Vehicle lamp imaging control method and system based on time domain memory feedback and storage medium

Abstract

The application discloses a time-domain memory feedback-based vehicle lamp imaging control method and system and a storage medium, and the method comprises the following steps: constructing a cascaded time-domain memory feedback network model according to projection data of an intelligent vehicle lamp; preliminarily optimizing the cascaded time-domain memory feedback network model by using stability determination; secondarily training the cascaded time-domain memory feedback network model by using a margin learning algorithm to obtain an optimal cascaded time-domain memory feedback network model; and adjusting the intelligent vehicle lamp in real time according to the optimal cascaded time-domain memory feedback network model to realize vehicle lamp imaging control. The application solves the problem that current vehicle lamp imaging control technology is difficult to be integrated with other intelligent systems, thereby leading to poor collaboration.

Description

Technical Field

[0001] This invention belongs to the field of automotive lighting technology, and in particular relates to an automotive lighting imaging control method, system, and storage medium based on time-domain memory feedback. Background Technology

[0002] With the increasing intelligence and automation of automobiles, intelligent headlights, as key components related to driving safety and experience, are becoming increasingly important. Traditional headlights offer limited illumination and cannot adapt to complex road conditions. Although advanced technologies such as adaptive high beam switching and cornering headlights have been implemented, there is still significant room for improvement in the precision of imaging control and the level of intelligence.

[0003] With the development of autonomous driving technology, intelligent headlights need to work in conjunction with vehicle sensor systems and autonomous driving algorithms to provide environmental information for autonomous driving. However, current headlight imaging control technology is difficult to deeply integrate with other intelligent systems and lacks more accurate, intelligent and effective collaborative intelligent headlight imaging control methods, which limits the development of autonomous driving technology.

[0004] To address the problem of poor coordination caused by the difficulty in integrating current vehicle lighting imaging control technology with other intelligent systems, a vehicle lighting imaging control method, system, and storage medium based on time-domain memory feedback are proposed. Summary of the Invention

[0005] This invention proposes a vehicle headlight imaging control device, method, system, and storage medium based on time-domain memory feedback, in order to at least solve the problem of poor coordination caused by the difficulty of integrating current vehicle headlight imaging control technology with other intelligent systems.

[0006] According to an embodiment of the present invention, a vehicle headlight imaging control method based on time-domain memory feedback is provided, comprising the steps of:

[0007] A cascaded time-domain memory feedback network model is constructed based on the projection data of the intelligent vehicle lights;

[0008] The cascaded time-domain memory feedback network model was initially optimized using stability criteria.

[0009] The optimal cascaded time-domain memory feedback network model is obtained by second-time training of the cascaded time-domain memory feedback network model using the residual learning algorithm.

[0010] The intelligent vehicle lights are adjusted in real time based on the optimal cascaded time-domain memory feedback network model to achieve vehicle light imaging control.

[0011] Optionally, the projection data includes any one or more combinations of the following: the projected area of ​​the vehicle headlights, the brightness and contrast measured at the front edge of the projected vehicle headlights, the brightness and contrast measured at the leftmost edge of the projected vehicle headlights, the brightness and contrast measured at the rightmost edge of the projected vehicle headlights, and the brightness and contrast measured at the center of the projected vehicle headlights.

[0012] Optionally, the step of constructing a cascaded time-domain memory feedback network model based on the projection data of the smart vehicle lights includes the following steps:

[0013] Construct a basic single-layer temporal memory feedback network model;

[0014] Construct a cascaded memory storage area based on a single-layer time-domain memory feedback network;

[0015] Weighted coupling is applied to the cascaded structure;

[0016] The residual learning algorithm is used to calculate the weight matrix of each feedback output layer;

[0017] By integrating the feedback output layers, a cascaded time-domain memory feedback network model is obtained.

[0018] Optionally, the step of constructing a cascaded memory storage area based on a single-layer time-domain memory feedback network includes the following steps:

[0019] Construct a memory storage area for a single-layer temporal memory feedback network to process projected area and global features;

[0020] Construct a memory storage area for a single-layer temporal memory feedback network to process edge brightness and contrast;

[0021] Construct a memory storage area for a single-layer temporal memory feedback network to process features in the central region;

[0022] The output of the previous single-layer time-domain memory feedback network is used as the input of the next single-layer time-domain memory feedback network to form a cascaded structure, resulting in a cascaded memory storage area.

[0023] Optionally, the weighted coupling of the cascaded structure includes the following steps:

[0024] Construct the coupling weight matrix of each sub-memory storage area in the cascaded memory storage area;

[0025] The real-time state of the lower-level sub-region is passed to the higher-level sub-region to obtain the cross-level connection weight matrix;

[0026] The state vector of each sub-region is calculated based on the internal connection weight matrix and the cross-level connection weight matrix, thereby achieving weight coupling.

[0027] Optionally, the preliminary optimization of the cascaded time-domain memory feedback network model using stability determination includes the following steps:

[0028] Obtain the internal recursive weight matrix of each sub-memory storage area;

[0029] Calculate the spectral radius of the internal recursive weight matrix;

[0030] When the spectral radius of the internal recursive weight matrix is ​​less than a preset threshold, the model is determined to be stable, and a preliminary optimized cascaded time-domain memory feedback network model is obtained; otherwise, the model is determined not to be stable, and the basic single-layer time-domain memory feedback network model is reconstructed for model iteration.

[0031] Optionally, the step of using the residual learning algorithm to train the cascaded time-domain memory feedback network model a second time to obtain the optimal cascaded time-domain memory feedback network model includes the following steps:

[0032] The historical input data is sequentially input into the cascaded temporal memory feedback network model for forward propagation, and the memory storage area state of the temporal memory feedback network layer is updated.

[0033] Calculate the output of the cascaded time-domain memory feedback network based on the state of the memory storage area;

[0034] Construct a loss function to measure the difference between the network's predicted output and the actual expected output;

[0035] The output of the residual connection is calculated using the residual learning algorithm;

[0036] The residual learning algorithm is used to calculate the gradient of the loss function with respect to the connection weight moments from the memory storage area to the feedback output layer;

[0037] Update the weight matrix using the residual gradient iteration method;

[0038] The optimal cascaded time-domain memory feedback network model is obtained when the loss function converges.

[0039] Optionally, the real-time adjustment of the intelligent vehicle lights based on the optimal cascaded time-domain memory feedback network model includes the following steps:

[0040] Input the real-time projection data of the vehicle lights into the optimal cascaded time-domain memory feedback network model;

[0041] The corrected voltage and offset angle of the vehicle headlights are output based on the cascaded time-domain memory feedback network model.

[0042] The headlights are adjusted in real time based on the predicted correction voltage and offset angle.

[0043] According to another embodiment of the present invention, a computer-readable storage medium is provided that stores a computer program for electronic data interchange, wherein the computer program causes a computer to perform the above-described method.

[0044] According to another embodiment of the present invention, a vehicle lighting imaging control system based on time-domain memory feedback is provided, comprising:

[0045] Headlight imaging module;

[0046] Environmental perception module;

[0047] Imaging control module;

[0048] Memory;

[0049] And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by the imaging control module, the programs causing the computer to perform the methods described above.

[0050] The advantages of the vehicle headlight imaging control method, system, and storage medium based on time-domain memory feedback of the present invention are:

[0051] (1) Construct a cascaded time-domain memory feedback network model based on the projection data of intelligent vehicle lights. Compared with the traditional vehicle light imaging technology solution, it can effectively improve the integration of vehicle light imaging control technology with other intelligent systems, and facilitate more accurate, intelligent and effective collaborative intelligent vehicle light imaging control.

[0052] (2) The cascaded time-domain memory feedback network model is initially optimized by stability judgment and the cascaded time-domain memory feedback network model is trained a second time by residual learning algorithm to obtain the optimal cascaded time-domain memory feedback network model. Compared with the traditional vehicle headlight projection imaging technology, it can effectively improve the control accuracy and effectiveness of the control model.

[0053] (3) The intelligent vehicle lights are adjusted in real time according to the optimal cascaded time-domain memory feedback network model to realize vehicle light imaging control. Compared with the traditional vehicle light imaging technology, the vehicle light imaging can be effectively controlled in real time by correcting the voltage and offset angle, which improves the accuracy and intelligence of vehicle light imaging. Attached Figure Description

[0054] Figure 1 This is a flowchart of the vehicle headlight imaging control method based on time-domain memory feedback according to an embodiment of the present invention;

[0055] Figure 2 This is a flowchart of step S01 of the intelligent imaging optimization method for vehicle lights according to an embodiment of the present invention;

[0056] Figure 3 This is a flowchart of sub-step S012 of the intelligent imaging optimization method for vehicle lights according to an embodiment of the present invention;

[0057] Figure 4 This is a flowchart of sub-step S013 of the intelligent imaging optimization method for vehicle lights according to an embodiment of the present invention;

[0058] Figure 5 This is a flowchart of step S02 of the intelligent imaging optimization method for vehicle lights according to an embodiment of the present invention;

[0059] Figure 6 This is a flowchart of step S03 of the intelligent imaging optimization method for vehicle lights according to an embodiment of the present invention;

[0060] Figure 7 This is a flowchart of step S04 of the intelligent imaging optimization method for vehicle lights according to an embodiment of the present invention;

[0061] Figure 8 This is a schematic diagram of the structure of the vehicle lighting imaging control system based on time-domain memory feedback according to an embodiment of the present invention. Detailed Implementation

[0062] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0063] The vehicle lights described in this embodiment include headlights and taillights for motor vehicles or non-motor vehicles.

[0064] According to an embodiment of the present invention, a vehicle headlight imaging control method based on time-domain memory feedback is provided, the flowchart of which is shown below. Figure 1 As shown, it includes:

[0065] Step S01: Construct a cascaded time-domain memory feedback network model based on the projection data of the intelligent vehicle lights;

[0066] Step S02: Use stability determination to preliminarily optimize the cascaded time-domain memory feedback network model;

[0067] Step S03: Use the residual learning algorithm to train the cascaded time-domain memory feedback network model a second time to obtain the optimal cascaded time-domain memory feedback network model;

[0068] Step S04: Adjust the intelligent vehicle lights in real time according to the optimal cascaded time-domain memory feedback network model to achieve vehicle light imaging control.

[0069] In one exemplary embodiment, the projection data includes any one or more combinations of the following: the projected area of ​​the vehicle headlights, the brightness and contrast measured at the front edge of the projected vehicle headlights, the brightness and contrast measured at the leftmost edge of the projected vehicle headlights, the brightness and contrast measured at the rightmost edge of the projected vehicle headlights, and the brightness and contrast measured at the center of the projected vehicle headlights. In this embodiment, the vehicle headlight data is sampled, including the projected area S of the vehicle headlights, the brightness L1 and contrast B1 measured at the front edge of the projected vehicle headlights, the brightness L2 and contrast B2 measured at the leftmost edge of the projected vehicle headlights, the brightness L3 and contrast B3 measured at the rightmost edge of the projected vehicle headlights, and the brightness L4 and contrast B4 measured at the center of the projected vehicle headlights.

[0070] In an exemplary embodiment, step S01, initializing the LED array parameters according to the requirements of the vehicle headlight projection system, is illustrated in the flowchart below. Figure 2 As shown, the steps include:

[0071] Step S011: Construct a basic single-layer temporal memory feedback network model;

[0072] Step S012: Construct a cascaded memory storage area based on a single-layer time-domain memory feedback network;

[0073] Step S013: Weight coupling is applied to the cascaded structure;

[0074] Step S014: Calculate the weight matrix of each feedback output layer using the residual learning algorithm;

[0075] Step S015: Integrate the feedback output layers to obtain the cascaded time-domain memory feedback network model.

[0076] In this embodiment, a basic temporal memory feedback network model is constructed, and a single-layer temporal memory feedback network is initialized, including the input weight matrix of the feature input layer. The internal weight matrix of the memory storage layer The output weight matrix of the feedback output layer ,in This represents the input dimension, which includes information such as the projected area S of the headlights, the brightness and contrast of each edge, etc. This indicates the number of neurons in the memory storage area. This indicates the output dimension, corresponding to the headlight power supply voltage, offset angle, etc.

[0077] Define an iteration count label t, with an initial value of 0;

[0078] Initialize a single-layer temporal memory feedback network layer with an initial iteration count of 1. If the maximum iteration count is not met, the iteration count will be continuously increased until the output iteration count is met.

[0079] Step S012 involves constructing a cascaded memory storage area based on a single-layer time-domain memory feedback network, as shown in the flowchart below. Figure 3 As shown, the steps include:

[0080] Step S0121: Construct the memory storage area for a single-layer temporal memory feedback network used to process projected area and global features;

[0081] Step S0122: Construct the memory storage area for a single-layer temporal memory feedback network used to process edge brightness and contrast;

[0082] Step S0123: Construct the memory storage area for a single-layer temporal memory feedback network used to process the features of the central region;

[0083] Step S0124: Use the output of the previous single-layer time-domain memory feedback network as the input of the next single-layer time-domain memory feedback network to form a cascaded structure and obtain the cascaded memory storage area.

[0084] In this embodiment, a memory storage area is constructed for a single-layer temporal memory feedback network used to process the projected area and global features, denoted as sub-region 1, and the input is... x 1 = [ S ] ; Construct a memory storage area for a single-layer temporal memory feedback network to handle edge brightness and contrast, denoted as sub-region 2, with the input being... x 2 = [ L 1 , B 1 , L 2 , B 2 , L 3 , B 3 ] ; Construct a memory storage area for a single-layer temporal memory feedback network to process features in the central region, denoted as sub-region 3, with the input being... x 3 = [ L 4 , B 4 ] .

[0085] By cascading the memory storage areas of multiple single-layer time-domain memory feedback networks into cascaded sub-memory storage areas, we have:

[0086] (1)

[0087] in, This represents the internal state of the time-domain memory feedback network. Indicates input, Indicates the output. This represents the internal connection weight matrix. This represents the input connection weight matrix. This indicates the output connection weight matrix. This represents the bias vector. and This represents the activation function.

[0088] In a cascaded time-domain memory feedback network, the output of the previous single-layer time-domain memory feedback network is used as the input of the next single-layer time-domain memory feedback network, forming a cascaded structure to obtain a cascaded memory storage area, i.e.:

[0089] (2)

[0090] in which, represents the number of cascaded layers.

[0091] In the step S013, weight coupling is performed on the cascaded structure, and the flowchart is as shown in Figure 4 , which includes the steps:

[0092] Step S0131: constructing a coupling weight matrix for each sub-memory storage area in the cascaded memory storage area;

[0093] Step S0132: transmitting the real-time state of the low-level sub-area to the high-level sub-area to obtain a cross-level connection weight matrix;

[0094] Step S0133: calculating the state vector of each sub-area according to the intra-sub-area connection weight matrix and the cross-level connection weight matrix to implement weight coupling.

[0095] In this embodiment, the coupling weight matrix of each sub-memory storage area in the cascaded memory storage area is constructed. If the cascaded memory storage area includes K sub-memory storage areas, the coupling weight matrix between the k-th sub-area and the m-th sub-area (m<k) is

[0096] (3)

[0097] transmitting the real-time state of the low-level sub-area (m) to the high-level sub-area (k) to implement cross-level feature fusion; the state update equation including the coupling term for the k-th sub-area is:[[]]

[0098] (4)

[0099] in which, represents the state vector of sub-area k at time t, represents a constant between 0 and 1, referred to as the leakage rate, which controls the forgetting speed of the current state to the past state, represents the intra-sub-area connection weight matrix of sub-area k, represents the input vector of sub-area k, represents the intra-sub-area recursive weight matrix, that is, the connection weight matrix from sub-area k-1 to sub-area k, represents the state vector of sub-area m at time t.

[0100] optimizing the output weight matrix of the control network through a residual learning algorithm ;

[0101] integrating each feedback output layer to obtain the cascaded time-domain memory feedback network model, which is:[[]]

[0102] y = W o u t b a s e ⋅ r c o n + W o u t m a r ⋅ [ r 1 ; r 2 ] + b o u t (5)

[0103] in, This represents the basic output weight matrix. This represents the concatenated vector of states from all sub-regions. This represents the weight matrix of the residual module. [ r 1 ; r 2 ] This represents the vertical concatenation of state vectors from different sub-memory storage areas. This represents the bias term of the feedback output layer.

[0104] Preferably, the activation function for the feedback output layer is the sigmoid function, and its formula is:

[0105] (6)

[0106] The output voltage mapping formula is:

[0107] (7)

[0108] Where V represents the output voltage, V max This indicates the maximum allowable output voltage, V. min This indicates the minimum allowable output voltage. The output weight matrix represents the voltage. Represents the output voltage vector. This term represents the bias of the output voltage.

[0109] The activation function for the offset angle is the tanh function, and its formula is:

[0110] (8)

[0111] The formula for offset angle mapping is:

[0112] (9)

[0113] in, Indicates the output voltage. This indicates the maximum allowable output voltage. This indicates the minimum allowable output voltage. The output weight matrix represents the voltage. Represents the offset angle vector. This term represents the bias of the output voltage.

[0114] In an exemplary embodiment, step S02, which involves preliminary optimization of the cascaded time-domain memory feedback network model using stability determination, is illustrated in the flowchart below. Figure 5 As shown, the steps include:

[0115] Step S021: Obtain the internal recursive weight matrix of each sub-memory storage area;

[0116] Step S022: Calculate the spectral radius of the internal recursive weight matrix;

[0117] Step S023: When the spectral radius of the internal recursive weight matrix is ​​less than the preset threshold, the model is determined to be stable, and a preliminary optimized cascaded time-domain memory feedback network model is obtained; otherwise, the model is determined not to be stable, and the basic single-layer time-domain memory feedback network model is reconstructed for model iteration.

[0118] In this embodiment, the internal recursive weight matrix of each sub-memory storage area is obtained. The spectral radius of the internal recursive weight matrix is ​​calculated as follows:

[0119] (10)

[0120] If the spectral radius of the internal recursive weight matrix is ​​less than a preset threshold (1 in this embodiment), the model is determined to be stable, and a preliminary optimized cascaded time-domain memory feedback network model is obtained; otherwise, the model is determined not to be stable, and the basic single-layer time-domain memory feedback network model is reconstructed for model iteration, i.e., return to step S012.

[0121] In an exemplary embodiment, step S03, which involves using a residual learning algorithm to train the cascaded time-domain memory feedback network model a second time to obtain the optimal cascaded time-domain memory feedback network model, is illustrated in the flowchart below. Figure 6 As shown, the steps include:

[0122] Step S031: Input the historical input data into the cascaded temporal memory feedback network model in sequence for forward propagation and update the memory storage area state of the temporal memory feedback network layer;

[0123] Step S032: Calculate the output of the cascaded time-domain memory feedback network based on the state of the memory storage area;

[0124] Step S033: Construct a loss function to measure the difference between the network's predicted output and the actual expected output;

[0125] Step S034: Calculate the output of the residual connection using the residual learning algorithm;

[0126] Step S035: Calculate the gradient of the loss function with respect to the connection weight moments from the memory storage area to the feedback output layer using the residual learning algorithm;

[0127] Step S036: Update the weight matrix using the residual gradient iteration method;

[0128] Step S037: When the loss function converges, the optimal cascaded time-domain memory feedback network model is obtained.

[0129] In this embodiment, the relevant parameters of the residual learning algorithm are initialized, including the key weight matrices of each layer of the cascaded temporal memory feedback network, and the number of sub-memory storage areas is [number missing]. The number of neurons is N, and the maximum number of iterations is T;

[0130] Define an iteration count label t, with an initial value of 0;

[0131] The initial iteration count is 1. If the maximum iteration count is not met, the iteration count will continue to increase until the output iteration count is met.

[0132] The prepared historical input data is sequentially fed into the constructed cascaded temporal memory feedback network model for forward propagation. For the k-th temporal memory feedback network layer, the memory storage state is updated according to the following formula:

[0133] (11)

[0134] in, Indicates the updated state of the memory storage area. This represents the input weight matrix of the k-th temporal memory feedback network layer. This represents the input of the k-th temporal memory feedback network layer at time t; This represents the internal weight matrix of the k-th temporal memory feedback network layer; This indicates the state of the memory storage area at the previous moment;

[0135] Calculate the output based on the state of the memory storage area:

[0136] (12)

[0137] in, This represents the output of the k-th temporal memory feedback network feature output layer. This represents the output weight matrix of the k-th temporal memory feedback network layer;

[0138] Ultimately, the output of the entire cascaded time-domain memory feedback network is:

[0139] (13)

[0140] The mean of squared deviations is used as the loss function to measure the difference between the network's predicted output and the actual expected output. The loss function is:

[0141] (14)

[0142] Where N represents the number of samples, This represents the actual output of the i-th sample. This represents the predicted output for the i-th sample. This represents the l2 norm.

[0143] Introducing a residual learning module, let the input of the residual learning module be... The output is The residual function is The output of the residual connection is:

[0144] (15)

[0145] Backpropagation is performed using the residual learning algorithm to update the connection weight matrix from the memory storage area to the feedback output layer. For example, according to the chain rule, the loss function right The gradient is:

[0146] ∂ J ∂ W out , re = − 2 N ∑ i = 1 N ( y true , i − y pred , i ) [ r re ( t ) ] T (16)

[0147] Where N represents the number of neurons, This represents the actual output of the i-th sample. This represents the predicted output for the i-th sample. This represents the updated memory storage state vector; similarly, the loss function can be calculated. and The gradient;

[0148] The weight matrix is ​​updated using the residual gradient iteration method as follows:

[0149] (17)

[0150] Among them, It is the learning rate, which controls the step size for each parameter update.

[0151] Determine if the current loss function has converged; if so, output the globally optimal position vector. This yields the optimal cascaded time-domain memory feedback network model.

[0152] In an exemplary embodiment, step S04, adjusting the intelligent vehicle lights in real time according to the optimal cascaded time-domain memory feedback network model, is illustrated in the flowchart below. Figure 7 As shown, the steps include:

[0153] Step S041: Input the real-time projection data of the vehicle lights into the optimal cascaded time-domain memory feedback network model;

[0154] Step S042: Output the corrected voltage and offset angle of the headlights according to the cascaded time-domain memory feedback network model;

[0155] Step S043: Adjust the headlights in real time based on the predicted correction voltage and offset angle.

[0156] In this embodiment, the real-time projection data of the vehicle lights is input into the optimal cascaded time-domain memory feedback network model;

[0157] A cascaded time-domain memory feedback network predicts the corrective voltage and offset angle of the headlights;

[0158] The headlights are controlled in real time based on the predicted correction voltage and offset angle. The correction voltage is converted into a drive signal, and the formula is as follows:

[0159] (18)

[0160] in, Indicates the driving voltage. Indicates the correction voltage. Indicates the maximum allowable voltage. Indicates the minimum allowable voltage;

[0161] The lens position is adjusted based on the offset angle, and the formula is as follows:

[0162] (19)

[0163] in, Indicates the number of steps of the stepper motor. Indicates the offset angle. This indicates the single-step angular resolution.

[0164] According to another embodiment of the present invention, a computer-readable storage medium is also provided, which stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the vehicle headlight imaging control method based on time-domain memory feedback described in the above embodiments.

[0165] According to another embodiment of the present invention, a vehicle lighting imaging control system based on time-domain memory feedback is also provided, the structural schematic diagram of which is shown below. Figure 8 As shown, it includes:

[0166] Headlight imaging module; used to control the brightness and range of the image.

[0167] Environmental perception module; used to collect the brightness and contrast of the projection area, and calculate the projection area in real time through camera and image processing algorithm;

[0168] Imaging control module; used to construct a cascaded memory storage area structure to predict headlight correction voltage and offset angle;

[0169] Memory;

[0170] And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by the imaging control module, the programs causing the computer to execute the vehicle headlight imaging control method based on time-domain memory feedback described in the above embodiments. The optical system optimization module includes an optical adjustment module and a processor.

[0171] The methods described above according to the invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA) for such software processing. It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods described herein. Furthermore, when a general-purpose computer accesses the code used to implement the processing shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the processing shown herein.

[0172] Of course, those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention and are not intended to limit the present invention. Any changes or modifications to the above embodiments that are within the scope of the present invention will fall within the protection scope of the present invention.

Claims

1. A vehicle headlight imaging control method based on time-domain memory feedback, characterized in that, include: A cascaded temporal memory feedback network model is constructed based on the projection data of intelligent vehicle lights. This construction includes the following steps: constructing a basic single-layer temporal memory feedback network model, initializing the single-layer temporal memory feedback network, including the input weight matrix of the feature input layer. The internal weight matrix of the memory storage layer The output weight matrix of the feedback output layer ,in This indicates the input dimension, which includes the projected area S of the headlights, and the brightness and contrast information of each edge. This indicates the number of neurons in the memory storage area. Indicates the output dimension; Define the iteration number label t, with an initial value of 0; initialize a single-layer temporal memory feedback network layer with an initial iteration number of 1. If the maximum iteration number is not met, the iteration number will continue to increase until the output iteration number is met. A cascaded memory storage area is constructed based on a single-layer temporal memory feedback network; the cascaded structure is weighted and coupled; the weight matrix of each feedback output layer is calculated using the residual learning algorithm; and the cascaded temporal memory feedback network model is obtained by integrating each feedback output layer. A stability criterion is used to preliminarily optimize the cascaded temporal memory feedback network model; the preliminary optimization of the cascaded temporal memory feedback network model using stability criterion includes the following steps: obtaining the internal recursive weight matrix of each sub-memory storage area; Calculate the spectral radius of the internal recursive weight matrix; when the spectral radius of the internal recursive weight matrix is ​​less than a preset threshold, the model is determined to be stable, and a preliminary optimized cascaded time-domain memory feedback network model is obtained. Otherwise, the model is deemed not to have reached stability, and the basic single-layer temporal memory feedback network model is reconstructed for model iteration; The optimal cascaded temporal memory feedback network (CTMFR) model is obtained by secondary training using the residual learning algorithm. This process includes the following steps: sequentially inputting historical input data into the CTMFR model for forward propagation and updating the memory storage state of the temporal memory feedback network layers; calculating the output of the CTMFR network based on the memory storage state; constructing a loss function to measure the difference between the network's predicted output and the actual expected output; and introducing a residual learning module, where the input to the residual learning module is... The output is The residual function is The output of the residual connection is The residual learning algorithm is used to calculate the output of the residual connection; the residual learning algorithm is used to calculate the gradient of the loss function with respect to the weight moments of the connection from the memory storage area to the feedback output layer; the weight matrix is ​​updated using the residual gradient iteration method; the optimal cascaded temporal memory feedback network model is obtained when the loss function converges. The intelligent vehicle lights are adjusted in real time based on the optimal cascaded time-domain memory feedback network model to achieve vehicle light imaging control.

2. The vehicle headlight imaging control method based on time-domain memory feedback according to claim 1, characterized in that, The projection data includes any one or more combinations of the following: the projected area of ​​the vehicle headlights, the brightness and contrast measured at the front edge of the projected vehicle headlights, the brightness and contrast measured at the leftmost edge of the projected vehicle headlights, the brightness and contrast measured at the rightmost edge of the projected vehicle headlights, and the brightness and contrast measured at the center of the projected vehicle headlights.

3. The vehicle headlight imaging control method based on time-domain memory feedback according to claim 1, characterized in that, The construction of a cascaded memory storage area based on a single-layer time-domain memory feedback network includes the following steps: Construct a memory storage area for a single-layer temporal memory feedback network to process projected area and global features; Construct a memory storage area for a single-layer temporal memory feedback network to process edge brightness and contrast; Construct a memory storage area for a single-layer temporal memory feedback network to process features in the central region; The output of the previous single-layer time-domain memory feedback network is used as the input of the next single-layer time-domain memory feedback network to form a cascaded structure, resulting in a cascaded memory storage area.

4. The vehicle headlight imaging control method based on time-domain memory feedback according to claim 1, characterized in that, The weighted coupling of the cascaded structure includes the following steps: Construct the coupling weight matrix of each sub-memory storage area in the cascaded memory storage area; The real-time state of the lower-level sub-region is passed to the higher-level sub-region to obtain the cross-level connection weight matrix; The state vector of each sub-region is calculated based on the internal connection weight matrix and the cross-level connection weight matrix, thereby achieving weight coupling.

5. The vehicle headlight imaging control method based on time-domain memory feedback according to claim 1, characterized in that, The real-time adjustment of intelligent vehicle lights based on the optimal cascaded time-domain memory feedback network model includes the following steps: Input the real-time projection data of the vehicle lights into the optimal cascaded time-domain memory feedback network model; The corrected voltage and offset angle of the vehicle headlights are output based on the cascaded time-domain memory feedback network model. The headlights are adjusted in real time based on the predicted correction voltage and offset angle.

6. A computer-readable storage medium storing a computer program for electronic data interchange, wherein, The computer program causes the computer to perform the method as described in any one of claims 1-5.

7. A vehicle headlight imaging control system based on time-domain memory feedback, characterized in that, include: Headlight imaging module; Environmental perception module; Imaging control module; Memory; And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by the imaging control module, the programs causing the computer to perform the method as described in any one of claims 1-5.

Images