuous set of parameters such as head position, size, and tilt angle. "Wake-up pose sequence" is a pre-stored specific head dynamic pattern used to trigger a wake-up operation. This system is typically deployed inside a laptop computer, utilizing its integrated camera for image acquisition and executing image processing and pattern recognition algorithms through its built-in processor.

[0136] Specifically, the intelligent wake-up system of this application includes the following main modules: The acquisition module 11 is used to acquire continuous first environmental images, which are images acquired by the camera in low-power operating mode. The acquisition module 11 can be a hardware unit integrating an image sensor and preliminary data buffering function, or a software component controlling the camera hardware to acquire images. In one implementation, the acquisition module 11 can control the camera to continuously acquire images at a frame rate of 5 frames per second and a resolution of 320x240 pixels, and transmit these images to the judgment module for preliminary analysis.

[0137] The judgment module 12 is used to determine whether the brightness difference of consecutive first environmental images exceeds a preset threshold. This judgment module 12 can be a dedicated digital signal processor (DSP) or a software algorithm on a general-purpose processor. For example, the judgment module 12 can calculate the average brightness difference between two consecutive frames of the first environmental images, or calculate the brightness change in a specific area of ​​the image. If the calculated brightness difference value is greater than the preset threshold (e.g., 10 gray levels), it is considered that a significant change has occurred in the environment, possibly indicating that a user is approaching or moving, and a signal is sent to the processing module.

[0138] Processing module 13 is configured to acquire consecutive second environmental images when the brightness difference exceeds a preset threshold. These second environmental images are acquired in the camera's active working mode, where the resolution and frame rate are higher than in the low-power working mode. Based on these consecutive second environmental images, a user's spatial pose sequence is extracted. This spatial pose sequence includes at least one head pose parameter: head position, head size, and head tilt angle. Processing module 13 can be a high-performance central processing unit (CPU) or graphics processing unit (GPU) running complex image processing and pattern recognition algorithms. In this case, the processing module instructs the acquisition module to switch the camera to active working mode to acquire images at a higher resolution and frame rate (e.g., 30 frames per second, 1280x720 pixels). Based on these consecutive second environmental images, processing module 13 extracts the user's spatial pose sequence. This spatial pose sequence may include at least one head pose parameter: head position, head size, and head tilt angle. For example, processing module 13 can determine the head position by analyzing the pixel distribution of the head region in the image, estimate the head size by the area of ​​the head region, and determine the head tilt angle by the angle between the head's main axis and the vertical direction.

[0139] The execution module 14 performs a wake-up operation when the similarity between the spatial pose sequence and a pre-stored wake-up pose sequence reaches a predetermined threshold. The execution module 14 can be a system control unit responsible for receiving the recognition results from the processing module and triggering corresponding system actions. For example, the execution module 14 can compare the currently extracted spatial pose sequence with multiple pre-stored wake-up pose sequences. The comparison can be performed by calculating the Euclidean distance or cosine similarity between the sequences. If the similarity reaches a preset threshold (e.g., 0.8), the execution module considers that the user is performing the expected wake-up action, thereby triggering the laptop to wake up, such as turning on the screen or unlocking the system.

[0140] The intelligent wake-up system of this application effectively solves the problem of low recognition rate in complex environments by coordinating the aforementioned modules. First, through the collaborative work of the acquisition and judgment modules, the system continuously monitors the environment in a low-power mode, switching to a high-power active working mode only when a significant brightness change is detected. This hierarchical wake-up mechanism avoids the camera remaining in a high-power state for extended periods, thus extending battery life. When the brightness difference exceeds a preset threshold, indicating potential user activity, the processing module instructs the acquisition module to switch to the active working mode to acquire a second environmental image at higher resolution and frame rate. This high-resolution and high-frame-rate image acquisition provides richer detail information, even under conditions of slight lens contamination or strong backlighting, laying the foundation for subsequent user spatial pose sequence extraction. By extracting the user's spatial pose sequence through the processing module, this application avoids the strong dependence of traditional face recognition methods on clear facial features. Even when the user is wearing a mask or their facial features are unclear, the system can still identify the user's intent through dynamic changes in head posture. Finally, the execution module compares the extracted spatial pose sequence with a pre-stored wake-up pose sequence. This recognition method based on dynamic posture patterns has higher robustness and anti-interference ability compared to static facial feature recognition. It can effectively cope with adverse conditions such as lens contamination, strong backlight, and users wearing masks, thereby significantly improving the success rate of smart wake-up and user experience.

[0141] The core innovation of this application lies in its hierarchical wake-up mechanism and pattern recognition method based on user spatial pose sequences, implemented in a modular system form. Compared with the closest existing technology, which typically relies directly on high-resolution images for face recognition, the recognition effect is significantly reduced when the ambient light is poor or the user's face is obscured. For example, when the camera lens is dusty or the user is in a strong backlight environment, the image quality acquired by traditional methods will be severely degraded, making it difficult to extract facial features and thus causing recognition failure. This application effectively filters out irrelevant wake-up attempts by introducing brightness difference judgment in a low-power working mode, and only initiates high-resolution image acquisition when necessary, thereby significantly reducing system power consumption while ensuring recognition effect. More importantly, this application extracts the user's spatial pose sequence through the processing module and compares it with the pre-stored wake-up pose sequence, avoiding over-reliance on clear facial features. This method can still identify the user's wake-up intention through dynamic changes in head posture even when the user is wearing a mask or their facial features are obscured, greatly improving the robustness and applicability of intelligent wake-up. For example, the user can trigger wake-up through specific head movements or nodding without removing the mask or adjusting the lighting. This innovative system enables the laptop's smart wake-up function to maintain high accuracy and reliability in various complex and adverse usage scenarios, providing users with a more convenient and intelligent interactive experience.

[0142] The embodiments described above are merely examples of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

Claims

1. A pattern recognition based intelligent wake-up method for a notebook computer, characterized in that, include: S1: Acquire a continuous first environmental image, wherein the continuous first environmental image is an image acquired by the camera in low-power working mode; S2: Determine whether the brightness difference of the consecutive first environmental images exceeds a preset threshold; S3: When the brightness difference exceeds a preset threshold, a continuous second environmental image is acquired. The second environmental image is an image acquired by the camera in the active working mode. The resolution and frame rate of the image acquired by the camera in the active working mode are higher than those in the low power working mode. The user's spatial posture sequence is extracted based on the continuous second environmental image. The user's spatial posture sequence includes at least one head posture parameter among head position, size, and tilt angle. S4: When the similarity between the spatial posture sequence and the pre-stored wake-up posture sequence reaches a predetermined threshold, a wake-up operation is performed.

2. The pattern recognition based smart wake up method for a notebook computer according to claim 1, wherein, The step of extracting the user's spatial pose sequence based on the continuous second environment image includes: S21: Perform grayscale processing on the consecutive second environment images respectively and identify and generate edge maps; S22: Obtain the user's head outline from the edge map; S23: Spatial pose data is calculated based on the user's head contour, and multiple consecutive spatial pose data constitute the user's spatial pose sequence.

3. The pattern recognition based smart wake up method for a notebook computer according to claim 2, wherein, In step S22, a closed contour is found from the edge map, and the contour with the largest area and the shape that meets the predetermined shape is selected as the user's head contour. The predetermined shape is ellipse or circle.

4. The pattern recognition based smart wake up method for a notebook computer according to claim 2, wherein, Step S23 specifically includes: S231: Extract multi-source visual motion features from the continuous second environment images based on the head contour, wherein the multi-source visual motion features include at least two of the following: motion centroid features, direction vector features, and patch center features; S232: The extracted at least two types of multi-source visual motion features are weighted and fused to obtain the user's head pose parameters; S233: Smooth the user's head posture parameters to obtain the user's spatial posture data.

5. The pattern recognition based smart wake up method for a notebook computer according to claim 4, wherein, The steps between S232 and S233 also include: S2331: Read data from the inertial measurement unit integrated inside the laptop computer; S2332: The jitter offset of the camera between consecutive second environmental image frames is estimated in real time using the data from the inertial measurement unit; S2333: The user's head posture parameters are compensated in reverse using the jitter offset to eliminate the influence of the camera's own jitter. In step S233, the compensated user's head pose parameters are smoothed to obtain the user's spatial pose data.

6. The pattern recognition based smart wake up method for a notebook computer according to claim 5, wherein, The data from the inertial measurement unit includes 3-axis acceleration and 3-axis angular velocity. In step S2332, the data from the inertial measurement unit is processed using a sensor fusion algorithm to estimate the angular change and / or linear displacement of the camera around its coordinate axes as the jitter offset. In step S2333, the jitter offset is subtracted from the user's head posture parameters to obtain the compensated user's head posture parameters.

7. The pattern recognition based smart wake up method for a notebook computer according to claim 4, wherein, Step S231 specifically includes: For each pair of consecutive second environmental images, the pixel brightness difference is calculated to obtain the brightness change value. Pixels whose brightness change value exceeds a predetermined threshold are marked as moving pixels. The continuous region composed of all the moving pixels is identified, and the region with the largest area and the longest duration is selected. Its geometric center is calculated as the moving centroid feature. and / or The continuous second environmental image is processed to obtain the pixel brightness gradient, and the histogram of the brightness gradient direction is statistically analyzed to determine the most dominant gradient direction, which is used as the direction vector feature. and / or Using a preset elliptical or circular template, template matching is performed on the continuous second environmental images to determine the center of the region with the highest matching score, which is used as the center feature of the patch.

8. The pattern recognition based smart wake up method for a notebook computer according to claim 1, wherein, Step S4 specifically includes: extracting head pose parameters that are the same as those in the pre-stored wake-up pose sequence from the spatial pose sequence, and calculating the Euclidean distance or cosine similarity between the same head pose parameters as the similarity for comparison.

9. The pattern recognition based smart wake up method for a notebook computer according to claim 1, wherein, The method further includes the user initially constructing a wake-up posture sequence for wake-up: capturing and recording the user's execution of a preset head dynamic sequence, and extracting the user's head posture parameters from it and storing them as the wake-up posture sequence.

10. A pattern recognition based intelligent wake-up system for a notebook computer, characterized by, include: The acquisition module is used to acquire continuous first environmental images, which are images acquired by the camera in low-power operation mode; The judgment module is used to determine whether the brightness difference of the consecutive first environmental images exceeds a preset threshold. The processing module is used to acquire a continuous second environmental image when the brightness difference exceeds a preset threshold. The second environmental image is an image acquired by the camera in the active working mode. The resolution and frame rate of the image acquired by the camera in the active working mode are higher than those in the low power working mode. Based on the continuous second environmental image, the user's spatial posture sequence is extracted. The user's spatial posture sequence includes at least one head posture parameter of head position, size and tilt angle. The execution module performs a wake-up operation when the similarity between the spatial pose sequence and the pre-stored wake-up pose sequence reaches a predetermined threshold.

Images