CN107796373A - A kind of distance-finding method of the front vehicles monocular vision based on track plane geometry model-driven - Google Patents

Abstract

The present invention provides a distance measuring method based on the monocular vision of the front vehicle driven by the plane geometric model of the lane, utilizes the CCD camera to collect the front target vehicle image, and utilizes the fusion Haar-like feature and Adaboost algorithm to identify the front target vehicle obtained in step 2 For the target vehicle ahead in the image, use the particle filter method to track the target vehicle ahead obtained in step 3, and construct the longitudinal distance measurement model of the lane plane geometry in the frame image according to the target vehicle ahead in each frame of image obtained above , to obtain the longitudinal perception distance y of the target source in the frame image, construct a vehicle ranging error dynamic compensation model according to the target vehicle ahead in each frame image obtained above, and obtain the longitudinal measurement error value z, according to the target target obtained in step 5 The source longitudinal perception distance y and the longitudinal measurement error value z calculate the longitudinal distance Y _W (P) between the front target vehicle and the own vehicle in the frame image.

Description

A distance measurement based on the monocular vision of the vehicle in front driven by the plane geometric model of the lane method

technical field

The invention belongs to the technical field of vehicle longitudinal safety assisted driving, and relates to a distance measuring method based on the monocular vision of a vehicle in front driven by a plane geometric model of a lane.

Background technique

Vehicle following is one of the most basic driving behaviors for drivers in traffic activities. The main threats faced by vehicles during the following phase are mainly longitudinal vehicle rear-end collisions, failure to maintain a certain safe distance between the self-vehicle and the vehicle in front, and the threat to the self-vehicle. Inaccurate determination of the speed of the vehicle in front and the vehicle in front leads to a rear-end collision of the vehicle. Accurately studying the distance value between the ego vehicle and the vehicle in front is of great significance for maintaining the distance between vehicles and vehicle collision warning.

The currently researched methods of vehicle distance measurement mainly include ultrasonic distance measurement, laser distance measurement, millimeter-wave radar distance measurement and machine vision distance measurement. Ultrasonic ranging is only suitable for short-distance ranging, while laser ranging and millimeter-wave radar ranging are too expensive to use. In contrast, machine vision ranging has a simple hardware structure, low cost, and rich and easy access to information. Therefore, the use of machine vision to measure vehicle distance has better practical value and application prospects.

In machine vision, various measurement methods are comprehensively compared. Because the data processing time of monocular vision measurement is short, it can meet the real-time performance of distance measurement, so the majority of vehicles use monocular vision to measure the distance between vehicles.

When performing automatic measurement of vehicle distance based on monocular vision, it is very important for the positioning of the vehicle in front, and the accuracy of positioning directly affects the accuracy of vehicle distance measurement. At present, the recognition method based on vehicle shadows is greatly affected by external light factors; the depth information of the image is obtained through the calibration of corresponding points based on the mathematical model method of the rear of the vehicle, but due to equipment limitations and calibration, it is impossible to obtain a high-precision coordinate system. Mutually transformed transformation matrices, the applicability is limited; when the front vehicle is recognized based on the monocular vision image grayscale processing, often only the rear profile of the front vehicle can be recognized, but due to the existence of the height of the rear overhang of the vehicle, it is bound to cause Large ranging error.

Contents of the invention

The purpose of the present invention is to provide a distance measuring method based on the monocular vision of the front vehicle driven by the plane geometric model of the lane, which solves the problems in the prior art such as the influence of light, the conversion of high-precision coordinate system, and the existence of the height of the vehicle's rear overhang. Problems and defects caused by ranging errors.

In order to achieve the above object, the technical scheme adopted in the present invention is:

The present invention provides a distance measuring method based on the monocular vision of the front vehicle driven by the plane geometric model of the lane, comprising the following steps:

Step 1, calibrate the CCD camera to obtain the effective focal length f, the height h of the CCD camera and the pitch angle θ of the CCD camera;

Step 2, using the CCD camera to collect the image of the target vehicle ahead;

Step 3, using the fusion Haar-like feature and Adaboost algorithm to identify the front target vehicle in the front target vehicle image obtained in step 2;

Step 4, using the particle filter method to track the front target vehicle acquired in step 3;

Step 5, constructing the longitudinal distance measurement model of the lane plane geometry in the frame image according to the front target vehicle in each frame image obtained above, and obtaining the longitudinal perception distance y of the target source in the frame image;

Step 6, constructing a vehicle ranging error dynamic compensation model based on the front target vehicle in each frame of image obtained above, and obtaining the longitudinal measurement error value z;

Step 7. Calculate the longitudinal distance Y _W (P) between the front target vehicle and your own vehicle in the frame image according to the target source longitudinal perception distance y obtained in step 5 and the longitudinal measurement error value z.

Preferably, in step 3, utilize fusion Haar-like feature and Adaboost algorithm to identify the specific method of the front target vehicle in the front target vehicle image that obtains in step 2:

S1, establish a sample set according to the image of the target vehicle ahead obtained in step 2, use the Adaboost algorithm to select the effective Haar-like features of the vehicle training samples in the sample set, each effective Haar-like feature generates a corresponding weak classifier, and classify the weak The weighted combination of classifiers becomes a strong classifier, and finally cascaded with waterfall classifiers to obtain a cascade classifier of feature samples;

S2, take a large number of vehicle training samples and extract the effective Haar-like features of the vehicle training samples, and then input the effective Haar-like features to the cascade classifier of the feature samples to detect the existence of the vehicle, and obtain the Adaboost cascade classifier ;

S3, according to the fixed position of the CCD camera, determine the region of interest in the image of the front target vehicle, use the Adaboost cascade classifier obtained in S2 to detect the presence of the vehicle in the region of interest, and finally obtain the front target in the image of the front target vehicle vehicle.

Preferably, in step 5, the CCD camera is obtained through calibration in step 1, the optical center of the CCD camera is set as point C in the longitudinal vehicle distance measurement model based on the plane geometry of the lane, and the projection point of the optical center C point on the road surface is The origin point O of the world coordinate system, the distance measurement feature point of the vehicle in front is P, the forward direction of the vehicle is the X _W axis of the world coordinate system, the Z _W axis of the world coordinate system is perpendicular to the road, and the imaging plane of the CCD camera is A'B'F'E', the plane of the far viewing angle is CEF, the plane where the center of the optical axis is located is CMN, and the plane where the ranging feature points are located is CC ₂ D, where the optical axis CC ₁ of the CCD camera and the imaging plane A'B' The intersection point of F'E' is C ₀ point, then CC ₀ is the focal length of the CCD sensor, that is, CC ₀ =f; the lower edge point A of the near-field image of the target vehicle in front of the image and the imaging plane A'B'F'E' Intersect at point G', the edge point B of the far field of view and the imaging plane A'B'F'E' intersect at point H; the point where the distance measurement feature point P of the vehicle in front is projected to the X _W axis of the world coordinate system is P', Among them, the longitudinal perception distance of the target source is the length of OP'.

Preferably, in step 5, the formula for calculating the longitudinal perception distance y of the target source is:

In the formula, v ₀ is the longitudinal image coordinate of the optical center, v(P ₀ ) is the longitudinal image coordinate of the feature point, and dy is the longitudinal length of the unit pixel.

Preferably, in step 6, the construction of the dynamic compensation model for the ranging error of the target vehicle in front specifically includes the following steps:

The first step is to fix the CCD vision sensor, and use the calibration method described in step 1 to calibrate and record the internal and external parameters of the CCD camera;

The second step is to fix the height of the target source from the ground, move the target source every 5m along the longitudinal direction of the road, so that it changes within the range of [10m, 100m] from the CCD visual sensor, and use the CCD camera to record the target source at each position Image;

The third step is to adjust the height of the target source from the ground so that it changes within [0.2m, 1m], and repeat the second step;

In the fourth step, use matlab to process the image obtained in the third step, and analyze the measurement error of the target source at different heights from the ground and different longitudinal positions.

Preferably, in step 6, the formula for calculating the longitudinal measurement error value z is:

z＝118-1124x-3.133y+3522x ² +34.6xy-0.0399y ² -4795x ³ -86.22x ² y-0.1845xy ² +0.0017y ³ +2676x ⁴ +98.62x ³ y+0.1428x ² y ² +0.001313 xy ³ -2.077e ^-5 y ⁴ +8.835e ^-8 y ⁵ -7.658e ^-6 xy ⁴ +0.0004x ² y ³ -0.1114x ³ y ² -38.23x ⁴ y-391.1x ⁵

In the formula, x is the height of the target source from the ground.

Preferably, in step 7, the formula for calculating the longitudinal distance Y _W (P) between the target vehicle ahead and the own vehicle is:

Compared with prior art, the beneficial effect of the present invention is:

The present invention provides a distance measuring method based on the monocular vision of the front vehicle driven by the plane geometric model of the lane. The CCD camera is used to collect the image of the front target vehicle, and the fusion of Haar-like features and Adaboost algorithm is used to identify the front target vehicle obtained in step 2. For the target vehicle ahead in the image, use the particle filter method to track the target vehicle ahead obtained in step 3, and construct the longitudinal distance measurement model of the lane plane geometry in the frame image according to the target vehicle ahead in each frame of image obtained above , to obtain the longitudinal perception distance y of the target source in the frame image, construct a vehicle ranging error dynamic compensation model according to the target vehicle ahead in each frame image obtained above, and obtain the longitudinal measurement error value z, according to the target target obtained in step 5 The source longitudinal perception distance y and the longitudinal measurement error value z calculate the longitudinal distance Y _W (P) between the front target vehicle and the own vehicle in the frame image; the present invention has the advantages of simple hardware structure, low cost and large software algorithm flexibility and It has the characteristics of higher measurement accuracy, and can avoid the influence of interference factors such as road shadows on both sides of the carriageway and other vehicles not in the lane, and improves the detection robustness of the system. The invention effectively solves the problem of vehicle rear suspension height The existence of , improves the identification efficiency of the longitudinal position relationship of the vehicle in front.

Description of drawings

Figure 1 is a schematic diagram of the installation of a CCD camera;

Fig. 2 is the flow process method of the distance measuring device of the present invention;

Figure 3 is a schematic diagram of the calibration of the internal parameters of the CCD camera;

Figure 4 is a schematic diagram of the calibration of the external parameters of the CCD camera;

Figure 5 is a structural diagram of a vehicle recognition algorithm based on Haar-like features and Adaboost;

Figure 6 is a schematic diagram of the construction of a feature sample cascade classifier;

Fig. 7 is a schematic diagram of target image vehicle contour extraction;

Fig. 8 is a diagram of geometric constraints of CCD camera imaging space;

Fig. 9 is a side view of the lane plane constrained ranging model;

Fig. 10 is a schematic diagram of analysis of sources of error in longitudinal vehicle distance measurement.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings.

As shown in Fig. 1, a kind of front vehicle monocular vision distance measuring device driven based on the plane geometric model of the lane provided by the present invention comprises a CCD camera 1, and the CCD camera 1 is fixed on the middle of the front windshield 2 of the vehicle using a suction cup. Location. Wherein, the CCD camera 1 is connected with the host computer through a BNC video cable and a video capture card.

As shown in Fig. 2, a kind of method for the monocular vision distance measurement of the front vehicle driven by the plane geometric model of the lane provided by the present invention, the specific steps are as follows:

Step 1. Calibration of the CCD camera:

The calibration of the visual sensor is a key problem to be solved in machine vision. The purpose of the calibration is to obtain the internal parameters and external parameters of the CCD camera to complete the conversion from the two-dimensional image to the three-dimensional scene in the subsequent steps. Specifically:

First, use the Camera Calibration Toolbox module in the MATLAB software and the plane target calibration board to calibrate to obtain the internal parameters of the CCD camera; Put it into the CameraCalibrationToolbox module in the MATLAB software to calculate the internal parameters of the CCD camera, and obtain the effective focal length f of the internal parameters of the CCD camera;

Secondly, the method based on the vanishing point of the road image is adopted, and the specific implementation method is as follows (as shown in Figure 4): on the plane target image acquired by the CCD camera, the left lane line and the right lane line are marked, and the left lane line and the right lane line are recorded at the same time The pixel coordinates of the intersection point O, and take a point on the left and right lane lines as far away from the intersection point as possible, record the pixel coordinates of the two points A and B, according to the calibration results of the internal parameters of the CCD camera and the two parallel lines in the three-dimensional world actual distance. The external parameters of the CCD camera 1 are obtained through calibration of the Calibration module in the calibration toolbox in the HALCON software. The external parameters of the CCD camera 1, the height h of the CCD camera and the pitch angle θ of the CCD camera, can be obtained through calculation.

Step 2. Acquisition and transmission of target vehicle images:

The CCD camera 1 collects the image of the target vehicle ahead, and transmits the collected image of the target vehicle ahead to the image processing software in the upper computer system through the BNC video cable and the video capture card, so as to obtain the image of the target vehicle ahead that can be analyzed and processed.

Step 3, using the fusion Haar-like feature and Adaboost algorithm to identify the front target vehicle in the front target vehicle image obtained in step 2, the specific process (as shown in Figure 5):

First, a sample set is established according to the image of the front target vehicle obtained in step 2, and the vehicle training samples and non-vehicle training samples in the sample set are preprocessed, that is, the Adaboost algorithm is used to select the effective Haar-like features in the vehicle training samples, each Effective features generate corresponding weak classifiers, and the weighted combination of weak classifiers becomes a strong classifier, and finally a cascade classifier of feature samples is constructed. The specific method is as follows:

The first step is to convert the image of the target vehicle ahead into an integral map: after converting the image of the target vehicle ahead using the integral map method, the integral map of the target vehicle image is obtained after conversion. Each point in the integral map represents the starting point of the upper left corner of the transformed image The sum of pixels within the rectangular area to that point, as shown in Equation 1.

Among them, i(x, y) is the pixel integral value of point (x, y) on the integral image, and i(x’, y’) is the pixel value in point (x, y) in the original image.

The second step is to use the Adaboost algorithm to perform fast and effective Haar-like feature extraction on the converted integral graph to generate a strong classifier with feature samples. Specifically: the Adaboost algorithm extracts a corresponding effective Haar-like feature through each cycle -like features, each effective feature generates a corresponding weak classifier, and the weighted combination of weak classifiers becomes a strong classifier;

The third step is to construct a cascade classifier of feature samples: usually most areas in the image to be detected do not contain the target vehicle, so the cascade classifier is used to quickly exclude non-vehicle areas and improve the speed of target detection. The present invention adopts classic waterfall classifiers to be cascaded, and each layer is an Adaboost strong classifier trained with the Adaboost algorithm, and each Adaboost strong classifier includes several weak classifiers, and the sample graphs to be identified are cascaded The classifier detects layer by layer. If it is judged as a negative sample at any layer, that is, a non-vehicle target image, the subsequent strong classifiers will not pass, which will allow the subsequent classifiers to have more time to identify positive samples. The sample window, that is, the image of the target vehicle, the specific process is shown in Figure 6. Considering the reliability of the vehicle target detection and the real-time performance of the algorithm, the cascade classifier used in the present invention adopts 8 layers of classifiers, the final detection rate is more than 0.9, and the false detection rate of each layer is 0.5, then each layer The detection rate is above 0.99.

Secondly, Haar-like feature extraction is performed on a large number of test samples, and the features are input to the Adaboost cascade classifier for vehicle presence detection to ensure the detection accuracy, detection speed and real-time performance of the algorithm. The above two steps build a fusion Haar-like Features and Adaboost's front vehicle recognition algorithm.

Finally, according to the fixed position of the CCD camera, it is determined that the ROI region of the front target vehicle image of the present invention is the lower half of the image, and the region of interest in the front target vehicle image is processed to obtain the vehicle recognition result. The specific method is: first use Integrating Haar-like features and Adaboost algorithm to process the region of interest (ROI) in the same way as the above training process, that is: perform image preprocessing and calculate the integral image on the region of interest on the target image; secondly, use the selected in the above training process Haar-like feature information, including structure, location, type, etc., is used to extract the Haar-like feature value of the region of interest to form a feature vector; finally, the Adaboost cascade classifier obtained by massive offline training is used to detect the presence of vehicles in the region of interest , and output the vehicle recognition result, as shown in Figure 7. While ensuring the detection accuracy and real-time performance of the algorithm, this method can effectively avoid the influence of interference factors such as road shadows on both sides of the lane and other vehicles not in the lane, and improve the detection robustness of the system.

Step 4, use the particle filter method to track the target vehicle in front:

When actually collecting images, due to the complex and diverse background images of the target vehicle, it is easy to cause problems such as missed detection by the algorithm or false alarms. Therefore, in order to ensure the real-time and robustness of the system, the present invention adopts the particle filter algorithm to track the vehicle in front, and the specific tracking steps are as follows: particle initialization, time update, observation update step, resampling and state update. Here, the particle number N of the particle filter is set to 100, and the sampling number of each iteration is 30. It can ensure that the tracking algorithm has high tracking accuracy and stability, and the tracking process takes an average of 20ms. The tracking algorithm has strong immunity to uncertain factors such as vehicle type, attitude change, and environmental interference, and can meet the real-time performance of the vehicle system. robustness requirements.

Step 5. Construct the longitudinal vehicle distance measurement model of the plane geometry of the lane:

Calculate the longitudinal vehicle distance measurement model based on the plane geometry of the lane for each frame of the processable image obtained above. First, establish the longitudinal distance measurement model based on the plane geometry of the lane according to the installation position of the CCD camera, as shown in Figure 8. The optical center of the CCD camera is set as point C, the projection point of the optical center C point on the road surface is the origin point O of the world coordinate system, the distance measurement feature point of the vehicle in front is P, and the forward direction of the vehicle is the point of the world coordinate system. The X _W axis, the Z _W axis of the world coordinate system are perpendicular to the road and face downward, the imaging plane of the CCD camera is A'B'F'E', the plane of the far viewing angle is CEF, and the plane where the center of the optical axis is located is the CMN and the distance measurement feature point. The plane is CC ₂ D, where the intersection of the optical axis CC ₁ of the CCD camera and the imaging plane A'B'F'E' is C ₀ , then CC ₀ is the focal length of the CCD sensor, that is, CC ₀ =f; in the image The lower edge point A of the near-field image of the target vehicle in front intersects with the imaging plane A'B'F'E' at point G', and the lower edge point B of the far-field image intersects with the imaging plane A'B'F'E' at point H; The point where the ranging feature point P of the vehicle in front is projected onto the X and _W axes of the world coordinate system is P', where the longitudinal perception distance of the target source is the length of OP'.

Secondly, the position of the front vehicle in the world coordinate system is derived according to the vehicle ranging model under the above-mentioned road plane constraints, as shown in Figure 9. Knowing C ₀ , CC ₀ , θ and the lengths of each side of the imaging plane A′B′F′E′, solve the ordinate value of the world coordinate of the distance measurement feature point P by the following formula, and measure the distance from the lane plane constraints in Figure 8 A side view of the model reveals:

CC ₀ ＝f,∠OC ₁ C＝θ, OC＝h (2)

In ΔC ₀ CP′ ₀ ,

Among them, P′ ₀ C ₀ can take positive and negative values, and:

In the formula, P′ ₀ C ₀ =y(C ₀ )-y(P′ ₀ )=[v ₀ -v(P′ ₀ )]×dy=[v ₀ -v(P ₀ )]×dy

In ΔOCP', OP'=OC×tan∠OCP'

It can be deduced that,

In the formula, v ₀ is the longitudinal image coordinate of the optical center, v(P ₀ ) is the longitudinal image coordinate of the feature point, dy is the longitudinal length of the unit pixel, and y is the longitudinal perception distance of the target source.

Step 6. Construct a vehicle ranging error dynamic compensation model:

As shown in Figure 10, the rear area of the target vehicle in front is divided into two parts, area A and area B, bounded by the lower edge of the vehicle outline. The vertical height of area A is the rear overhang height h _α of the target vehicle, and point D is the lower edge of the vehicle rear area. The midpoint, point P is the midpoint of the projection of the rear area of the vehicle on the road surface. Since the front vehicle identification algorithm can only detect the rear profile of the target vehicle in front, point D is selected as the longitudinal distance measurement feature point to make the longitudinal distance measurement value too large.

In order to reduce the error caused by the height of the rear overhang to the measurement of the longitudinal vehicle distance and make the measured value of the longitudinal vehicle distance as close as possible to the real value, the present invention uses a red target source with a specification of 2m×1m as the simulation object of the rear area of the vehicle in front. The source is fixed on the bracket and the height from the ground is adjustable. Through the analysis and statistics of a large number of vehicle samples, it is determined that the variation range of the rear overhang height of the vehicle is [0.2m, 1m].

Its specific implementation steps are:

The first step is to fix the CCD vision sensor, and use the calibration method described in step 1 to calibrate and record the internal and external parameters of the CCD camera;

The second step is to fix the height of the target source from the ground, move the target source every 5m along the longitudinal direction of the road, so that it changes within the range of [10m, 100m] from the CCD visual sensor, and use the CCD camera to record the target source at each position Image;

The third step is to adjust the height of the target source from the ground so that it changes within [0.2m, 1m], and repeat the second step;

The fourth step is to use matlab to process the image obtained in the third step, and analyze the measurement error of the target source at different heights from the ground and different longitudinal positions. According to the error data of the target source at different heights from the ground and different longitudinal positions, the regressive dynamic compensation model of longitudinal measurement error is:

According to the error data of the target source at different heights from the ground and different longitudinal positions, the regressive dynamic compensation model of longitudinal measurement error is:

Where: in the formula, x is the height of the target source from the ground, y is the longitudinal perception distance of the target source, and z is the longitudinal measurement error.

Step 7. Calculation of vehicle distance:

According to the construction of the longitudinal vehicle distance measurement model based on the plane geometry of the lane in step 5 and the construction of the dynamic compensation model of the vehicle ranging error in step 6, the vehicle distance is reconstructed, and the reconstructed longitudinal distance measurement model is:

Claims (7)

1. a distance measuring method based on the vehicle monocular vision of the front vehicle driven by the plane geometric model of the lane, it is characterized in that, comprising the following steps: Step 1, calibrate the CCD camera to obtain the effective focal length f, the height h of the CCD camera and the pitch angle θ of the CCD camera; Step 2, using the CCD camera to collect the image of the target vehicle ahead; Step 3, using the fusion Haar-like feature and Adaboost algorithm to identify the front target vehicle in the front target vehicle image obtained in step 2; Step 4, using the particle filter method to track the front target vehicle acquired in step 3; Step 5, constructing the longitudinal distance measurement model of the lane plane geometry in the frame image according to the front target vehicle in each frame image obtained above, and obtaining the longitudinal perception distance y of the target source in the frame image; Step 6, constructing a vehicle ranging error dynamic compensation model based on the front target vehicle in each frame of image obtained above, and obtaining the longitudinal measurement error value z; Step 7. Calculate the longitudinal distance Y _W (P) between the front target vehicle and your own vehicle in the frame image according to the target source longitudinal perception distance y obtained in step 5 and the longitudinal measurement error value z. 2. the distance measuring method of a kind of front vehicle monocular vision based on lane plane geometric model driving according to claim 1, it is characterized in that: in step 3, utilize fusion Haar-like feature and Adaboost algorithm recognition step 2 to obtain The specific method of the front target vehicle in the front target vehicle image: S1, establish a sample set according to the image of the target vehicle ahead obtained in step 2, use the Adaboost algorithm to select the effective Haar-like features of the vehicle training samples in the sample set, each effective Haar-like feature generates a corresponding weak classifier, and classify the weak The weighted combination of classifiers becomes a strong classifier, and finally cascaded with waterfall classifiers to obtain a cascade classifier of feature samples; S2, take a large number of vehicle training samples and extract the effective Haar-like features of the vehicle training samples, and then input the effective Haar-like features to the cascade classifier of the feature samples to detect the existence of the vehicle, and obtain the Adaboost cascade classifier ; S3, according to the fixed position of the CCD camera, determine the region of interest in the image of the front target vehicle, use the Adaboost cascade classifier obtained in S2 to detect the presence of the vehicle in the region of interest, and finally obtain the front target in the image of the front target vehicle vehicle. 3. the distance measuring method based on the vehicle monocular vision of a kind of front vehicle monocular vision driven by a lane plane geometric model according to claim 1, characterized in that: in step 5, by step 1, the CCD camera is calibrated and gained, based on the lane plane In the geometric longitudinal vehicle distance measurement model, the optical center of the CCD camera is set as point C, the projection point of the optical center C point on the road surface is the origin point O of the world coordinate system, the distance measurement feature point of the vehicle in front is P, and the vehicle is moving forward. The direction is the X _W axis of the world coordinate system, the Z _W axis of the world coordinate system is perpendicular to the road facing downward, the imaging plane of the CCD camera is A′B′F′E′, the plane of the far viewing angle is CEF, and the plane where the center of the optical axis is located is The plane where the CMN and the ranging feature points are located is CC ₂ D, where the intersection point of the optical axis CC ₁ of the CCD camera and the imaging plane A′B′F′E′ is point C ₀ , then CC ₀ is the focal length of the CCD sensor, that is CC ₀ = f; the lower edge point A of the near-field image of the target vehicle ahead in the image intersects with the imaging plane A'B'F'E' at point G', and the lower edge point B of the far-field image intersects with the imaging plane A'B'F'E' intersects at point H; the point where the ranging feature point P of the vehicle in front is projected onto the X and _W axes of the world coordinate system is P', where the longitudinal perception distance of the target source is the length of OP'. 4. a kind of distance measuring method based on the monocular vision of the front vehicle driven by the lane plane geometric model according to claim 3, characterized in that: in step 5, the calculation formula of the target target source longitudinal perception distance y is: <mrow><mi>y</mi><mo>=</mo><mi>h</mi><mo>&amp;times;</mo><mi>t</mi><mi>a</mi><mi>n</mi><mo>&amp;lsqb;</mo><mfrac><mi>&amp;pi;</mi><mn>2</mn></mfrac><mo>-</mo><mi>&amp;theta;</mi><mo>+</mo><mi>arctan</mi><mrow><mo>(</mo><mfrac><mrow><mo>&amp;lsqb;</mo><msub><mi>v</mi><mn>0</mn></msub><mo>-</mo><mi>v</mi><mrow><mo>(</mo><msub><mi>P</mi><mn>0</mn></msub><mo>)</mo></mrow><mo>&amp;rsqb;</mo><mo>&amp;times;</mo><mi>d</mi><mi>y</mi></mrow><mi>f</mi></mfrac><mo>)</mo></mrow><mo>&amp;rsqb;</mo></mrow> In the formula, v ₀ is the longitudinal image coordinate of the optical center, v(P ₀ ) is the longitudinal image coordinate of the feature point, and dy is the longitudinal length of the unit pixel. 5. according to claim 3 or 4, a kind of distance measuring method based on the vehicle monocular vision of the front vehicle driven by the plane geometric model of the lane, it is characterized in that: in step 6, the construction of the dynamic compensation model of the distance measurement error of the front target vehicle, Specifically include the following steps: The first step is to fix the CCD vision sensor, and use the calibration method described in step 1 to calibrate and record the internal and external parameters of the CCD camera; The second step is to fix the height of the target source from the ground, move the target source every 5m along the longitudinal direction of the road, so that it changes within the range of [10m, 100m] from the CCD visual sensor, and use the CCD camera to record the target source at each position Image; The third step is to adjust the height of the target source from the ground so that it changes within [0.2m, 1m], and repeat the second step; In the fourth step, use matlab to process the image obtained in the third step, and analyze the measurement error of the target source at different heights from the ground and different longitudinal positions. 6. a kind of distance measuring method based on the monocular vision of the front vehicle driven by the lane plane geometric model according to claim 5, characterized in that: in step 6, the calculation formula of the longitudinal measurement error value z is: z＝118-1124x-3.133y+3522x ² +34.6xy-0.0399y ² -4795x ³ -86.22x ² y-0.1845xy ² +0.0017y ³ +2676x ⁴ +98.62x ³ y+0.1428x ² y ² +0.001313 xy ³ -2.077e ^-5 y ⁴ +8.835e ^-8 y ⁵ -7.658e ^-6 xy ⁴ +0.0004x ² y ³ -0.1114x ³ y ² -38.23x ⁴ y-391.1x ⁵ In the formula, x is the height of the target source from the ground. 7. A method for measuring distance based on the monocular vision of the front vehicle driven by the plane geometric model of the lane according to claim 6, characterized in that: in step 7, the longitudinal vehicle distance Y between the front target vehicle and the own vehicle The formula for calculating _W (P) is: <mrow><msub><mi>Y</mi><mi>W</mi></msub><mrow><mo>(</mo><mi>P</mi><mo>)</mo></mrow><mo>=</mo><mi>h</mi><mo>&amp;times;</mo><mi>t</mi><mi>a</mi><mi>n</mi><mo>&amp;lsqb;</mo><mfrac><mi>&amp;pi;</mi><mn>2</mn></mfrac><mo>-</mo><mi>&amp;theta;</mi><mo>+</mo><mi>arctan</mi><mrow><mo>(</mo><mfrac><mrow><mo>&amp;lsqb;</mo><msub><mi>v</mi><mn>0</mn></msub><mo>-</mo><mi>v</mi><mrow><mo>(</mo><msub><mi>P</mi><mn>0</mn></msub><mo>)</mo></mrow><mo>&amp;rsqb;</mo><mo>&amp;times;</mo><mi>d</mi><mi>y</mi></mrow><mi>f</mi></mfrac><mo>)</mo></mrow><mo>&amp;rsqb;</mo><mo>-</mo><mi>z</mi><mo>.</mo></mrow>