A feedback tracking control method based on scalarized visual frame pixels

Abstract

This invention provides a feedback tracking control method based on scalarized visual frame pixels. First, a dynamic state-space model of a vehicle following system is established and the system state variables are determined. Then, a state feedback tracking controller is designed for the dynamic state-space model. The original image of the leading vehicle is acquired and input into an image signal synthesizer. After processing by the image signal synthesizer, a background image and a standard image are obtained. Next, a binarized digital image pixel mask method is used to scalarize the high-dimensional features of the original image, background image, and standard image of the leading vehicle to obtain the number of pixels of the leading vehicle and the image error, which has an implicit functional relationship with the system state variables. Finally, based on the functional relationship between the image error and the system state variables, the image error is directly fed back to the state feedback tracking controller to control the throttle of the following vehicle, achieving fixed-distance following control.

Description

Technical Field

[0001] This application relates to the field of vehicle tracking control technology, and in particular to a feedback tracking control method based on scalarized visual frame pixels. Background Technology

[0002] In recent years, research problems in the field of control have become increasingly complex. Traditional control is based on a feedback framework, making the accurate acquisition of system state essential and important. In complex application scenarios, system state is often difficult to measure directly and can only be estimated from high-dimensional measurement data from sensors. Perception-based control methods have shown significant advantages and have been widely applied in research related to autonomous driving, drones, and robot control and navigation.

[0003] Autonomous driving control is one of the fields with high requirements for perception-based control. In this field, perception-based control essentially needs to extract effective information from high-dimensional measurement signals acquired by sensors and generate low-level vehicle control commands. Currently, the mainstream methods in existing technologies are modular pipelines and end-to-end control. Existing technology "CN113911139A" discloses a vehicle control method, device, and electronic device. This method first acquires environmental perception data and the current driving data of the autonomous vehicle. Then, based on the environmental perception data and the current driving data, it determines the target time delay of each stage of the end-to-end control mechanism, including the perception and localization stage, the tracking and prediction stage, the decision-making stage, and the control stage. Finally, based on the target time delay and time delay threshold of each stage, it determines the target processing mechanism from simple and complex processing mechanisms for each stage and controls the autonomous driving system of the autonomous vehicle to work based on the corresponding target processing mechanism at each stage.

[0004] However, both the modular pipeline and end-to-end control methods mentioned above are difficult to directly integrate high-dimensional sensing data, and still lack in terms of interpretability and reliability. Summary of the Invention

[0005] To overcome the technical problems of insufficient interpretability and reliability of existing perception-based control technologies, this invention provides a feedback tracking control method based on scalarized visual frame pixels.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] A feedback tracking control method based on scalarized visual frame pixels, the method comprising the following steps:

[0008] S1: Establish a dynamic state-space model of the vehicle following system and determine the system state variables; design a state feedback tracking controller for the dynamic state-space model; the vehicle following system includes at least one leading vehicle and at least one following vehicle.

[0009] S2: Acquire the raw image with the lead vehicle and input it into an image signal synthesizer with an encoder-dual decoder architecture to obtain the background image and the standard image;

[0010] S3: The high-dimensional features of the original image, background image, and standard image are scalarized using a digital image pixel mask algorithm based on binarization to obtain the image error; the high-dimensional feature is the number of pixels of the leading vehicle;

[0011] S4: Based on the preset mapping function between the image error and the system state variables, the image error is fed back to the state feedback tracking controller. The speed of the following vehicle is controlled according to the output of the state feedback tracking controller to achieve vehicle following control.

[0012] Preferably, in step S1, the dynamic state-space model is specifically as follows:

[0013]

[0014] in, Let be the system state vector, represented as , This represents the derivative of the system state vector with respect to time. and Let these be the first and second system state variables, respectively, defined as follows:

[0015]

[0016] in, and These are the real-time speeds of the lead vehicle and the following vehicles, respectively. It is the steady-state driving force that follows the vehicle. To follow the vehicle's real-time driving force; Indicates real-time vehicle distance. This indicates the preset desired following distance. Indicates the control input for following the vehicle. This indicates the control gain set according to specific needs, to meet... ;

[0017] Preferably, in step S1, the functional form of the state feedback tracking controller is: The state feedback tracking controller is based on system state variables. and Calculate control quantity Then, the real-time driving force required to follow the vehicle is calculated based on the following formula. :

[0018] .

[0019] Preferably, in step S2, the image signal synthesizer with an encoder-dual decoder architecture specifically comprises:

[0020] The encoder consists of six convolutional modules and a Flatten layer connected in sequence. Each convolutional module consists of a convolutional layer, a batch normalization layer and a ReLU activation layer connected in sequence.

[0021] The dual decoder includes two parallel decoder branches. Each decoder branch includes a Reshape layer and six transposed convolutional modules connected in sequence. Each transposed convolutional module consists of a transposed convolutional layer, a batch normalization layer, and a ReLU activation layer connected in sequence.

[0022] Preferably, the original image is input into an encoder for feature extraction to obtain a feature vector;

[0023] The feature vector is passed through the first decoder branch to output the background image appearance flow field. The background image appearance flow field and the original image are jointly subjected to bilinear sampling to obtain the background image.

[0024] After performing a vector concatenation operation between the feature vector and the preset expected following distance, the standard image appearance flow field is output through the second decoder branch. The standard image appearance flow field is then combined with the original image to perform bilinear sampling to obtain the standard image.

[0025] Preferably, in step S3, the digital image pixel mask algorithm based on binarization specifically includes:

[0026]

[0027] in, Indicates following the vehicle to capture the view ahead. Original image of the leader vehicle of the Mi. This indicates that the image signal synthesizer is based on the original image. The estimated background image, This indicates that the image signal synthesizer is based on the original image. and expected following distance The estimated standard image, These represent the characteristic parameters of the leadership vehicle. Characteristic parameters representing environmental background; A custom gain greater than 0; Indicates taking the absolute value; Represents a symbolic function; This represents the summation function.

[0028] Preferably, in step S4, the preset mapping function is specifically:

[0029]

[0030] in, Indicates a This is the calibration function for the independent variable.

[0031] Preferably, the mapping function has the following properties, including:

[0032] Monotonicity: For any two system state variable values and If a relationship exists Then the following relationship holds:

[0033]

[0034] in, Represents the value of system state variables The corresponding image error value is defined as

[0035] ; Represents the value of system state variables The corresponding image error value is defined as ;

[0036] Same number:

[0037]

[0038] Preferably, in step S4, the observed first system state variable is replaced with image error. The feedback is directly sent to the state feedback tracking controller, and the function form of the replaced state feedback tracking controller is as follows: .

[0039] Preferably, in step S2, the camera installed on the following vehicle is used to acquire the original image of the leading vehicle, and the camera is pointed directly forward.

[0040] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0041] This invention first establishes a dynamic state-space model of a vehicle following system and determines the system state variables, then designs a state feedback tracking controller for the dynamic state-space model. The vehicle following system includes at least one leader vehicle and at least one follower vehicle. Next, an original image with the leader vehicle is acquired and input into an image signal synthesizer with an encoder-dual decoder architecture to obtain a background image and a standard image. Then, a binary-based digital image pixel mask algorithm is used to scalarize the high-dimensional features of the original image, background image, and standard image to obtain the image error. The high-dimensional feature is the number of pixels in the leader vehicle. Finally, based on a preset mapping function between the image error and the system state variables, the image error is fed back to the state feedback tracking controller. The speed of the follower vehicle is controlled according to the output of the state feedback tracking controller to achieve vehicle following control.

[0042] This invention designs a state feedback controller based on a two-vehicle following system, observing system state variables from visual frame data. Compared to end-to-end and modular pipeline methods, it features lightweight and low latency, while providing provable stability guarantees. The proposed binarization-based digital image pixel mask method enhances vehicle pixel signals, eliminates the most numerous but useless background pixels, significantly reduces the complexity of high-dimensional data, and minimizes the impact of environmental factors on observation errors. The proposed image error scalarization effectively bridges the gap between high-dimensional sensor measurement data and low-level control output signals from the controller, providing a new solution to the difficulty of directly integrating high-dimensional sensing data using traditional model-based methods. The image error has a mapping relationship with the system state variables, ensuring that its direct integration into the feedback controller does not affect system convergence. Attached Figure Description

[0043] Figure 1 This is a flowchart of a feedback tracking control method based on scalarized visual frame pixels in Example 1;

[0044] Figure 2 This is a schematic diagram of a vehicle following system based on a feedback tracking control method using scalarized visual frame pixels, as described in Example 1.

[0045] Figure 3 This is a structural diagram of an image signal synthesizer for a feedback tracking control method based on scalarized visual frame pixels in Example 1;

[0046] Figure 4 The input image is used for offline testing of a feedback tracking control method based on scalarized visual frame pixels in Example 1.

[0047] Figure 5This is an output image of an image signal synthesizer used in an offline test of a feedback tracking control method based on scalarized visual frame pixels, as shown in Example 1.

[0048] Figure 6 This is a flowchart of a binary-based digital image pixel mask method for a feedback tracking control method based on scalarized visual frame pixels, as described in Example 1.

[0049] Figure 7 The graph shows the mapping relationship between image error and system state variables in a feedback tracking control method based on scalarized visual frame pixels in Example 1.

[0050] Figure 8 The CARLA simulation results of a feedback tracking control method based on scalarized visual frame pixels in Example 1 are shown in Figure 1.

[0051] Figure 9 This is a system structure diagram of a feedback tracking control method based on scalarized visual frame pixels in Example 2;

[0052] Figure 10 This is a structural diagram of a feedback tracking control method based on scalarized visual frame pixels in Example 3. Detailed Implementation

[0053] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this application.

[0054] To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions;

[0055] It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0056] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0057] Example 1

[0058] like Figure 1 As shown, this embodiment provides a feedback tracking control method based on scalarized visual frame pixels, the method including the following steps:

[0059] S1: Establish a dynamic state-space model of the vehicle following system and determine the system state variables; design a state feedback tracking controller for the dynamic state-space model; the vehicle following system includes at least one leading vehicle and at least one following vehicle.

[0060] S2: Acquire the raw image with the lead vehicle and input it into an image signal synthesizer with an encoder-dual decoder architecture to obtain the background image and the standard image;

[0061] S3: The high-dimensional features of the original image, background image, and standard image are scalarized using a digital image pixel mask algorithm based on binarization to obtain the image error; the high-dimensional feature is the number of pixels of the leading vehicle;

[0062] S4: Based on the preset mapping function between the image error and the system state variables, the image error is fed back to the state feedback tracking controller. The speed of the following vehicle is controlled according to the output of the state feedback tracking controller to achieve vehicle following control.

[0063] In practical implementation, the vehicle following system, such as Figure 2 As shown, the vehicle following system under autonomous driving has the following dynamic state-space model:

[0064]

[0065] The state vector is defined as follows: , The derivative of the system state variable with respect to time is given. The system state variable and control input are defined as follows:

[0066]

[0067] in, and These are the real-time speeds of the lead vehicle and the following vehicles, respectively. It is the steady-state driving force that follows the vehicle; Indicates real-time vehicle distance; This indicates the preset expected following distance, which is a preset fixed value. In one implementation, This indicates that the preset expected following distance is 10.0 meters; This indicates the control input for following the vehicle, where This indicates the control gain set according to specific needs.

[0068] Seeking and The time derivative is obtained as follows:

[0069]

[0070] The following expression for the vehicle's velocity can be obtained from Newton's laws of motion:

[0071]

[0072] in, and These represent the mass of the lead vehicle and the following vehicles, respectively. It is the drag coefficient following the vehicle; and These are the real-time driving forces of the leading vehicle and the following vehicles, respectively.

[0073] Substituting (6) into (5) yields:

[0074]

[0075] From (4), we can obtain:

[0076]

[0077] Substituting (3) and (8) into (7), the dynamic state-space model in step S1 is as follows:

[0078]

[0079] A state feedback tracking controller is designed using the backstepping method. The functional form of the controller is expressed as follows: According to the status and Calculate control quantity Then, based on equation (8), the required power of the vehicle is calculated, so as to reasonably control the throttle, so as to accelerate or decelerate the vehicle, thereby following the vehicle's speed. Track the speed of the leader's vehicle Distance between the two vehicles Tracking Expectations .when and At that time, it was considered that the dynamic state-space model had reached asymptotic stability.

[0080] Subsequently, the original image of the leader vehicle is acquired from the following vehicle equipped with an RGB camera, and the original image is input into an image signal synthesizer with an encoder-dual decoder CNN architecture. The structure diagram of the image signal synthesizer is shown below. Figure 3 As shown, the background image and the standard image are estimated by this neural network. The offline testing method for image signal synthesis is to... Figure 4 The image shows five images of the leadership vehicle representing different distances, which are used as inputs to the image signal synthesizer. The outputs are as follows: Figure 5 As shown, where Figure 4 In, (a) s=7.0m, (b) s=10.0m, (c) s=15.0m, (d) s=20.0m, (e) s=25.0m.

[0081] The encoder-dual-decoder CNN structure in this embodiment is as follows: it includes one encoder and two decoders. The encoder comprises six convolutional modules and one Flatten operation. Each convolutional module contains one convolutional layer (kernel size / strike / padding: 3×3 / 2 / 1), one batch normalization layer, and one ReLU activation function. The dual decoder comprises two decoder branches, each of which includes one Reshape operation and six transposed convolutional modules. Each transposed convolutional module includes one transposed convolutional layer and one batch normalization layer. The transposed convolutional layers of the first five transposed convolutional blocks are set to: kernel size / strike / padding: 4×4 / 2 / 1, using the ReLU activation function. The transposed convolutional layers of the sixth transposed convolutional block are set to: kernel size / strike / padding: 3×3 / 1 / 1, using the Tanh activation function. The original image is input into the encoder, where feature maps are extracted through six convolutional layers. These feature maps are then flattened one-dimensionally to obtain feature vectors. These feature vectors are input into the first decoder branch, which outputs the background image's appearance flow field. This background image is then subjected to bilinear sampling in conjunction with the input image to obtain the background image. The feature vectors and the desired distance are concatenated and input into the second decoder branch, which outputs the standard image's appearance flow field. This standard image is then subjected to bilinear sampling in conjunction with the input image to obtain the standard image.

[0082] like Figure 6 The original image, background image, and standard image are processed using a binary-based digital image pixel masking method to scalarize the high-dimensional features in the image, namely the number of pixels of the leading vehicle. Figure 4 The process of this method is clearly expressed, and it can also be expressed by the following formula:

[0083] in, This indicates that the RGB camera on the vehicle is capturing the front. Image of the leader's vehicle; This indicates that the image signal synthesizer is based on the original image. The estimated pure background image, i.e. There were no vehicles belonging to leaders, only environmental background information; This indicates that the image signal synthesizer is based on the original image. and expected vehicle distance The estimated standard image, Indicates the characteristic parameters of the leadership vehicle; Characteristic parameters representing environmental background; For custom gain; for multidimensional real arrays , Indicates to Take the absolute value of all elements in the array. Indicates to Each element in the expression is subjected to a sign function operation, the sign function being defined as:

[0084]

[0085] in Represents array Elements in; The result of the operation is a pair The sum of all elements in the expression.

[0086] The result after scalarization is called image error. With system state variables It possesses an implicit functional relationship, which can be visualized through experiments, such as... Figure 7 As shown. This functional relationship can be expressed mathematically as:

[0087]

[0088] in, Indicates a The implicit mapping function for the independent variable has a non-unique expression and can be calibrated according to actual needs. In this embodiment, the mapping function can be fitted through multiple trials. The mapping function has the following properties:

[0089] Monotonicity: For any two system state variable values and If a relationship exists Then the following relationship holds:

[0090]

[0091] in, , .

[0092] Same number: and The positive and negative values ​​remain consistent, which can be expressed as:

[0093]

[0094] Subsequently, based on the mapping relationship between image error and system state variables, the image error replaces the observed system state and is directly fed back to the designed feedback tracking controller. Therefore, the feedback tracking controller is designed as follows: The state feedback controller only needs the system state. The original image is sufficient to achieve good tracking and control of the system state variables on the target trajectory.

[0095] This embodiment uses image error. Instead of the observed system state Therefore, the state feedback tracking controller only needs the raw images from the onboard RGB camera to achieve tracking and control of the system state variables, without the need for other sensors to accurately measure the system state variables. The effectiveness of this method can be proven using Lyapunov stability theory.

[0096] The simulation results conducted in the CARLA autonomous driving simulator are as follows: Figure 8 As shown in the experiment, the expected vehicle distance was fixed at 20.0m before 55 seconds of simulation time and at 10.0m after 55 seconds. Before 110 seconds of simulation time, the speed of the leading vehicle varied sinusoidally for three cycles; after 110 seconds, the speed of the leading vehicle was fixed at 0. Simulation results show that the distance between the following vehicle and the leading vehicle tracks the expected distance well, with an error of no more than 3%. The speed of the following vehicle also tracks the speed of the leading vehicle well. The system state variables are eventually uniformly bounded; when the prediction error of the image synthesizer approaches 0, the system state variables will approach asymptotic stability from eventually uniformly bounded. Simulation results demonstrate the effectiveness of the proposed vision-based feedback tracking method using scalarized visual frame pixel data.

[0097] In this embodiment, a suitable feedback controller is designed for the established state-space equation model of the vehicle following system. An RGB camera installed on the following vehicle continuously captures and outputs images of the leading vehicle. The image signal synthesizer estimates the background image and the standard image based on the original image output by the RGB camera. An image error is constructed from the three image signals using the proposed binary-based digital image pixel mask method. This image error is input into the feedback controller and replaces the system state variables observed from the image error. Without the need for precise measurement using other sensors, the purpose of observing the system state variables through the output image of the RGB camera is achieved. This also enables the various system state variables to track the target trajectory, making the system asymptotically stable.

[0098] This embodiment designs a state feedback controller based on a two-vehicle following system, observing system state variables from visual frame data. Compared to end-to-end and modular pipeline methods, it features lightweight and low latency, while providing provable stability guarantees. The proposed binary-based digital image pixel mask method significantly reduces the complexity of high-dimensional image data and minimizes the impact of environmental factors on observation errors. The proposed image error scalarization of high-dimensional image features effectively bridges the gap between high-dimensional sensor measurement data and low-level control output signals from the controller, providing a new solution to the difficulty of directly integrating high-dimensional sensing data using traditional model-based methods. The image error has a mapping relationship with the system state variables, ensuring that its direct integration into the feedback controller does not affect system convergence.

[0099] Example 2

[0100] like Figure 9 As shown, this embodiment provides a feedback tracking control system based on scalarized visual frame pixels, including: an image acquisition unit 201, an image estimation unit 202, an image processing unit 203, and a tracking control unit 204;

[0101] The image acquisition unit 201 is used to acquire the original image;

[0102] The image estimation unit 202 is used to input the original image acquired by the image acquisition unit 201 into an image signal synthesizer with an encoder-dual decoder architecture CNN to estimate the background image and the standard image;

[0103] The image processing unit 203 is used to scalarize the high-dimensional features of the original image, background image, and standard image using a binarization-based digital image pixel mask method to obtain the number of pixels of the leading vehicle and to obtain the image error;

[0104] The tracking control unit 204 is used to feed the image error directly to the pre-designed tracking controller based on the mapping relationship between image error and system state variables, thereby controlling the speed of the following vehicle and realizing fixed-distance following control.

[0105] In specific implementation, the image acquisition unit 201 includes a leading vehicle and a following vehicle equipped with an RGB camera. The leading vehicle is located in front of the following vehicle. In one embodiment, both the leading vehicle and the following vehicle are located in the same lane. When there are no lane markings, the following vehicle is generally located directly behind the leading vehicle. The following vehicle is equipped with an RGB camera that can acquire images of the leading vehicle. In this embodiment, the camera of the following vehicle can be mounted on the roof or the hood. The camera is directed directly in front of the following vehicle and is used to continuously capture original images of the leading vehicle located at a certain distance in front of the following vehicle and traveling in the same direction as the following vehicle.

[0106] Subsequently, the image estimation unit 202 inputs the original image into an image signal synthesizer with an encoder-dual decoder CNN architecture. The original image first passes through the encoder, where feature maps are extracted using six convolutional layers. These feature maps are then flattened in one dimension using a Flatten operation to obtain feature vectors. These feature vectors are input into the first decoder branch, which outputs the background image appearance flow field. This flow field is then combined with the input original image to perform bilinear sampling, resulting in the background image. The feature vectors and the desired distance are concatenated and input into the second decoder branch, which outputs the standard image appearance flow field. This standard image is then combined with the input original image to perform bilinear sampling, resulting in the standard image.

[0107] Next, the image processing unit 203 uses a binary-based digital image pixel mask method to process the image, scalarizing the high-dimensional features in the image. The resulting image error can be expressed by the following formula:

[0108]

[0109] in, This indicates that the RGB camera on the vehicle is capturing the front. Image of the leader's vehicle; This indicates that the image signal synthesizer is based on the original image. The estimated pure background image, i.e. There were no vehicles belonging to leaders, only environmental background information; This indicates that the image signal synthesizer is based on the original image. and expected vehicle distance The estimated standard image, Indicates the characteristic parameters of the leadership vehicle; Characteristic parameters representing environmental background; For custom gain; for multidimensional real arrays , Indicates to Take the absolute value of all elements in the array. Indicates to Each element in the expression is subjected to a sign function operation, the sign function being defined as: ,in Represents array Elements in; The result of the operation is a pair The sum of all elements in the expression.

[0110] Finally, the tracking control unit 204, based on the implicit functional relationship between the image error and the system state variables, directly feeds back the image error to the pre-designed tracking controller, controlling the throttle of the following vehicle to achieve fixed-distance following control. The functional relationship can be mathematically expressed as:

[0111]

[0112] in, Indicates a The implicit mapping function for the independent variable has a non-unique expression and can be calibrated according to actual needs. In this embodiment, the mapping function can be fitted through multiple trials. The mapping function has the following properties:

[0113] Monotonicity: For any two system state variable values and If a relationship exists Then the following relationship holds:

[0114]

[0115] in, , .

[0116] Same number: and The positive and negative values ​​remain consistent, which can be expressed as:

[0117]

[0118] Subsequently, based on the mapping relationship between image error and system state variables, the image error replaces the observed system state and is directly fed back to the designed feedback tracking controller. Therefore, the feedback tracking controller is designed as follows: The state feedback controller only needs the system state. The original image is sufficient to achieve good tracking and control of the system state variables on the target trajectory.

[0119] Example 3

[0120] like Figure 10As shown, this embodiment provides a feedback tracking control device 31 for scalarized visual frame pixels. The feedback tracking control device 31 for scalarized visual frame pixels includes a processor 32, a memory 33, and a computer program stored in the memory 33 and executable on the processor 32. When the processor 32 executes the computer program, it implements the steps in the above method embodiments. Alternatively, when the processor 32 executes the computer program, it implements the functions of each module / unit in the above device embodiments.

[0121] For example, the computer program can be divided into one or more modules / units, which are stored in the memory 33 and executed by the processor 32 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the feedback tracking control device 31 that scalarizes visual frame pixels. For example, the computer program can be divided into the modules shown in Embodiment 2. The specific functions of each module are described in the working process of the device described in the above embodiments, and will not be repeated here.

[0122] The feedback tracking control device 31 for scalarized visual frame pixels may include, but is not limited to, a processor 32 and a memory 33. Those skilled in the art will understand that the schematic diagram is merely an example of the feedback tracking control device 31 for scalarized visual frame pixels and does not constitute a limitation on the device. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the feedback tracking control device 31 for scalarized visual frame pixels may also include input / output devices, network access devices, buses, etc.

[0123] The processor 32 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 32 is the control center of the feedback tracking control device 31 for scalarized visual frame pixels, connecting all parts of the feedback tracking control device 31 for scalarized visual frame pixels through various interfaces and lines.

[0124] The memory 33 can be used to store the computer program and / or modules. The processor 32 implements various functions of the feedback tracking control device 31 for scalarized visual frame pixels by running or executing the computer program and / or modules stored in the memory 33 and calling the data stored in the memory 33. The memory 33 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 33 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0125] The module / unit integrated by the feedback tracking control device 31 for scalarized visual frame pixels, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 32, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0126] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0127] The same or similar labels correspond to the same or similar parts;

[0128] The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this application.

[0129] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A feedback tracking control method based on scalarized visual frame pixels, characterized in that, The method comprises the following steps: S1: establishing a dynamic state space model of a vehicle following system and determining system state variables, designing a state feedback tracking controller for the dynamic state space model; the vehicle following system comprises at least one leader vehicle and at least one follower vehicle; S2: acquiring an original image with a leader vehicle and inputting the original image into an image signal synthesizer with an encoder-double decoder architecture to obtain a background image and a standard image; The image signal synthesizer with the encoder-double decoder architecture is specifically: The encoder is composed of 6 convolution modules and a Flatten layer connected in sequence, and each convolution module is composed of a convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence; The double decoder comprises two parallel decoder branches, each of which comprises a Reshape layer and 6 transpose convolution modules connected in sequence, and each transpose convolution module is composed of a transpose convolution layer, a batch normalization layer and a ReLU activation layer connected in sequence; After the original image is input into the encoder for feature extraction, a feature vector is obtained; the feature vector is input through the first decoder branch to output a background image appearance flow field, the background image appearance flow field is combined with the original image to perform bilinear sampling to obtain a background image; after the feature vector and a preset expected following vehicle distance are subjected to vector connection operation, the second decoder branch is used to output a standard image appearance flow field, the standard image appearance flow field is combined with the original image to perform bilinear sampling to obtain a standard image; S3: quantizing high-dimensional features of the original image, the background image and the standard image by using a binary-based digital image pixel mask algorithm to obtain an image error; the high-dimensional features are pixel numbers of the leader vehicle; S4: based on a preset mapping function between the image error and the system state variables, the image error is fed back to the state feedback tracking controller, the speed of the follower vehicle is controlled according to the output of the state feedback tracking controller, and following control is realized.

2. The feedback tracking control method based on quantized visual frame pixels according to claim 1, characterized in that, In the step S1, the dynamic state space model is specifically: wherein is the system state vector, denoted as , denotes the derivative of the system state vector with respect to time, and are the first and second system state variables, respectively, defined as wherein, and are real-time speeds of the lead vehicle and the following vehicle, respectively, is a steady-state driving force of the following vehicle, is a real-time driving force of the following vehicle; denotes a real-time vehicle distance, denotes a preset desired following vehicle distance, denotes a control input of the following vehicle, denotes a control gain set according to specific requirements, satisfying .

3. The feedback tracking control method based on quantized visual frame pixels according to claim 1, wherein, The function form of the state feedback tracking controller in the step S1 is , the state feedback tracking controller calculates the control variable according to the system state variable and , and calculates the real-time driving force required by the following vehicle based on the following formula : 。 4. The feedback tracking control method based on quantized visual frame pixels according to claim 1, wherein, In the step S3, the binary-based digital image pixel mask algorithm is specifically: wherein represents a raw image of a leading vehicle in front of a following vehicle, represents a raw image of a leading vehicle in front of a following vehicle, represents a raw image of a leading vehicle in front of a following vehicle, represents an estimated background image, represents an estimated background image, represents an estimated background image, represents an estimated background image, represents a characteristic parameter of a leading vehicle, represents a characteristic parameter of an environmental background; is a user-defined gain greater than 0; represents taking the absolute value; represents the sign function; represents the summation function.

5. The feedback tracking control method based on quantized visual frame pixels according to claim 1, wherein, In the step S4, the preset mapping function is specifically: wherein represents a calibration function with as argument.

6. The feedback tracking control method based on quantized visual frame pixels according to claim 5, wherein, The mapping function has the following properties, including: Monotonicity: For any two system state variable values and if the relation exists, then the following relation holds: wherein, represents a system state variable value a corresponding image error value, defined as ; representing system state variable values corresponding image error values, defined as ; Same sign property: 。 7. The feedback tracking control method based on quantized visual frame pixels according to claim 1, wherein, In the step S4, the observed first system state variable is replaced by the image error which is fed back directly into the state feedback tracking controller, which takes the form after replacement.

8. The feedback tracking control method based on quantized visual frame pixels according to any one of claims 1-7, characterized in that, In the step S2, the original image with the leader vehicle is acquired by using a camera installed on the follower vehicle, and the camera direction points to the front.

Images