Front vehicle following method and system based on artificial intelligence
By using an AI-based vehicle following method, which utilizes vehicle bounding box recognition networks and image processing technology, the interference problem of electromagnetic waves or infrared light when following a vehicle in traffic jams is solved, resulting in better following performance and safety.
Patent Information
- Application Number
- CN202311292764.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-08
- Publication Date
- 2026-01-27
AI Technical Summary
Existing electromagnetic wave or infrared vehicle following methods have poor following performance in traffic jams, are easily interfered with by the signals of surrounding vehicles, affect the vehicle's ability to follow the vehicle in front, and pose a risk of traffic accidents.
An AI-based vehicle following method is adopted. By acquiring the driving record image of the vehicle itself, the bounding box recognition network is used to determine the bounding box and fuzzy distance of the target vehicle. The driving speed is adjusted in combination with the vehicle speed to maintain a suitable distance.
It improves the ability to follow the vehicle in front in traffic jams, enhances safety between vehicles, and reduces the risk of traffic accidents.
Smart Images
Figure CN121415482A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, and specifically to a method and system for following a vehicle based on artificial intelligence. Background Technology
[0002] With the improvement of people's living standards and the continuous increase in the number of private cars, traffic accidents such as vehicle collisions and rear-end collisions are becoming more frequent and serious. In order to reduce the frequency of such traffic accidents and to avoid malicious lane-changing and traffic jams from vehicles behind, maintaining a suitable driving distance from the target vehicle in front is very important. Therefore, it is urgent to determine an intelligent following method.
[0003] Current methods of following other vehicles often use electromagnetic waves or infrared rays for assistance. However, these methods are less effective in traffic jams. If other vehicles in the vicinity also have similar systems, their signals can easily interfere with each other. Therefore, this method not only affects the vehicle's ability to follow other vehicles but also poses a certain risk of traffic accidents during the entire following process. Summary of the Invention
[0004] To address the problem of poor following ability of vehicles using electromagnetic waves or infrared radiation, the present invention aims to provide a vehicle following method and system based on artificial intelligence.
[0005] This invention provides a method for following a vehicle based on artificial intelligence, comprising the following steps:
[0006] Acquire the driving record image of the vehicle at the previous moment and the driving record image at the current moment during the current driving process;
[0007] Based on the driving record images from the previous moment and the current moment during the current driving process of this vehicle, determine the relevant images from the driving record images from the previous moment and the relevant images from the driving record images from the current moment.
[0008] Based on the driving record image from the previous moment, the driving record image from the current moment, and the relevant images from the driving record images from these two moments, the target vehicle and its bounding box in the driving record image from the previous moment, the driving record image from the current moment, and the relevant images from the driving record images from these two moments are determined, and then the bounding box area of the target vehicle in the driving record image from the previous moment and the bounding box area of the target vehicle in the driving record image from the current moment are determined.
[0009] Based on the driving record image from the previous moment, the driving record image from the current moment, and the bounding box and area of the target vehicle in the driving record images from these two moments, determine the fuzzy distance between the current vehicle and the target vehicle from the previous moment and the fuzzy distance between the current vehicle and the target vehicle.
[0010] The average speed of the target vehicle at the current moment is determined based on the vehicle's current speed, the fuzzy distance between the vehicle and the target vehicle at the previous moment, the fuzzy distance between the vehicle and the target vehicle at the current moment, and the time interval between the driving record image at the previous moment and the driving record image at the current moment.
[0011] Based on the fuzzy distance between the current vehicle and the target vehicle at the current moment, and the average speed of the target vehicle at the current moment, determine the speed of the current vehicle between the current moment and the next moment.
[0012] Furthermore, the steps for determining the relevant images from the dashcam footage include:
[0013] Based on the dashcam image, the dashcam image is processed and then input into the vehicle bounding box recognition network to determine the bounding box of each vehicle in the processed dashcam image.
[0014] Based on the number of bounding boxes for each vehicle in the processed dash cam image, it is determined whether the number of bounding boxes for each vehicle in the processed dash cam image meets the processing termination condition. If not, the processed dash cam image is processed again, and the reprocessed dash cam image is input into the vehicle bounding box recognition network to determine the bounding boxes for each vehicle in the reprocessed dash cam image. Based on the number of bounding boxes for each vehicle in the reprocessed dash cam image, it is determined whether the number of bounding boxes for each vehicle in the reprocessed dash cam image meets the processing termination condition. The above steps are repeated until the number of bounding boxes for each vehicle in the dash cam image meets the processing termination condition.
[0015] Each processed driving record image corresponding to the condition that the number of bounding boxes of a vehicle does not meet the processing termination condition is taken as the relevant image of that driving record image.
[0016] Further steps in processing dashcam images include:
[0017] The driving recorder image is downsampled to obtain the downsampled driving recorder image. The downsampled driving recorder image is then upsampled to obtain the upsampled driving recorder image.
[0018] The dashcam image is subjected to histogram equalization to obtain the dashcam image after histogram equalization.
[0019] The edge pixels of each vehicle in the histogram equalization-processed driving record image are determined, and the position of the edge pixels of each vehicle in the upsampled driving record image is determined based on the position of the edge pixels of each vehicle in the histogram equalization-processed driving record image.
[0020] Based on the gray values and positions of the edge pixels of each vehicle in the histogram equalization-processed dash cam image and the positions of the edge pixels of each vehicle in the upsampled dash cam image, the gray values of the edge pixels of each vehicle in the upsampled dash cam image are updated, and the updated upsampled dash cam image is used as the final processed dash cam image.
[0021] Further steps in determining the bounding box area of the target vehicle in the dashcam image include:
[0022] Based on the bounding box of the target vehicle in the dashcam image, determine the area of the first bounding box of the target vehicle in the dashcam image; based on the bounding box of the target vehicle in each related image of the dashcam image, determine the area of the second bounding box of the target vehicle in each related image of the dashcam image.
[0023] The bounding box area of the target vehicle in the dashcam image is determined based on the first bounding box area of the target vehicle in the dashcam image and the second bounding box area of the target vehicle in each related image of the dashcam image.
[0024] Furthermore, the formula for calculating the bounding box area of the target vehicle in the dashcam image is as follows:
[0025]
[0026] Where s is the bounding box area of the target vehicle in the dashcam image, s0 is the first bounding box area of the target vehicle in the dashcam image, and s i is the area of the second bounding box of the target vehicle in the i-th related image of the dashcam image, and k-1 is the number of related images of the dashcam image.
[0027] Further steps in determining the fuzzy distance between the current vehicle and the target vehicle include:
[0028] Based on the dashcam image and the bounding box of the target vehicle in the dashcam image, determine the perpendicular distance from the center point of the bounding box of the target vehicle in the dashcam image to the bottom edge of the dashcam image.
[0029] Obtain the correlation coefficient of the perpendicular distance of the bounding box of the target vehicle and the correlation coefficient of the bounding box area of the target vehicle. Based on the correlation coefficient of the perpendicular distance of the target vehicle, the correlation coefficient of the bounding box area of the target vehicle, and the perpendicular distance and bounding box area of the bounding box of the target vehicle in the driving record image, determine the fuzzy distance between the current vehicle and the target vehicle.
[0030] Furthermore, the steps for determining the correlation coefficient between the perpendicular distance of the target vehicle's bounding box and the correlation coefficient between the area of the target vehicle's bounding box include:
[0031] Acquire the driving record images of this vehicle at each moment during its historical driving process, the position of the front of this vehicle in the driving record images at each moment, and the position of the bounding box of the target vehicle;
[0032] Based on the position of the front of the vehicle and the position of the bounding box of the target vehicle in the driving record image at each moment, filter each target driving record image and obtain the actual distance between the vehicle and the target vehicle in each target driving record image;
[0033] Based on each target driving record image, determine the bounding box of the target vehicle in each target driving record image;
[0034] Based on each target driving record image, the actual distance between the vehicle and the target vehicle in each target driving record image, and the bounding box of the target vehicle in each target driving record image, determine the correlation coefficient of the vertical distance of the bounding box of the target vehicle and the correlation coefficient of the area of the bounding box of the target vehicle.
[0035] Furthermore, the steps for determining the average speed of the target vehicle at the current moment include:
[0036] The difference between the current fuzzy distance between the vehicle and the target vehicle and the previous fuzzy distance between the vehicle and the target vehicle is calculated. Based on this difference and the time interval between the driving record image of the previous moment and the driving record image of the current moment, the relative speed between the vehicle and the target vehicle at the current moment is determined.
[0037] Determine the average speed of the target vehicle at the current moment based on the vehicle's current speed and the relative speed between the vehicle and the target vehicle.
[0038] Furthermore, the steps for determining the vehicle's speed between the current moment and the next moment include:
[0039] If the fuzzy distance between the current vehicle and the target vehicle is less than the first preset travel distance value, then control the vehicle's travel speed between the current time and the next time to be lower than the average travel speed of the target vehicle at the current time.
[0040] If the fuzzy distance between the current vehicle and the target vehicle is greater than the first preset travel distance value and less than the second preset travel distance value, then control the vehicle's travel speed between the current time and the next time to be equal to the average travel speed of the target vehicle at the current time.
[0041] If the fuzzy distance between the current vehicle and the target vehicle is greater than the second preset travel distance value, then the vehicle's travel speed between the current time and the next time will be controlled to be higher than the average travel speed of the target vehicle at the current time.
[0042] The present invention also provides an artificial intelligence-based vehicle following system, including a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement an artificial intelligence-based vehicle following method.
[0043] The present invention has the following beneficial effects:
[0044] This invention acquires driving record images from the previous moment and the current moment during the vehicle's current driving process, determines relevant images from both the previous and current moments, and then identifies the target vehicle and its bounding box within these images, as well as the relevant images from both moments. This allows for the determination of the bounding box area of the target vehicle in the driving record images from both moments. Considering that special weather conditions can affect the size of the bounding box area of the target vehicle in the driving record images, this invention comprehensively acquires and processes the bounding boxes of the target vehicle from all relevant images in the driving record images when determining the bounding box area. This enhances the accuracy of the bounding box area of the target vehicle in the driving record images from both moments, increases the precision of the fuzzy distance between the vehicle and the target vehicle, and thus improves the vehicle's ability to follow the vehicle in front.
[0045] This invention determines the fuzzy distance between the vehicle and the target vehicle at the previous moment and the current moment by using the dashcam images from the previous moment, the current moment, and the bounding boxes and areas of the target vehicle in these two moment images. It then determines the average speed of the target vehicle at the current moment, and consequently, the vehicle's speed between the current and next moment. By using the fuzzy distance and the target vehicle's average speed at the current moment, this invention adjusts the vehicle's speed in real time between the current and next moment, ensuring that the fuzzy distance between the vehicle and the target vehicle remains within a suitable range, effectively improving the vehicle's following ability. Attached Figure Description
[0046] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0047] Figure 1 This is a flowchart of a vehicle following method based on artificial intelligence according to the present invention. Detailed Implementation
[0048] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solution proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0050] This embodiment provides an artificial intelligence-based vehicle following method. The application scenario of this method is: when a vehicle is driving on the road, the vehicle's vehicle following system assists the driver in following the vehicle, such as... Figure 1 As shown, the steps of this method include:
[0051] (1) Obtain the driving record image of the previous moment and the driving record image of the current moment during the current driving process of this vehicle.
[0052] In this embodiment, the dashcam installed in the vehicle can capture driving record images at every moment during the vehicle's current driving process, with each driving record image being m×n in size. To facilitate subsequent image processing, this embodiment acquires the driving record images from the previous moment and the current moment during the vehicle's current driving process, and then performs grayscale processing on both images to obtain grayscale-processed driving record images from the previous moment and the current moment.
[0053] It should be noted that the vehicle following method in this embodiment is based on the driving record image of the previous moment and the driving record image of the current moment. It uses image processing technology to enable the vehicle to follow the target vehicle in front. Compared with the existing vehicle following method, the vehicle following method based on image processing technology in this embodiment will not be affected by the surrounding vehicles when in traffic jams, and the vehicle following effect of this method will be better.
[0054] (2) Based on the driving record images from the previous moment and the current moment during the vehicle's current driving process, determine the relevant images from the driving record images from the previous moment and the relevant images from the current moment. The steps include:
[0055] (2-1) Based on the driving record image of the previous moment and the driving record image of the current moment, process the driving record images of the two moments, and input the processed driving record images of the two moments into the vehicle bounding box recognition network to determine the bounding box of each vehicle in the processed driving record images of the two moments.
[0056] In this embodiment, the driving record images from the previous moment and the current moment are processed. The details of the driving record image processing are described in steps (2-1-1) to (2-1-4), and will not be described again here. The processed driving record images from the previous moment and the current moment are sequentially input into a pre-built and trained vehicle bounding box recognition network to obtain the bounding boxes of each vehicle in the driving record image from the previous moment and the bounding boxes of each vehicle in the driving record image from the current moment.
[0057] It should be noted that the vehicle bounding box recognition network in this embodiment adopts an Encoder-Decoder structure. This network first encodes the input dashcam image and then decodes it to obtain the bounding boxes of each vehicle in the dashcam image and the corresponding confidence scores of each bounding box. Based on the confidence scores of each vehicle's bounding boxes, the bounding box with the highest confidence score for each vehicle is obtained. It is then determined whether the confidence score of the bounding box with the highest confidence score for each vehicle is greater than a set confidence threshold. In this embodiment, the confidence threshold is set to 0.85. If the confidence score of the bounding box with the highest confidence score for a certain vehicle is not greater than the set confidence threshold, it means that the vehicle has no bounding box. If the confidence score of the bounding box with the highest confidence score for a certain vehicle is greater than the set confidence threshold, the network outputs the bounding box with the highest confidence score for that vehicle, thus obtaining the bounding boxes of each vehicle in the dashcam image. The process of constructing and training the vehicle bounding box recognition network is prior art and is not within the scope of protection of this invention, and will not be described in detail here.
[0058] The clarity of dashcam images captured during vehicle operation is affected by weather conditions. For example, if the weather is cloudy when the dashcam image is captured, the clarity of the dashcam image will be poor. In other words, the image portions of each vehicle in the dashcam image will be similar in color to the background color. As a result, the edges of the image portions of each vehicle in the grayscale dashcam image will be blurry, leading to inaccurate bounding boxes for each vehicle output by the vehicle bounding box recognition network.
[0059] To mitigate the impact of this situation on the bounding box size of each vehicle in the subsequently determined driving record images, this embodiment performs relevant processing on the driving record images from the previous moment and the current moment. The processing steps include:
[0060] (2-1-1) The driving record images from the previous moment and the current moment are downsampled to obtain the downsampled driving record images from the previous moment and the current moment. Then, the downsampled driving record images from the previous moment and the current moment are upsampled to obtain the upsampled driving record images from the previous moment and the current moment. The downsampling and upsampling processes are existing technologies and are not within the scope of this invention; therefore, they will not be described in detail here.
[0061] In this embodiment, the downsampled dash cam image is 1 / 4 the size of the original dash cam image. For example, if the original dash cam image is 1024×1024, after downsampling it becomes 512×512, and after further downsampling it becomes 256×256, and so on. Downsampling will result in the loss of some image information, making the entire dash cam image blurry, but the basic shape of the entire dash cam image remains unchanged, and it also reduces most of the noise in the dash cam image.
[0062] In addition, to ensure that the size of the downsampled dashcam image is consistent with the size of the dashcam image before downsampling, this embodiment performs upsampling on the downsampled dashcam image.
[0063] (2-1-2) The driving record images from the previous moment and the current moment are subjected to histogram equalization processing to obtain the histogram equalized driving record images from the previous moment and the current moment. Histogram equalization processing is prior art and is not within the scope of protection of this invention, and will not be described in detail here.
[0064] This embodiment performs histogram equalization on the driving record images from the previous moment and the current moment, which enhances the contrast of the driving record images and strengthens the edge information of each vehicle in the driving record images.
[0065] (2-1-3) Determine the edge pixels of each vehicle in the previous time-of-flight driving record image and the current time-of-flight driving record image after histogram equalization, and determine the position of the edge pixels of each vehicle in the previous time-of-flight driving record image and the current time-of-flight driving record image after upsampling based on the position of the edge pixels of each vehicle in the previous time-of-flight driving record image and the current time-of-flight driving record image after histogram equalization.
[0066] It should be noted that each pixel in the previous and current driving record images after histogram equalization is compared one-to-one with each pixel in the previous and current driving record images after upsampling.
[0067] This embodiment obtains the edge pixels of each vehicle in the driving record images at two time points after histogram equalization processing through edge detection, and then obtains the position of the edge pixels of each vehicle. Based on the position of the edge pixels of each vehicle in the driving record images at two time points after histogram equalization processing, the position of the edge pixels of each vehicle in the driving record images at two time points after upsampling processing is determined.
[0068] (2-1-4) Based on the gray values and positions of the edge pixels of each vehicle in the previous and current driving record images after histogram equalization and the positions of the edge pixels of each vehicle in the previous and current driving record images after upsampling, update the gray values of the edge pixels of each vehicle in the previous and current driving record images after upsampling. If the gray values of the edge pixels of each vehicle in the previous and current driving record images after upsampling are equal to the gray values of the edge pixels of each vehicle in the previous and current driving record images after histogram equalization, use the updated upsampling driving record images of the previous and current driving record images as the final processed driving record images of the previous and current driving record images.
[0069] (2-2) Based on the number of bounding boxes of each vehicle in the processed previous moment's driving record image, determine whether the number of bounding boxes of each vehicle in the processed previous moment's driving record image meets the processing termination condition. If not, process the processed previous moment's driving record image again and input the reprocessed previous moment's driving record image into the vehicle bounding box recognition network to determine the bounding boxes of each vehicle in the reprocessed previous moment's driving record image. Based on the number of bounding boxes of each vehicle in the reprocessed previous moment's driving record image, determine whether the number of bounding boxes of each vehicle in the reprocessed previous moment's driving record image meets the processing termination condition. Repeat the above steps until the number of bounding boxes of each vehicle in the previous moment's driving record image meets the processing termination condition.
[0070] Based on the number of bounding boxes for each vehicle in the processed dashcam image at the current moment, it is determined whether the number of bounding boxes for each vehicle in the processed dashcam image at the current moment meets the processing termination condition. If not, the processed dashcam image at the current moment is processed again, and the reprocessed dashcam image at the current moment is input into the vehicle bounding box recognition network to determine the bounding boxes for each vehicle in the reprocessed dashcam image at the current moment. Based on the number of bounding boxes for each vehicle in the reprocessed dashcam image at the current moment, it is determined whether the number of bounding boxes for each vehicle in the reprocessed dashcam image at the current moment meets the processing termination condition. The above steps are repeated until the number of bounding boxes for each vehicle in the dashcam image at the current moment meets the processing termination condition.
[0071] It should be noted that the processing termination condition mentioned above refers to the fact that the number of bounding boxes of each vehicle in the processed dashcam image is less than the number of bounding boxes of each vehicle in the unprocessed dashcam image.
[0072] (2-3) Take the driving record images of the previous time when the number of bounding boxes of the vehicles does not meet the processing termination condition as the relevant images of the driving record images of the previous time. Take the driving record images of the current time when the number of bounding boxes of the vehicles does not meet the processing termination condition as the relevant images of the driving record images of the current time. That is, take the driving record images in the processed driving record images where the number of bounding boxes of each vehicle is equal to the number of bounding boxes of each vehicle in the driving record images before processing as the relevant images of the driving record images.
[0073] At this point, through steps (2-1) to (2-3), the relevant images of the driving record image at the previous moment and the relevant images of the driving record image at the current moment are determined respectively.
[0074] (3) Based on the driving record image from the previous moment, the driving record image from the current moment, and the relevant images from both moments, determine the target vehicle and its bounding box in the driving record image from the previous moment, the driving record image from the current moment, and the relevant images from both moments, and then determine the bounding box area of the target vehicle in the driving record image from the previous moment and the bounding box area of the target vehicle in the driving record image from the current moment. The steps include:
[0075] (3-1) Based on the driving record image from the previous moment, the driving record image from the current moment, and the relevant images from both moments, determine the target vehicle and its bounding box in the driving record image from the previous moment, the driving record image from the current moment, and the relevant images from both moments. The steps include:
[0076] (3-1-1) Construct a coordinate system on the driving record image of the previous moment, the driving record image of the current moment, and the relevant images of the driving record image of the previous moment and the driving record image of the current moment.
[0077] In this embodiment, taking the construction of a coordinate system on the current moment's driving record image as an example, the coordinate system is constructed with the lower left corner of the current moment's driving record image as the origin, and the center point of the bounding box of each vehicle in the current moment's driving record image is denoted as h(x1). z ,y1 zLet h(x10, y10) be the center point of the bottom edge of the current dashcam image. Following the method of constructing a coordinate system on the current dashcam image, construct coordinate systems on the previous dashcam image, the previous dashcam image, and the relevant images of the current dashcam image. Mark the center points of the bounding boxes of each vehicle in the previous dashcam image, the previous dashcam image, and the relevant images of the current dashcam image, as well as the center point of the bottom edge of their respective dashcam images.
[0078] It should be noted that in this embodiment, the vehicle's dashcam is installed in the center of the front seat. The dashcam in this position does not obstruct the view ahead and has the best monitoring field of view. Furthermore, the front of the vehicle in the dashcam image captured by the dashcam in this position is symmetrically distributed in the image.
[0079] (3-1-2) Based on the driving record images of the previous moment and the current moment, and the positions of the bounding boxes of each vehicle in the driving record images of the previous moment and the current moment, determine the Euclidean distance and the difference in the horizontal coordinate of each vehicle in the driving record images of the previous moment and the current moment, and then determine the target vehicle and the bounding box of the target vehicle in the driving record images of the previous moment and the current moment.
[0080] This embodiment takes the process of determining the target vehicle and its bounding box in the current driving record image as an example. It uses the center point h(x1) within the bounding box of each vehicle in the current driving record image as an example. z ,y1 z Given the center point h(x10, y10) of the bottom edge of the current dash cam image, calculate the Euclidean distance between the center point within the bounding box of each vehicle in the current dash cam image and the center point of the bottom edge of the current dash cam image. The calculation formula is as follows:
[0081]
[0082] Where d1 is the Euclidean distance between the center point of the bounding box of each vehicle in the current driving record image and the center point of the bottom edge of the current driving record image, and x1 z Let y1 be the x-coordinate of the center point of the bounding box of each vehicle in the current driving record image. z x10 is the ordinate of the center point of the bounding box of each vehicle in the current driving record image, x10 is the abscissa of the center point of the bottom edge of the current driving record image, and y10 is the ordinate of the center point of the bottom edge of the current driving record image.
[0083] The x-coordinate difference between the center point of the bounding box of each vehicle in the current dashcam image and the center point of the bottom edge of the current dashcam image is calculated using the following formula:
[0084] d2=|x1 z -x10|
[0085] Where d2 is the difference in x-coordinate between the center point of the bounding box of each vehicle in the current driving record image and the center point of the bottom edge of the current driving record image, and x1 z x10 is the x-coordinate of the center point of the bounding box of each vehicle in the current driving record image, and x10 is the x-coordinate of the center point of the bottom edge of the current driving record image.
[0086] It should be noted that since the target vehicle being followed by this vehicle is one located approximately directly in front of it, this embodiment only needs to consider vehicles in front of this vehicle and not vehicles on either side. Therefore, this embodiment sets a suitable range for the difference d2 between the x-coordinate of the center point of the bounding box of each vehicle in the current driving record image and the center point of the bottom edge of the current driving record image. n is the length of the dashcam image.
[0087] according to From the bounding boxes of each vehicle in the current driving record image, filter out the bounding boxes of each vehicle that meet the specified range, and then obtain the Euclidean distance d1 corresponding to the bounding boxes of each vehicle that meets the specified range. Sort the Euclidean distances d1 in a certain order, and obtain the bounding box of the vehicle corresponding to the smallest Euclidean distance d1. The bounding box of this vehicle is the bounding box of the target vehicle in the current driving record image.
[0088] The target vehicle and its bounding box in the dashcam image at the previous moment are obtained by referring to the method used to determine the target vehicle and its bounding box in the dashcam image at the current moment.
[0089] (3-1-3) Based on the positions of the bounding boxes of each vehicle in the relevant images of the driving record images of the previous and current times, determine the Euclidean distance and the difference in the horizontal coordinate of each vehicle in the relevant images of the driving record images of the previous and current times, and then determine the target vehicle and the bounding box of the target vehicle in the relevant images of the driving record images of the previous and current times.
[0090] In this embodiment, the method of determining the target vehicle and the bounding box of the target vehicle in each related image of the driving recorder image at the previous moment and the current moment is the same as the method of determining the target vehicle and the bounding box of the target vehicle in the driving recorder image at the current moment in step (3-1-2). Referring to the method of determining the target vehicle and the bounding box of the target vehicle in the driving recorder image at the current moment, the target vehicle and the bounding box of the target vehicle in each related image of the driving recorder image at the previous moment and the current moment are obtained.
[0091] (3-2) Based on the vehicle recording image from the previous moment, the vehicle recording image from the current moment, and the relevant images from both moments, determine the bounding box area of the target vehicle in the vehicle recording image from the previous moment and the bounding box area of the target vehicle in the vehicle recording image from the current moment. The steps include:
[0092] (3-2-1) Based on the bounding boxes of the target vehicle in the driving record image of the previous moment and the driving record image of the current moment, determine the area of the first bounding box of the target vehicle in the driving record image of the previous moment and the driving record image of the current moment. Based on the bounding boxes of the target vehicle in each related image of the driving record image of the previous moment and the driving record image of the current moment, determine the area of the second bounding box of the target vehicle in each related image of the driving record image of the previous moment and the driving record image of the current moment.
[0093] In this embodiment, based on the positions of the four corner points of the bounding box of the target vehicle in the previous moment's driving record image and the current moment's driving record image, the length and width of the bounding box of the target vehicle in the previous moment's driving record image and the current moment's driving record image are determined. The length and width of the bounding box of the target vehicle in the previous moment's driving record image and the current moment's driving record image are multiplied to obtain the area of the first bounding box of the target vehicle in the previous moment's driving record image and the current moment's driving record image. The area of the first bounding box is denoted as s0.
[0094] Based on the positions of the four corner points of the target vehicle's bounding box in each relevant image from the previous and current time-series dashcam images, the length and width of the target vehicle's bounding box in each relevant image are determined. The length and width of the target vehicle's bounding box in each relevant image are then multiplied to obtain the area of the second bounding box of the target vehicle in each relevant image from the previous and current time-series dashcam images. This second bounding box area is denoted as s. i .
[0095] (3-2-2) Based on the area of the first bounding box of the target vehicle in the driving record image of the previous moment and the driving record image of the current moment, and the area of the second bounding box of the target vehicle in each related image of the driving record image of the previous moment and the driving record image of the current moment, determine the area of the bounding box of the target vehicle in the driving record image of the previous moment and the driving record image of the current moment.
[0096] In this embodiment, taking the calculation of the bounding box area of the target vehicle in the current driving record image as an example, the calculation formula is as follows:
[0097]
[0098] Where s is the bounding box area of the target vehicle in the current driving record image, s0 is the first bounding box area of the target vehicle in the current driving record image, and s i Let k be the area of the second bounding box of the target vehicle in the i-th related image of the current driving record image, and k-1 be the number of related images of the current driving record image.
[0099] The bounding box area of the target vehicle in the dashcam image at the previous moment is obtained by referring to the method used to determine the bounding box area of the target vehicle in the dashcam image at the current moment.
[0100] (4) Based on the driving record image from the previous moment, the driving record image from the current moment, and the bounding box and area of the target vehicle in the driving record images from these two moments, determine the fuzzy distance between the current vehicle and the target vehicle from the previous moment and the fuzzy distance between the current vehicle and the target vehicle from the previous moment. The steps include:
[0101] (4-1) Based on the vehicle recording image of the previous moment, the vehicle recording image of the current moment, and the bounding box of the target vehicle in the vehicle recording images of the two moments, determine the perpendicular distance from the center point of the bounding box of the target vehicle in the vehicle recording image of the previous moment and the vehicle recording image of the current moment to the bottom edge of the vehicle recording image.
[0102] In this embodiment, a perpendicular line is drawn from the center point of the bounding box of the target vehicle in the previous moment's driving record image and the current moment's driving record image to the bottom edge of the entire driving record image. This perpendicular line distance is obtained from the center point of the bounding box of the target vehicle in the previous moment's driving record image and the current moment's driving record image to the bottom edge of the driving record image. This perpendicular line distance can measure the blur distance between the current vehicle and the target vehicle. The larger the perpendicular line distance, the larger the blur distance between the current vehicle and the target vehicle.
[0103] (4-2) Obtain the correlation coefficient of the perpendicular distance of the bounding box of the target vehicle and the correlation coefficient of the bounding box area of the target vehicle. Based on the correlation coefficient of the perpendicular distance of the target vehicle, the correlation coefficient of the bounding box area of the target vehicle, and the perpendicular distance and bounding box area of the bounding box of the target vehicle in the driving record image of the previous moment and the driving record image of the current moment, determine the fuzzy distance between the current vehicle and the target vehicle in the previous moment and the current moment.
[0104] This embodiment takes determining the fuzzy distance between the current vehicle and the target vehicle as an example, and the calculation formula is as follows:
[0105]
[0106] Where l is the fuzzy distance between the current vehicle and the target vehicle at the current moment, a1 is the correlation coefficient of the perpendicular distance of the bounding box of the target vehicle in the current driving record image, a2 is the correlation coefficient of the area of the bounding box of the target vehicle in the current driving record image, l0 is the perpendicular distance of the bounding box of the target vehicle in the current driving record image, and s is the area of the bounding box of the target vehicle in the current driving record image.
[0107] It should be noted that, based on the principle of perspective in cameras, the farther away a target vehicle is, the smaller its bounding box; conversely, the closer the target vehicle is, the larger its bounding box. Therefore, the area of the target vehicle's bounding box can measure the fuzzy distance between the two vehicles; the larger the area of the target vehicle's bounding box, the smaller the fuzzy distance between them. Referring to the formula for calculating the fuzzy distance between the two vehicles at the current moment, the fuzzy distance between the two vehicles at the previous moment can be obtained.
[0108] In addition, the correlation coefficient a1 of the perpendicular distance of the target vehicle's bounding box and the correlation coefficient a2 of the area of the target vehicle's bounding box are predetermined, and the determination steps include:
[0109] (4-2-1) Obtain the driving record image of the vehicle at each moment during the historical driving process, the position of the front of the vehicle in the driving record image at each moment, and the position of the bounding box corresponding to the target vehicle.
[0110] This embodiment acquires driving record images of the vehicle at each moment during a certain historical driving process from the vehicle's dashcam memory card, and inputs each moment's driving record image into a pre-constructed vehicle front recognition network and a vehicle bounding box recognition network to obtain the vehicle's front portion and the bounding box of the target vehicle in the driving record image at each moment, thereby obtaining the position of the vehicle's front and the position of the bounding box corresponding to the target vehicle in the driving record image at each moment.
[0111] It should be noted that, in this embodiment, the front of the vehicle in the driving record image at each moment during a certain historical driving process is symmetrical in the image. This type of driving record image is helpful for accurately determining the fuzzy distance between the vehicle and the target vehicle.
[0112] (4-2-2) Based on the position of the front of the vehicle and the position of the bounding box of the target vehicle in the driving record image at each time moment, filter out each target driving record image in the driving record image at each time moment, and obtain the actual distance between the vehicle and the target vehicle in each target driving record image.
[0113] In this embodiment, based on the position of the vehicle's front end in the driving record image at each time moment, edge detection is performed on the vehicle's front end in the driving record image at each time moment to obtain a fitted curve of the vehicle's front end in the driving record image at each time moment. The fitted curve is then reflected on the image to obtain the curve function in the driving record image at each time moment, and the curve function is denoted as y1=f1(x). Based on the position of the bounding box of the target vehicle in the driving record image at each time moment, the position of each pixel on the bottom edge of the bounding box of the target vehicle at each time moment is determined, and then the linear equation of the bottom edge of the bounding box of the target vehicle in the driving record image at each time moment is determined, and the linear equation is denoted as y=f(x).
[0114] In this embodiment, the center point coordinates h(x10, y10) of the bottom edge of the driving record image at each time moment are obtained by step (3-1-1). A perpendicular line is drawn from the center point of the bottom edge of the driving record image at each time moment to the bottom edge of the driving record image at each time moment. This perpendicular line intersects the curve function and the straight line equation in the driving record image at each time moment at a point. The abscissa of the perpendicular line is substituted into the curve function and the straight line equation in the driving record image at each time moment to obtain the intersection point coordinates h(x10, f1(x10)) of the curve function and the intersection point coordinates h(x10, f(x10)) of the straight line equation in the driving record image at each time moment.
[0115] According to the intersection coordinates h(x10, f1(x10)) of the curve function and the intersection coordinates h(x10, f(x10)) of the straight line equation in the driving record image at each moment, each target driving record image in the driving record image at each moment is screened. The detailed screening steps are as follows: If the distance between the intersection coordinates h(x10, f1(x10)) of the curve function and the intersection coordinates h(x10, f(x10)) of the straight line equation in the driving record image at a certain moment is within the preset distance range, then the driving record image at this moment is the target driving record image, so as to obtain the actual distances between the vehicle itself and the target vehicle in the driving record image at this moment. In this embodiment, the preset distance range is f(x10) - 5 < f1(x10) < f(x10) + 5. According to the above screening steps, each target driving record image in the driving record image at each moment and the actual distances between the vehicle itself and the target vehicle in each of these target driving record images are obtained.
[0116] It should be noted that the actual distances between the vehicle itself and the target vehicle in each target driving record image are determined when the driving record images at each moment in the historical driving process are collected, and the actual distances between each vehicle itself and the target vehicle are not unique, that is, each target driving record image has a corresponding actual distance between the vehicle itself and the target vehicle.
[0117] (4-2-3) According to each target driving record image, the actual distances between the vehicle itself and the target vehicle in each target driving record image, and the bounding boxes of the target vehicles in each target driving record image, respectively determine the correlation coefficient of the perpendicular distance of the bounding box corresponding to the target vehicle and the correlation coefficient of the area of the bounding box of the target vehicle.
[0118] In this embodiment, according to each target driving record image and the bounding boxes of the target vehicles in each target driving record image, and referring to step (4-1) and step (3-2), respectively determine the perpendicular distance from the center point inside the bounding box of the target vehicle in each target driving record image to the bottom edge of the driving record image and the area of the bounding box of the target vehicle in each target driving record image.
[0119] According to the perpendicular distance from the center point inside the bounding box of the target vehicle in each target driving record image to the bottom edge of the driving record image and the actual distances between the vehicle itself and the target vehicle in each target driving record image, determine the correlation coefficient of the perpendicular distance of the bounding box corresponding to the target vehicle. The calculation formula is as follows:
[0120]
[0121] Among them, a1 is the correlation coefficient of the perpendicular distance of the bounding box of the target vehicle, l′ l0 represents the actual distance between the vehicle and the target vehicle in each target driving record image, l0 represents the perpendicular distance from the center point of the bounding box of the target vehicle in each target driving record image to the bottom edge of the driving record image, and mean() is the mean function.
[0122] Based on the bounding box area of the target vehicle in each target dashcam image and the actual distance between the current vehicle and the target vehicle in each target dashcam image, the correlation coefficient of the bounding box area of the target vehicle is determined. The calculation formula is as follows:
[0123] a2=mean(l ′ mean(s)
[0124] Where a2 is the correlation coefficient of the bounding box area of the target vehicle, l ′ Let be the actual distance between the vehicle and the target vehicle in each target driving record image, s be the bounding box area of the target vehicle in each target driving record image, and mean() be the mean function.
[0125] (5) Based on the current vehicle speed, the fuzzy distance between the vehicle and the target vehicle at the previous moment, the fuzzy distance between the vehicle and the target vehicle at the current moment, and the time interval between the driving record image at the previous moment and the driving record image at the current moment, determine the average speed of the target vehicle at the current moment. The steps include:
[0126] (5-1) Subtract the fuzzy distance between the current vehicle and the target vehicle from the fuzzy distance between the current vehicle and the target vehicle at the previous moment. Based on this difference and the time interval between capturing the driving record image at the previous moment and capturing the driving record image at the current moment, determine the relative speed between the current vehicle and the target vehicle. The calculation formula is as follows:
[0127]
[0128] Where v0 is the relative speed between the current vehicle and the target vehicle at the current moment, l11 is the fuzzy distance between the current vehicle and the target vehicle at the previous moment, l21 is the fuzzy distance between the current vehicle and the target vehicle at the current moment, and Δt is the time interval between the driving record image at the previous moment and the driving record image at the current moment when the image is captured.
[0129] (5-2) Based on the current speed of the vehicle and the relative speed between the vehicle and the target vehicle, determine the average speed of the target vehicle at the current moment. The calculation formula is as follows:
[0130] v20 = v10 - v0
[0131] Where v20 is the average speed of the target vehicle at the current moment, v10 is the speed of the current vehicle, and v0 is the relative speed between the current vehicle and the target vehicle.
[0132] (6) Determine the speed of the vehicle between the current time and the next time based on the fuzzy distance between the vehicle and the target vehicle at the current time and the average speed of the target vehicle at the current time.
[0133] First, it should be noted that the purpose of determining the vehicle's speed between the current moment and the next moment is to control the fuzzy distance between the vehicle and the target vehicle, keeping the fuzzy distance between them within a suitable range. This includes the following details:
[0134] (6-1) If the fuzzy distance between the current vehicle and the target vehicle is less than the first travel distance preset value L1, then control the travel speed of the current vehicle between the current time and the next time to be lower than the average travel speed of the target vehicle at the current time, so that the fuzzy distance between the current vehicle and the target vehicle gradually increases and tends to a suitable travel distance.
[0135] (6-2) If the fuzzy distance between the current vehicle and the target vehicle is greater than the first preset travel distance value L1 and less than the second preset travel distance value L2, then control the travel speed of the current vehicle between the current time and the next time to be equal to the average travel speed of the target vehicle at the current time, so that the fuzzy distance between the current vehicle and the target vehicle is kept within a suitable travel distance range.
[0136] (6-3) If the fuzzy distance between the current vehicle and the target vehicle is greater than the second travel distance preset value L2, then control the travel speed of the current vehicle between the current time and the next time to be higher than the average travel speed of the target vehicle at the current time, so that the fuzzy distance between the current vehicle and the target vehicle gradually shortens and tends to a suitable travel distance.
[0137] It should be noted that since the average speed of the target vehicle at the current moment is obtained by combining the dashcam images from the previous moment and the current moment, the speed of the vehicle corresponding to the first dashcam image is determined in this embodiment by using the blurred distance between the vehicle and the target vehicle in the first dashcam image and the vehicle's speed at the current moment. The details include:
[0138] If the blurred distance between the vehicle and the target vehicle in the first driving record image is less than the first preset driving distance value L1, then the driving speed of the vehicle between the current moment and the next moment will be controlled to be lower than the driving speed of the vehicle at the current moment.
[0139] If the blurred distance between the vehicle and the target vehicle in the first driving record image is greater than the first preset driving distance value L1 and less than the second preset driving distance value L2, then the driving speed of the vehicle between the current moment and the next moment is controlled to be equal to the driving speed of the vehicle at the current moment.
[0140] If the blurred distance between the vehicle and the target vehicle in the first driving record image is greater than the second driving distance preset value L2, then the driving speed of the vehicle between the current moment and the next moment will be controlled to be higher than the driving speed of the vehicle at the current moment.
[0141] This embodiment also provides an AI-based vehicle following system, including a processor and a memory. The processor processes instructions stored in the memory to implement the AI-based vehicle following method described above. This AI-based vehicle following system acquires image information via a dashcam, sends the image information to the processor, and uses the instructions stored in the memory to process the image information in the processor, thereby enabling the vehicle to follow the target vehicle and expanding the functionality of the dashcam.
[0142] This invention improves the accuracy of the bounding box area of the target vehicle in the driving record image by determining the target vehicle and its bounding box in each related image of the driving record image. This enhances the accuracy of the fuzzy distance between the vehicle and the target vehicle, ensures the correctness of the vehicle's speed between the current moment and the next moment, and ultimately improves the vehicle's ability to follow the vehicle in front.
[0143] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0144] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0145] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for following a vehicle based on artificial intelligence, characterized in that, Includes the following steps: Acquire the driving record image of the vehicle at the previous moment and the driving record image at the current moment during the current driving process; Based on the driving record images from the previous moment and the current moment during the current driving process of this vehicle, determine the relevant images from the driving record images from the previous moment and the relevant images from the driving record images from the current moment. Based on the driving record image from the previous moment, the driving record image from the current moment, and the relevant images from the driving record images from these two moments, the target vehicle and its bounding box in the driving record image from the previous moment, the driving record image from the current moment, and the relevant images from the driving record images from these two moments are determined, and then the bounding box area of the target vehicle in the driving record image from the previous moment and the bounding box area of the target vehicle in the driving record image from the current moment are determined. Based on the driving record image from the previous moment, the driving record image from the current moment, and the bounding box and area of the target vehicle in the driving record images from these two moments, determine the fuzzy distance between the current vehicle and the target vehicle from the previous moment and the fuzzy distance between the current vehicle and the target vehicle. The average speed of the target vehicle at the current moment is determined based on the vehicle's current speed, the fuzzy distance between the vehicle and the target vehicle at the previous moment, the fuzzy distance between the vehicle and the target vehicle at the current moment, and the time interval between the driving record image at the previous moment and the driving record image at the current moment. Based on the fuzzy distance between the current vehicle and the target vehicle at the current moment, and the average speed of the target vehicle at the current moment, determine the speed of the current vehicle between the current moment and the next moment.
2. The method for following a vehicle based on artificial intelligence according to claim 1, characterized in that, The steps for determining the relevant images in a dashcam recording include: Based on the dashcam image, the dashcam image is processed and then input into the vehicle bounding box recognition network to determine the bounding box of each vehicle in the processed dashcam image. Based on the number of bounding boxes for each vehicle in the processed dash cam image, it is determined whether the number of bounding boxes for each vehicle in the processed dash cam image meets the processing termination condition. If not, the processed dash cam image is processed again, and the reprocessed dash cam image is input into the vehicle bounding box recognition network to determine the bounding boxes for each vehicle in the reprocessed dash cam image. Based on the number of bounding boxes for each vehicle in the reprocessed dash cam image, it is determined whether the number of bounding boxes for each vehicle in the reprocessed dash cam image meets the processing termination condition. The above steps are repeated until the number of bounding boxes for each vehicle in the dash cam image meets the processing termination condition. Each processed driving record image corresponding to the condition that the number of bounding boxes of a vehicle does not meet the processing termination condition is taken as the relevant image of that driving record image.
3. The method for following a vehicle based on artificial intelligence according to claim 2, characterized in that, The steps for processing dashcam images include: The driving recorder image is downsampled to obtain the downsampled driving recorder image. The downsampled driving recorder image is then upsampled to obtain the upsampled driving recorder image. The dashcam image is subjected to histogram equalization to obtain the dashcam image after histogram equalization. The edge pixels of each vehicle in the histogram equalization-processed driving record image are determined, and the position of the edge pixels of each vehicle in the upsampled driving record image is determined based on the position of the edge pixels of each vehicle in the histogram equalization-processed driving record image. Based on the gray values and positions of the edge pixels of each vehicle in the histogram equalization-processed dash cam image and the positions of the edge pixels of each vehicle in the upsampled dash cam image, the gray values of the edge pixels of each vehicle in the upsampled dash cam image are updated, and the updated upsampled dash cam image is used as the final processed dash cam image.
4. The method for following a vehicle based on artificial intelligence according to claim 1, characterized in that, The steps for determining the bounding box area of a target vehicle in a dashcam image include: Based on the bounding box of the target vehicle in the dashcam image, determine the area of the first bounding box of the target vehicle in the dashcam image; based on the bounding box of the target vehicle in each related image of the dashcam image, determine the area of the second bounding box of the target vehicle in each related image of the dashcam image. The bounding box area of the target vehicle in the dashcam image is determined based on the first bounding box area of the target vehicle in the dashcam image and the second bounding box area of the target vehicle in each related image of the dashcam image.
5. The method for following a vehicle based on artificial intelligence according to claim 4, characterized in that, Formula for calculating the bounding box area of a target vehicle in a dashcam image: Where s is the bounding box area of the target vehicle in the dashcam image, s0 is the first bounding box area of the target vehicle in the dashcam image, and s i is the area of the second bounding box of the target vehicle in the i-th related image of the dashcam image, and k-1 is the number of related images of the dashcam image.
6. The method for following a vehicle based on artificial intelligence according to claim 1, characterized in that, The steps to determine the fuzzy distance between this vehicle and the target vehicle include: Based on the dashcam image and the bounding box of the target vehicle in the dashcam image, determine the perpendicular distance from the center point of the bounding box of the target vehicle in the dashcam image to the bottom edge of the dashcam image. Obtain the correlation coefficient of the perpendicular distance of the bounding box of the target vehicle and the correlation coefficient of the bounding box area of the target vehicle. Based on the correlation coefficient of the perpendicular distance of the target vehicle, the correlation coefficient of the bounding box area of the target vehicle, and the perpendicular distance and bounding box area of the bounding box of the target vehicle in the driving record image, determine the fuzzy distance between the current vehicle and the target vehicle.
7. The method for following a vehicle based on artificial intelligence according to claim 6, characterized in that, The steps for determining the correlation coefficient between the perpendicular distance of the target vehicle's bounding box and the correlation coefficient between the area of the target vehicle's bounding box include: Acquire the driving record images of this vehicle at each moment during its historical driving process, the position of the front of this vehicle in the driving record images at each moment, and the position of the bounding box of the target vehicle; Based on the position of the front of the vehicle and the bounding box of the target vehicle in the dashcam images at each moment, filter each target dashcam image and obtain the actual distance between the vehicle and the target vehicle in each target dashcam image; Based on each target driving record image, determine the bounding box of the target vehicle in each target driving record image; Based on each target driving record image, the actual distance between the vehicle and the target vehicle in each target driving record image, and the bounding box of the target vehicle in each target driving record image, determine the correlation coefficient of the vertical distance of the bounding box of the target vehicle and the correlation coefficient of the area of the bounding box of the target vehicle.
8. The method for following a vehicle based on artificial intelligence according to claim 1, characterized in that, The steps to determine the average speed of the target vehicle at the current moment include: The difference between the current fuzzy distance between the vehicle and the target vehicle and the previous fuzzy distance between the vehicle and the target vehicle is calculated. Based on this difference and the time interval between the driving record image of the previous moment and the driving record image of the current moment, the relative speed between the vehicle and the target vehicle at the current moment is determined. Determine the average speed of the target vehicle at the current moment based on the vehicle's current speed and the relative speed between the vehicle and the target vehicle.
9. The method for following a vehicle based on artificial intelligence according to claim 1, characterized in that, The steps to determine the vehicle's speed between the current moment and the next moment include: If the fuzzy distance between the current vehicle and the target vehicle is less than the first preset travel distance value, then control the vehicle's travel speed between the current time and the next time to be lower than the average travel speed of the target vehicle at the current time. If the fuzzy distance between the current vehicle and the target vehicle is greater than the first preset travel distance value and less than the second preset travel distance value, then control the vehicle's travel speed between the current time and the next time to be equal to the average travel speed of the target vehicle at the current time. If the fuzzy distance between the current vehicle and the target vehicle is greater than the second preset travel distance value, then the vehicle's travel speed between the current time and the next time will be controlled to be higher than the average travel speed of the target vehicle at the current time.
10. A vehicle-following system based on artificial intelligence, characterized in that, It includes a processor and a memory, the processor being used to process instructions stored in the memory to implement an AI-based vehicle following method as described in any one of claims 1-9.