Unmanned aerial vehicle visual information frame residual distribution prediction model construction method and prediction method

Abstract

The invention provides an unmanned aerial vehicle visual information frame residual distribution prediction model construction method and a prediction method. The method comprises the following steps: obtaining a frame residual distribution standard deviation actual value corresponding to a plurality of coding parameter groups after coding an unmanned aerial vehicle visual information frame; performing first-level linear regression on the quantization step size and the frame residual distribution standard deviation actual value in each coding parameter group; determining a first-level estimation error actual value between the first-level frame residual distribution standard deviation predicted value and the first-level frame residual distribution standard deviation actual value; performing second-level linear regression on the search range and the first-level estimation error actual value in each coding parameter group; determining a second-level estimation error actual value between the first-level estimation error predicted value and the first-level estimation error actual value; performing three-level linear regression on the related frame number and the second-level estimation error actual value in each coding parameter group; the frame residual distribution prediction model is obtained according to the three linear regression equations, the calculation complexity is reduced, and the model construction speed and the distribution prediction precision are improved.

Description

Technical Field

[0001] The present invention belongs to the field of electronic information, and in particular relates to a method for constructing a residual distribution prediction model of a drone visual information frame, a method for predicting the residual distribution of a drone visual information frame, a device for constructing a residual distribution prediction model of a drone visual information frame, a device for predicting the residual distribution of a drone visual information frame, an electronic device and a storage medium. Background Art

[0002] UAV visual information is usually collected by airborne visual acquisition equipment and encoded into video form for transmission and display. The original visual information takes up a lot of storage space, which is not conducive to transmission. Therefore, it is necessary to encode the visual information frame to reduce the space occupation. The commonly used encoding standard is H.264, which controls the encoding rate by configuring different encoding parameters. During the encoding process, the frame residual distribution directly affects the output bit rate and visual information distortion. Accurately estimating and predicting the frame residual distribution is of great significance to optimizing the visual information frame encoding control.

[0003] In related technologies, exponential models are usually used to estimate and predict the residual distribution of visual information frames. However, due to the limited computing power of the processors onboard drones and the high latency requirements for real-time visual information transmission, this method is limited in its application on drone flight platforms, and there are technical problems such as high computational complexity, slow model building speed, and insufficient distribution prediction accuracy. Summary of the invention

[0004] In a first aspect, an embodiment of the present invention provides a method for constructing a prediction model for residual distribution of a drone visual information frame, comprising: obtaining multiple coding parameter groups, and determining the actual value of the frame residual distribution standard deviation corresponding to each coding parameter group after encoding the drone visual information frame, wherein the coding parameter group includes a quantization step size, a search range, and a related number of frames; performing linear regression on the quantization step size and the actual value of the frame residual distribution standard deviation in each coding parameter group to obtain a first-order linear regression equation; calculating the first-order frame residual distribution standard deviation prediction value corresponding to each quantization step size according to the first-order linear regression equation, and determining the first-order frame residual distribution standard deviation prediction value and the frame residual. The actual value of the first-level estimation error between the actual value of the distribution standard deviation; perform linear regression on the search range in each coding parameter group and the actual value of the first-level estimation error to obtain a second-level linear regression equation; calculate the first-level estimation error prediction value corresponding to each search range according to the second-level linear regression equation, and determine the actual value of the second-level estimation error between the first-level estimation error prediction value and the actual value of the first-level estimation error; perform linear regression on the relevant frame number in each coding parameter group and the actual value of the second-level estimation error to obtain a third-level linear regression equation; determine the frame residual distribution prediction model according to the first-level linear regression equation, the second-level linear regression equation and the third-level linear regression equation.

[0005] In some embodiments, after determining the frame residual distribution prediction model, it also includes: calculating the frame residual distribution standard deviation prediction value corresponding to each encoding parameter group according to the frame residual distribution prediction model; and constructing an error distribution estimation model corresponding to the frame residual distribution prediction model according to the frame residual distribution standard deviation prediction value and the actual value of the frame residual distribution standard deviation.

[0006] In some embodiments, the error distribution estimation model corresponding to the frame residual distribution prediction model is constructed based on the frame residual distribution standard deviation prediction value and the frame residual distribution standard deviation actual value, including: determining the final error value between the frame residual distribution standard deviation prediction value and the frame residual distribution standard deviation actual value; determining an initial mixed Gaussian model; using an expectation maximization algorithm to iteratively optimize the initial mixed Gaussian model based on the final error value to obtain a final mixed Gaussian model, and the final mixed Gaussian model is used as the error distribution estimation model.

[0007] In some embodiments, the obtaining of multiple coding parameter groups includes: obtaining a coding parameter set, the coding parameter set including a quantization step parameter set, a search range parameter set and a related frame number parameter set; randomly selecting a quantization step, a search range and a related frame number from the coding parameter set to form a coding parameter group; repeatedly performing the steps of randomly selecting a quantization step, a search range and a related frame number from the coding parameter set to obtain multiple coding parameter groups.

[0008] In a second aspect, an embodiment of the present invention provides a method for predicting the residual distribution of a frame of a UAV visual information, which is applied to a frame residual distribution prediction model constructed by the method for constructing a frame residual distribution prediction model of a UAV visual information as described in any one of the first aspects, and the method comprises: obtaining a group of coding parameters to be predicted for the UAV visual information, the coding parameter group comprising a quantization step size, a search range and a related number of frames; inputting the group of coding parameters to be predicted into the frame residual distribution prediction model to obtain a standard deviation value of the frame residual distribution of the UAV visual information.

[0009] In some embodiments, after obtaining the standard deviation value of the frame residual distribution of the drone visual information, it also includes: when obtaining the error distribution estimation model corresponding to the frame residual distribution prediction model, randomly determining the error value of the frame residual distribution standard deviation according to the error distribution estimation model; determining the final value of the frame residual distribution standard deviation according to the frame residual distribution standard deviation value and the error value.

[0010] In a third aspect, an embodiment of the present invention provides a device for constructing a residual distribution prediction model for a drone visual information frame, comprising: a first acquisition module, for acquiring multiple coding parameter groups, and determining the actual value of the frame residual distribution standard deviation corresponding to each coding parameter group after encoding the drone visual information frame, wherein the coding parameter group includes a quantization step size, a search range, and a related number of frames; a first-level regression module, for performing a linear regression on the quantization step size and the actual value of the frame residual distribution standard deviation in each coding parameter group to obtain a first-level linear regression equation; calculating the first-level frame residual distribution standard deviation prediction value corresponding to each quantization step size according to the first-level linear regression equation, and determining the first-level frame residual distribution standard deviation prediction value and the frame residual distribution standard deviation. The first-level estimation error actual value between the actual values; the second-level regression module, which is used to perform linear regression on the search range in each coding parameter group and the first-level estimation error actual value to obtain a second-level linear regression equation; the first-level estimation error prediction value corresponding to each search range is calculated according to the second-level linear regression equation, and the second-level estimation error actual value between the first-level estimation error prediction value and the first-level estimation error actual value is determined; the third-level regression module, which is used to perform linear regression on the relevant frame number in each coding parameter group and the second-level estimation error actual value to obtain a third-level linear regression equation; the prediction model construction module, which is used to determine the frame residual distribution prediction model according to the first-level linear regression equation, the second-level linear regression equation and the third-level linear regression equation.

[0011] In a fourth aspect, an embodiment of the present invention provides a device for predicting the residual distribution of a frame of a UAV visual information, characterized in that it is applied to a frame residual distribution prediction model constructed by the method for constructing a frame residual distribution prediction model of a UAV visual information as described in any one of the first aspects, and the device includes: a second acquisition module, used to obtain a group of coding parameters to be predicted for the UAV visual information, the coding parameter group including a quantization step size, a search range and a related number of frames; a prediction module, used to input the group of coding parameters to be predicted into the frame residual distribution prediction model to obtain a standard deviation value of the frame residual distribution of the UAV visual information.

[0012] In a fifth aspect, an embodiment of the present invention provides an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to implement the steps of the method for constructing a residual distribution prediction model of a drone visual information frame as described in any one of the first aspect or the method for predicting the residual distribution of a drone visual information frame as described in the second aspect when executing the program stored in the memory.

[0013] In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for constructing a residual distribution prediction model of a drone visual information frame as described in any one of the first aspect or the steps of the method for predicting the residual distribution of a drone visual information frame as described in the second aspect.

[0014] The beneficial effects brought by the present invention are as follows:

[0015] It can be seen from the above scheme that the method for constructing a prediction model for residual distribution of unmanned aerial vehicle visual information frame and the prediction method provided by the embodiment of the present invention first perform a first-level linear regression between the quantization step size and the standard deviation of the residual distribution of the frame transform, and obtain the first-level estimation error based on the difference between the predicted value and the actual value of the first-level linear regression; then, a second-level linear regression is performed between the first-level estimation error and the search range to obtain the second-level estimation error; then, a third-level linear regression is performed between the relevant frame number and the second-level estimation error; finally, the results of these three linear regressions are linearly combined to complete the construction of the prediction model for residual distribution of visual information frame transform. This method reduces the computational complexity by performing three-level linear regression modeling on the coding parameters such as the quantization step size, the search range, and the relevant frame number, and linearly combines the results, which effectively improves the model construction speed and distribution prediction accuracy, and provides a solution for the bit rate control and quality assessment of unmanned aerial vehicle visual information coding. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a method for constructing a residual distribution prediction model of a UAV visual information frame provided by an embodiment of the present invention;

[0017] Figure 2 A schematic flow chart of another method for constructing a residual distribution prediction model of a UAV visual information frame provided by an embodiment of the present invention;

[0018] Figure 3 A schematic diagram of a flow chart of a method for predicting residual distribution of a UAV visual information frame provided by an embodiment of the present invention;

[0019] Figure 4 A flowchart of a method for constructing a residual distribution prediction model and prediction method for a UAV visual information frame provided by an embodiment of the present invention;

[0020] Figure 5 A schematic diagram of the structure of a device for constructing a residual distribution prediction model of a UAV visual information frame provided by an embodiment of the present invention;

[0021] Figure 6 A schematic diagram of the structure of a device for predicting residual distribution of a UAV visual information frame provided by an embodiment of the present invention;

[0022] Figure 7 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0024] In the related technology, an exponential model is usually used to model the standard deviation of the visual information frame residual distribution. The model construction time is long and the computational complexity is high, and it has poor applicability on UAV flight platforms with limited computing resources. In addition, the UAV is affected by airflow and environment during flight, which causes the collected visual information to often jitter and shake violently, which has a great impact on the prediction of the frame residual distribution. The frame residual standard deviation model based on the exponential model is prone to overfitting, which in turn reduces the model prediction accuracy.

[0025] In response to the above technical problems, the technical concept of the present invention is to adopt a three-level linear regression method to model the coding parameters such as quantization step size, search range, and related frame number respectively, and then linearly add the results of these three linear regressions to complete the construction of the visual information frame residual distribution prediction model.

[0026] Figure 1 A flowchart of a method for constructing a residual distribution prediction model of a UAV visual information frame provided by an embodiment of the present invention. Figure 1 As shown, the construction method includes:

[0027] Step S101, obtaining multiple coding parameter groups, and determining the actual value of the frame residual distribution standard deviation corresponding to each coding parameter group after encoding the drone visual information frame, wherein the coding parameter group includes a quantization step size, a search range, and a related frame number.

[0028] Specifically, assume that M (M>2) coding parameter groups are obtained, each coding parameter group includes a quantization step size Q, a search range λ, and a related frame number θ, which can be recorded as {Q, λ, θ} i , i = 1, 2, ... M; traverse all encoding parameter groups and encode, that is, use {Q, λ, θ} i Encode the UAV visual information frame, obtain and store the actual value of the standard deviation of the frame transformation residual distribution of the corresponding encoding parameter group σ i .

[0029] In some embodiments, obtaining multiple coding parameter groups in step S101 includes: obtaining a coding parameter set, the coding parameter set including a quantization step parameter set, a search range parameter set and a related frame number parameter set; randomly selecting a quantization step, a search range and a related frame number from the coding parameter set to form a coding parameter group; repeating the step of randomly selecting a quantization step, a search range and a related frame number from the coding parameter set to obtain multiple coding parameter groups.

[0030] Specifically, a coding parameter set is set, including a quantization step parameter set, a search range parameter set and a related frame number parameter set; a quantization step Q is randomly selected from the quantization step parameter set, a search range λ is randomly selected from the search range parameter set, and a related frame number θ is randomly selected from the related frame number parameter set to form a coding parameter group; the above steps are repeated to construct multiple coding parameter groups.

[0031] Step S102: Perform linear regression on the actual values ​​of the quantization step size and the frame residual distribution standard deviation in each coding parameter group to obtain a first-order linear regression equation.

[0032] Specifically, for all the quantization step sizes Q in the coding parameter groups and their corresponding actual values ​​of the standard deviation of the frame transform residual σ i Perform linear regression and obtain the first-order linear regression equation as shown below:

[0033]

[0034] in, It represents the predicted value of the standard deviation of the first-level frame transform residual corresponding to the quantization step size, Q represents the quantization step size, and a and b represent the parameters of the first-level linear regression equation.

[0035] It should be noted that experiments have shown that the quantization step size has the greatest impact on the distribution of visual information frame transformation residuals among the coding parameters, so in this embodiment, a first-order linear regression equation for the quantization step size is first constructed.

[0036] Step S103, calculating the predicted value of the first-level frame residual distribution standard deviation corresponding to each quantization step size according to the first-level linear regression equation, and determining the actual value of the first-level estimation error between the predicted value of the first-level frame residual distribution standard deviation and the actual value of the frame residual distribution standard deviation.

[0037] Specifically, for each encoding parameter group corresponding to the quantization step size Q i , put it into formula (1), and get the corresponding standard deviation prediction value of the primary frame residual distribution and the actual value of the standard deviation of the frame transform residual distribution σ i By comparison, we can get the actual value of the first-level estimation error e Q,i; After traversing the quantization step lengths in all coding parameter groups, the set of actual values ​​of the first-level estimation error {e Q}.

[0038] Step S104: perform linear regression on the search range in each coding parameter group and the actual value of the first-level estimation error to obtain a second-level linear regression equation.

[0039] Specifically, a linear regression is performed on the search ranges in all coding parameter groups and their corresponding actual values ​​of the first-level estimation errors to construct a second-level linear regression equation, as shown below:

[0040]

[0041] in, represents the predicted value of the first-level estimation error, λ represents the search range, and c and d represent the parameters of the second-level linear regression equation.

[0042] Step S105, calculating the first-level estimation error prediction value corresponding to each search range according to the second-level linear regression equation, and determining the second-level estimation error actual value between the first-level estimation error prediction value and the first-level estimation error actual value.

[0043] Specifically, for each encoding parameter group, the search range λ i , put it into formula (2), and get the corresponding first-level estimation error prediction value and the actual value of the first-level estimation error e Q,i By comparison, we can get the actual value of the secondary estimation error e Q,λ,i After traversing the search range in all coding parameter groups, the actual value set of the secondary estimation error {e Q,λ,i}.

[0044] Step S106: Perform linear regression on the relevant frame numbers in each encoding parameter group and the actual value of the secondary estimation error to obtain a third-level linear regression equation.

[0045] Specifically, a linear regression is performed on the relevant frame numbers in all coding parameter groups and the corresponding actual values ​​of the secondary estimation errors to construct a three-level linear regression equation, as shown below:

[0046]

[0047] in, represents the predicted value of the secondary estimation error, θ represents the number of related frames, and g and h represent the parameters of the third-level linear regression equation.

[0048] Step S107, determining a frame residual distribution prediction model according to the first-level linear regression equation, the second-level linear regression equation and the third-level linear regression equation.

[0049] Specifically, formulas (1), (2), and (3) are linearly superimposed to complete the construction of the frame residual distribution prediction model (i.e., the frame residual distribution standard deviation model), as shown below:

[0050]

[0051] in, Represents the predicted value of the standard deviation of the frame transform residual distribution.

[0052] The method for constructing a residual distribution prediction model of a drone visual information frame provided by an embodiment of the present invention first performs a first-level linear regression between a quantization step size and a standard deviation of a residual distribution of a frame transformation, obtains a first-level estimation error based on the difference between the predicted value of the first-level linear regression and the actual value, then performs a second-level linear regression between a search range and the first-level estimation error, compares the predicted value of the second-level linear regression with the first-level estimation error, obtains a second-level estimation error; then performs a third-level linear regression between the relevant frame number and the second-level estimation error; finally, linearly adds the three linear regressions to complete the construction of a prediction model for the residual distribution of a visual information frame transformation; that is, this embodiment solves the problems of high computational complexity and long computational time of a traditional residual distribution prediction model for a visual information frame transformation, as well as the problems of overfitting and insufficient precision in the process of being applied to a drone flight platform by adopting a third-level linear regression method.

[0053] Based on the above embodiments, Figure 2 A flowchart of another method for constructing a residual distribution prediction model of a UAV visual information frame provided by an embodiment of the present invention. Figure 2 As shown, the construction method includes:

[0054] Step S201, obtaining multiple coding parameter groups, and determining the actual value of the frame residual distribution standard deviation corresponding to each coding parameter group after encoding the drone visual information frame, wherein the coding parameter group includes a quantization step size, a search range, and a related frame number.

[0055] Step S202: Perform linear regression on the actual values ​​of the quantization step size and the frame residual distribution standard deviation in each coding parameter group to obtain a first-order linear regression equation.

[0056] Step S203, calculating the predicted value of the first-level frame residual distribution standard deviation corresponding to each quantization step size according to the first-level linear regression equation, and determining the actual value of the first-level estimation error between the predicted value of the first-level frame residual distribution standard deviation and the actual value of the frame residual distribution standard deviation.

[0057] Step S204: perform linear regression on the search range in each coding parameter group and the actual value of the first-level estimation error to obtain a second-level linear regression equation.

[0058] Step S205, calculating the first-level estimation error prediction value corresponding to each search range according to the second-level linear regression equation, and determining the second-level estimation error actual value between the first-level estimation error prediction value and the first-level estimation error actual value.

[0059] Step S206: Perform linear regression on the relevant frame numbers in each encoding parameter group and the actual value of the secondary estimation error to obtain a third-level linear regression equation.

[0060] Step S207, determining a frame residual distribution prediction model according to the first-level linear regression equation, the second-level linear regression equation and the third-level linear regression equation.

[0061] Step S208: Calculate the predicted value of the frame residual distribution standard deviation corresponding to each coding parameter group according to the frame residual distribution prediction model.

[0062] Step S209: construct an error distribution estimation model corresponding to the frame residual distribution prediction model according to the frame residual distribution standard deviation prediction value and the frame residual distribution standard deviation actual value.

[0063] It should be noted that the implementation of steps S201-S207 in this embodiment is similar to that of steps S101-S107 in the aforementioned embodiment, and will not be repeated here.

[0064] The difference from the previous embodiment is that, in order to further improve the prediction accuracy of the frame residual distribution of unmanned visual information, in this embodiment, the frame residual distribution standard deviation prediction value corresponding to each coding parameter group is calculated according to the frame residual distribution prediction model; and the error distribution estimation model corresponding to the frame residual distribution prediction model is constructed according to the frame residual distribution standard deviation prediction value and the actual value of the frame residual distribution standard deviation.

[0065] Specifically, for each encoding parameter set {Q, λ, θ} i , put it into formula (4), and get the corresponding frame transform residual distribution standard deviation prediction value according to The actual value of the standard deviation of the residual distribution of the frame transform σ i Construct an error distribution estimation model corresponding to the standard deviation model of the frame residual distribution.

[0066] In some embodiments, the step S209 includes: determining the final error value between the predicted value of the frame residual distribution standard deviation and the actual value of the frame residual distribution standard deviation; determining an initial mixed Gaussian model; using an expectation maximization algorithm to iteratively optimize the initial mixed Gaussian model according to the final error value to obtain a final mixed Gaussian model, wherein the final mixed Gaussian model is the error distribution estimation model.

[0067] Specifically, the three-level linear regression results in the error of the frame residual distribution prediction model being a mixed Gaussian distribution. With σ i By comparison, the final error value e is calculated. σ ; Then, determine the initial mixed Gaussian model, that is, set the initial value μ of the mixed Gaussian model parameters k and σ k , μ k represents the initial value of the mean of the mixed Gaussian model, σ k represents the initial value of the variance of the mixed Gaussian model, and sets the maximum number of iterations N; then, the expectation maximization algorithm (EM algorithm) is used to analyze the error of the standard deviation model of the frame residual distribution, and the converged final mixed Gaussian model (μ c , σ c ).

[0068] Error estimation is performed according to the error distribution estimation model, and the error estimation result is combined with the three-level linear regression model to realize the residual distribution estimation of visual information frame transformation.

[0069] On the basis of the foregoing embodiments, the frame residual distribution standard deviation prediction value corresponding to each coding parameter group is calculated according to the frame residual distribution prediction model; the error distribution estimation model corresponding to the frame residual distribution prediction model is constructed according to the frame residual distribution standard deviation prediction value and the actual value of the frame residual distribution standard deviation; that is, for the linear regression model estimation error, a machine learning method is used to further improve its accuracy.

[0070] Figure 3 A flow chart of a method for predicting residual distribution of a UAV visual information frame provided by an embodiment of the present invention is applied to the frame residual distribution prediction model constructed by the method for constructing a UAV visual information frame residual distribution prediction model in the aforementioned embodiment, such as Figure 3 As shown, the prediction method includes:

[0071] Step S301: Obtain a group of coding parameters to be predicted for drone visual information, where the group of coding parameters includes a quantization step size, a search range, and a related number of frames.

[0072] Step S302: input the to-be-predicted coding parameter group into the frame residual distribution prediction model to obtain a frame residual distribution standard deviation value of the UAV visual information.

[0073] Specifically, for the coding parameter group used for the drone visual information that needs to predict the frame residual distribution standard deviation, it is substituted into the frame residual distribution standard deviation model corresponding to formula (4) to obtain the frame residual distribution standard deviation prediction value. The frame residual distribution of the drone visual information can be analyzed and predicted based on the frame residual distribution standard deviation prediction value.

[0074] Continue to refer Figure 3 As shown, in some embodiments, after step S302, the following steps are also included:

[0075] Step S303: when the error distribution estimation model corresponding to the frame residual distribution prediction model is obtained, the error value of the frame residual distribution standard deviation is randomly determined according to the error distribution estimation model.

[0076] Step S304: determining a final value of the frame residual distribution standard deviation according to the frame residual distribution standard deviation value and the error value.

[0077] Specifically, the error estimation distribution model is the converged final mixed Gaussian model (μ c , σ c ), from which the error e is randomly selected c , e c Substitute into formula (5) and calculate the final value of the standard deviation of the residual of the visual information frame transformation:

[0078]

[0079] On the basis of the foregoing embodiments, by obtaining a group of coding parameters to be predicted for the UAV visual information, the coding parameter group includes a quantization step size, a search range and a related number of frames; the group of coding parameters to be predicted is input into the frame residual distribution prediction model to obtain the frame residual distribution standard deviation value of the UAV visual information, and the three-level linear regression model adopted improves the calculation efficiency and reduces the calculation complexity; and by randomly determining the error value of the frame residual distribution standard deviation according to the error distribution estimation model when obtaining the error distribution estimation model corresponding to the frame residual distribution prediction model; determining the final value of the frame residual distribution standard deviation according to the frame residual distribution standard deviation value and the error value, and by combining the error estimation result with the three-level linear regression model, the accuracy of the visual information frame transformation residual distribution prediction is further improved.

[0080] Figure 4 The present invention provides a flowchart of a method for constructing a residual distribution prediction model and a prediction method for a UAV visual information frame. Figure 4 The embodiments of the present invention are described in detail:

[0081] (1) Set the encoding parameter set: including the quantization step parameter set, the search range parameter set and the related frame number parameter set.

[0082] (2) Randomly select coding parameters to construct multiple coding parameter groups: randomly select a quantization step from the quantization step parameter set, randomly select a search range from the search range parameter set, and randomly select a related frame number from the related frame number parameter set to form a coding parameter group. Repeat the above process to construct multiple coding parameter groups.

[0083] (3) Traverse all coding parameter groups for encoding and obtain the actual value of the standard deviation of the frame transform residual distribution: Use each coding parameter group to encode the drone information frame, determine the difference between the drone information frame before and after encoding, and further obtain the actual value of the standard deviation of the frame transform residual distribution.

[0084] (4) First-level linear regression: Perform linear regression on the quantization step size and the actual value of the frame residual distribution standard deviation in all coding parameter groups to obtain a first-level linear regression equation; substitute the quantization step size in each coding parameter group into the first-level linear regression equation to obtain the corresponding first-level frame residual distribution standard deviation prediction value, and determine the actual value of the first-level estimation error between the first-level frame residual distribution standard deviation prediction value and the actual value of the frame residual distribution standard deviation.

[0085] (5) Secondary linear regression: Perform linear regression on the search ranges in all coding parameter groups and the actual values ​​of the first-level estimation errors to obtain a secondary linear regression equation; substitute the search range in each coding parameter group into the secondary linear regression equation to obtain the corresponding first-level estimation error prediction value, and determine the actual value of the second-level estimation error between the first-level estimation error prediction value and the actual value of the first-level estimation error.

[0086] (6) Three-level linear regression: Linear regression is performed on the relevant frame numbers and the actual values ​​of the secondary estimation errors in all coding parameter groups to obtain a three-level linear regression equation.

[0087] (7) Superposition to complete model construction: linearly add the first-level linear regression equation, the second-level linear regression equation, and the third-level linear regression equation to obtain the standard deviation model of the frame residual distribution.

[0088] (8) Initialize the error Gaussian distribution and EM iterative algorithm parameters: set the initial values ​​of the mean and variance of the mixed Gaussian model, and the maximum number of iterations of the EM algorithm.

[0089] (9) Iterative solution to generate error Gaussian distribution: each coding parameter group is introduced into the frame residual distribution standard deviation model to obtain the predicted value of the frame residual distribution standard deviation, and compared with the actual value of the frame residual distribution standard deviation to determine the error value; the EM algorithm is used to iteratively solve the error value to obtain a convergent error Gaussian distribution.

[0090] (10) Randomly generate error distribution based on error Gaussian distribution: randomly select errors from the converged error Gaussian distribution.

[0091] (11) Combining model construction and error estimation to achieve the prediction of the standard deviation of the visual information frame transformation residual: input the coding parameters to be predicted into the constructed frame residual distribution prediction model to obtain the frame residual distribution standard deviation value, and combine the frame residual distribution standard deviation value with the error in (10) to obtain the final value of the frame transformation residual standard deviation.

[0092] In summary, this embodiment adopts a three-level linear regression method to model coding parameters such as quantization step size, search range, and related frame number, that is, first, a first-level linear regression is performed between the quantization step size and the standard deviation of the frame transform residual distribution, and a first-level estimation error is obtained based on the difference between the predicted value and the actual value of the first-level linear regression; then, a second-level linear regression is performed between the first-level estimation error and the search range to obtain a second-level estimation error; then, a third-level linear regression is performed between the related frame number and the second-level estimation error; finally, the results of these three linear regressions are linearly combined to complete the construction of the visual information frame transform residual distribution prediction model; in addition, the model error caused by the three-level linear regression is a mixed Gaussian distribution, and the expectation maximization algorithm is used to iteratively solve the model error, and finally an error distribution estimation model is obtained; error estimation is performed according to the error distribution estimation model, and the error estimation result is combined with the three-level linear regression model to realize the visual information frame transform residual distribution estimation, which reduces the computational complexity, improves the modeling speed and distribution prediction accuracy, and is suitable for UAV flight platforms.

[0093] Figure 5 A schematic diagram of a structure of a device for constructing a residual distribution prediction model of a UAV visual information frame provided by an embodiment of the present invention, such as Figure 5 As shown, the construction device includes:

[0094] The first acquisition module 501 is used to acquire multiple coding parameter groups and determine the actual value of the frame residual distribution standard deviation corresponding to each coding parameter group after encoding the drone visual information frame, wherein the coding parameter group includes a quantization step size, a search range, and a related frame number;

[0095] A first-level regression module 502 is used to perform linear regression on the actual values ​​of the quantization step size and the frame residual distribution standard deviation in each coding parameter group to obtain a first-level linear regression equation;

[0096] Calculating a predicted value of a first-level frame residual distribution standard deviation corresponding to each quantization step size according to the first-level linear regression equation, and determining an actual value of a first-level estimation error between the predicted value of the first-level frame residual distribution standard deviation and an actual value of the frame residual distribution standard deviation;

[0097] A secondary regression module 503 is used to perform a linear regression on the search range in each coding parameter group and the actual value of the primary estimation error to obtain a secondary linear regression equation;

[0098] Calculating a first-order estimation error prediction value corresponding to each search range according to the second-order linear regression equation, and determining a second-order estimation error actual value between the first-order estimation error prediction value and the first-order estimation error actual value;

[0099] A three-level regression module 504 is used to perform linear regression on the relevant frame number and the actual value of the secondary estimation error in each encoding parameter group to obtain a three-level linear regression equation;

[0100] The prediction model building module 505 is used to determine the frame residual distribution prediction model according to the first-level linear regression equation, the second-level linear regression equation and the third-level linear regression equation.

[0101] In some embodiments, the apparatus further comprises an error model construction module 506, which is used to calculate a predicted value of a frame residual distribution standard deviation corresponding to each coding parameter group according to the frame residual distribution prediction model;

[0102] An error distribution estimation model corresponding to the frame residual distribution prediction model is constructed according to the frame residual distribution standard deviation prediction value and the frame residual distribution standard deviation actual value.

[0103] In some embodiments, the error model building module 506 is specifically used to:

[0104] Determining a final error value between the predicted value of the frame residual distribution standard deviation and the actual value of the frame residual distribution standard deviation;

[0105] Determine the initial Gaussian mixture model;

[0106] The initial mixed Gaussian model is iteratively optimized according to the final error value using an expectation maximization algorithm to obtain a final mixed Gaussian model, and the final mixed Gaussian model is used as the error distribution estimation model.

[0107] In some embodiments, the first acquisition module 501 is specifically used to:

[0108] Acquire a coding parameter set, wherein the coding parameter set includes a quantization step parameter set, a search range parameter set, and a related frame number parameter set;

[0109] Randomly selecting a quantization step size, a search range and a related frame number from the encoding parameter set to form an encoding parameter group;

[0110] The step of randomly selecting a quantization step size, a search range and a related frame number from the encoding parameter set is repeatedly performed to obtain multiple encoding parameter groups.

[0111] Technicians in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process and corresponding beneficial effects of the above-described drone visual information frame residual distribution prediction model construction device can refer to the corresponding process in the aforementioned drone visual information frame residual distribution prediction model construction method example, and will not be repeated here.

[0112] Figure 6 The schematic diagram of the structure of a UAV visual information frame residual distribution prediction device provided by an embodiment of the present invention is applied to the frame residual distribution prediction model constructed by the UAV visual information frame residual distribution prediction model construction method in the aforementioned embodiment, such as Figure 6 As shown, the construction device includes:

[0113] The second acquisition module 601 is used to obtain a group of coding parameters to be predicted for the UAV visual information, where the group of coding parameters includes a quantization step size, a search range, and a related frame number;

[0114] The prediction module 602 is used to input the encoding parameter group to be predicted into the frame residual distribution prediction model to obtain the standard deviation value of the frame residual distribution of the drone visual information.

[0115] In some embodiments, the prediction module 602 is further configured to:

[0116] When an error distribution estimation model corresponding to the frame residual distribution prediction model is obtained, an error value of a frame residual distribution standard deviation is randomly determined according to the error distribution estimation model;

[0117] A final value of the frame residual distribution standard deviation is determined according to the frame residual distribution standard deviation value and the error value.

[0118] Technicians in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process and corresponding beneficial effects of the drone visual information frame residual distribution prediction device described above can refer to the corresponding process in the aforementioned drone visual information frame residual distribution prediction method example, and will not be repeated here.

[0119] like Figure 7 As shown, an embodiment of the present invention provides an electronic device, including a processor 701, a communication interface 702, a memory 703 and a communication bus 704, wherein the processor 701, the communication interface 702, and the memory 703 communicate with each other through the communication bus 704.

[0120] Memory 703, used for storing computer programs;

[0121] In one embodiment of the present invention, the processor 701 is used to execute the program stored in the memory 703 to implement the steps of the method for constructing a drone visual information frame residual distribution prediction model or the method for predicting the residual distribution of a drone visual information frame provided in any of the aforementioned method embodiments.

[0122] The implementation principle and technical effect of the electronic device provided by the embodiment of the present invention are similar to those of the above embodiment and will not be described in detail here.

[0123] The above-mentioned memory 703 can be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk or a ROM. The memory 703 has a storage space for program codes for executing any method steps in the above-mentioned method. For example, the storage space for program codes may include various program codes for implementing various steps in the above method respectively. These program codes can be read from or written into one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards or floppy disks. Such computer program products are usually portable or fixed storage units. The storage unit may have a storage segment or storage space arranged similarly to the memory 703 in the above-mentioned electronic device. The program code can be compressed, for example, in an appropriate form. Generally, the storage unit includes a program for executing the method steps according to an embodiment of the present invention, that is, a code that can be read by a processor such as 701, which, when run by an electronic device, causes the electronic device to execute various steps in the method described above.

[0124] An embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for constructing a residual distribution prediction model of a drone visual information frame or the method for predicting residual distribution of a drone visual information frame are implemented as described above.

[0125] The computer-readable storage medium may be included in the device / apparatus described in the above embodiment; or it may exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0126] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus, or device.

[0127] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0128] The above are preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for constructing a residual distribution prediction model for a UAV visual information frame, characterized in that: include: Acquire multiple encoding parameter groups, and determine the actual value of the frame residual distribution standard deviation corresponding to each encoding parameter group after encoding the drone visual information frame, wherein the encoding parameter group includes a quantization step size, a search range, and a related number of frames; Performing linear regression on the actual values ​​of the quantization step size and the standard deviation of the frame residual distribution in each coding parameter group to obtain a first-order linear regression equation; Calculating a predicted value of a first-level frame residual distribution standard deviation corresponding to each quantization step size according to the first-level linear regression equation, and determining an actual value of a first-level estimation error between the predicted value of the first-level frame residual distribution standard deviation and an actual value of the frame residual distribution standard deviation; Performing linear regression on the search range in each encoding parameter group and the actual value of the first-level estimation error to obtain a second-level linear regression equation; Calculating a first-order estimation error prediction value corresponding to each search range according to the second-order linear regression equation, and determining a second-order estimation error actual value between the first-order estimation error prediction value and the first-order estimation error actual value; Performing linear regression on the relevant frame number and the actual value of the secondary estimation error in each encoding parameter group to obtain a third-level linear regression equation; A frame residual distribution prediction model is determined according to the first-level linear regression equation, the second-level linear regression equation and the third-level linear regression equation.

2. The method according to claim 1, characterized in that: After determining the frame residual distribution prediction model, the method further includes: Calculate the frame residual distribution standard deviation prediction value corresponding to each coding parameter group according to the frame residual distribution prediction model; An error distribution estimation model corresponding to the frame residual distribution prediction model is constructed according to the frame residual distribution standard deviation prediction value and the frame residual distribution standard deviation actual value.

3. The method according to claim 2, characterized in that The step of constructing an error distribution estimation model corresponding to a frame residual distribution prediction model according to the frame residual distribution standard deviation prediction value and the frame residual distribution standard deviation actual value includes: Determining a final error value between the predicted value of the frame residual distribution standard deviation and the actual value of the frame residual distribution standard deviation; Determine the initial Gaussian mixture model; The initial mixed Gaussian model is iteratively optimized according to the final error value using an expectation maximization algorithm to obtain a final mixed Gaussian model, and the final mixed Gaussian model is used as the error distribution estimation model.

4. The method according to any one of claims 1 to 3, characterized in that: The obtaining of multiple encoding parameter groups includes: Acquire a coding parameter set, wherein the coding parameter set includes a quantization step parameter set, a search range parameter set, and a related frame number parameter set; Randomly selecting a quantization step size, a search range and a related frame number from the encoding parameter set to form an encoding parameter group; The step of randomly selecting a quantization step size, a search range and a related frame number from the encoding parameter set is repeatedly performed to obtain multiple encoding parameter groups.

5. A method for predicting residual distribution of UAV visual information frames, characterized in that: The frame residual distribution prediction model constructed by the method for constructing a UAV visual information frame residual distribution prediction model according to any one of claims 1 to 4 is applied, and the method comprises: Obtaining a to-be-predicted coding parameter group of the UAV visual information, wherein the coding parameter group includes a quantization step size, a search range, and a related frame number; The encoding parameter group to be predicted is input into the frame residual distribution prediction model to obtain the frame residual distribution standard deviation value of the UAV visual information.

6. The method according to claim 5, characterized in that After obtaining the standard deviation value of the frame residual distribution of the UAV visual information, the method further includes: When an error distribution estimation model corresponding to the frame residual distribution prediction model is obtained, an error value of a frame residual distribution standard deviation is randomly determined according to the error distribution estimation model; A final value of the frame residual distribution standard deviation is determined according to the frame residual distribution standard deviation value and the error value.

7. A device for constructing a residual distribution prediction model for a drone visual information frame, characterized in that: include: A first acquisition module is used to acquire multiple coding parameter groups and determine an actual value of a frame residual distribution standard deviation corresponding to each coding parameter group after encoding the drone visual information frame, wherein the coding parameter group includes a quantization step size, a search range, and a related frame number; A first-level regression module is used to perform linear regression on the actual values ​​of the quantization step size and the frame residual distribution standard deviation in each encoding parameter group to obtain a first-level linear regression equation; Calculating a predicted value of a first-level frame residual distribution standard deviation corresponding to each quantization step size according to the first-level linear regression equation, and determining an actual value of a first-level estimation error between the predicted value of the first-level frame residual distribution standard deviation and an actual value of the frame residual distribution standard deviation; A secondary regression module is used to perform linear regression on the search range in each encoding parameter group and the actual value of the primary estimation error to obtain a secondary linear regression equation; Calculating a first-order estimation error prediction value corresponding to each search range according to the second-order linear regression equation, and determining a second-order estimation error actual value between the first-order estimation error prediction value and the first-order estimation error actual value; A three-level regression module is used to perform linear regression on the relevant frame number and the actual value of the secondary estimation error in each encoding parameter group to obtain a three-level linear regression equation; The prediction model building module is used to determine the frame residual distribution prediction model according to the first-level linear regression equation, the second-level linear regression equation and the third-level linear regression equation.

8. A device for predicting residual distribution of unmanned aerial vehicle visual information frames, characterized in that: The device is applied to the frame residual distribution prediction model constructed by the method for constructing the UAV visual information frame residual distribution prediction model according to any one of claims 1 to 4, and comprises: A second acquisition module is used to obtain a to-be-predicted coding parameter group of the UAV visual information, wherein the coding parameter group includes a quantization step size, a search range, and a related frame number; The prediction module is used to input the encoding parameter group to be predicted into the frame residual distribution prediction model to obtain the frame residual distribution standard deviation value of the UAV visual information.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the steps of the method for constructing a residual distribution prediction model of a UAV visual information frame as described in any one of claims 1 to 4 or the method for predicting residual distribution of a UAV visual information frame as described in claim 5 or 6 when executing a program stored in a memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a residual distribution prediction model of a UAV visual information frame as described in any one of claims 1 to 4 or the method for predicting residual distribution of a UAV visual information frame as described in claim 5 or 6 are implemented.

Images