WO2023020004A1

WO2023020004A1 - Vehicle distance detection method and system, and device and medium

Info

Publication number: WO2023020004A1
Application number: PCT/CN2022/089625
Authority: WO
Inventors: 韩毅; 田迪; 关甜; 张平
Original assignee: 长安大学
Priority date: 2021-08-16
Filing date: 2022-04-27
Publication date: 2023-02-23
Also published as: CN113657265A; CN113657265B

Abstract

Disclosed in the present invention are a vehicle distance detection method and system, and a device and a medium. The method comprises: S1, acquiring an image directly ahead of a driven vehicle during the travelling process of the driven vehicle; S2, performing vehicle detection on the acquired image by using a deep learning algorithm YOLOv4, and according to the size of a cross section of a vehicle, classifying all vehicles into three categories, i.e., small vehicles, passenger vehicles and trucks; S3, determining the relationships, respectively corresponding to the three categories, i.e., the small vehicles, the passenger vehicles and the trucks, between a distance and the number of pixel points; S4, selecting a vehicle directly ahead in the current lane of the driven vehicle; and S5, for the vehicle directly ahead of the driven vehicle, determining the category and the number of pixel points of the vehicle, and obtaining the distance between the vehicle and the driven vehicle by means of the relationship between the determined category and a distance. By means of the present application, the distance between the current vehicle and a vehicle ahead can be precisely estimated, thereby providing assistance for safe driving.

Description

A vehicle distance detection method, system, device and medium

technical field

The invention belongs to the field of automobile safety, and relates to a vehicle distance detection method, system, equipment and medium.

Background technique

According to the latest data, the number of motor vehicles in my country has reached 360 million, of which the number of cars is 270 million, which has led to a series of problems such as road congestion. Due to the large number of vehicles, it is inevitable to be on the same road with other vehicles during the normal driving of the vehicle. When an emergency occurs, if the distance is too close, accidents will easily occur. Accurately estimating the distance between oneself and the vehicle in front can help the driver judge whether to slow down to maintain a safe distance. A sufficient safe distance can give the driver sufficient time to act in an emergency and effectively avoid accidents.

technical problem

The purpose of the present invention is to overcome the above-mentioned shortcomings of the prior art, and provide a vehicle distance detection method, system, equipment and medium, which can accurately estimate the distance between the self-vehicle and the vehicle in front, and provide assistance for safe driving.

technical solution

In order to achieve the above object, the present invention adopts the following technical solutions to achieve:

A vehicle distance detection method, comprising the following processes:

S1, during the process of driving the vehicle, acquire the image directly in front of the driving vehicle;

S2, use the deep learning algorithm YOLOv4 to detect vehicles on the acquired images, and divide all vehicles into three categories: cars, passenger cars and trucks according to the size of the cross-section of the vehicle;

S3, determine the relationship between the distance and the number of pixels corresponding to the three types of cars, passenger cars and trucks;

S4, selecting the vehicle directly ahead in the current lane of the driving vehicle;

S5, for the vehicle directly in front of the driving vehicle, judge the category and number of pixels of the vehicle, and obtain the distance between the vehicle and the driving vehicle through the relationship between the determined category and the distance.

Preferably, the specific process of using the deep learning algorithm YOLOv4 for vehicle detection is: collecting a data set containing three types of targets to be detected: car, passenger car, and truck, labeling the data in the data set in detail, and using this data set to train the YOLOv4 algorithm; first The convolutional neural network is used for feature extraction, and then the gradient descent algorithm is used to train the model. Finally, the NMS algorithm is used to eliminate the overlapping bounding boxes of the same target, and a detection model that can accurately detect the above three types of targets is obtained.

Preferably, the specific process of S3 is: collecting the corresponding data between the distance and the number of pixels for three types of cars, passenger cars and trucks for multiple times, and fitting the expression between the distance and the number of pixels.

Preferably, the specific process of S4 is: according to the principle of minimum difference between the abscissa of the center point of the predicted frame and the abscissa of the center point of the overall image, select the vehicle existing in the current lane and exclude the vehicles in other adjacent lanes.

Preferably, the specific process of S5 is: for the vehicle in the current lane, use the deep learning algorithm to generate the target frame to obtain the current target pixel points, and select the corresponding expression of the car, bus or truck according to the specific target category information to calculate the current distance.

Preferably, after obtaining the distance between the driving vehicle and the vehicle in front, if the distance is less than the safety distance, an early warning message is sent to the driver.

Further, the relationship between the vehicle speed and the safety distance is determined, and if the distance is less than the minimum safety distance under the current vehicle speed, an early warning message is sent to the driver.

A vehicle distance detection system, comprising the following processes:

The image acquisition module is used to acquire the image directly in front of the driving vehicle during the driving process;

The vehicle classification module is used to detect vehicles using the deep learning algorithm YOLOv4 on the acquired images, and classify all vehicles into three categories: car, passenger car and truck according to the size of the cross-section of the vehicle;

The module for determining the relationship between the distance and the number of pixels is used to determine the relationship between the distance and the number of pixels corresponding to the three types of cars, passenger cars and trucks;

The target screening module is used to select the vehicle directly ahead in the current lane of the driving vehicle;

The distance calculation module is used for judging the category and number of pixels of the vehicle directly in front of the driving vehicle, and obtaining the distance between the vehicle and the driving vehicle through the relationship between the category and distance determined.

A computer device, including a memory, a processor, and a computer program stored in the memory and operable on the processor, when the processor executes the computer program, the vehicle distance as described in any one of the above is realized. The steps of the detection method.

A computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the vehicle distance detection method described in any one of the above items are realized.

Beneficial effect

Compared with the prior art, the present invention has the following beneficial effects:

The present invention classifies vehicles through the deep learning algorithm YOLOv4, and performs distance and pixel point fitting for different types of vehicles to obtain the relationship between the distance and pixel points of different types of vehicles, thereby classifying the vehicle directly in front of the driving vehicle, and then Bring in the relationship between the corresponding distance and the number of pixels to obtain the precise distance between the driving vehicle and the vehicle directly in front, and provide assistance for safe driving.

Furthermore, through the relationship between the vehicle speed and the safe distance, the current speed can be used to determine whether the current distance is safe, and it can adapt to different vehicle speeds, making the distance warning more intelligent.

Description of drawings

Fig. 1 is the image obtained during the traveling of the present invention;

Fig. 2 is the vehicle detection result in the photographed image of the present invention;

Fig. 3 is the vehicle exclusion result of other lanes of the present invention.

BEST MODE FOR CARRYING OUT THE INVENTION

The present invention is described in further detail below in conjunction with accompanying drawing:

The vehicle distance detection method of the present invention fixes the angle of the camera, places it in the middle of the front end of the vehicle, and places the on-board computer at the rear end of the vehicle. The camera collects images in the direction of the vehicle, and the on-board computer processes the collected images to determine whether there is a vehicle in front of the vehicle according to a specific algorithm. When there is a vehicle, the vehicle distance is calculated according to a specific algorithm.

Step 1, image capture: When the vehicle is running, the camera captures the image in front of the vehicle on the road for algorithm processing, and the obtained image to be processed is shown in Figure 1. Since the position of the camera is fixed, and the vehicles that will affect the driving safety of the current vehicle must be in the same lane, the target vehicle whose distance should be measured should be in a specific area of the image.

Step 2, image processing: use the deep learning algorithm YOLOv4 to detect the vehicle on the image captured by the camera. Because different vehicles have different cross-sectional sizes at the same distance, the patent of the present invention roughly divides the vehicles into three categories according to the size of the cross-sectional area: small cars, passenger cars, and trucks. In the YOLOv4 algorithm training, the three types of cars, passenger cars, and trucks are trained separately, so that these three types of targets can be distinguished during detection. First, collect a data set containing three types of targets to be detected: car, passenger car, and truck, and label the data in the data set in detail. Use this data set to train the YOLOv4 algorithm required for this patent on a professional computer. First, use the convolutional neural network for feature extraction, then use the gradient descent algorithm to train the model, and finally use the NMS algorithm to eliminate overlapping bounding boxes of the same target, and obtain accurate results. A detection model that detects the above three types of targets. Further transplant the trained model to the on-board computer for real-time target detection. The result of processing Figure 1 is shown in Figure 2.

Step 3, target determination, there are often multiple vehicles in one image, including not only the vehicle in front of the current lane, but also the image in front of the adjacent lane. Because there is no interference between vehicles in adjacent lanes and this vehicle, the distance does not need to be measured. Because the camera is installed at the front of the vehicle, the current lane is generally in the center of the image. According to the principle of minimum difference between the abscissa of the center point of the prediction frame and the abscissa of the center point of the overall image, the vehicles existing in the current lane are selected and the vehicles in other adjacent lanes are excluded. As shown in Figure 2, the upper left corner of the image is taken as the origin, the horizontal axis is the X axis, and the vertical axis is the Y axis. The horizontal length of the image is 473 pixels, and the vertical height is 355 pixels. Therefore, the abscissa of the image center point is 236. In Fig. 3, the abscissa of the center point of the leftmost vehicle prediction frame is 122, the abscissa of the center point of the middle vehicle prediction frame is 160, and the abscissa of the center point of the rightmost vehicle prediction frame is 247. Therefore, according to the principle of minimum difference between the abscissa of the center point of the prediction frame and the abscissa of the center point of the overall image, it is determined that the target existing in the current lane is the rightmost vehicle. When driving on a curve, when an emergency accident occurs and the vehicle needs to be braked, the vehicle in front is still the vehicle that causes the greatest potential safety hazard, so the detection target is still determined according to the above process. The results after excluding the objects of other lanes are shown in Fig. 3.

Step 4, data processing: take pictures in advance according to the specific installation position of the camera and calculate the number of pixels of the target at a certain distance. For example, when the separation distance is 1000 cm, the pixel points of the vehicle target frame in front are 8000, and when the separation distance is 2000 cm, the pixel points of the target frame of the vehicle ahead are 5000. Using the above scheme, the corresponding data between the interval distance and the pixel points is collected multiple times, and the expression between the two is fitted. The above data processing method is carried out separately for the three types of cars, passenger cars, and trucks, and the relationship expressions between the pixels of the prediction frame and the distance corresponding to the three types of vehicles can be obtained.

Step 5, distance calculation: For any target in the current lane, use the deep learning algorithm to generate the target frame to obtain the number of pixels of the current target. In addition, according to the specific target category information, select an expression of car, bus, and truck to calculate the current distance. In this way, The separation distance can be calculated according to any number of target pixels. For example, for the target information in Figure 1, the result in Figure 2 can be obtained by using the deep learning algorithm to generate the target frame, and the prediction result in Figure 3 can be obtained according to the target determination method in step three. In Figure 3, the detection result is bus, and the YOLOv4 algorithm can output the coordinate information of the prediction frame, calculate the number of pixels contained in the prediction frame, use the obtained pixel point information to select the pixel point and distance relationship expression corresponding to the bus, and calculate the current The distance to the target vehicle.

Step six, information early warning: according to the specific characteristics of the current vehicle, the relationship between the vehicle speed and the safety distance is determined in advance. During the driving process, the algorithm proposed by this patent is used to help the driver determine the accurate distance between vehicles ahead. If the calculated distance is less than the minimum safe distance at the current speed, an early warning message will be issued to remind the driver to slow down and increase the distance to avoid braking in emergency situations. Insufficient moving distance leads to accidents.

Example:

Step 1, image capture: When the vehicle is running, the camera captures the image in front of the vehicle on the road for algorithm processing, and the obtained image to be processed is shown in Figure 1.

Step 2, image processing: Collect a data set containing three types of targets to be detected: car, passenger car, and truck in advance, and label the data in the data set in detail. Use this data set to train the YOLOv4 algorithm required for this patent on a professional computer. First, use the convolutional neural network for feature extraction, then use the gradient descent algorithm to train the model, and finally use the NMS algorithm to eliminate overlapping bounding boxes of the same target, and obtain accurate results. A detection model that detects the above three types of targets. Further transplant the trained model to the on-board computer for real-time target detection. The result of processing Figure 1 is shown in Figure 2.

Step 3, target determination: There are often multiple vehicles in an image, including not only the vehicle in front of the current lane, but also the image in front of the adjacent lane. Because there is no interference between vehicles in adjacent lanes and this vehicle, the distance does not need to be measured. Because the camera is installed at the front of the vehicle, the current lane is generally in the center of the image. When driving on a curve, when an emergency accident occurs and the vehicle needs to be braked, the vehicle directly ahead is still the vehicle that causes the greatest safety hazard, so the detection target directly ahead is still considered. According to the principle of minimum difference between the abscissa of the center point of the prediction frame and the abscissa of the center point of the overall image, the vehicles existing in the current lane are selected and the vehicles in other adjacent lanes are excluded. As shown in Figure 2, the upper left corner of the image is taken as the origin, the horizontal axis is the X axis, and the vertical axis is the Y axis. In Fig. 2, the abscissa of the center point of the leftmost vehicle prediction frame is 122, the abscissa of the center point of the middle vehicle prediction frame is 160, and the abscissa of the center point of the rightmost vehicle prediction frame is 247. Therefore, according to the principle of minimum difference between the abscissa of the center point of the prediction frame and the abscissa of the center point of the overall image, it is determined that the target existing in the current lane is the rightmost vehicle. The results after excluding the objects of other lanes are shown in Fig. 3.

Step 4, data processing: take pictures in advance according to the specific installation position of the camera and calculate the number of pixels of the target at a certain distance.

As for the bus target:

When the separation distance is 1000 cm, the pixels of the target frame of the vehicle in front are 8000.

When the separation distance is 2000 cm, the pixels of the target frame of the vehicle in front are 5000.

When the separation distance is 3000 cm, the pixels of the target frame of the vehicle in front are 3000.

When the separation distance is 4000 cm, the pixels of the target frame of the vehicle in front are 1500.

When the separation distance is 5000 cm, the pixel points of the vehicle target frame in front are 600.

When the separation distance is 6000 cm, the pixels of the target frame of the vehicle in front are 250.

When the separation distance is 7000 cm, the pixel points of the target frame of the vehicle in front are 100.

Using the above data, taking the pixels of the target frame as the independent variable x and the interval distance as y, the expression between the two is fitted as y=9.812e-5*x ² -1.442x+6390.

Step 5, distance calculation: Based on the detection results determined in step 3, assuming that the number of pixels in the target frame of the bus is 4000, then using the formula obtained in step 4, the separation distance can be calculated as 2191.92 cm.

Step 6, information early warning: if the safe vehicle speed is 20km/h in the above calculation conditions, and the self-driving speed is 50km/h, it can be judged that the separation distance is insufficient and accidents are likely to occur in an emergency state, and the driver will be given an information early warning. Remind the driver to increase the driving distance.

The vehicle distance detection system of the present invention includes the following processes:

The image acquisition module is used to acquire the image directly in front of the driving vehicle during the driving process.

The vehicle classification module is used to detect vehicles using the deep learning algorithm YOLOv4 on the acquired images, and classify all vehicles into three categories: car, passenger car and truck according to the size of the cross-section of the vehicle.

The module for determining the relationship between the distance and the number of pixels is used to determine the relationship between the distance and the number of pixels corresponding to the three types of cars, passenger cars and trucks.

The target screening module is used to select the vehicle directly ahead in the current lane of the driving vehicle.

The computer device according to the present invention includes a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the computer program, the vehicle distance detection method as described above is realized. A step of.

In the computer-readable storage medium of the present invention, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned vehicle distance detection method are realized.

The above content is only to illustrate the technical ideas of the present invention, and cannot limit the protection scope of the present invention. Any changes made on the basis of the technical solutions according to the technical ideas proposed in the present invention shall fall within the scope of the claims of the present invention. within the scope of protection.

Claims

A vehicle distance detection method, characterized in that, comprising the following process:

S1, during the process of driving the vehicle, acquire the image directly in front of the driving vehicle;

S2, use the deep learning algorithm YOLOv4 to detect vehicles on the acquired images, and divide all vehicles into three categories: cars, passenger cars and trucks according to the size of the cross-section of the vehicle;

S3, determine the relationship between the distance and the number of pixels corresponding to the three types of cars, passenger cars and trucks;

S4, selecting the vehicle directly ahead in the current lane of the driving vehicle;

S5, for the vehicle directly in front of the driving vehicle, judge the category and number of pixels of the vehicle, and obtain the distance between the vehicle and the driving vehicle through the relationship between the determined category and the distance.
The vehicle distance detection method according to claim 1, wherein the specific process of using the deep learning algorithm YOLOv4 for vehicle detection is: collecting data sets including three types of targets to be detected: cars, passenger cars, and trucks, and the data in the data set Make detailed annotations and use this data set to train the YOLOv4 algorithm; first use the convolutional neural network for feature extraction, then use the gradient descent algorithm to train the model, and finally use the NMS algorithm to eliminate overlapping bounding boxes of the same target, and obtain the above three types of targets that can be accurately detected detection model.
The vehicle distance detection method according to claim 1, characterized in that, the specific process of S3 is: collecting the corresponding data between the distance and the number of pixels for three types of cars, passenger cars and trucks for multiple times, and fitting the distance and the number of pixels expressions between.
The vehicle distance detection method according to claim 1, wherein the specific process of S4 is: according to the principle of minimum difference between the center point abscissa of the prediction frame and the center point abscissa of the overall image, the vehicles existing in the current lane are selected and excluded. other vehicles in adjacent lanes.
The vehicle distance detection method according to claim 1, characterized in that, the specific process of S5 is: for the vehicle in the current lane, use the deep learning algorithm to generate the target frame, obtain the current target pixel points, and select the car according to the specific target category information, The expression corresponding to passenger car or truck calculates the current distance.
The vehicle distance detection method according to claim 1, characterized in that, after obtaining the distance between the driving vehicle and the vehicle in front, if the distance is less than a safe distance, an early warning message is sent to the driver.
The vehicle distance detection method according to claim 6, characterized in that the relationship between the vehicle speed and the safety distance is determined, and if the distance is smaller than the minimum safety distance at the current vehicle speed, an early warning message is sent to the driver.
A vehicle distance detection system, characterized in that it comprises the following processes:

The image acquisition module is used to acquire the image directly in front of the driving vehicle during the driving process;

The vehicle classification module is used to detect vehicles using the deep learning algorithm YOLOv4 on the acquired images, and classify all vehicles into three categories: car, passenger car and truck according to the size of the cross-section of the vehicle;

The module for determining the relationship between the distance and the number of pixels is used to determine the relationship between the distance and the number of pixels corresponding to the three types of cars, passenger cars and trucks;

The target screening module is used to select the vehicle directly ahead in the current lane of the driving vehicle;

The distance calculation module is used for judging the category and number of pixels of the vehicle directly in front of the driving vehicle, and obtaining the distance between the vehicle and the driving vehicle through the relationship between the category and distance determined.
A computer device, comprising a memory, a processor, and a computer program stored in the memory and operable on the processor, characterized in that, when the processor executes the computer program, the computer program according to claims 1 to 1 is implemented. 7. The steps of any one of the vehicle distance detecting methods.
A computer-readable storage medium, the computer-readable storage medium stores a computer program, characterized in that, when the computer program is executed by a processor, the vehicle distance detection method according to any one of claims 1 to 7 is realized step.