Video playback method and device

By fusing signals from visual, distance, and positioning sensors, the system automatically senses the user's intention to leave and transfers playback status data, solving the problem of operational lag in cross-device video playback and achieving seamless playback, thus improving the continuity and intelligence of the viewing experience.

CN123123628APending Publication Date: 2026-11-17HUNAN HAPPLY SUNSHINE INTERACTIVE ENTERTAINMENT MEDIA CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611381585.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-07
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing cross-device video playback solutions cannot automatically sense user movement and predict the timing of transitions, resulting in operational delays, a strong sense of disconnect, and difficulties in synchronizing playback control status.

Method used

By fusing signals from visual, distance, and positioning sensors, the system determines the confidence level of the user's intention to leave. When the confidence level reaches a threshold, it automatically collects and streams playback status data to the target device, enabling seamless playback continuation.

Benefits of technology

It eliminates the sense of disconnect caused by operational lag and improves the continuity and intelligence of the cross-device viewing experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123628A_ABST
    Figure CN123123628A_ABST
Patent Text Reader

Abstract

This invention provides a video playback method and apparatus. Based on visual sensor signals, distance sensor signals, and positioning sensor signals, the confidence level of the user's intention to leave is determined. If the confidence level of the user's intention to leave is greater than the transition trigger threshold, the playback status data of the first playback device is automatically collected and sent to the second playback device, so that the second playback device can seamlessly continue playback. In this process, no user intervention is required, realizing the automatic transition of video playback tasks between playback devices, eliminating the sense of disconnect caused by operation lag, and significantly improving the continuity and intelligence level of cross-device viewing experience.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of video playback technology, and in particular to a video playback method and apparatus. Background Technology

[0002] With the widespread adoption of smart TVs and smartphones, cross-device video playback has become a common need in home settings.

[0003] Currently, if users briefly leave the TV viewing area to use the restroom or answer a phone call, they typically need to manually search for content and re-enter the playback progress on their mobile phones, a cumbersome and disjointed experience. Although some platforms support "screen casting" or cross-device playback continuation functions, existing solutions rely on user-initiated actions, such as clicking a screen casting button or scanning a QR code to bind the device. They cannot automatically detect the user's movement the moment they leave the TV and initiate playback preparation in advance, resulting in a significant time lag between the user's intention to leave and the video resuming playback on the target device. Furthermore, existing technologies struggle to proactively detect user movement within the home space and cannot predict the timing of playback transitions, essentially remaining a "passive response" model. Regarding status synchronization, existing solutions require users to manually locate the playback progress after switching devices, and playback control states such as comment context, volume settings, and audio track selection often fail to migrate synchronously, resulting in a strong sense of interruption. Summary of the Invention

[0004] In view of this, the present invention provides a video playback method and apparatus that eliminates the sense of disconnect caused by operation lag and significantly improves the continuity and intelligence of the cross-device viewing experience.

[0005] The first aspect of this invention provides a video playback method, comprising:

[0006] Acquire signals from the vision sensor, distance sensor, and positioning sensor;

[0007] Based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal, determine the confidence level of the user's intention to leave;

[0008] If the confidence level of the user's intention to leave is greater than the flow trigger threshold, the playback status data of the first playback device is collected.

[0009] The playback status data of the first playback device is transferred to the second playback device; wherein the second playback device continues playback according to the playback status data of the first playback device.

[0010] Optionally, determining the confidence level of the user's intention to leave based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal includes:

[0011] Based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal, the observation vector at the current moment is determined;

[0012] Based on the state estimation vector of the previous moment and the preset state transition matrix, determine the prior state estimation vector of the current moment;

[0013] Based on the observation vector at the current moment, the prior state estimation vector at the current moment, and the preset observation matrix, the observation residual is determined;

[0014] The dynamic fusion gain is determined based on the error covariance matrix of the prior state estimation vector at the current moment, the preset observation matrix, and the preset observation noise covariance matrix.

[0015] The prior state estimation vector at the current moment is corrected using the observation residual and the dynamic fusion gain to obtain the posterior state estimation vector at the current moment;

[0016] Based on the posterior state estimation vector at the current moment, the confidence level of the user's intention to leave is determined.

[0017] Optionally, determining the dynamic fusion gain based on the error covariance matrix of the prior state estimation vector at the current time, the preset observation matrix, and the preset observation noise covariance matrix includes:

[0018] The residual covariance is determined based on the error covariance matrix of the prior state estimation vector at the current moment, the preset observation matrix, and the preset observation noise covariance matrix.

[0019] The dynamic fusion gain is determined based on the error covariance matrix, the preset observation matrix, and the residual covariance.

[0020] Optionally, determining the observation vector at the current moment based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal includes:

[0021] Based on the visual sensor signals, the user's posture feature vector is determined, and based on the user's posture feature vector, reference posture center vector, and posture weight matrix, a standing discrimination score is determined.

[0022] The rate of change of distance between the user and the first playback device is determined based on the distance sensor signal.

[0023] Based on the positioning sensor signal, the relative position information between the user and the second playback device is determined;

[0024] Based on the standing discrimination score, the distance change rate, and the relative position information, the observation vector for the current moment is generated.

[0025] Optionally, the video playback method further includes:

[0026] The distance acceleration is determined based on the rate of change of distance.

[0027] The confidence multiplier corresponding to the distance channel in the preset observation matrix is ​​determined based on the magnitude of the distance acceleration; wherein, the confidence multiplier is used to adjust the coefficients corresponding to the distance channel in the preset observation matrix.

[0028] Optionally, the video playback method further includes:

[0029] If the aspect ratios of the first playback device and the second playback device are inconsistent, saliency detection is performed on the current video frame to obtain saliency detection results; wherein, the saliency detection includes at least two of the following: face saliency detection, subject object saliency detection, and global semantic saliency detection;

[0030] The weights corresponding to each saliency map in the saliency detection results are dynamically assigned based on the face detection results.

[0031] Based on the saliency maps and their corresponding weights in the saliency detection results, a comprehensive saliency map is determined.

[0032] The optimal cropping window position is determined based on the comprehensive saliency map, and the current video frame is cropped according to the optimal cropping window position to obtain a video frame with an aspect ratio adapted to the second playback device.

[0033] Optionally, the video playback method further includes:

[0034] Obtain the feature vector returned by the user;

[0035] The user-returned feature vector is input into the confidence evaluation model to obtain the user-returned confidence score.

[0036] When the user's confidence level reaches the return threshold, the playback status data of the second playback device is collected and transferred to the first playback device; wherein, the first playback device continues playback according to the playback status data of the second playback device.

[0037] Optionally, the video playback method further includes:

[0038] Record each accidental touch of a label by a user that triggers a flow event;

[0039] The user's accidental touch rate is determined based on the accidental touch label;

[0040] Based on the false touch rate, a personalized false touch rate estimate for each user is determined;

[0041] The flow trigger threshold and the return threshold are adjusted based on the personalized false touch rate estimate.

[0042] Optionally, the video playback method further includes:

[0043] Record the feature vector of each time a user triggers a flow event;

[0044] A linear discriminant model is trained based on the feature vectors of the aforementioned flow events, and feature weight vectors are extracted from the trained linear discriminant model.

[0045] The feature weight vector is mapped to the diagonal elements of the preset observation matrix.

[0046] A second aspect of the present invention provides a video playback device, comprising:

[0047] The sensor signal acquisition unit is used to acquire visual sensor signals, distance sensor signals, and positioning sensor signals;

[0048] The departure intention confidence determination unit is used to determine the user's departure intention confidence based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal;

[0049] The first acquisition unit is used to acquire playback status data of the first playback device when the confidence level of the user's intention to leave is greater than the flow trigger threshold.

[0050] The first transfer unit is used to transfer the playback status data of the first playback device to the second playback device; wherein the second playback device continues playback according to the playback status data of the first playback device.

[0051] Optionally, the departure intention confidence determination unit includes:

[0052] An observation vector determination unit is used to determine the observation vector at the current moment based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal.

[0053] The prior state estimation vector determination unit is used to determine the prior state estimation vector at the current moment based on the state estimation vector at the previous moment and the preset state transition matrix.

[0054] The observation residual determination unit is used to determine the observation residual based on the observation vector at the current time, the prior state estimation vector at the current time, and the preset observation matrix.

[0055] The dynamic fusion gain determination unit is used to determine the dynamic fusion gain based on the error covariance matrix of the prior state estimation vector at the current time, the preset observation matrix, and the preset observation noise covariance matrix.

[0056] The correction unit is used to correct the prior state estimation vector at the current time using the observation residual and the dynamic fusion gain to obtain the posterior state estimation vector at the current time.

[0057] The departure intention confidence determination subunit is used to determine the user's departure intention confidence based on the state posterior estimation vector at the current moment.

[0058] Optionally, the dynamic fusion gain determination unit includes:

[0059] The residual covariance determination unit is used to determine the residual covariance based on the error covariance matrix of the prior state estimation vector at the current time, the preset observation matrix, and the preset observation noise covariance matrix.

[0060] The dynamic fusion gain determination subunit is used to determine the dynamic fusion gain based on the error covariance matrix, the preset observation matrix, and the residual covariance.

[0061] Optionally, the observation vector determination unit includes:

[0062] The standing discrimination score determination unit is used to determine the user's posture feature vector based on the visual sensor signal, and to determine the standing discrimination score based on the user's posture feature vector, reference posture center vector and posture weight matrix;

[0063] The distance change rate determination unit is used to determine the distance change rate between the user and the first playback device based on the distance sensor signal.

[0064] The relative position information determination unit is used to determine the relative position information between the user and the second playback device based on the positioning sensor signal.

[0065] The observation vector determination subunit is used to generate the observation vector at the current moment based on the standing discrimination score, the distance change rate, and the relative position information.

[0066] Optionally, the video playback device further includes:

[0067] A distance acceleration determination unit is used to determine the distance acceleration based on the distance change rate;

[0068] The confidence multiplier determination unit is used to determine the confidence multiplier corresponding to the distance channel in the preset observation matrix based on the amplitude of the distance acceleration; wherein the confidence multiplier is used to adjust the coefficients corresponding to the distance channel in the preset observation matrix.

[0069] Optionally, the video playback device further includes:

[0070] A saliency detection unit is used to perform saliency detection on the current video frame and obtain a saliency detection result when the aspect ratios of the first playback device and the second playback device are inconsistent; wherein, the saliency detection includes at least two of face saliency detection, subject object saliency detection and global semantic saliency detection;

[0071] The weight allocation unit is used to dynamically allocate the weights corresponding to each saliency map in the saliency detection results based on the face detection results.

[0072] The comprehensive saliency map determination unit is used to determine the comprehensive saliency map based on each saliency map in the saliency detection result and the weight corresponding to each saliency map;

[0073] The cropping unit is used to determine the optimal cropping window position based on the comprehensive saliency map, and crop the current video frame according to the optimal cropping window position to obtain a video frame with an aspect ratio adapted to the second playback device.

[0074] Optionally, the video playback device further includes:

[0075] Return to the feature vector acquisition unit, used to obtain the feature vector returned by the user;

[0076] The input unit is used to input the user-returned feature vector back into the confidence evaluation model to obtain the user-returned confidence score.

[0077] The second acquisition unit is used to acquire playback status data of the second playback device when the user's return confidence level reaches the return threshold.

[0078] The second transfer unit is used to transfer the playback status data of the second playback device to the first playback device; wherein the first playback device continues playback according to the playback status data of the second playback device.

[0079] Optionally, the video playback device further includes:

[0080] The first recording unit is used to record each accidental touch of the label by the user during a flow event;

[0081] A false touch rate determination unit is used to determine the user's false touch rate based on the false touch tag;

[0082] A personalized accidental touch rate estimation unit is used to determine a user's personalized accidental touch rate estimate based on the accidental touch rate.

[0083] An adjustment unit is used to adjust the flow trigger threshold and the return threshold based on the personalized false touch rate estimate.

[0084] Optionally, the video playback device further includes:

[0085] The second recording unit is used to record the feature vector of each time a user triggers a flow event;

[0086] The feature weight vector determination unit is used to train a linear discriminant model based on the feature vector of the circulation event, and extract the feature weight vector from the trained linear discriminant model.

[0087] A mapping unit is used to map the feature weight vector to the diagonal elements of the preset observation matrix.

[0088] A third aspect of the present invention provides an electronic device, comprising:

[0089] One or more processors;

[0090] A storage device on which one or more programs are stored;

[0091] When the one or more programs are executed by the one or more processors, the one or more processors implement the video playback method as described in any one of the first aspects.

[0092] A fourth aspect of the present invention provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the video playback method as described in any one of the first aspects.

[0093] As can be seen from the above solutions, the present invention provides a video playback method and apparatus. By determining the confidence level of the user's intention to leave based on visual sensor signals, distance sensor signals, and positioning sensor signals, if the confidence level of the user's intention to leave is greater than the flow trigger threshold, the playback status data of the first playback device is automatically collected and sent to the second playback device, so that the second playback device can seamlessly continue playback. In this process, no user intervention is required, realizing the automatic flow of video playback tasks between playback devices, eliminating the sense of disconnect caused by operation lag, and significantly improving the continuity and intelligence level of cross-device viewing experience. Attached Figure Description

[0094] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0095] Figure 1 A flowchart of a video playback method provided in an embodiment of the present invention;

[0096] Figure 2A flowchart of a method for determining the confidence level of a user's intention to leave, provided in another embodiment of the present invention;

[0097] Figure 3 A flowchart illustrating a method for determining the observation vector at the current moment, as provided in another embodiment of the present invention;

[0098] Figure 4 A flowchart of a video adaptive re-cropping method provided in another embodiment of the present invention;

[0099] Figure 5 A flowchart of a video playback reverse streaming method provided in another embodiment of the present invention;

[0100] Figure 6 A flowchart illustrating a method for adjusting a flow trigger threshold and a return threshold, as provided in another embodiment of the present invention;

[0101] Figure 7 A flowchart illustrating a method for personalized allocation of sensor channel trust levels according to another embodiment of the present invention;

[0102] Figure 8 This is a schematic diagram of a video playback device provided in another embodiment of the present invention. Detailed Implementation

[0103] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0104] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0105] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this invention are all information and data authorized by the user or fully authorized by all parties.

[0106] It should be noted that the concepts of "first" and "second" mentioned in this invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0107] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0108] This invention provides a video playback method, such as... Figure 1 As shown, it specifically includes:

[0109] S101. Acquire visual sensor signals, distance sensor signals, and positioning sensor signals.

[0110] The visual sensor signals can be, but are not limited to, image frame sequences captured by a camera mounted on the first playback device (such as a TV, computer, or tablet), used to sense the user's posture and movements. The distance sensor signals can be, but are not limited to, spatial distance measurements between the user and the screen of the first playback device output by a time-of-flight (TOF) sensor or infrared distance sensor mounted on the first playback device. The positioning sensor signals can be, but are not limited to, Bluetooth signal strength indicators and wireless network round-trip time values ​​collected by the second playback device (such as a mobile phone or tablet), used to determine the spatial position of the first playback device relative to the second playback device.

[0111] In the practical application of this invention, the visual sensor signal can be continuously acquired by the camera of the first playback device at a preset frame rate. The distance sensor signal is continuously acquired by the TOF or infrared distance sensor of the first playback device at a preset sampling frequency, acquiring the distance value between the user and the screen. The positioning sensor signal is acquired by the second playback device through signal interaction with the first playback device via a Bluetooth module to obtain the received signal strength indication value, and simultaneously by obtaining the round-trip delay value between the second playback device and the first playback device through a wireless network protocol. The above three signals are generated independently on the time axis at their respective inherent sampling frequencies, forming the original data basis for subsequent fusion and inference of the user's departure intention confidence.

[0112] S102. Based on the visual sensor signal, distance sensor signal and positioning sensor signal, determine the confidence level of the user's intention to leave.

[0113] First, a state vector can be predefined to describe the user's motion state at the current moment. Its components include the user's spatial position estimate p_user, movement speed v_user, acceleration a_user, and the probability of leaving θ_intent.

[0114] Specifically, the state vector x(t) can be represented as: x(t) = [p_user(t), v_user(t), a_user(t), θ_intent(t)] T .

[0115] The uniform acceleration model and intention decay characteristics of user motion are described by the state transition equation x(t+1)=F·x(t)+B·u(t)+w(t), where F is the preset state transition matrix, B is the preset control input matrix, u(t) is the control input vector, representing the user's external motion input known at time t, specifically including the user's intentional movement acceleration components along the horizontal and vertical directions, which is obtained by second-order difference of the distance sensor signal; w(t) is the process noise vector, representing the random disturbances that cannot be captured, including the user's unconscious small movements, sensor quantization errors, etc. Its statistical characteristics are modeled as a zero-mean Gaussian distribution w(t)~N(0,Q), where Q is the process noise covariance matrix, and the diagonal elements of Q are calibrated according to the physical range and historical statistical variance of each state dimension.

[0116] Optionally, in another embodiment of the present invention, one implementation of step S102 is as follows: Figure 2 As shown, it includes:

[0117] S201. Determine the observation vector at the current moment based on the visual sensor signal, distance sensor signal, and positioning sensor signal.

[0118] Optionally, in another embodiment of the present invention, one implementation of step S201 is as follows: Figure 3 As shown, it includes:

[0119] S301. Based on the visual sensor signal, determine the user's posture feature vector, and based on the user's posture feature vector, reference posture center vector and posture weight matrix, determine the standing discrimination score.

[0120] In the practical application of this invention, the coordinate information of 17 key points of the human skeleton can be obtained from the image frames acquired by the visual sensor through the skeleton key point extraction network. Then, based on these coordinate information, the difference in the vertical coordinates of the shoulder and hip, Δh_sh (to represent the degree of upper body straightness), the vertical velocity of the midpoint of the spine, v_spine, and the trunk tilt angle, θ_torso, are calculated. The above three features are combined into the user posture feature vector vs(t)=[Δh_sh(t), v_spine(t), θ_torso(t)] at the current time t.

[0121] Subsequently, based on the user's posture feature vector vs(t), the reference posture center vector μ_standing, and the posture weight matrix W_standing, the standing discrimination score D_standing(t) is determined, which can be calculated using the following formula:

[0122] D_standing(t)=‖W_standing·(vs(t)-μ_standing)‖2;

[0123] Where μ_standing = [μ_Δh, μ_v, μ_θ]^T, is obtained by calculating the mean from the labeled standing samples. μ_Δh is the sample mean of the shoulder-hip height difference in the standing state, μ_v is the sample mean of the spinal velocity at the moment of standing transition, and μ_θ is the mean of the standing trunk angle. The posture weight matrix W_standing = diag(w_1, w_2, w_3) can be determined by offline learning through maximizing the inter-class / intra-class scatter ratio S_B / S_W using the Fisher linear discriminant criterion, but not limited to this. The three component weights reflect the discriminative contribution of each feature to the sitting / standing classification. The larger the D_standing, the closer the current posture is to the standing / standing state.

[0124] In the practical application of this invention, when the camera is blocked or there is insufficient light, causing the average confidence value of the skeleton key points to be lower than the confidence threshold of the skeleton key points, the visual channel is automatically marked as invalid.

[0125] S302. Determine the rate of change of distance between the user and the first playback device based on the distance sensor signal.

[0126] In the practical application of this invention, after obtaining the instantaneous distance value between the user and the first playback device based on the distance sensor signal, a first-order difference operation is performed on the instantaneous distance value to obtain the distance change rate v_d(t)=(d(t)-d(t-1))×fs, where fs is the sampling frequency Hz.

[0127] S303. Determine the relative position information between the user and the second playback device based on the positioning sensor signal.

[0128] The positioning sensor signals include Bluetooth signal strength indication and wireless network round-trip time.

[0129] In the practical application of this invention, the Bluetooth received signal strength indicator value is subjected to exponential smoothing filtering to remove high-frequency noise from the signal, resulting in a smoothed signal strength value. This smoothed signal strength value is then combined with the wireless network round-trip time delay value for joint positioning to determine the relative position information of the user and the playback device.

[0130] It should be noted that when the Bluetooth signal reception is affected by environmental interference, the weight of the wireless network round-trip time delay in the joint positioning is automatically increased to ensure positioning accuracy.

[0131] Specifically, the variance of Bluetooth RSSI, σ²_RSSI, is monitored in real time. When σ²_RSSI exceeds a preset interference threshold σ²_threshold, the weights of the joint positioning are adaptively adjusted from the default (ω_BT=0.6, ω_WIFI=0.4) to (ω_BT=0.2, ω_WIFI=0.8). The weight adjustment formula is: ω_WIFI=0.4+0.4×max(0,(σ²_RSSI-σ²_threshold) / (σ²_max-σ²_threshold)), where σ²_max is the maximum allowable interference variance, ω_BT is the weight of the Bluetooth received signal in the joint positioning, and ω_WIFI is the weight of the wireless network round-trip delay in the joint positioning. This adaptive weight adjustment mechanism ensures that the system maintains positioning accuracy even in environments where Bluetooth signals are blocked or multipath effects are severe.

[0132] S304. Based on the standing discrimination score, distance change rate, and relative position information, generate the observation vector for the current moment.

[0133] In practical applications of this invention, the standing discrimination score, distance change rate, and relative position information can be combined into the observation vector at the current moment. Specifically, the observation vector at the current moment can be expressed as: y(t) = [D_standing(t), v_d(t), d_rtt(t)] T Alternatively, the smoothed signal intensity value can be added as an information dimension in the observation vector, such as y(t)=[D_standing(t), v_d(t), r_smooth(t), d_rtt(t)]. T .

[0134] Where r_smooth(t) is the smoothing signal intensity value, and d_rtt(t) is the relative position information.

[0135] In the practical application of this invention, the observations of each sensor can be associated with the state vector by y(t+1)=H·x(t+1)+v(t), where H is a preset observation matrix and v(t) is an observation noise vector, representing random errors in sensor measurements, including attitude estimation jitter of the vision sensor, ranging accuracy error of the distance sensor, signal fluctuation of the positioning sensor, etc. Its statistical characteristics are modeled as a zero-mean Gaussian distribution v(t)~N(0,R), where R is the observation noise covariance matrix. The diagonal elements of R are determined according to the measurement variance of each sensor under the calibration environment, and the off-diagonal elements are set to zero when the noise between sensors is uncorrelated.

[0136] S202. Based on the state estimation vector of the previous moment and the preset state transition matrix, determine the prior state estimation vector of the current moment.

[0137] Specifically, based on the state estimation vector from the previous time step Given (t-1|t-1) and a preset state transition matrix F, predict the prior state estimate vector at the current time step. (t|t-1), (t|t-1)=F· (t-1|t-1).

[0138] S203. Based on the observation vector at the current time, the prior state estimation vector at the current time, and the preset observation matrix, determine the observation residual.

[0139] The observation residual represents the deviation between the observation vector at the current moment and the prior state estimation vector based on the current moment. The preset observation matrix H is used to map the prior state estimation vector at the current moment from the state space to the observation space, establishing a linear relationship between each state component and each observation component.

[0140] Specifically, observation residuals =y(t)-H· (t|t-1).

[0141] S204. Determine the dynamic fusion gain based on the error covariance matrix of the prior state estimation vector at the current moment, the preset observation matrix, and the preset observation noise covariance matrix.

[0142] Optionally, in another embodiment of the present invention, one implementation of step S204 includes:

[0143] Based on the error covariance matrix, preset observation matrix, and preset observation noise covariance matrix of the prior state estimation vector at the current moment, the residual covariance is determined; based on the error covariance matrix, preset observation matrix, and residual covariance, the dynamic fusion gain is determined.

[0144] In the practical application of this invention, the residual covariance S(t) can be calculated as follows:

[0145] S(t) = H·P(t|t-1)·H T +R;

[0146] Where P(t|t-1) is the error covariance matrix of the prior state estimation vector at the current time, and R is the preset observation noise covariance matrix, used to describe the measurement noise level of each sensor channel. The smaller the diagonal element value, the higher the measurement accuracy of the corresponding channel. P(t|t-1)=F·P(t-1|t-1)·F T +Q, where Q is the process noise covariance matrix, used to quantify the uncertainty of the state transition model itself.

[0147] In the practical application of this invention, the dynamic fusion gain K(t) can be calculated as follows:

[0148] K(t) = P(t|t-1)·H T ·S(t) -1 ;

[0149] The dynamic fusion gain in this invention is an adaptively adjusted coefficient matrix, which assigns different trust weights to each sensor channel during the calibration step. Channels with lower measurement noise are given higher gain, while channels with higher measurement noise or continuous failures are given lower gain.

[0150] In the practical application of this invention, the posterior covariance P(t|t) = (IK(t)·H)·P(t|t-1), where I is the identity matrix, used to maintain dimensional consistency and achieve element-wise scaling, and (IK(t)·H) represents subtracting the product of the gain and the observation matrix from the identity matrix, which is equivalent to reducing each state dimension to different degrees. It reduces the uncertainty of state estimation through observation information. The more reliable the observation (the greater the gain), the greater the corresponding reduction and the smaller the posterior covariance.

[0151] S205. The prior state estimation vector at the current time is corrected using the observation residual and dynamic fusion gain to obtain the posterior state estimation vector at the current time.

[0152] Specifically, the predicted state is corrected by using the observation residual as the correction amount and the dynamic fusion gain as the weighting coefficient, thereby obtaining the optimal state estimate after fusing multi-source information.

[0153] In the practical application of this invention, the posterior state estimation vector at the current moment can be calculated using the following formula. (t|t):

[0154] (t|t)= (t|t-1)+K(t)· .

[0155] S206. Based on the posterior state estimation vector at the current moment, determine the confidence level of the user's intention to leave.

[0156] Specifically, the user's intention to leave probability component is extracted from the state posterior estimation vector, and this probability component is mapped to a confidence interval of 0 to 1 through a probability compression function to obtain the confidence of the user's intention to leave.

[0157] Optionally, in another embodiment of the present invention, one implementation of the video playback method further includes:

[0158] The distance acceleration is determined based on the rate of change of distance; the confidence multiplier corresponding to the distance channel in the preset observation matrix is ​​determined based on the magnitude of the distance acceleration.

[0159] The confidence multiplier is used to adjust the coefficients corresponding to the distance channels in the preset observation matrix.

[0160] In the practical application of this invention, the distance change rate v_d(t) can be subjected to a second-order difference operation to obtain the distance acceleration a_d(t) = (v_d(t) - v_d(t-1)) × fs.

[0161] In the practical application of this invention, the confidence multiplier γ_d(t) can be divided into three levels according to the amplitude of distance acceleration. For example: when |a_d(t)| is greater than or equal to 0.4 m / s² (accelerating away), γ_d = 1.0; when |a_d(t)| < 0.4 m / s² (moving away at a constant speed), γ_d = 0.7; when |a_d(t)| < 0.1 m / s² (moving away at a deceleration), γ_d = 0.4.

[0162] This invention integrates three heterogeneous signals—visual, distance, and positioning—in a unified manner. The dynamic fusion gain mechanism enables the system to adaptively adjust the trust weights based on the real-time prediction accuracy and observation noise of each channel. When a channel fails continuously, its observation noise is set to a maximum value, thereby making the fusion gain approach zero and achieving automatic weight reduction. The weights are gradually restored after the channel is restored.

[0163] S103. If the confidence level of the user's intention to leave is greater than the flow trigger threshold, collect the playback status data of the first playback device.

[0164] Among them, playback status data refers to the collection of all status parameters of the video playback session currently running on the first playback device.

[0165] Optionally, in another embodiment of the present invention, the specific implementation of collecting playback status data can be as follows (steps A1 to A3):

[0166] Step A1: Get the current playback progress.

[0167] The playback progress is measured in milliseconds and can be taken from the playback device's internal timer, but is not limited to this.

[0168] Optionally, in another embodiment of the present invention, in order to improve the accuracy of progress synchronization, the original progress value needs to be compensated for a time delay.

[0169] Specifically, the network transmission one-way latency Δt_network is obtained, which is half of the round-trip latency; the estimated decoding latency of the target device Δt_decode is obtained; and the buffer pre-fill time Δt_buffer is obtained. All three compensation values ​​are updated using the sliding average of the previous stream's measured values. The original playback progress progress_raw is added to the above three compensation values ​​to obtain the compensated playback progress progress_compensated = progress_raw + Δt_network + Δt_decode + Δt_buffer.

[0170] Step A2: Collect volume parameters, bullet screen data, video identification information, authentication credentials, subtitle information, audio track information, and image quality parameters.

[0171] The bullet screen data includes at least the N most recently posted bullet screen comments and their timestamps. Video identification information uniquely identifies the currently playing video content. Authentication credentials are used by the target device to verify user identity and playback permissions when resuming playback.

[0172] Step A3: Serialize all the data collected above to generate a broadcast control status data packet.

[0173] S104. Transfer the playback status data stream from the first playback device to the second playback device.

[0174] The second playback device continues playback based on the playback status data of the first playback device.

[0175] In the practical application of this invention, the transmission of playback status data can be accomplished through a direct communication link between devices, which is not limited here.

[0176] Optionally, in another embodiment of the present invention, the data transfer can be implemented as follows:

[0177] During the initial playback, a data packet containing all playback control status data is sent. In subsequent playbacks, differential incremental data packets are sent, containing only the difference fields between the current playback control status and the previous confirmed status. The receiving end merges these differential incremental data packets into its local buffer, restoring the complete playback status.

[0178] Optionally, in another embodiment of the present invention, each data entry may also include a cyclic redundancy check value for the receiving end to verify the integrity of the data transmission. If the verification fails, the receiving end requests a retransmission; after the number of retransmissions exceeds a preset number, the system switches to a cloud relay mode.

[0179] Optionally, in another embodiment of the present invention, the transmission channel preferentially establishes an encrypted direct connection through the local area network device self-discovery protocol, and falls back to the end-to-end encrypted cloud relay mode when the direct connection fails to be established.

[0180] Optionally, in another embodiment of the present invention, one implementation of the video playback method is as follows: Figure 4 As shown, it also includes:

[0181] S401. If the aspect ratios of the first playback device and the second playback device are inconsistent, perform saliency detection on the current video frame and obtain the saliency detection result.

[0182] The saliency detection includes at least two of the following: face saliency detection, subject object saliency detection, and global semantic saliency detection.

[0183] For example, when switching from a 16:9 landscape screen on a TV to a 9:16 portrait screen on a mobile phone, the video needs to be adaptively re-cropped to retain the main content of the video to the greatest extent possible in the portrait display area.

[0184] Specifically, face saliency detection, subject saliency detection, and global semantic saliency detection are performed on the current video frame to generate face saliency map, subject saliency map, and global semantic saliency map, respectively.

[0185] In the practical application of this invention, face saliency detection obtains the position bounding box and detection confidence of each face in the current frame through a multi-stage cascaded face detection network. A face saliency map S_face(x,y) is generated using a Gaussian kernel with the center of each face as the center. When multiple faces exist, the maximum value of each pixel is taken.

[0186] The subject saliency detection uses a lightweight object detector to output bounding boxes for various objects. These bounding boxes are then weighted according to the product of category priority weights and detection confidence scores to generate a subject saliency map S_obj(x,y). The category priority weights can be set so that people have the highest weight, followed by animals, then text, and other categories have the lowest weight. For example: people 1.0, animals 0.8, text 0.7, and others 0.4.

[0187] Global semantic saliency detection generates a normalized heatmap S_sal(x,y) from the pixel-level semantic saliency map generation network. This heatmap can capture semantically important regions outside the bounding box.

[0188] S402. Dynamically assign weights to each saliency map in the saliency detection results based on the face detection results.

[0189] In the practical application of this invention, the dynamic allocation strategy can be as follows: when a face is detected, the face saliency map is assigned a weight of 0.45, i.e., ω_face=0.45, the subject saliency map is assigned a weight of 0.30, i.e., ω_obj=0.30, and the global semantic saliency map is assigned a weight of 0.25, i.e., ω_sal=0.25; when no face is detected but a subject is detected, the face saliency map is assigned a weight of 0, i.e., ω_face=0, the subject saliency map is assigned a weight of 0.55, i.e., ω_obj=0.55, and the global semantic saliency map is assigned a weight of 0.45, i.e., ω_sal=0.45; when neither a face nor a subject is detected, the global semantic saliency map is assigned all weights, i.e., ω_sal=1, thereby ensuring effective saliency output in any scenario.

[0190] S403. Based on the saliency maps and their corresponding weights in the saliency detection results, determine the comprehensive saliency map.

[0191] In the practical application of this invention, the comprehensive saliency map S_final can be calculated using the following formula:

[0192] S_final=ω_face×S_face+ω_obj×S_obj+ω_sal×S_sal.

[0193] S404. Determine the optimal cropping window position based on the comprehensive saliency map, and crop the current video frame according to the optimal cropping window position to obtain a video frame with an aspect ratio adapted to the second playback device.

[0194] In the practical application of this invention, the greedy sliding window search method can be accelerated by using integral images, and the search can be carried out in steps of four pixels to find the clipping position with the largest sum of comprehensive saliency within the window.

[0195] An integral map can be constructed based on a comprehensive saliency map. In the integral map, the value at each position is the sum of all saliency values ​​within the rectangular area from the top left corner of the comprehensive saliency map to that position. Using this integral map, during the greedy sliding window search process, the sum of saliency values ​​within any candidate clipping window can be quickly obtained through four table lookups without pixel-by-pixel accumulation, thus accelerating the search.

[0196] Optionally, in another embodiment of the present invention, to maintain visual stability of the image between frames, the cropping window positions of adjacent video frames are smoothed using an exponential moving average, with a smoothing coefficient set to 0.3, while constraining the maximum displacement of the cropping positions of adjacent frames to no more than 16 pixels. When a hard scene change is detected, the smoothing process is reset to avoid cross-scene cropping position trailing.

[0197] Optionally, in another embodiment of the present invention, when the user manually adjusts the cropping frame, the current scene features and the user offset are recorded for subsequent personalized cropping correction.

[0198] Optionally, in another embodiment of the present invention, one implementation of the video playback method further includes a reverse flow step, such as... Figure 5 As shown, it also includes:

[0199] S501. Obtain the feature vector returned by the user.

[0200] The user-returned feature vector includes at least one of the following: location signal strength value, wireless network round-trip time distance value, near-field sensor distance value, approach velocity value, approach acceleration value, and near-field condition duration frame number.

[0201] The positioning signal strength value is the result of smoothing and filtering the Bluetooth received signal strength indication value. The wireless network round-trip time delay distance value is obtained by converting the round-trip time delay. The near field sensor distance value is the near field distance of people measured by the TOF sensor. The approach velocity value is the first difference of this distance value (negative values ​​indicate proximity). The approach acceleration value is the second difference. The near field condition duration frame number is the number of consecutive frames in which each signal simultaneously meets the near field condition.

[0202] S502. Input the user-returned feature vector into the confidence evaluation model to obtain the user-returned confidence score.

[0203] The confidence assessment model can be a lightweight linear model trained by logistic regression, whose output is mapped to the 0 to 1 interval via a sigmoid function.

[0204] S503. When the user's confidence level reaches the return threshold, collect the playback status data of the second playback device.

[0205] In practical applications of this invention, a preset condition can also be set. This preset condition can be set to a confidence level reaching a threshold and lasting for more than three seconds, using a sliding window to allow a 20% jitter tolerance. After triggering the reverse flow event, a confirmation pop-up appears on the second playback device. After a three-second timeout, confirmation is automatically performed. If the user cancels, the accidental touch is recorded and the accidental touch tag count is updated.

[0206] S504. Transfer the playback status data stream from the second playback device to the first playback device.

[0207] The first playback device continues playback based on the playback status data of the second playback device.

[0208] In the actual application of this invention, in the reverse flow of playback status data merging, the playback progress takes the current value of the second playback device as the authoritative source, the bullet screen data is taken as the union of the second playback device and the first playback device set, and the most recent 30 bullet screens are retained after deduplication by timestamp, and the volume retains the current adjustment value of the second playback device.

[0209] In the actual application of this invention, during the reverse flow of the screen adaptation, the center horizontal coordinate of the current display screen of the second playback device is used as the reference, and several columns of pixels are extended to the left and right sides. The extended area is taken from the original 16:9 pixels stored in the preloaded buffer before reverse flow to ensure geometric alignment and no interpolation distortion.

[0210] Optionally, in another embodiment of the present invention, when the target playback device (the target playback device is the second playback device when the flow is forward, and the target playback device is the first playback device when the flow is reverse) starts to continue playback, the current playback device (the current playback device is the first playback device when the flow is forward, and the current playback device is the second playback device when the flow is reverse) may also stop playback, which is not limited here.

[0211] In the practical application of this invention, the specific implementation method for stopping playback is as follows.

[0212] The current playback device and the target playback device synchronize their clocks using a network time protocol or a precise time protocol, aligning the system clocks at both ends with a synchronization accuracy controlled within 20 milliseconds. Based on the synchronized clock, the transition time is determined.

[0213] During the transition, the current playback device is controlled to perform a volume fade-out operation for a preset fade-out duration. Simultaneously, the target playback device is controlled to perform a volume fade-in operation for the same preset fade-out duration. For example, the fade-out duration can be set to 300 milliseconds. Within this time, the volume of the current playback device linearly decreases from its current value to zero, while the volume of the target playback device linearly increases from zero to the target volume value. The combined audio power of both devices does not exceed the target volume value plus M decibels to prevent popping sounds.

[0214] Optionally, in another embodiment of the present invention, one implementation of the video playback method is as follows: Figure 6 As shown, it also includes:

[0215] S601, Record each accidental touch of a label by the user that triggers a flow event.

[0216] S602. Determine the user's accidental touch rate based on accidental touch tags.

[0217] In the practical application of this invention, the accidental touch tag for accidental touch operation can be recorded as 1, and the accidental touch tag for non-accidental touch operation can be recorded as 0, so as to determine the user's accidental touch rate based on all accidental touch tags.

[0218] S603. Determine the user's personalized accidental touch rate estimate based on the accidental touch rate.

[0219] Specifically, the false touch rate can be used as the observed variable, and the posterior probability can be updated based on the conjugate prior distribution to obtain a personalized estimate of the false touch rate for each user. Specifically, the conjugate prior of the false touch rate π is set as B(…). , After observing the number of accidental touches in several cycles, the posterior distribution remains B( +k, +nk), and its posterior mean π_post is the personalized false touch rate estimate.

[0220] S604. Adjust the flow trigger threshold and return threshold based on the personalized accidental touch rate estimate.

[0221] In the practical application of this invention, the intent threshold can be increased to reduce sensitivity when π_post>0.15, and the threshold can be decreased to increase sensitivity when π_post<0.05. The magnitude of the change is constrained by the width of the posterior confidence interval (the magnitude is reduced when convergence is not achieved).

[0222] Optionally, in another embodiment of the present invention, one implementation of the video playback method is as follows: Figure 7 As shown, it also includes:

[0223] S701, Record the feature vector of each time a user triggers a flow event.

[0224] The feature vector includes one or more of the following: standing discrimination score, distance change rate, Bluetooth signal strength value, round-trip time distance value, intent confidence, confirmation delay, and accidental tag touch, without limitation here.

[0225] S702. Train a linear discriminant model based on the feature vectors of the flow events, and extract the feature weight vector from the trained linear discriminant model.

[0226] Specifically, the feature weights are trained using a linear discriminant model P(confirm=1|Φ)=σ(w^T·Φ+b) and an L2 regularization term is introduced. The feature weight vector is then extracted from the trained linear discriminant model.

[0227] S703. Map the feature weight vector to the diagonal elements of the preset observation matrix.

[0228] Continuing with the above example, after normalizing the first few dimensions of the feature weight vector, the normalized values ​​of each dimension are assigned to the corresponding diagonal elements of the preset observation matrix H, thereby realizing the personalized allocation of trust levels for different sensor channels.

[0229] As can be seen from the above scheme, the present invention provides a video playback method. By determining the confidence level of the user's intention to leave based on visual sensor signals, distance sensor signals, and positioning sensor signals, if the confidence level of the user's intention to leave is greater than the flow trigger threshold, the playback status data of the first playback device is automatically collected and sent to the second playback device, so that the second playback device can seamlessly continue playback. In this process, no user intervention is required, realizing the automatic flow of video playback tasks between playback devices, eliminating the sense of disconnect caused by operation lag, and significantly improving the continuity and intelligence level of cross-device viewing experience.

[0230] Another embodiment of the present invention provides a video playback device, such as... Figure 8 As shown, it specifically includes:

[0231] The sensor signal acquisition unit 801 is used to acquire visual sensor signals, distance sensor signals and positioning sensor signals.

[0232] The departure intention confidence determination unit 802 is used to determine the user's departure intention confidence based on visual sensor signals, distance sensor signals, and positioning sensor signals.

[0233] The first acquisition unit 803 is used to acquire playback status data of the first playback device when the confidence level of the user's intention to leave is greater than the flow trigger threshold.

[0234] The first transfer unit 804 is used to transfer the playback status data of the first playback device to the second playback device.

[0235] The second playback device continues playback based on the playback status data of the first playback device.

[0236] For details on the specific operation of the units disclosed in the above embodiments of the present invention, please refer to the corresponding method embodiments, such as... Figure 1 As shown, it will not be elaborated further here.

[0237] As can be seen from the above solution, the present invention provides a video playback device that determines the confidence level of the user's intention to leave based on visual sensor signals, distance sensor signals, and positioning sensor signals. If the confidence level of the user's intention to leave is greater than the flow trigger threshold, the device automatically collects the playback status data of the first playback device and sends it to the second playback device, thereby enabling the second playback device to seamlessly continue playback. In this process, no user intervention is required, realizing the automatic flow of video playback tasks between playback devices, eliminating the sense of disconnect caused by operation lag, and significantly improving the continuity and intelligence level of cross-device viewing experience.

[0238] Another embodiment of the present invention provides an electronic device, comprising:

[0239] One or more processors.

[0240] A storage device on which one or more programs are stored.

[0241] When the one or more programs are executed by the one or more processors, the one or more processors implement the video playback method as described in the above embodiments.

[0242] Another embodiment of the present invention provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the video playback method as described in the above embodiments.

[0243] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0244] It should be noted that the computer-readable medium described above in this invention can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0245] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0246] Another embodiment of the present invention provides a computer program product, which, when executed, is used to perform the video playback method described above.

[0247] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, it performs the functions defined in the methods of the embodiments of the present invention.

[0248] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in this invention is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely exemplary forms for implementing the invention.

[0249] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of the invention. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0250] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to the specific combination of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with technical features of the present invention (but not limited to) that have similar functions.

Claims

1. A video playback method, characterized in that, include: Acquire signals from the vision sensor, distance sensor, and positioning sensor; Based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal, determine the confidence level of the user's intention to leave; If the confidence level of the user's intention to leave is greater than the flow trigger threshold, the playback status data of the first playback device is collected. The playback status data of the first playback device is transferred to the second playback device; wherein the second playback device continues playback according to the playback status data of the first playback device.

2. The video playback method according to claim 1, characterized in that, Determining the confidence level of the user's intention to leave based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal includes: Based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal, the observation vector at the current moment is determined; Based on the state estimation vector of the previous moment and the preset state transition matrix, determine the prior state estimation vector of the current moment; Based on the observation vector at the current moment, the prior state estimation vector at the current moment, and the preset observation matrix, the observation residual is determined; The dynamic fusion gain is determined based on the error covariance matrix of the prior state estimation vector at the current moment, the preset observation matrix, and the preset observation noise covariance matrix. The prior state estimation vector at the current moment is corrected using the observation residual and the dynamic fusion gain to obtain the posterior state estimation vector at the current moment; Based on the posterior state estimation vector at the current moment, the confidence level of the user's intention to leave is determined.

3. The video playback method according to claim 2, characterized in that, The step of determining the dynamic fusion gain based on the error covariance matrix of the prior state estimation vector at the current time, the preset observation matrix, and the preset observation noise covariance matrix includes: The residual covariance is determined based on the error covariance matrix of the prior state estimation vector at the current moment, the preset observation matrix, and the preset observation noise covariance matrix. The dynamic fusion gain is determined based on the error covariance matrix, the preset observation matrix, and the residual covariance.

4. The video playback method according to claim 2, characterized in that, Determining the observation vector at the current moment based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal includes: Based on the visual sensor signals, the user's posture feature vector is determined, and based on the user's posture feature vector, reference posture center vector, and posture weight matrix, a standing discrimination score is determined. The rate of change of distance between the user and the first playback device is determined based on the distance sensor signal. Based on the positioning sensor signal, the relative position information between the user and the second playback device is determined; Based on the standing discrimination score, the distance change rate, and the relative position information, the observation vector for the current moment is generated.

5. The video playback method according to claim 4, characterized in that, Also includes: The distance acceleration is determined based on the rate of change of distance. The confidence multiplier corresponding to the distance channel in the preset observation matrix is ​​determined based on the magnitude of the distance acceleration; wherein, the confidence multiplier is used to adjust the coefficients corresponding to the distance channel in the preset observation matrix.

6. The video playback method according to claim 1, characterized in that, Also includes: If the aspect ratios of the first playback device and the second playback device are inconsistent, saliency detection is performed on the current video frame to obtain saliency detection results; wherein, the saliency detection includes at least two of the following: face saliency detection, subject object saliency detection, and global semantic saliency detection; The weights corresponding to each saliency map in the saliency detection results are dynamically assigned based on the face detection results. Based on the saliency maps and their corresponding weights in the saliency detection results, a comprehensive saliency map is determined. The optimal cropping window position is determined based on the comprehensive saliency map, and the current video frame is cropped according to the optimal cropping window position to obtain a video frame with an aspect ratio adapted to the second playback device.

7. The video playback method according to claim 1, characterized in that, Also includes: Obtain the feature vector returned by the user; The user-returned feature vector is input into the confidence evaluation model to obtain the user-returned confidence score. When the user's confidence level reaches the return threshold, the playback status data of the second playback device is collected and transferred to the first playback device; wherein, the first playback device continues playback according to the playback status data of the second playback device.

8. The video playback method according to claim 7, characterized in that, Also includes: Record each accidental touch of a label by a user that triggers a flow event; The user's accidental touch rate is determined based on the accidental touch label; Based on the false touch rate, a personalized false touch rate estimate for each user is determined; The flow trigger threshold and the return threshold are adjusted based on the personalized false touch rate estimate.

9. The video playback method according to claim 2, characterized in that, Also includes: Record the feature vector of each time a user triggers a flow event; A linear discriminant model is trained based on the feature vectors of the aforementioned flow events, and feature weight vectors are extracted from the trained linear discriminant model. The feature weight vector is mapped to the diagonal elements of the preset observation matrix.

10. A video playback device, characterized in that, include: The sensor signal acquisition unit is used to acquire visual sensor signals, distance sensor signals, and positioning sensor signals; The departure intention confidence determination unit is used to determine the user's departure intention confidence based on the visual sensor signal, the distance sensor signal, and the positioning sensor signal; The first acquisition unit is used to acquire playback status data of the first playback device when the confidence level of the user's intention to leave is greater than the flow trigger threshold. The first transfer unit is used to transfer the playback status data of the first playback device to the second playback device; wherein the second playback device continues playback according to the playback status data of the first playback device.