Automatic early warning steering method and system

Abstract

The invention relates to the technical field of vehicle intelligent control, in particular to an automatic early warning steering method and system, and the method comprises the steps: pre-collecting a plurality of sample images on a pre-driving path of a vehicle hub, and generating a training data set based on the plurality of sample images; constructing a recognition network architecture, and training the recognition network architecture by adopting the training data set to obtain a tiny target recognition network model; acquiring a real-time image on a pre-driving path of a vehicle hub in real time, and performing feature extraction on the real-time image by adopting the tiny target identification network model to obtain target information and target position information corresponding to the real-time image; and the target information and the target position information are processed to obtain a steering angle, and vehicle steering is controlled based on the steering angle to change the driving path of the vehicle hub. According to the invention, detail detection and early-warning steering of the pre-driving path of the vehicle hub can be realized at least.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle intelligent control, and in particular relates to an automatic warning steering method and system. Background Art

[0002] With the gradual popularization of intelligent driving, more and more intelligent vehicle control strategies are emerging, targeting collision avoidance, pedestrian detection, dangerous object detection, etc., but they are all intelligent control technologies for large targets. In addition, the common dynamic ranging technology, image-based target recognition technology, and radar-based target recognition technology involved in the existing intelligent control technology all have a pain point, that is, slow processing speed, which still brings certain safety risks.

[0003] Existing intelligent control technologies specifically include automatic emergency braking technology, lane departure warning technology, blind spot monitoring technology, traffic signal recognition technology, driver monitoring technology, rear traffic crossing prompt technology, and vehicle door opening warning technology. Among them, automatic emergency braking technology uses millimeter wave radar for ranging, which can effectively measure targets within 170 meters in front of the vehicle. During adaptive cruise and normal driving, the vehicle calculates the emergency braking distance based on the status of the vehicle in front, its own vehicle speed and acceleration status. If it approaches the critical braking distance, the vehicle will automatically perform emergency braking to prevent traffic accidents and improve driving safety; lane departure warning technology mainly uses cameras to detect lane auxiliary lines for lane departure warning. This technology is developed and implemented using traditional algorithms (non-deep learning algorithms); blind spot detection technology, traffic light recognition technology, driver monitoring technology, rear traffic crossing prompt technology, and vehicle door opening warning technology are all implemented using cameras equipped with target recognition intelligent algorithms. Different training sets are used for training in different scenarios to achieve application purposes in different scenarios.

[0004] However, the prior art does not have an intelligent control technology for performing detailed detection on the pre-travel area of ​​the vehicle wheel hub. Summary of the invention

[0005] In view of this, the present invention aims to provide an automatic warning steering method and system, which is at least conducive to the detailed detection of the pre-driving path of the vehicle wheel hub and the warning steering.

[0006] To achieve the above object, the technical solution created by the present invention is implemented as follows:

[0007] The invention provides an automatic warning steering method, comprising: pre-collecting a plurality of sample images on a pre-driving path of a vehicle wheel hub, generating a training data set based on the plurality of sample images; constructing a recognition network architecture, and using the training data set to train the recognition network architecture to obtain a small target recognition network model; real-time collection of real-time images on the pre-driving path of the vehicle wheel hub, and using the small target recognition network model to perform feature extraction on the real-time images to obtain target information and target position information corresponding to the real-time images; processing the target information and target position information to obtain a steering angle, and controlling the vehicle steering based on the steering angle to change the driving path of the vehicle wheel hub.

[0008] Furthermore, generating a training data set based on multiple sample images includes: pre-collecting N sample images on a pre-driving path of a vehicle wheel hub, selecting M sample images from the N sample images for manual target labeling operations to obtain the M labeled sample images; training a recognition model based on the labeled M sample images, and using the recognition model to label the remaining (NM) sample images to obtain the labeled N sample images; and establishing a training data set based on the labeled N sample images.

[0009] Furthermore, establishing a training data set based on the labeled N sample images includes: performing image transformation on the labeled N sample images to expand the labeled N sample images to xN sample images, and establishing a training data set including xN sample images, where x is an integer greater than 1.

[0010] Furthermore, the small target recognition network model is a deep convolutional neural network model, which includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer and a fully connected layer.

[0011] Furthermore, the recognition network architecture is trained using a training data set to obtain a small target recognition network model, including: inputting sample images in the training data set into the recognition network architecture, constructing a loss function, and adjusting the model parameters of the recognition network architecture through an optimization algorithm until the loss function is minimized to obtain a small target recognition network model.

[0012] Furthermore, the loss function is as follows:

[0013]

[0014] Where loss represents the loss function, y∈{1,0}, y represents the true target category, p∈[0,1], p represents the estimated probability when y=1, (1-p) represents the estimated probability when y=0, γ represents an adjustable static factor, β represents an adjustable dynamic factor, and e p Represents the amplification method of probability.

[0015] Furthermore, the target position information includes lateral position information and distance information, and the target information includes target type; processing the target information and target position information to obtain the steering angle includes: using a steering coefficient formula to calculate the steering coefficient according to the vehicle speed, vehicle stability, distance information, target type and lateral position information, and obtaining the steering angle through the steering coefficient.

[0016] Furthermore, the steering coefficient formula is as follows:

[0017]

[0018] Among them, K represents the steering coefficient, V represents the current vehicle speed, and V max represents the maximum design speed of the vehicle, S represents the stability score of the vehicle, S max represents the maximum value of the vehicle's stability score, D represents the distance between the target and the vehicle, and D min represents the minimum distance for triggering an early warning, C represents the target type coefficient, and C max represents the maximum value of the target type coefficient, P represents the lateral position coefficient, P max Represents the maximum value of the lateral position coefficient, and a, b, c, d and e are all weight factors.

[0019] Further, obtaining the steering angle through the steering coefficient includes: calculating the steering angle using a steering angle formula, the steering angle formula is as follows:

[0020] θ=K×t 2 ;

[0021] Among them, K represents the steering coefficient and t represents the remaining steering time.

[0022] On the other hand, the present invention provides an automatic warning steering system, including: a data capture layer, the data capture layer is used to collect real-time images on the pre-driving path of the vehicle wheel hub; a strategy customization layer, the strategy customization layer includes a small target recognition network model, the small target recognition network model is used to extract features from the real-time image to obtain target information and target position information corresponding to the real-time image; the real-time operation layer, the real-time operation layer is used to process the target information and target position information to obtain a steering angle, and control the vehicle steering based on the steering angle to change the driving path of the vehicle wheel hub.

[0023] Compared with the prior art, the invention can achieve the following beneficial effects: the embodiment of the invention uses a camera to capture real-time images of the pre-driving path of the vehicle wheel hub, and then uses a small target recognition network model with a high processing speed to process the real-time image, so as to identify small dangerous objects on the pre-driving path of the vehicle wheel hub in real time, and change the driving path of the vehicle wheel hub according to the recognition result, so as to prevent the vehicle from having a tire blowout during driving, thereby avoiding personal loss caused by a tire blowout. The significance of the automatic warning steering method provided by the embodiment of the invention is that: since the driver will not observe the small targets on the road ahead in detail when the vehicle is driving at high speed, based on this, the embodiment of the invention provides an automatic warning steering method, which can realize real-time monitoring of the path that the wheel is about to pass. Once an object that may pose a threat to the wheel is detected, the warning mechanism will be immediately triggered to automatically adjust the driving direction of the vehicle, so as to prevent potential dangerous situations and ensure the safety of passengers and pedestrians. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings constituting part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments and descriptions of the present invention are used to explain the present invention and do not constitute an improper limitation on the present invention. In the drawings:

[0025] Figure 1 A schematic diagram of the network structure of a small target recognition network model according to an embodiment of the present invention;

[0026] Figure 2 A schematic diagram of a usage scenario of the automatic warning steering method described in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the invention more clear, the invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described here are only used to explain the invention and do not constitute a limitation of the invention.

[0028] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0029] In the description of the invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the invention, unless otherwise specified, the meaning of "multiple" is two or more.

[0030] In the description of the invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the invention can be understood according to specific circumstances.

[0031] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0032] The invention provides an automatic warning steering method, comprising: pre-collecting a plurality of sample images on a pre-driving path of a vehicle wheel hub, generating a training data set based on the plurality of sample images; constructing a recognition network architecture, and using the training data set to train the recognition network architecture to obtain a small target recognition network model; real-time collection of real-time images on the pre-driving path of the vehicle wheel hub, and using the small target recognition network model to perform feature extraction on the real-time images to obtain target information and target position information corresponding to the real-time images; processing the target information and target position information to obtain a steering angle, and controlling the vehicle steering based on the steering angle to change the driving path of the vehicle wheel hub.

[0033] It should be noted that the targets involved in the embodiments of the present invention are relatively small dangerous objects, which may be nails, glass, blades, stones or bricks, etc.

[0034] In some embodiments of the present invention, a camera is used to capture sample images and real-time images. In some examples, the camera can be a visible light camera. In some examples, the camera can be placed at the location of the two headlights of the vehicle, and the second is placed at the location of the rearview mirror of the vehicle. In other examples, the camera can also be placed at other locations on the front end of the vehicle, as long as the camera can capture the image of the pre-driving path of the vehicle wheel hub.

[0035] In some embodiments of the present invention, the number of cameras is usually two, for example Figure 2 The first camera and the second camera shown in .

[0036] In some embodiments of the present invention, a camera may be used to collect images at a distance of 19 m to 20 m in front of the wheel hub of the vehicle.

[0037] Furthermore, generating a training data set based on multiple sample images includes: pre-collecting N sample images on a pre-driving path of a vehicle wheel hub, selecting M sample images from the N sample images to perform a manual target labeling operation, that is, manually labeling smaller dangerous items in the sample images to obtain the labeled M sample images; training a recognition model based on the labeled M sample images, and using the recognition model to label the remaining (NM) sample images to obtain the labeled N sample images; and establishing a training data set based on the labeled N sample images.

[0038] In some embodiments of the present invention, the specific steps of pre-collecting N sample images on the pre-driving path of the vehicle wheel hub may be as follows: using a visible light camera to capture images on the pre-driving path in front of the vehicle wheel hub in real time, i.e., sample images. Before capturing, various smaller dangerous items may be placed on the pre-driving path in front of the vehicle wheel hub, and then sample images may be collected and stored. In some examples, 1,000 sample image data may be collected, i.e., N=1,000, and then the sample graphics may be manually annotated with targets offline. 100 sample images may be annotated, i.e., M=100, and then the recognition model may be trained with the data set of the 100 manually annotated sample images, and then the trained recognition model may be used to annotate the remaining 900 sample images, and the 100 manually annotated sample images may be merged with the 900 sample images annotated by the recognition model to obtain a data set of 1,000 annotated sample images, each annotated sample image corresponds to corresponding image data, and the 1,000 sample images and their corresponding image data constitute a training data set.

[0039] Furthermore, establishing a training data set based on the labeled N sample images includes: performing image transformation on the labeled N sample images to expand the labeled N sample images to xN sample images, and establishing a training data set including xN sample images, where x is an integer greater than 1.

[0040] In some embodiments of the present invention, in order to increase the diversity and scale of the training data set, image transformation technology can be used to transform the N labeled sample images. The image transformation may include horizontal mirror inversion, position translation adjustment, angle rotation transformation, region cropping processing and image compression operations. Through these image transformation operations, the capacity of the training data set can be expanded to 8 times the original capacity, thereby generating a richer final training data set that is more suitable for model training.

[0041] For further reference, Figure 1 The trained small target recognition network model is a deep convolutional neural network model, which includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, and a fully connected layer. The deep convolutional neural network model uses convolution kernels of different sizes to extract features of different scales from sample images. In some examples, an image of 640×640 pixels can be input into the deep convolutional neural network model, and the first convolutional layer is used to perform convolution operations on the image with a step size of 2 and 3×3×50 to obtain a feature map of 320×320×50. The first pooling layer is used to perform 2×2 pooling on the feature map output by the first convolutional layer to obtain a feature map of 160×160×50. Then, the second convolutional layer is used to perform convolution operations on the feature map output by the first pooling layer with a step size of 2 and 3×3×100 to obtain a feature map of 81×81×100. The second pooling layer is used to perform 2×2 pooling on the feature map output by the first pooling layer to obtain a feature map of 160×160×50. The feature map output by the second convolutional layer is pooled 2×2 to obtain a feature map of 41×41×100. The third convolutional layer is used to perform convolution operations with a step size of 2 and 3×3×200 on the feature map output by the second pooling layer to obtain a feature map of 21×21×200. The third pooling layer is used to perform 2×2 pooling on the feature map output by the third convolutional layer to obtain a feature map of 11×11×200. The feature map is then passed through the fully connected layer to obtain a feature map of 24200×100, that is, 100 data are obtained, including the coordinate positions of 20 targets (lateral position information and distance information), target lengths of 20 targets, target widths of 20 targets, and target types of 20 targets.

[0042] The network architecture of the small target recognition network model provided by the embodiment of the present invention is simple, and is simpler than the existing YOLO (You Only Look Once) algorithm, SSD (Single Shot MultiBox Detector) algorithm, Faster RCNN algorithm and NanoDet algorithm. The small target recognition network model provided by the embodiment of the present invention can realize rapid detection of smaller targets, because it only extracts three features and performs one full connection, so as to achieve the detection of smaller dangerous targets.

[0043] Furthermore, the recognition network architecture is trained using a training data set to obtain a small target recognition network model, including: inputting sample images in the training data set into the recognition network architecture, and constructing a loss function, adjusting the model parameters of the recognition network architecture through an optimization algorithm until the loss function is minimized, so as to obtain a small target recognition network model. In other words, by constructing a deep feature learning framework (recognition network architecture), the deep feature learning framework uses a variety of convolutional filters to extract features from image data, and fuses these features for collaborative learning, thereby generating a feature representation model for a specific target, and by repeatedly training and iterating this deep feature learning framework, optimizing and determining the network parameters that can track the target features, i.e., weights, to obtain a small target recognition network model.

[0044] Furthermore, the loss function is as follows:

[0045]

[0046] Where loss represents the loss function, y∈{1,0}, y represents the true type of the target, if the target is a real small dangerous object, then y=1, if the target is a fake small dangerous object, then y=0, p∈[0,1], p represents the estimated probability when y=1, (1-p) represents the estimated probability when y=0, γ represents an adjustable static factor, β represents an adjustable dynamic factor, e p Represents the amplification method of probability.

[0047] It should be noted that the loss function loss is a standard cross entropy, which is -log(p t ) and a dynamic scaling factor (1-p t ) γ The product of the probability weights and β(p t +e p ) is used to represent the loss function provided in the embodiment of the present invention. The loss function provided in the embodiment of the present invention adds the probability weight and β(p t +e p), which is conducive to increasing the detection accuracy of smaller dangerous targets. After calculation, the average index of average precision can be increased by 2 points.

[0048] The loss function loss is used to find the difference between the real target data and the training target data. The target data mainly includes the horizontal and vertical coordinates of the target coordinate frame (horizontal position information and distance information), target length, target width and target type. After the loss is calculated by the loss function, the weights in the model can be updated by the reverse gradient descent algorithm. Then the collected real-time image is input into the trained small target recognition network model to obtain the target information. Combined with the shooting distance of the camera, the target position information can be calculated. The target information and target position information are processed to obtain the steering angle, and the corresponding steering angle is transmitted to the vehicle control system to control the vehicle steering so that the vehicle's wheel hub avoids smaller dangerous objects.

[0049] Furthermore, the target position information includes lateral position information and distance information; the target information includes target type, and processing the target information and target position information to obtain the steering angle includes: using the steering coefficient formula to calculate the steering coefficient according to the vehicle speed, vehicle stability, distance information, target type and lateral position information, and obtaining the steering angle through the steering coefficient.

[0050] Furthermore, the steering coefficient formula is as follows:

[0051]

[0052] Where K represents the steering coefficient, V represents the current vehicle speed, and the unit of the current vehicle speed is m / s or km / h. max represents the maximum design speed of the vehicle. The maximum design speed and the current speed must be in the same unit. S represents the stability score of the vehicle. S is a dimensionless value between 0 and 1, where 1 represents complete stability and S max Represents the maximum value of the vehicle's stability score, S max =1, D represents the distance between the target and the vehicle (distance information), the unit of D can be m, D min represents the minimum distance for triggering an early warning, and its unit is m. C represents the target category coefficient. The target category coefficient C is a weight value given according to the target category. It can be a dimensionless value between 0 and 1, or a value of a specific category. C max represents the maximum value of the target category coefficient, which can be 1 or the maximum weight value. P represents the lateral position coefficient, which is a weight value given according to the target lateral position information. It can be a dimensionless value between 0 and 1, or a value of a specific classification. maxRepresents the maximum value of the lateral position coefficient, which can be 1 or the maximum weight value. a, b, c, d and e are all weight factors. a, b, c, d and e are used to adjust the influence of each parameter on the steering coefficient, which can be optimized through experiments or simulations.

[0053] Further, obtaining the steering angle through the steering coefficient includes: calculating the steering angle using a steering angle formula, the steering angle formula is as follows:

[0054] θ=K×t 2 ;

[0055] Where K represents the steering coefficient and t represents the remaining steering time. The remaining steering time refers to the remaining steering time from the start of steering to 2 seconds if the vehicle is deemed to complete steering within 2 seconds. For example, when steering begins, the remaining steering time is 2, and the required steering angle is 4K. As time changes, the remaining steering time gradually decreases, and the corresponding steering angle also gradually decreases.

[0056] In some embodiments of the present invention, K = 5 × 10 -6 .

[0057] The automatic warning steering method provided by the embodiment of the present invention can quickly and accurately detect smaller dangerous objects during the driving process of the vehicle, and can realize real-time monitoring of the path that the wheels are about to pass. Once an object that may pose a threat to the wheels is detected, the warning mechanism will be immediately triggered to automatically adjust the driving direction of the vehicle, thereby preventing potential dangerous situations and ensuring the safety of passengers and pedestrians.

[0058] On the other hand, the present invention provides an automatic warning steering system for implementing the automatic warning steering method of the above-mentioned embodiment, the automatic warning steering system includes: a data capture layer, the data capture layer is used to collect real-time images on the pre-driving path of the vehicle wheel hub; a strategy customization layer, the strategy customization layer includes a small target recognition network model, the small target recognition network model is used to extract features of the real-time image to obtain target information and target position information corresponding to the real-time image; the real-time operation layer, the real-time operation layer is used to process the target information and target position information to obtain a steering angle, and control the vehicle steering based on the steering angle to change the driving path of the vehicle wheel hub.

[0059] The automatic warning steering system provided by the present invention includes a data capture layer, a strategy customization layer and a real-time operation layer. The data capture layer may include a visible light camera arranged at the front end of the vehicle, the strategy customization layer may be an intelligent processing module of the vehicle, and the real-time operation layer may be a steering control system of the vehicle.

[0060] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the disclosure of the present invention can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0061] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An automatic warning steering method, characterized in that: include: Pre-collecting a plurality of sample images on a pre-travel path of a wheel hub of a vehicle, and generating a training data set based on the plurality of sample images; Constructing a recognition network architecture, and using the training data set to train the recognition network architecture to obtain a small target recognition network model; Real-time collection of a real-time image on the pre-travel path of the vehicle wheel hub, and feature extraction of the real-time image using the small target recognition network model to obtain target information and target position information corresponding to the real-time image; The target information and the target position information are processed to obtain a steering angle, and the vehicle steering is controlled based on the steering angle to change the driving path of the vehicle wheel hub.

2. The automatic warning steering method according to claim 1, characterized in that: Generating a training data set based on the plurality of sample images comprises: Pre-collecting N sample images on a pre-travel path of a wheel hub of a vehicle, and selecting M sample images from the N sample images to perform a manual target marking operation to obtain the M marked sample images; A recognition model is trained based on the M labeled sample images, and the remaining (NM) sample images are labeled using the recognition model to obtain N labeled sample images; The training data set is established according to the labeled N sample images.

3. The automatic warning steering method according to claim 2, characterized in that: Establishing the training data set according to the labeled N sample images includes: performing image transformation on the labeled N sample images to expand the labeled N sample images to xN sample images, and establishing the training data set including the xN sample images, wherein x is an integer greater than 1.

4. The automatic warning steering method according to claim 1, characterized in that: The small target recognition network model is a deep convolutional neural network model, which includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer and a fully connected layer.

5. The automatic warning steering method according to claim 4, characterized in that: Using the training data set to train the recognition network architecture to obtain a small target recognition network model includes: inputting sample images in the training data set into the recognition network architecture, constructing a loss function, and adjusting the model parameters of the recognition network architecture through an optimization algorithm until the loss function is minimized to obtain the small target recognition network model.

6. The automatic warning steering method according to claim 5, characterized in that: The loss function is as follows: Where loss represents the loss function, y∈{1,0}, y represents the true target category, p∈[0,1], p represents the estimated probability when y=1, (1-p) represents the estimated probability when y=0, γ represents an adjustable static factor, β represents an adjustable dynamic factor, and e p Represents the amplification method of probability.

7. The automatic warning steering method according to claim 1, characterized in that: The target position information includes lateral position information and distance information, and the target information includes target type; Processing the target information and the target position information to obtain the steering angle includes: using a steering coefficient formula to calculate a steering coefficient according to the vehicle speed, the vehicle stability state, the distance information, the target type and the lateral position information, and obtaining the steering angle through the steering coefficient.

8. The automatic warning steering method according to claim 7, characterized in that: The steering coefficient formula is as follows: Among them, K represents the steering coefficient, V represents the current vehicle speed, and V max represents the maximum design speed of the vehicle, S represents the stability score of the vehicle, S max represents the maximum value of the vehicle's stability score, D represents the distance between the target and the vehicle, and D min represents the minimum distance for triggering an early warning, C represents the target type coefficient, and C max represents the maximum value of the target type coefficient, P represents the lateral position coefficient, P max Represents the maximum value of the lateral position coefficient, and a, b, c, d and e are all weight factors.

9. The automatic warning steering method according to claim 7, characterized in that: Acquiring the steering angle through the steering coefficient includes: calculating the steering angle using a steering angle formula, and the steering angle formula is as follows: θ=K×t 2 ; Among them, K represents the steering coefficient and t represents the remaining steering time.

10. An automatic warning steering system, characterized in that: include: A data capture layer, the data capture layer is used to collect real-time images on the pre-travel path of the vehicle wheel hub; A strategy customization layer, the strategy customization layer includes a small target recognition network model, the small target recognition network model is used to extract features from the real-time image to obtain target information and target position information corresponding to the real-time image; A real-time operation layer is used to process the target information and the target position information to obtain a steering angle, and control the vehicle steering based on the steering angle to change the driving path of the vehicle wheel hub.

Images