Dynamic eyeball tracking method for naked eye 3D and tablet computer

Abstract

The invention belongs to the field of 3D display, and provides a dynamic eyeball tracking method for naked eye 3D and a tablet personal computer, and the method comprises the steps: obtaining a video frame shot by the tablet personal computer, and carrying out the detection of eye and nose regions of the video frame based on an error learning framework; performing content classification on the eye and nose areas, extracting SIFT features in the eye and nose areas based on a scale invariant feature transform algorithm, and selecting a corresponding supervised descent model according to a content classification result to obtain first coordinates of pupil centers of human eyes; estimating a three-dimensional coordinate of the pupil center according to the 3D face model and the first coordinate, and dynamically adjusting a pixel arrangement mode of a screen of the tablet computer and the position of a cylindrical lens grating based on the three-dimensional coordinate, so that a left eye image and a right eye image projected by the tablet computer are matched with the three-dimensional coordinate of the pupil center; according to the invention, the problem that the naked-eye 3D effect cannot be achieved when the existing naked-eye 3D screen generates ghosting when the position of an observer is not correct enough or moves or the deviation exceeds a certain degree can be solved.

Description

Technical Field

[0001] The present invention relates to the field of 3D display, and in particular to a dynamic eye tracking method and a tablet computer for naked-eye 3D. Background Art

[0002] There are two main technical routes for naked-eye 3D screen technology at present, namely parallax barrier 3D technology and cylindrical lens 3D technology. Parallax barrier 3D technology uses a shielding plate to display the images that the left and right eyes should see on the adjacent pixels on the left and right. Through the propagation of light along a straight line, one side is blocked to enter the left eye, and the other side is blocked to enter the right eye, so that the two eyes can see two different images on one screen, achieving the naked-eye 3D effect. Cylindrical lens 3D technology uses a cylindrical lens to display the images that the left and right eyes should see on the adjacent pixel grids on the left and right. Through the principle of light refraction, one side is refracted to enter the left eye, and the other side is refracted to enter the right eye, so that the two eyes can see two different images on one screen, achieving the naked-eye 3D effect.

[0003] Both of the above solutions have problems: the distance between the eyes and the screen and the angle and position of the eyes must be exactly at the position of the left and right eye light after blocking / refraction. If there is any deviation, ghosting will occur. When the deviation exceeds a certain degree, the normal image cannot be seen. Summary of the invention

[0004] In view of the above technical problems, the present invention provides a dynamic eye tracking method and a tablet computer for naked-eye 3D to solve the problem that the existing naked-eye 3D screen produces ghosting when the observer's position is not correct or the observer moves, and the naked-eye 3D effect cannot be achieved when the observer is offset to more than a certain extent.

[0005] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0006] According to one aspect of the present invention, a dynamic eye tracking method for naked-eye 3D is disclosed, the method comprising: Obtain a video frame shot by a tablet computer, and detect eye and nose regions of the video frame based on an error learning framework, wherein the detector of the error learning framework includes using an Adaboost algorithm with a local binary pattern; Performing content classification on the eye and nose region, wherein the conditions for content classification include whether glasses are worn, the type of lighting environment, glasses reflection, and thick glasses; extracting SIFT features from the eye and nose region based on a scale-invariant feature transformation algorithm; selecting a corresponding supervised descent model according to the result of the content classification; training multiple descent directions for the SIFT features to minimize the average value of the nonlinear square function of each landmark point, so that the SIFT features perform regression-based landmark point alignment, so that the eye and nose shapes of the eye and nose region are aligned, and obtaining the first coordinate of the pupil center of the human eye; According to a universal 3D face model and the first coordinates, the three-dimensional coordinates of the pupil center are estimated, and based on the three-dimensional coordinates, the pixel arrangement of the screen of the tablet computer and the position of the cylindrical lens grating are dynamically adjusted so that the left eye image and the right eye image projected by the tablet computer are coordinated with the three-dimensional coordinates of the pupil center.

[0007] Furthermore, the error learning framework includes an early stage, a mid-stage, and a mature stage during training, wherein: In an early stage, local texture information of nose and eyes of a human face is detected from a first sample image set based on the local binary pattern of the Adaboost algorithm, a first cascade classifier is trained based on the local texture information, the first cascade classifier is applied to a second sample image set, and sample images that are incorrectly classified by the first cascade classifier are collected, wherein the first sample image set includes positive sample images and negative sample images, wherein the positive sample images are images with human faces, and the negative samples are images without human faces; In the middle stage, based on the sample images that were misclassified in the early stage, setting and adjusting to obtain a second cascade classifier, using the second cascade classifier to classify the sample images that were misclassified in the early stage, and collecting the sample images that were misclassified again; In the mature stage, a threshold of classification accuracy is set, and the second cascade classifier is continuously adjusted until the classification accuracy of the second cascade classifier for the sample image that is misclassified again reaches the threshold.

[0008] Furthermore, the method further comprises: After obtaining the first coordinate of the pupil center, checking the aligned eye and nose region based on a tracking checker to check whether the alignment result correctly includes the pupil, the tracking checker including SIFT features, a pre-trained support vector machine classifier, and an Adaboost algorithm; When the tracking checker checks that the alignment result is incorrect, calculating the first coordinate of the pupil center of a subsequent frame of the video frame; When the tracking checker checks that the alignment result is correct, the alignment result is maintained, and the first coordinate is used as the starting coordinate, and based on the tracking checker, the pupil center is tracked and checked in subsequent frames of the video frame.

[0009] Furthermore, the tracking checker includes: Extracting the SIFT features from the aligned eye and nose regions based on the support vector machine classifier, and classifying the extracted SIFT features to determine whether they include the pupil; Capturing local texture features in the eye and nose region based on the local binary pattern in the Adaboost algorithm to determine whether the eye and nose region contains a pupil; The judgment results of the support vector machine classifier and the Adaboost algorithm are weighted and fused to obtain the final inspection result.

[0010] Furthermore, when training the tracking checker, the judgment result of the Adaboost algorithm is identified, and the samples with errors identified by the Adaboost algorithm are input into the support vector machine classifier to train the support vector machine classifier.

[0011] Furthermore, when estimating the three-dimensional coordinates of the pupil center, the three-dimensional coordinates of the pupil center are calculated along the facial normal direction and combined with the fixed pupil distance to obtain the three-dimensional coordinates of the pupil center. The facial normal direction is estimated by aligning the 3D face model with preset landmark points in the eye and nose area.

[0012] Furthermore, the method further comprises: The tablet computer shoots the video frame with a near infrared camera, and during content classification, if the content classification result includes eyeglass reflection, the reflection area in the eye and nose area is repaired.

[0013] Furthermore, the repairing of the reflective area in the eye and nose area includes: Based on pixel gradient information, detecting the reflection area in the eye and nose area; Performing edge minimization on the reflection area, wherein the edge minimization includes a smoothing method and an interpolation method; Based on the pixel gradient information, the gradient difference of the reflection area is minimized.

[0014] According to another aspect of the present invention, a dynamic eye tracking tablet computer for naked eye 3D is disclosed, the tablet computer comprising: An eye and nose detection module, used to obtain a video frame captured by a front camera, and detect the eye and nose regions of the video frame based on an error learning framework, wherein the detector of the error learning framework includes using an Adaboost algorithm with a local binary pattern; An eye-nose alignment module is used to perform content classification on the eye-nose region, where the conditions for content classification include whether glasses are worn, the type of lighting environment, glasses reflection, and thick glasses. SIFT features are extracted from the eye-nose region based on a scale-invariant feature transformation algorithm. According to the result of content classification, a corresponding supervised descent model is selected to train multiple descent directions for the SIFT features to minimize the average value of the nonlinear square function of each landmark point, so that the SIFT features perform regression-based landmark point alignment, so that the eye-nose shapes of the eye-nose region are aligned, and the first coordinate of the pupil center of the human eye is obtained; A control display module is used to estimate the three-dimensional coordinates of the pupil center according to a universal 3D face model and the first coordinates, and dynamically adjust the pixel arrangement of the screen and the position of the cylindrical lens grating based on the three-dimensional coordinates, so that the left eye image and the right eye image projected by the screen are coordinated with the three-dimensional coordinates of the pupil center.

[0015] The technical solution disclosed in this disclosure has the following beneficial effects: The present disclosure is based on content classification and can meet the tracking of pupil centers under various environmental conditions. The classification, detection, and tracking methods adopted only require CPU resources and do not rely on GPU computing, so that they can be executed under low system resource conditions; The present disclosure can dynamically adjust the pixel arrangement of the video frame and the projected left-eye image and right-eye image by real-time identification of the pupil position, so that the observer can obtain a good naked-eye 3D experience when the position changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of a dynamic eye tracking method for naked-eye 3D in an embodiment of this specification; Figure 2 This is a structural block diagram of a dynamic eye tracking tablet computer for naked-eye 3D in an embodiment of this specification. DETAILED DESCRIPTION

[0017] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0018] In addition, the accompanying drawings are only schematic illustrations of the present disclosure. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0019] In one embodiment, if Figure 1 As shown, this specification provides a dynamic eye tracking method for naked-eye 3D, and the execution subject of the method can be a computer, a server, a mobile phone, a tablet computer, etc. The method can specifically include the following steps S101~S103: In step S101, a video frame shot by a tablet computer is obtained, and eye and nose regions of the video frame are detected based on an error learning framework, wherein the detector of the error learning framework includes using an Adaboost algorithm with a local binary pattern.

[0020] The captured video frame can be captured by a visible light camera or an infrared camera. The video frame is one or more frames in the captured video stream, such as a certain starting frame. The eye and nose regions are then detected on the video frame. If the eye and nose regions cannot be detected through the error learning framework, the process jumps to the next frame or other preset frames until the eye and nose regions are detected from the frame.

[0021] In step S102, content classification is performed on the eye and nose region, and conditions for content classification include whether glasses are worn, lighting environment type, glasses reflection, and thick glasses. SIFT features are extracted from the eye and nose region based on a scale-invariant feature transformation algorithm. According to the result of content classification, a corresponding supervised descent model is selected, and multiple descent directions are trained for the SIFT features to minimize the average value of the nonlinear square function of each landmark point, so that the SIFT features perform regression-based landmark point alignment, so that the eye and nose shapes of the eye and nose region are aligned, and the first coordinate of the pupil center of the human eye is obtained.

[0022] Among them, since the supervised descent method (SDM) is a regression-based method that solves the shape alignment problem as a general optimization problem, a single supervised descent model cannot be easily optimized globally on images under various conditions. In order to solve this problem, the eye and nose regions are content classified, and the corresponding supervised descent model is applied to align the eye and nose shapes according to the image quality and conditions (whether glasses are worn, the type of lighting environment is dark or bright, infrared photos, visible light photos, glasses reflection, thick glasses). Thick glasses refer to glasses with lenses that are too large and too thick, which are identified by preset classification conditions. Content classification can be specifically based on machine algorithms such as support vector machines SVM, decision trees, random forests, etc. When establishing a classifier, select appropriate features to represent the image content. These features can effectively distinguish different image content categories, such as brightness, contrast, edges, etc. Select a suitable machine learning algorithm to train the classifier, and then use the labeled image dataset to train the classifier to ensure that there are enough samples for each type of image. The classifier is trained through training data so that it can learn the feature representation of different image contents. Then, the trained classifier is applied to the new video frames with the eye and nose areas divided to classify these images. According to the output of the classifier, the content category of each image is obtained, such as "bright image", "dark image", "glasses reflection image", etc. These categories can be cross-existing with each other. Finally, according to the content category of the image, the corresponding SDM aligner is selected. For example, for "dark image", a supervised descent model optimized for "dark image" is selected to process images under low light conditions.

[0023] In addition, in the alignment of the eye and nose regions, the scale-invariant feature transform (SIFT) algorithm is an algorithm for image feature detection and description, which can effectively identify key points in the image and extract descriptors of these key points. The SIFT feature has good invariance to illumination changes, rotation, scaling, etc. When extracting SIFT features, the key points in the image are detected through a Gaussian pyramid. These key points are significant feature points in the image, such as edges and corners. Then, for each detected key point, the descriptor of its local neighborhood is calculated. These descriptors contain the gradient information of the area around the key point to form a high-dimensional feature vector, namely the SIFT feature. The supervised descent method is a regression-based method for shape alignment problems. It learns the changes from the initial estimate to the target shape to minimize the objective function, thereby performing efficient shape alignment. In this embodiment, when the supervised descent model is trained, an initial estimate is first made, that is, the initial alignment result of the eye and nose region obtained from the error learning framework is used as the input of the model, and then a regression model is constructed to learn the movement of the key points. Through multiple iterations, the model trains a series of descent directions to adjust the key point positions, and sets the objective function as the nonlinear square error of each key point. The goal of the model is to minimize the error between the predicted coordinates and the true coordinates by optimizing this objective function. Specifically, in the alignment, the landmark points are first aligned. The landmark points refer to the positions of the key points (such as the center of the pupil, the corner of the eye, the tip of the nose, etc.) in the eye and nose region. The SIFT features extracted from the eye and nose region are used as input to provide information about the positions of the key points. The supervised descent model gradually optimizes the positions of the landmark points according to the key point positions output by the SIFT features through regression learning to improve the alignment accuracy. In each iteration, the supervised descent model calculates the error according to the current landmark point positions, and then adjusts these positions to reduce the error, forming a series of descent directions. The rules are updated at each descent, and the update rules are implemented by the nonlinear least squares method. Each iterative optimization is carried out in the direction of reducing the nonlinear square function of each landmark point. After multiple iterations and optimizations, the supervised descent model adjusts the position of the landmark points to the optimal state, thereby extracting the precise coordinates of the pupil center. The final output results include the coordinates of the key points in the eye and nose area, as well as the calculated precise coordinates of the pupil center.

[0024] In step S103, the three-dimensional coordinates of the pupil center are estimated according to the universal 3D face model and the first coordinates, and the pixel arrangement of the screen of the tablet computer and the position of the cylindrical lens grating are dynamically adjusted based on the three-dimensional coordinates, so that the left eye image and the right eye image projected by the tablet computer are coordinated with the three-dimensional coordinates of the pupil center.

[0025] Among them, the general 3D face model can be Candide-3. When estimating the three-dimensional coordinates of the pupil center, the three-dimensional coordinates of the pupil center are calculated along the facial normal direction and combined with the fixed pupil distance to obtain the three-dimensional coordinates of the pupil center. The facial normal direction is estimated by aligning the 3D face model with preset landmarks in the eye and nose area.

[0026] Additionally, the required pixel arrangement is calculated, including: dynamically adjusting the pixel arrangement on the tablet screen according to the three-dimensional coordinates of the pupil center, specifically, the pixel arrangement of the screen can be repositioned by calculating the deviation between the current pupil position and the ideal observation point, for example, if the pupil center is offset, the virtual pixel position of the screen is adjusted to ensure that the images of the left eye and the right eye overlap correctly; the position of the cylindrical lens grating is adjusted, including: adjusting the position of the cylindrical lens grating according to the three-dimensional coordinates of the pupil center to ensure that the grating can properly guide light at different viewing angles so that the image seen by each eye meets the stereoscopic visual effect, and by calculating the refraction and perspective relationship of the light at different coordinates, it is ensured that the cylindrical lens can effectively separate the images of the left eye and the right eye.

[0027] In one embodiment, in the prior art, the detection of the eye and nose regions mostly relies on neural networks, which consumes a lot of computing resources. Here, this embodiment adopts an error-based learning framework (Error-Based Learning, EBL), which is a heuristic learning method that can imitate the process of human learning from errors to improve the performance of machine learning models. The error-based learning framework includes an early stage, a mid-stage, and a mature stage during training, and the training steps include S201-203: S201: In an early stage, local texture information of nose and eyes of a human face is detected from a first sample image set based on the local binary pattern of the Adaboost algorithm, a first cascade classifier is trained based on the local texture information, the first cascade classifier is applied to a second sample image set, and sample images incorrectly classified by the first cascade classifier are collected, wherein the first sample image set includes positive sample images and negative sample images, the positive sample images are images with human faces, and the negative samples are images without human faces.

[0028] Among them, the Adaboost algorithm framework is used for basic training, which extracts features from images through local binary patterns (LBP) and then trains a first cascade classifier based on these features. The first cascade classifier is trained on a smaller first sample image set. After training, the first cascade classifier is applied to a larger second sample image set to identify those images that are misclassified. These misclassified images will become training samples for the next mid-term stage, while those correctly classified images are deleted.

[0029] S202: In the mid-term stage, based on the sample images that were misclassified in the early stage, setting and adjusting to obtain a second cascade classifier, using the second cascade classifier to classify the sample images that were misclassified in the early stage, and collecting the sample images that were misclassified again.

[0030] The training focus is shifted to samples that were misclassified in the early stages. These samples are usually difficult to classify and are the most challenging to improve the model. Then, based on these erroneous samples, the classifier is redesigned and adjusted to obtain the second cascade classifier, which further improves its ability to classify complex situations. This process is equivalent to a more targeted optimization of the classifier, so that it gradually learns to correctly handle the difficulties that were not correctly identified in the early stages.

[0031] S203: In the mature stage, a threshold of classification accuracy is set, and the second cascade classifier is continuously adjusted until the classification accuracy of the sample image that is misclassified again by the second cascade classifier reaches the threshold. The number of samples in the first sample image set is much less than that in the second sample image set, such as a difference of 10 times.

[0032] In the mature stage, the error learning framework only focuses on samples that are still misclassified in the mid-term stage. By continuing to train these key samples, the final second cascade classifier is further refined to improve its accuracy for all input data. Training at this stage not only reduces training time, but also achieves a higher classification accuracy through repeated screening and adjustment.

[0033] In one embodiment, the method further comprises: After obtaining the first coordinate of the pupil center, checking the aligned eye and nose region based on a tracking checker to check whether the alignment result correctly includes the pupil, the tracking checker including SIFT features, a pre-trained support vector machine classifier, and an Adaboost algorithm; When the tracking checker checks that the alignment result is incorrect, calculating the first coordinate of the pupil center of a subsequent frame of the video frame; When the tracking checker checks that the alignment result is correct, the alignment result is maintained, and the first coordinate is used as the starting coordinate, and based on the tracking checker, the pupil center is tracked and checked in subsequent frames of the video frame.

[0034] Specifically, during inspection, the tracking inspector includes: extracting the SIFT features of the aligned eye and nose area based on the support vector machine classifier, and determining whether the extracted SIFT features include the pupil after classification; capturing the local texture features in the eye and nose area based on the local binary patterns in the Adaboost algorithm to determine whether the eye and nose area contains the pupil; and weightedly fusing the determination results of the support vector machine classifier and the Adaboost algorithm to obtain the final inspection result.

[0035] As a supplement, when training the tracking checker, the judgment result of the Adaboost algorithm is identified, and the samples with errors identified by the Adaboost algorithm are input into the support vector machine classifier to train the support vector machine classifier.

[0036] In one embodiment, the method further includes: the tablet computer shoots the video frame with a near infrared camera, and when classifying the content, if the content classification result includes eyeglass reflection, the reflection area in the eye and nose area is repaired. Specifically, the repairing of the reflection area in the eye and nose area includes: based on pixel gradient information, detecting the reflection area in the eye and nose area; minimizing the edge of the reflection area, and performing the edge minimization includes smoothing method and interpolation method; based on the pixel gradient information, minimizing the gradient difference of the reflection area.

[0037] Based on the same idea, Figure 2 As shown, an exemplary embodiment of the present disclosure further provides a dynamic eye tracking system for naked-eye 3D, wherein the tablet computer comprises: The eye and nose detection module 201 is used to obtain a video frame captured by a front camera, and detect the eye and nose regions of the video frame based on an error learning framework, wherein the detector of the error learning framework includes using an Adaboost algorithm with a local binary pattern; The eye-nose alignment module 202 is used to perform content classification on the eye-nose region, where the content classification conditions include whether glasses are worn, the type of lighting environment, glasses reflection, and thick glasses. SIFT features are extracted from the eye-nose region based on a scale-invariant feature transformation algorithm. According to the result of the content classification, a corresponding supervised descent model is selected to train multiple descent directions for the SIFT features to minimize the average value of the nonlinear square function of each landmark point, so that the SIFT features perform regression-based landmark point alignment, so that the eye-nose shapes of the eye-nose region are aligned, and the first coordinate of the pupil center of the human eye is obtained; The control display module 203 is used to estimate the three-dimensional coordinates of the pupil center according to the general 3D face model and the first coordinates, and dynamically adjust the pixel arrangement of the screen and the position of the cylindrical lens grating based on the three-dimensional coordinates, so that the left eye image and the right eye image projected by the screen are coordinated with the three-dimensional coordinates of the pupil center.

[0038] In the above embodiment, based on content classification, the tracking of the pupil center under various environmental conditions can be met. The classification, detection, and tracking methods adopted only require CPU resources and do not rely on GPU calculations, so that they can be executed under lower system resource conditions. By identifying the pupil position in real time, the pixel arrangement of the video frame can be dynamically adjusted, and the projected left-eye image and right-eye image can be adjusted in real time, so that the observer can obtain a good naked-eye 3D experience when the position changes.

[0039] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the exemplary implementation of the present disclosure.

[0040] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiment of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0041] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. A dynamic eye tracking method for naked eye 3D, It is characterized in that The method comprises: Obtain a video frame shot by a tablet computer, and detect eye and nose regions of the video frame based on an error learning framework, wherein the detector of the error learning framework includes using an Adaboost algorithm with a local binary pattern; Performing content classification on the eye and nose region, wherein the conditions for content classification include whether glasses are worn, the type of lighting environment, glasses reflection, and thick glasses; extracting SIFT features from the eye and nose region based on a scale-invariant feature transformation algorithm; selecting a corresponding supervised descent model according to the result of the content classification; training multiple descent directions for the SIFT features to minimize the average value of the nonlinear square function of each landmark point, so that the SIFT features perform regression-based landmark point alignment, so that the eye and nose shapes of the eye and nose region are aligned, and obtaining the first coordinate of the pupil center of the human eye; According to a universal 3D face model and the first coordinates, the three-dimensional coordinates of the pupil center are estimated, and based on the three-dimensional coordinates, the pixel arrangement of the screen of the tablet computer and the position of the cylindrical lens grating are dynamically adjusted so that the left eye image and the right eye image projected by the tablet computer are coordinated with the three-dimensional coordinates of the pupil center.

2. A dynamic eye tracking method for naked eye 3D according to claim 1, It is characterized in that The error learning framework includes an early stage, a mid-stage, and a mature stage during training, wherein: In an early stage, local texture information of nose and eyes of a human face is detected from a first sample image set based on the local binary pattern of the Adaboost algorithm, a first cascade classifier is trained based on the local texture information, the first cascade classifier is applied to a second sample image set, and sample images that are incorrectly classified by the first cascade classifier are collected, wherein the first sample image set includes positive sample images and negative sample images, wherein the positive sample images are images with human faces, and the negative samples are images without human faces; In the middle stage, based on the sample images that were misclassified in the early stage, setting and adjusting to obtain a second cascade classifier, using the second cascade classifier to classify the sample images that were misclassified in the early stage, and collecting the sample images that were misclassified again; In the mature stage, a threshold of classification accuracy is set, and the second cascade classifier is continuously adjusted until the classification accuracy of the second cascade classifier for the sample image that is misclassified again reaches the threshold.

3. The dynamic eye tracking method for naked eye 3D according to claim 1, It is characterized in that The method further comprises: After obtaining the first coordinate of the pupil center, checking the aligned eye and nose region based on a tracking checker to check whether the alignment result correctly includes the pupil, the tracking checker including SIFT features, a pre-trained support vector machine classifier, and an Adaboost algorithm; When the tracking checker checks that the alignment result is incorrect, calculating the first coordinate of the pupil center of a subsequent frame of the video frame; When the tracking checker checks that the alignment result is correct, the alignment result is maintained, and the first coordinate is used as the starting coordinate, and based on the tracking checker, the pupil center is tracked and checked in subsequent frames of the video frame.

4. The method for dynamic eye tracking for naked eye 3D according to claim 3, It is characterized in that The tracking checker includes: Extracting the SIFT features from the aligned eye and nose regions based on the support vector machine classifier, and classifying the extracted SIFT features to determine whether they include the pupil; Capturing local texture features in the eye and nose region based on the local binary pattern in the Adaboost algorithm to determine whether the eye and nose region contains a pupil; The judgment results of the support vector machine classifier and the Adaboost algorithm are weighted and fused to obtain the final inspection result.

5. A dynamic eye tracking method for naked eye 3D according to claim 4, It is characterized in that When training the tracking checker, the judgment result of the Adaboost algorithm is identified, and the samples with errors identified by the Adaboost algorithm are input into the support vector machine classifier to train the support vector machine classifier.

6. The dynamic eye tracking method for naked eye 3D according to claim 1, It is characterized in that When estimating the three-dimensional coordinates of the pupil center, the three-dimensional coordinates of the pupil center are calculated along the facial normal direction and combined with the fixed pupil distance to obtain the three-dimensional coordinates of the pupil center. The facial normal direction is estimated by the alignment of the 3D face model and the preset landmark points of the eye and nose area.

7. The dynamic eye tracking method for naked eye 3D according to claim 1, It is characterized in that The method further comprises: The tablet computer shoots the video frame with a near infrared camera, and during content classification, if the content classification result includes eyeglass reflection, the reflection area in the eye and nose area is repaired.

8. A dynamic eye tracking method for naked eye 3D according to claim 7, It is characterized in that The repairing of the reflective area in the eye and nose area comprises: Based on pixel gradient information, detecting the reflection area in the eye and nose area; Performing edge minimization on the reflection area, wherein the edge minimization includes a smoothing method and an interpolation method; Based on the pixel gradient information, the gradient difference of the reflection area is minimized.

9. A dynamic eye tracking tablet for naked eye 3D, It is characterized in that The tablet computer comprises: An eye and nose detection module, used to obtain a video frame captured by a front camera, and detect the eye and nose regions of the video frame based on an error learning framework, wherein the detector of the error learning framework includes using an Adaboost algorithm with a local binary pattern; An eye-nose alignment module is used to perform content classification on the eye-nose region, where the conditions for content classification include whether glasses are worn, the type of lighting environment, glasses reflection, and thick glasses. SIFT features are extracted from the eye-nose region based on a scale-invariant feature transformation algorithm. According to the result of content classification, a corresponding supervised descent model is selected to train multiple descent directions for the SIFT features to minimize the average value of the nonlinear square function of each landmark point, so that the SIFT features perform regression-based landmark point alignment, so that the eye-nose shapes of the eye-nose region are aligned, and the first coordinate of the pupil center of the human eye is obtained; A control display module is used to estimate the three-dimensional coordinates of the pupil center according to a universal 3D face model and the first coordinates, and dynamically adjust the pixel arrangement of the screen and the position of the cylindrical lens grating based on the three-dimensional coordinates, so that the left eye image and the right eye image projected by the screen are coordinated with the three-dimensional coordinates of the pupil center.

Images