CN106407951A - Monocular vision-based nighttime front vehicle detection method - Google Patents

Abstract

The invention discloses a monocular vision-based nighttime front vehicle detection method. The method includes the following steps that: 1, images are acquired, and vehicle lamp detection is performed on a nighttime front vehicle based on a CenSurE operator, so that strong corner points can be obtained; 2, nighttime front vehicle segmentation is carried out based on vehicle lamp color information, so that segmented regions can be obtained; 3, a region with the most strong corner points is selected from segmented regions obtained in the step 2, so that a detection region can be obtained; and 4, vehicle lamp matching is performed on the detection region, so that the location of a target vehicle can be determined. According to the method of the invention, the CenSurE operator under a plurality of scales is calculated, the results of the CenSurE operator are detected, and vehicle matching operation is carried out, and therefore, the problems of large matching error and inaccurate target detection of an existing night nighttime front vehicle detection method can be solved. The monocular vision-based nighttime front vehicle detection method has a high application value.

Description

A night-time front vehicle detection method based on monocular vision

technical field

The invention belongs to the technical field of night target detection methods, and in particular relates to a nighttime front vehicle detection method based on monocular vision.

Background technique

The rapid growth of automobiles has aggravated the occurrence of road traffic accidents, especially the frequent occurrence of traffic accidents at night, which has brought huge losses to people's lives and properties. Vision-based target detection technology provides the possibility for target detection in night traffic scenes. Since the matching of feature points between adjacent frames of the multi-eye (including binocular) vision system has a large amount of calculation, when the number of vehicles is large, the real-time performance is poor. Therefore, based on monocular Vision vehicle detection technology came into being. However, most of the existing vehicle detection technologies based on monocular vision are suitable for the daytime traffic environment. Due to the worsening driving environment at night, traffic accidents at night are more frequent than during the daytime. Therefore, research on the detection method of vehicles ahead at night based on monocular vision It is of great significance to improve the driving environment and reduce traffic accidents.

The light conditions at night are poor, and the appearance characteristics of the vehicle are difficult to be detected. The most prominent feature of the vehicle at night is the high-brightness headlights. Therefore, most nighttime vehicle detection methods detect the vehicle by detecting the headlights. Junbin Guo et al. counted the distribution of taillight brightness in 300 different environments, and used the maximum inter-class variance method (Otsu) to determine the optimal segmentation threshold, and eliminated non-taillight targets according to the red threshold of the HSV color space, and finally based on position, area, etc. Taillight pairing with prior knowledge.

Based on the light scattering attenuation model, Wei Zhang et al. obtained the reflection grayscale map and reflection suppression map of the headlights as the input vector of the Markov random field, but if the vehicle is very close to the camera, the reflection coefficient is difficult to calculate.

Jiann-Der Lee et al. used the LoG operator and light scattering model to obtain the headlight area and used the optical flow method to track the vehicle. Using the LoG operator, the problem of difficult detection of close-range vehicles was solved.

O'Malley et al. proposed a red threshold segmentation method based on the HSV color space, and verified the symmetry of the lights according to the cross-correlation coefficient and used the Kalman filter for tracking, but only paired the lights according to the cross-correlation coefficient, and the error was large .

Naoya Kosaka et al. used a double-layer center-surround filter to approximate the LoG operator, screened out the feature points with high response values, and then used the support vector machine (SVM) to classify the feature points, and eliminated noise through lane line detection and motion trajectory point, the detection accuracy rate of headlights is high, but the pairing is only based on the principle of consistent filter response values, resulting in a large pairing error.

Hulin Kuang et al. used EdgeBoxes to find a region of interest (ROI) with a higher score in the multi-scale Retinex enhanced image. After extracting 5 features of the ROI, the SVM trained the weight of each feature and modified the final score. The ROI with a high score is vehicle. This method does not require car light pairing, which reduces some errors, but for dim traffic scenes, the enhancement algorithm is not very effective, and the accuracy of the ROI obtained also decreases.

Contents of the invention

The purpose of the present invention is to provide a nighttime vehicle detection method based on monocular vision, which solves the problems of relatively large matching errors and inaccurate detection targets in the existing nighttime vehicle detection method.

A kind of technical scheme adopted in the night vehicle detection method based on monocular vision of the present invention is, comprises the following steps:

Step 1, collect images, and detect the headlights of the vehicle ahead at night based on the CenSurE operator to obtain strong corner points;

Step 2: Segment the vehicle ahead at night based on the light color information to obtain the segmented area;

Step 3, select the region that occupies the most strong corner points in step 1 in the segmented region in step 2, and obtain the detection region;

Step 4, perform vehicle light pairing on the detection area in step 3, and determine the position of the target vehicle.

The present invention is also characterized in that,

The operation steps of step 1 are as follows:

Step 1.1, take pictures and collect images, calculate the value of the corresponding integral map according to the collected image, the value of any point I(x, y) in the integral map is the sum of all values in the upper left corner of the corresponding position in the original image, such as formula (1 ),

Step 1.2, construct a CenSurE filter to sample the integral image on a logarithmic scale,

The scale space represented by the value of any point I(x, y) is divided into three groups, in which the size of the CenSurE filter kernel of each layer in the first group is increased by 2 in turn, and the size of the CenSurE filter kernel of each layer in the second group is in turn Increase by 4, the CenSurE filter kernel size of each layer in the third group is increased by 8 in turn, and 5 layers of scale images are selected in each group, and the CenSurE filter outer kernel size is also calculated according to the above method;

That is, the kernel size of the CenSurE filter should satisfy (2n+1)×(2n+1), and the outer kernel size should satisfy (4n+1)×(4n+1). In order to make the DC response of the filter zero, the scale Space normalization, then the weight coefficient I _n of the kernel should satisfy formula (2),

The weight coefficient O _n of the outer core should satisfy formula (3)

When the sum of the pixel values contained in the outer core is out_value and the sum of the pixel values contained in the inner core is in_value, the pixel filter response value L satisfies formula (4),

L=O _n ·out_value-I _n ·in_value, (4)

Step 1.3, perform extreme value detection on the scale space in step 1.2,

With the image processed in step 1.2, calculate the pixel filter response value of each scale space in the image according to formula (4), then perform non-maximum value suppression in the scale space, and record the extreme point;

Step 1.4, filter the extreme points in step 1.3 to remove unstable feature points,

L _x and L _y are the partial derivatives of the pixel filter response value L in the x and y directions. Gaussian filtering is performed on L _x , L _y , and L _x L _y to obtain the eigenvalues of the Harris matrix. If the smaller eigenvalues are larger than the adaptive Threshold t, strong corner points are obtained.

The operation steps of step 2 are as follows: convert the image collected in step 1 from RGB space to HSV color space, use the red threshold to segment all the regions, use the white threshold to segment all the regions located in the left 1/3 region, Together as a result of HSV color space segmentation,

H means hue, where H≥340° or H≤30° for red threshold, H for white threshold is 0°～360°; S means saturation, S≤30 for red threshold, S≤20 for white threshold; V means The lightness of the color, the red threshold and the white threshold are all taken as 80≤V≤100.

The operation steps of headlight pairing in step 4 are specifically as follows, assuming that L _i and L _j are two candidate lights, the areas are A _i and A _j respectively, and the image coordinates of the center of the lights are (x _i , y _i ), (x _j ,y _j ), the pairing constraints are as follows:

a. When the heights of the two candidate lights are the same, the vertical coordinates of the two lights should satisfy the formula (5)

|y _i -y _j |<Δh, (5)

b. When the horizontal distance between the two candidate lights is within a certain range, formula (6) should be satisfied

Δw ₁ <| _xi -x _j |<Δw ₂ , (6)

c. When the areas of the two candidate lights are the same, formula (7) should be satisfied

|A _i -A _j |<ΔA, (7)

In formula (5), Δh is the height difference threshold, in formula (6), Δw ₁ and Δw ₂ are the level difference thresholds, in formula (7), ΔA is the area difference threshold, which satisfies the pairing constraint conditions, and after the pairing is completed, the corresponding The circumscribed rectangular frame should satisfy the aspect ratio within a certain range and satisfy formula (8),

Among them, x _i,left , x _j,right are the leftmost and rightmost coordinates of the area, y _i,top , y _j,bottom are the uppermost and lowermost coordinates of the area, and Δration is the aspect ratio threshold of the box.

The specific steps of non-maximum value suppression in step 1.3 are: each point in the scale space and its 26 adjacent points, where the 26 adjacent points include the detection point in the middle and its 8 adjacent points of the same scale , and the 9×2 points corresponding to the upper and lower adjacent scales are compared, and then the extreme points are recorded.

The adaptive threshold in step 1.4 is obtained using the multi-level Otsu method, and the specific steps are:

In the gray histogram, let f _i be the number of pixels with gray level i, and N be the total number of pixels, then N satisfies formula (9)

N=f ₀ +f ₁ +...+f _l-1 , (9)

Where l is the number of histograms, l=1, 2, 3, 4...,

Then the distribution probability P _i of the number f _i of pixels with gray level i is formula (10),

Use k thresholds T={t ₁ ,...,t _n ,...,t _k } to divide the image into k+1 categories, and the inter-class variance V _BC (T) is formula (11)

Wherein, formula (11) μ _n is the gray scale mean value when k=n, μ _T is the overall gray scale mean value, the values of w _n and μ _n are as formula (12),

The intra-class variance v _WC (T) is formula (13)

Among them, σ _n in formula (13) is the gray variance when k=n, and the values of w _n and σ _n are as in formula (14)

Combining formula (9) to formula (14), the total variance v _T of the image and the total mean value μ _T of the image can be obtained, which is formula (15)

Define the segmentation factor SF of the image as formula (16),

When SF>0.9, the classification is stopped, and t _k at this time is taken as the adaptive threshold.

The beneficial effect of the present invention is that: a nighttime front vehicle detection method based on monocular vision of the present invention can carry out the operation of vehicle pairing by using the CenSurE operator to calculate multi-scales, and then detect the results, which not only solves the problem of night vehicle In the light detection method, the pairing error is large, and the detection target is not accurate, and it has a good application value.

Description of drawings

Fig. 1 is the structural diagram of the inner core and outer core of CenSurE filter of the present invention;

Figure 2 is a schematic diagram of area pixels and calculations.

detailed description

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

A kind of nighttime front vehicle detection method based on monocular vision of the present invention, comprises the following steps:

Step 1, collect images, and detect the headlights of the vehicle ahead at night based on the CenSurE operator to obtain strong corner points;

Step 2: Segment the vehicle ahead at night based on the light color information to obtain the segmented area;

Step 3, select the region that occupies the most strong corner points in step 1 in the segmented region in step 2, and obtain the detection region;

Step 4, perform vehicle light pairing on the detection area in step 3, and determine the position of the target vehicle.

The operation steps of step 1 are as follows:

Step 1.1, take pictures and collect images, calculate the value of the corresponding integral map according to the collected image, the value of any point I(x, y) in the integral map is the sum of all values in the upper left corner of the corresponding position in the original image, such as formula (1 ),

In step 1.2, construct a CenSurE filter, and sample the integral image on a logarithmic scale. The purpose is to improve the stability of local extremum points. As shown in Figure 1, the CenSurE filter uses a square kernel, which has the highest calculation efficiency and satisfies the Real-time requirements, after the integral map is constructed, the cumulative sum of pixels in any matrix area in the image can be obtained through simple operations, as shown in Figure 2,

The scale space represented by the value of any point I(x, y) is divided into three groups, and 5 layers of scale images are selected in each group, among which, the kernel size of the CenSurE filter of each layer in the first group is increased by 2 in turn, and the filter is selected filter kernel, its size is 3×3, 5×5, 7×7, 9×9 and 11×11 in turn, the size of the CenSurE filter kernel in each layer in the second group is increased by 4 in sequence, and the filter kernel is selected, and its size The order is 7×7, 11×11, 15×15, 19×19 and 23×23, the size of the CenSurE filter kernel of each layer in the third group is increased by 8 in turn, and the filter kernel is selected, and its size is 15×15 in turn , 23×23, 31×31, 39×39, 47×47, the size of the outer core of the CenSurE filter is also calculated according to the above method;

That is, the kernel size of the CenSurE filter should satisfy (2n+1)×(2n+1), and the outer kernel size should satisfy (4n+1)×(4n+1). In order to make the DC response of the filter zero, the scale Space normalization, then the weight coefficient I _n of the kernel should satisfy formula (2),

The weight coefficient O _n of the outer core should satisfy formula (3)

When the sum of the pixel values contained in the outer core is out_value and the sum of the pixel values contained in the inner core is in_value, the pixel filter response value L satisfies formula (4),

L=O _n ·out_value-I _n ·in_value, (4)

Step 1.3, perform extreme value detection on the scale space in step 1.2,

With the image processed in step 1.2, calculate the pixel filter response value of each scale space in the image according to formula (4), then perform non-maximum suppression in the scale space, and record the extreme points;

Step 1.4, filter the extreme points in step 1.3 to remove unstable feature points,

In order to obtain stable feature points, it is not enough to filter out weak response points based on the threshold value, because the filter has a relatively strong response value to the edge of the image, and once the feature points fall on the edge of the image, these points are very unstable. Since the feature points on the edge or line have a large principal curvature in the parallel direction and a small principal curvature in the vertical direction, the scale-adaptive Harris method is used to calculate the principal curvature ratio H to remove the unstable response points, as shown in the following formula ,

L _x and L _y are the partial derivatives of the pixel filter response value L in the x and y directions. Gaussian filtering is performed on the principal curvature ratio H to obtain the eigenvalues of the Harris matrix. If the smaller eigenvalues are greater than the adaptive threshold t, a strong corner.

Since the headlights and taillights highlight white and red respectively, it is necessary to segment the red and white areas of the image, the operation steps of step 2 are as follows: convert the image collected in step 1 from the RGB space to the HSV color space, and use the threshold Segment the red and white areas. From experience, the vehicles coming from the opposite direction generally appear on the left side of the image, detect the headlights, detect the tail lights of the vehicles traveling in the same direction, and segment all the areas separated by the red threshold and the white threshold. The area located in the left 1/3 of all areas of , together as the result of HSV color space segmentation,

H means hue, where H≥340° or H≤30° for red threshold, H for white threshold is 0°～360°; S means saturation, S≤30 for red threshold, S≤20 for white threshold; V means The lightness of the color, the red threshold and the white threshold are all taken as 80≤V≤100.

The operation steps of headlight pairing in step 4 are specifically as follows, assuming that L _i and L _j are two candidate lights, the areas are A _i and A _j respectively, and the image coordinates of the center of the lights are (x _i , y _i ), (x _j ,y _j ), the pairing constraints are as follows:

a. When the heights of the two candidate lights are the same, the vertical coordinates of the two lights should satisfy the formula (5)

|y _i -y _j |<Δh, (5)

b. When the horizontal distance between the two candidate lights is within a certain range, formula (6) should be satisfied

Δw ₁ <| _xi -x _j |<Δw ₂ , (6)

c. When the areas of the two candidate lights are the same, formula (7) should be satisfied

|A _i -A _j |<ΔA, (7)

In formula (5), Δh is the height difference threshold, in formula (6), Δw ₁ and Δw ₂ are the level difference thresholds, in formula (7), ΔA is the area difference threshold, which satisfies the pairing constraint conditions, and after the pairing is completed, the corresponding The circumscribed rectangular frame should satisfy the aspect ratio within a certain range and satisfy formula (8),

Among them, x _i,left , x _j,right are the leftmost and rightmost coordinates of the area, y _i,top , y _j,bottom are the uppermost and lowermost coordinates of the area, and Δration is the aspect ratio threshold of the box.

In formulas (7)-(8), due to daily life experience, Δh is taken as 10 pixels, Δw ₁ is taken as 20 pixels, Δw ₂ is taken as 50 pixels, ΔA is taken as 30 pixels, and Δration is taken as 10 pixels for calculation and matching.

The specific steps of non-maximum value suppression in step 1.3 are: each point in the scale space and its 26 adjacent points, where the 26 adjacent points include the detection point in the middle and its 8 adjacent points of the same scale , and the 9×2 points corresponding to the upper and lower adjacent scales are compared, and then the extreme points are recorded.

The adaptive threshold in step 1.4 is obtained using the multi-level Otsu method, and the specific steps are:

In the gray histogram, let f _i be the number of pixels with gray level i, and N be the total number of pixels, then N satisfies formula (9)

N=f ₀ +f ₁ +...+f _l-1 , (9)

Where l is the number of histograms, l=1, 2, 3, 4...,

Then the distribution probability P _i of the number f _i of pixels with gray level i is formula (10),

Use k thresholds T={t ₁ ,...,t _n ,...,t _k } to divide the image into k+1 categories, and the inter-class variance V _BC (T) is formula (11)

Wherein, formula (11) μ _n is the gray scale mean value when k=n, μ _T is the overall gray scale mean value, the values of w _n and μ _n are as formula (12),

The intra-class variance v _WC (T) is formula (13)

Among them, σ _n in formula (13) is the gray variance when k=n, and the values of w _n and σ _n are as in formula (14)

Combining formula (9) to formula (14), the total variance v _T of the image and the total mean value μ _T of the image can be obtained, which is formula (15)

Define the segmentation factor SF of the image as formula (16),

When SF>0.9, the classification is stopped, and t _k at this time is taken as the adaptive threshold.

The beneficial effect of the present invention is: the present invention can carry out the calculation of vehicle pairing by adopting the CenSurE operator under multi-scale calculation, and then detecting the result, which not only solves the problem of relatively large pairing error in the detection method of vehicle lights at night, and the detection target Inaccurate problem, but also very good application value.

Claims (6)

1. a night vehicle detection method based on monocular vision, is characterized in that, comprises the following steps: Step 1, collect images, and detect the headlights of the vehicle ahead at night based on the CenSurE operator to obtain strong corner points; Step 2: Segment the vehicle ahead at night based on the light color information to obtain the segmented area; Step 3, select the region that occupies the most strong corner points in step 1 in the segmented region in step 2, and obtain the detection region; Step 4, perform vehicle light pairing on the detection area in step 3, and determine the position of the target vehicle. 2. a kind of night vehicle detection method based on monocular vision according to claim 1, is characterized in that, the operation step of described step 1 is specifically: Step 1.1, take pictures and collect images, calculate the value of the corresponding integral map according to the collected image, the value of any point I(x, y) in the integral map is the sum of all values in the upper left corner of the corresponding position in the original image, such as formula (1 ), $\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}{\underset{{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; x</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$ Step 1.2, construct a CenSurE filter to sample the integral image on a logarithmic scale, The scale space represented by the value of any point I(x, y) is divided into three groups, in which the size of the CenSurE filter kernel of each layer in the first group is increased by 2 in turn, and the size of the CenSurE filter kernel of each layer in the second group is in turn Increase by 4, the CenSurE filter kernel size of each layer in the third group is increased by 8 in turn, and 5 layers of scale images are selected in each group, and the CenSurE filter outer kernel size is also calculated according to the above method; That is, the kernel size of the CenSurE filter should satisfy (2n+1)×(2n+1), and the outer kernel size should satisfy (4n+1)×(4n+1). In order to make the DC response of the filter zero, the scale Space normalization, then the weight coefficient I _n of the kernel should satisfy formula (2), ${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}\frac{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 3...</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$ The weight coefficient O _n of the outer core should satisfy formula (3) ${\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}\frac{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 3...</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$ When the sum of the pixel values contained in the outer core is out_value and the sum of the pixel values contained in the inner core is in_value, the pixel filter response value L satisfies formula (4), L=O _n ·out_value-I _n ·in_value, (4) Step 1.3, perform extreme value detection on the scale space in step 1.2, With the image processed in step 1.2, calculate the pixel filter response value of each scale space in the image according to formula (4), then perform non-maximum suppression in the scale space, and record the extreme points; Step 1.4, filter the extreme points in step 1.3 to remove unstable feature points, L _x and L _y are the partial derivatives of the pixel filter response value L in the x and y directions. Gaussian filtering is performed on L _x , L _y , and L _x L _y to obtain the eigenvalues of the Harris matrix. If the smaller eigenvalues are larger than the adaptive Threshold t, strong corner points are obtained. 3. a kind of nighttime front vehicle detection method based on monocular vision according to claim 1, is characterized in that, the operation step of described step 2 is specifically, the image that step 1 collects is converted into HSV color by RGB space Space, all regions segmented using the red threshold value, and the left 1/3 region of all regions segmented using the white threshold value are collectively the result of HSV color space segmentation. H means hue, where H≥340° or H≤30° for red threshold, H for white threshold is 0°～360°; S means saturation, S≤30 for red threshold, S≤20 for white threshold; V means The lightness of the color, the red threshold and the white threshold are all taken as 80≤V≤100. 4. A night-time front vehicle detection method based on monocular vision according to claim 1, characterized in that, the operation step of the headlight pairing in step 4 is specifically assuming that L _i and L _j are two candidates Car lights, the areas are A _i , A _j , the image coordinates of the center of the car lights are ( _xi , y _i ), (x _j , y _j ), and the pairing constraints are as follows: a. When the heights of the two candidate lights are the same, the vertical coordinates of the two lights should satisfy the formula (5) |y _i -y _j |<Δh, (5) b. When the horizontal distance between the two candidate lights is within a certain range, formula (6) should be satisfied Δw ₁ <| _xi -x _j |<Δw ₂ , (6) c. When the areas of the two candidate lights are the same, formula (7) should be satisfied |A _i -A _j |<ΔA, (7) In formula (5), Δh is the height difference threshold, in formula (6), Δw ₁ and Δw ₂ are the level difference thresholds, in formula (7), ΔA is the area difference threshold, which satisfies the pairing constraint conditions, and after the pairing is completed, the corresponding The circumscribed rectangular frame should satisfy the aspect ratio within a certain range and satisfy formula (8), $\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; t</span>}}}{\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$ Among them, x _i,left , x _j,right are the leftmost and rightmost coordinates of the area, y _i,top , y _j,bottom are the uppermost and lowermost coordinates of the area, and Δration is the aspect ratio threshold of the box. 5. a kind of nighttime front vehicle detection method based on monocular vision according to claim 2, is characterized in that, the concrete step of the non-maximum suppression in step 1.3 is: each point in the scale space and its 26 Adjacent points, among which 26 adjacent points include the detection point in the middle and 8 adjacent points of the same scale, and 9×2 points corresponding to the upper and lower adjacent scales for comparison, and then record the extreme points. 6. a kind of night front vehicle detection method based on monocular vision according to claim 2, is characterized in that, the self-adaptive threshold in described step 1.4 uses multistage Otsu method to obtain, and its concrete steps are: In the gray histogram, let f _i be the number of pixels with gray level i, and N be the total number of pixels, then N satisfies formula (9) N=f ₀ +f ₁ +...+f _l-1 , (9) Where l is the number of histograms, l=1, 2, 3, 4..., Then the distribution probability P _i of the number f _i of pixels with gray level i is formula (10), $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; N</span>}\\ \underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right.<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 4...</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$ Use k thresholds T={t ₁ ,...,t _n ,...,t _k } to divide the image into k+1 categories, and the inter-class variance V _BC (T) is formula (11) ${\mathrm{<span\; class="notranslate">\; V</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 3...</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; ...</span>}<span\; class="notranslate">\; ;</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$ Wherein, formula (11) μ _n is the gray scale mean value when k=n, μ _T is the overall gray scale mean value, the values of w _n and μ _n are as formula (12), $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 12</span>}<span\; class="notranslate">\; )</span>$ The intra-class variance v _WC (T) is formula (13) ${\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; W</span>}\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 13</span>}<span\; class="notranslate">\; )</span>$ Among them, σ _n in formula (13) is the gray variance when k=n, and the values of w _n and σ _n are as in formula (14) $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}\end{array}\right.<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; )</span>$ Combining formula (9) to formula (14), the total variance v _T of the image and the total mean value μ _T of the image can be obtained, which is formula (15) $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; IP</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\end{array}\right.<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 15</span>}<span\; class="notranslate">\; )</span>$ Define the segmentation factor SF of the image as formula (16), $\mathrm{<span\; class="notranslate">\; S</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 16</span>}<span\; class="notranslate">\; )</span>$ When SF>0.9, the classification is stopped, and t _k at this time is taken as the adaptive threshold.