A lane line detection method and device based on vision

Abstract

A lane line detection method and device based on vision. The detection method includes six steps: the first step is a preprocessing module, the second step is a feature extraction module, the third step is a model fitting module, the fourth step is a lane line refinement processing module, the fifth step is a tracking prediction module, and the sixth step is a lane line detection output parameter structuring. The present invention is used for unmanned autonomous driving and safety assisted driving.

Description

Technical field:

[0001] The invention relates to a vision-based lane line detection method and device. Background technology:

[0002] If you want to achieve assisted driving or unmanned autonomous driving, you must first learn to observe the road from a visual perspective. The most direct and useful method is to detect lane lines. Lane line detection is relatively easy to detect compared to other traffic signs. However, due to image blur caused by vehicle speed, changes in lighting, object occlusion, stains, etc., the image quality is affected, or the acquired sign line will be severely deformed due to changes in the camera shooting angle, thus affecting the accuracy of lane line recognition.

[0003] Lane detection methods are mainly divided into region-based detection methods, edge-based detection methods, and deep learning-based detection methods. The region-based method defines the lane recognition problem as a classification problem, that is, dividing the road image into lane line parts and non-lane line parts. It must overcome the existence of noise, such as shadows, water accumulation, and road stains, and the time spent by the classifier is generally long. This method calculates the model parameters after analyzing the target information in the image on the basis of establishing a road parameter model; the advantage of this method is that it is not affected by the ground conditions, but the disadvantage is that the calculation complexity is high, the algorithm has a large time overhead, and the real-time performance is poor. The edge-based method is to obtain the edge map of the road scene, and then use two-dimensional or three-dimensional curves to establish a lane line model for the strong geometric features of the structured road; in specific structured road scenes, this method can often achieve satisfactory results. In the face of complex driving environments, the detection effect is often unstable, and missed detection and false detection are common. The detection method based on deep learning is a relatively cutting-edge scientific research technology. Establishing a deep learning model will greatly solve the problem of unstable feature extraction in complex lane line scenes. However, how to establish a suitable model and subsequent processing is still a big problem, and this method cannot meet the requirements of vehicle real-time performance. Summary of the invention:

[0004] The purpose of the present invention is to provide a vision-based lane line detection method and device which fully utilizes the advantages of a monocular camera in visual detection, can identify lane lines of different curvatures, and combines regional and edge methods.

[0005] The above purpose is achieved through the following technical solutions:

[0006] A lane line detection method based on vision includes six steps: a first step of a preprocessing module, a second step of a feature extraction module, a third step of a model fitting module, a fourth step of a lane line refinement processing module, a fifth step of a tracking prediction module, and a sixth step of a lane line detection output parameter structuring.

[0007] In the vision-based lane line detection method, the first step preprocessing module includes image graying, image smoothing and enhancement processing technology. The image graying technology utilizes the common characteristics of the lane line to highlight the gray value of the lane line after graying to the greatest extent; the weighted average algorithm that conforms to people's visual cognition is used to process the corresponding relationship between the color RGB image and the grayscale image, which is expressed as formula (4-1);

[0008] G ray =R×0.299+G×0.587+B×0.114 (Formula 4-1)

[0009] In the smoothing and enhancement processing of the grayscale image, it is performed to obtain an image with obvious grayscale contrast, dim background, and clear lane lines, thereby reducing the complexity of the image processing process; and the noise is removed by using median filtering.

[0010] The lane line detection method based on vision, the second step feature extraction module is that the lane line mark has obvious rising edge and falling edge, and when extracting feature points, it is necessary to extract a point on the rising edge and the falling edge to form a point pair; the lane line feature point is a jump point with a large difference in grayscale value with adjacent pixels, and the local gradient method is used to extract the ground mark feature point, first calculate the mean value near the horizontal line of the pixel point, and evaluate the intensity level of the pixel grayscale value. When the grayscale intensity near the pixel is too high or too low, the grayscale gradient between the lane line and the ground will become smaller; the grayscale average value is calculated as shown in formula (4-2),

[0011]

[0012] Then calculate the edge rise point e p and the falling point e v , satisfying the conditions of formula (4-3),

[0013]

[0014] It is required that the ground identification feature points are pairs of points consisting of adjacent rising edge points and falling edge points, and that they meet a certain distance and satisfy formula (4-4).

[0015] Δw=e p (x)-e v (x)>W (Formula 4-4)

[0016] Define the lane feature points f(x) as pairs of e p (x) and e v The average value of (x), that is, formula (4-5),

[0017] f(x)=(e p(x)+e v (x)) / 2 (Formula 4-5).

[0018] In the lane line detection method based on vision, the third step model fitting module is to establish a lane line model based on the structured road constraint hypothesis, and divide the lane line field of view into three parts: near field of view, middle field of view and far field of view. Among them, the near field of view and the middle field of view area are matched with a straight line model, and the far field of view area adopts a weighted hyperbola fitting model. In the straight line fitting model, a straight line detection method combining the advantages of two algorithms is designed; first, the approximate range of the straight line is determined by using the Hough transform, and then the improved least squares method is used to obtain the precise straight line parameters for the feature point set in each area, and before the curve fitting, a curve target pixel point set is established; each pixel in the target pixel set is matched with a straight line model; All the pixels should meet the following two conditions: first, they should be on the lane line as much as possible, and second, the number of points in the set should be sufficient. Only with sufficient data can the interference caused by error pixels be reduced. Based on the parameters of line detection, the accuracy of lane line target pixel extraction in the far field of view is estimated by relying on the high accuracy of line detection. The pixel point at the end point of the line is used as the starting seed point of the search, and the gradient value correlation and gray value correlation scanning and searching are performed in the image starting from the starting seed point. The direction of the lane line is first determined according to the gradient value correlation, the search area is defined, and then the gray value correlation is used for local search. A linear hyperbola model is used for fitting in the target pixel set, as shown in Formula 4-5.

[0019]

[0020] In the lane line detection method based on vision, the lane line extraction processing module in the fourth step is to extract the lane line after fitting the lane line model, screen out reliable lane lines, and remove false detection marking lines; perform two steps of processing, one is the vanishing point constraint relationship, and the other is the parallel relationship of multiple lane lines;

[0021] In the vanishing point constraint relationship, the error tolerance region is specified in the horizontal and vertical directions respectively; assuming that the vanishing point position of the previous frame is p(x,y), the error tolerance region of the vanishing point p′(x′,y′) is formula (4-6)

[0022]

[0023] In the parallel relationship of multiple lane lines, three control points are specified for each lane line, and these three control points are designated as the intersection points of the vision line and the lane line. According to the parallel relationship of lane lines, the lane lines with the same control point number should have similar distance, slope and angle.

[0024] The lane line detection method based on vision, the fifth step tracking prediction module is in the actual acquisition system and most intelligent vehicle systems, the on-board camera directly obtains video stream information, and the vehicle movement is continuous in time and space; by using the Kalman filter to predict the movement state of each lane in real time, observe the match between the movement state and the current one.

[0025] In the vision-based lane line detection method, the sixth step of structuring the lane line detection output parameters is to perform unified structured processing on the lane line detection results.

[0026] A device for a lane line detection method based on vision, comprising: a front of a car, the front of the car having a front windshield, a reinforcement plate connected below the front windshield, the reinforcement plate connected to the sheet metal of the front of the car and fixed with bolts, the bottom of the reinforcement plate fixedly and vertically connected to a support plate, the top of the reinforcement plate connected to an adjustment cover plate via a built-in hinge, the left side of the reinforcement plate fixedly connected to a left baffle, the right side of the reinforcement plate fixedly connected to a right baffle, the support plate supporting the bottom of a camera base, the camera base connected to the reinforcement plate and fixed with screws, and the camera base connected to a monocular camera.

[0027] The device of the lane line detection method based on vision, the camera base is inserted between the support plate, the left baffle plate, and the right baffle plate, the left baffle plate is provided with an arc-shaped left adjustment opening, the left adjustment rod is inserted into the arc-shaped left adjustment opening in a tight fit, the outer end of the left adjustment rod is connected to the left fixed limit plate, the left fixed limit plate is blocked on the outer side of the left baffle plate, the left adjustment rod is connected to the left pull rod, and the left pull rod is connected to the left side of the adjustment cover plate, the right baffle plate is provided with an arc-shaped right adjustment opening, the right adjustment rod is inserted into the arc-shaped right adjustment opening in a tight fit, the outer end of the right adjustment rod is connected to the right fixed limit plate, the right fixed limit plate is blocked on the outer side of the right baffle plate, the right adjustment rod is connected to the right pull rod, and the right pull rod is connected to the right side of the adjustment cover plate; the left adjustment rod and the right adjustment rod control the inclination angle of the adjustment cover plate.

[0028] Beneficial effects:

[0029] 1. The present invention can ensure the real-time requirements of lane line detection in various scenarios and can provide reliable lane line information.

[0030] 2. The present invention is applicable to lane line detection with different curvatures and is applicable when the road line is blocked or partially damaged.

[0031] 3. The lane line detection of the present invention is based on monocular vision and can meet the needs of camera installation in different positions.

[0032] 4. The present invention can firmly fix the camera base on the front of the car, and the fixation is very firm and very safe when the car is driving. Description of the drawings:

[0033] Attached Figure 1 This is the overall flow chart of the lane line detection method.

[0034] Attached Figure 2 It is the brightness change map of a certain line scan in the lane line image.

[0035] Attached Figure 3 This is the lane line feature extraction result image.

[0036] Attached Figure 4 It is a schematic diagram of the lane line fitting model.

[0037] Attached Figure 5 This is the lane line straight line model fitting flow chart.

[0038] Attached Figure 6 This is a schematic diagram of the parallel relationship of lane lines.

[0039] Attached Figure 7 This is the lane line detection result map.

[0040] Attached Figure 8 This is the lane line parameter debugging interface.

[0041] Attached Fig. 9 It is a structural diagram of this product.

[0042] Attached Fig.10 Yes Fig. 9 Rear view of. Specific implementation method:

[0043] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention.

[0044] Embodiment 1:

[0045] A lane line detection method based on vision includes six steps: a first step of a preprocessing module, a second step of a feature extraction module, a third step of a model fitting module, a fourth step of a lane line refinement processing module, a fifth step of a tracking prediction module, and a sixth step of a lane line detection output parameter structuring.

[0046] Embodiment 2:

[0047] In the vision-based lane line detection method described in Example 1, the first step preprocessing module includes image graying, image smoothing and enhancement processing technology. The image graying technology utilizes the common characteristics of the lane line to highlight the gray value of the lane line after graying to the greatest extent; the weighted average algorithm that conforms to people's visual cognition is used to process the corresponding relationship between the color RGB image and the grayscale image, which is expressed as formula (4-1);

[0048] Gra y =R×0.299+G×0.587+B×0.114 (Formula 4-1)

[0049] Among them, R, G, and B represent the red, green, and blue color components of the color RGB image respectively. ray Represents the pixel value of a grayscale image;

[0050] In the smoothing and enhancement of grayscale images, appropriate optimization is performed to obtain images with obvious grayscale contrast, dim background, and clearer lane line contours, so as to improve the reliability of the algorithm and reduce the complexity of the image processing process; of course, not all collected images need to undergo the above preprocessing process, and it needs to be judged based on the quality of the image. Among them, what is necessary in the experiment is grayscale processing, and the noise is removed by using median filtering.

[0051] Embodiment 3:

[0052] In the vision-based lane line detection method described in Example 1, the second step feature extraction module is that the lane line mark has obvious rising and falling edges. When extracting feature points, it is necessary to extract one point on the rising and falling edges to form a point pair; the lane line feature point belongs to a jump point with a large difference in grayscale value with adjacent pixels, such as Figure 2 The figure shows the brightness change of a certain line in the lane line image. The local gradient method is used to extract the ground feature points. First, the mean value near the horizontal line of the pixel point is calculated to evaluate the intensity level of the pixel gray value. When the gray intensity near the pixel is too high or too low, the gray gradient between the lane line and the ground will become smaller. The gray average value is calculated as shown in formula (4-2).

[0053]

[0054] (x, y) is the column and row of the pixel in the image coordinate system, and t represents the area in the row. In the experiment, taking t=5 can achieve good results.

[0055] Then calculate the edge rise point e p and the falling point e v , satisfying the conditions of formula (4-3),

[0056]

[0057] It is required that the ground identification feature points are pairs of points consisting of adjacent rising edge points and falling edge points, and that they meet a certain distance and satisfy formula (4-4).

[0058] Δw=e p (x)-e v (x)>W (Formula 4-4)

[0059] e p (x) and e v (x) is the column pixel coordinate of the rising and falling points, Δw is the width of the rising and falling points, and W is the maximum number of pixels occupied by the lane line in the image.

[0060] Define the lane feature points f(x) as pairs of e p (x) and e v The average value of (x), that is, formula (4-5), such as Figure 3 The figure shows the result of feature point extraction.

[0061] f(x)=(e p (x)+e v (x)) / 2 (Formula 4-5).

[0062] Embodiment 4:

[0063] In the vision-based lane detection method described in Example 1, the third step model fitting module establishes a lane model based on the structured road constraint hypothesis, and divides the lane field of view into three parts: near field of view, middle field of view and far field of view. The near field of view and middle field of view are matched with a straight line model, and the far field of view is matched with a weighted hyperbola fitting model, such as Figure 4 The lane line fitting model is shown in Figure 1. In the straight line fitting model, we consider the advantages and disadvantages of Hough transform and least squares method and design a straight line detection method that combines the advantages of the two algorithms. First, we use Hough transform to determine the approximate range of the straight line, and then use the improved least squares method to obtain the precise straight line parameters for the feature point set in each area. The flowchart is shown in Figure 1. Figure 5As shown. Before curve fitting, a set of target pixel points of the curve is established; each pixel point in the target pixel set should meet the following two conditions: first, as many as possible are on the lane line, and second, the number of points in the set is sufficient. Only sufficient data can reduce the interference caused by error pixels; based on the parameters of straight line detection, the accuracy of lane line target pixel extraction in the far field of view is estimated by relying on the high accuracy of straight line detection; the pixel point at the end point of the straight line is used as the starting seed point of the search, and the gradient value correlation and gray value correlation scanning and searching are performed in the image starting from the starting seed point; the direction of the lane line is first determined according to the gradient value correlation, the search area is delineated, and then the gray value correlation is used for local search; the linear hyperbola model is used for fitting in the target pixel set, as shown in Formula 4-5, which is also very applicable when the road line is blocked or partially damaged.

[0064]

[0065] The parameter h in the formula represents the longitudinal position of the vanishing point of the road plane on the image plane, that is, the position of the vanishing line; k represents the curvature of the road marking; b l 、b r The values ​​of represent the distances of the left and right lane lines relative to the road markings. l =0 represents the lane marking on the right, when equation b r =0 represents the lane marking on the left; u H Indicates the distance from the lane line to the longitudinal axis.

[0066] Embodiment 5:

[0067] In the vision-based lane line detection method described in Example 1, the fourth step lane line extraction processing module is to extract the lane lines after the lane line model is fitted, mainly to screen out reliable lane lines and eliminate false detection marking lines; we mainly perform two steps of processing, one is the vanishing point constraint relationship, and the other is the parallel relationship of multiple lane lines;

[0068] In the vanishing point constraint relationship, the error tolerance region is specified in the horizontal and vertical directions respectively; assuming that the vanishing point position of the previous frame is p(x,y), the error tolerance region of the vanishing point p′(x′,y′) is formula (4-6)

[0069]

[0070] T x and T y They are the range of movement in the horizontal and vertical directions respectively.

[0071] In the parallel relationship of multiple lane lines, at least three control points are specified for each lane line, and these three control points are designated as the intersection points of the vision line and the lane line. Figure 6 The figure shows a schematic diagram of multi-lane line control points. i-j The information of the lane control point includes the position information, the slope and angle information of the lane at that point, where i represents the lane number and j represents the control point number of each lane. According to the parallel relationship of lanes, lanes with the same control point number should have similar distance, slope and angle. For example, S 1-1 , S 2-1 , S 3-1 , S 4-1 The position distances between them are close, and the changes in slope difference and angle difference are small.

[0072] Embodiment 6:

[0073] In the lane detection method based on vision described in Example 1, the fifth step tracking prediction module is in the actual acquisition system and most intelligent vehicle systems. The vehicle camera directly obtains video stream information, and there is often a large redundancy between two adjacent frames in the video stream. The vehicle movement is continuous in time and space; the Kalman filter is used to predict the motion state of each lane in real time, and the motion state is observed to match the current state.

[0074] Embodiment 7:

[0075] In the vision-based lane detection method described in Example 1, the sixth step of structuring the lane detection output parameters is to display the lane detection results on different platforms in a better manner and to perform unified structural processing on the lane detection results. Specifically, it includes the following contents:

[0076] ①Model degree (model level, 0- straight line, 1- quadratic curve, 2- cubic curve).

[0077] ②Quality (lane line quality, the higher the value, the more reliable it is).

[0078] ③Lanetype (lane line type, 0-solid white, 1-solid yellow, 2-dashed white, 3-dashed yellow, 4-road edge, 5-unspecified).

[0079] ④PositionParameter (position parameters C3, C2, C1, C0, the equation represented is the cubic curve equation x = C3y 3 +C2y 2 +C1y+C0).

[0080] Embodiment 8:

[0081] The method for lane line detection based on vision described in Example 1 mainly includes the following steps:

[0082] (1) Obtain image stream; when designing the system, in order to be compatible with the online data collection and offline data of the smart car camera, the image acquisition module is separated; online data needs to be saved to the memory in the form of a file stream, and only the latest file is operated in subsequent operations; offline data supports two types of image data, jpg and png. If it is video data, it needs to be converted into an image first; the monocular camera is installed in the center of the car's front windshield, about 1 meter away from the ground, to ensure that the camera's optical axis is parallel to the plane of the vehicle chassis and faces the front of the vehicle.

[0083] (2) Lane line parameter debugging. In the process of this invention, in order to intuitively display the parameters in the algorithm process, such as Figure 8 The interface shown here uses the three-line calibration method to intuitively display the parameters such as field of view partition, region of interest, line width, Hough transform, etc. in lane line detection; the parameters mainly include:

[0084] ①imageSize (processing image resolution, the default is 800pix*608pix);

[0085] ② The vertical coordinates of the divided visual field, Y1 (the vertical coordinate of the upper end of the middle field of view), Y2 (the vertical coordinate of the upper end of the near field of view), Y3 (the vertical coordinate of the lower end of the near field of view);

[0086] ③ Hough transform parameters, minLength (minimum length of a straight line), minVote (minimum number of votes for a straight line), maxGap (distance tolerance between points on the same straight line);

[0087] ④minLaneWidth (determines and displays the minimum distance between two lane lines); the parameters are saved in the config.txt file.

[0088] (3) Lane line detection: Place the parameter file and the lane line detection program in the same directory and start the program to detect lane lines in the image in real time.

[0089] (4) Lane detection output parameter structuring: The lane detection results are structured and the position of each lane is output as a cubic curve equation.

[0090] (5) Output of lane line detection results: The parameter file can be set to display the lane line detection results, g_maxLeftCount (the maximum number of left lane lines that can be displayed), g_maxRightCount (the maximum number of right lane lines that can be displayed), and can also be set to display the intermediate results of all steps.

[0091] Embodiment 9:

[0092] A device for a lane line detection method based on vision, comprising: a car front 1, the car front having a front windshield 2, a reinforcement plate 3 connected below the front windshield, the reinforcement plate connected to the sheet metal of the car front and fixed with bolts, the bottom of the reinforcement plate fixedly and vertically connected to a support plate 4, the top of the reinforcement plate connected to an adjustment cover plate 5 via a built-in hinge, the left side of the reinforcement plate fixedly connected to a left baffle 6, the right side of the reinforcement plate fixedly connected to a right baffle 7, the support plate supporting the bottom of a camera base 8, the camera base connected to the reinforcement plate and fixed with screws, and the camera base connected to a monocular camera 9.

[0093] Embodiment 10:

[0094] In the device of a lane line detection method based on vision described in Example 8, the camera base is inserted between the support plate, the left baffle plate, and the right baffle plate. The left baffle plate is provided with an arc-shaped left adjustment opening 10, and the left adjustment rod 11 is inserted into the arc-shaped left adjustment opening in a tight fit. The outer end of the left adjustment rod is connected to a left fixed limit plate 12, and the left fixed limit plate is blocked on the outer side of the left baffle plate. The left adjustment rod is connected to a left pull rod 13, and the left pull rod is connected to the left side of the adjustment cover plate. The right baffle plate is provided with an arc-shaped right adjustment opening 14, and the right adjustment rod 15 is inserted into the arc-shaped right adjustment opening in a tight fit. The outer end of the right adjustment rod is connected to a right fixed limit plate 16, and the right fixed limit plate is blocked on the outer side of the right baffle plate. The right adjustment rod is connected to a right pull rod 17, and the right pull rod is connected to the right side of the adjustment cover plate. The left adjustment rod and the right adjustment rod control the tilt angle of the adjustment cover plate. Adjusting the adjustment cover plate can well protect the monocular camera from being rained on, and can extend the service life of the monocular camera.

Claims

1. A lane line detection method based on vision, characterized by : This detection method includes six steps: the first step is a preprocessing module, the second step is a feature extraction module, the third step is a model fitting module, the fourth step is a lane line refinement processing module, the fifth step is a tracking prediction module, and the sixth step is a lane line detection output parameter structuring; The first step preprocessing module includes image smoothing and enhancement processing technology, and uses a weighted average algorithm that conforms to people's visual cognition to process the corresponding relationship between the color RGB image and the grayscale image, which is expressed as formula (4-1); G ray =R×0.299+G×0.587+B×0.114 (Formula 4-1) Among them, R, G, and B represent the red, green, and blue color components of the color RGB image respectively. ray Represents the pixel value of a grayscale image; In the smoothing and enhancement processing of the grayscale image, an image with obvious grayscale contrast, dim background and clear lane line is obtained to reduce the complexity of the image processing process; the noise is removed by median filtering; and the lane line detection method based on vision is used for detection; The second step feature extraction module is that the lane line mark has obvious rising and falling edges. When extracting feature points, it is necessary to extract a point on the rising and falling edges to form a point pair; the lane line feature point belongs to the jump point with a large difference in grayscale value with the adjacent pixel. The local gradient method is used to extract the ground mark feature point. First, the mean value near the horizontal line of the pixel point is calculated to evaluate the intensity level of the pixel grayscale value. When the grayscale intensity near the pixel is too high or too low, the grayscale gradient between the lane line and the ground will become smaller; the calculation formula of the grayscale average value is shown in (4-2), (x, y) is the column and row of the pixel in the image coordinate system, t represents the area in the row, and in the experiment, t = 5 can achieve good results; Then calculate the edge rise point e p and the falling point e v , satisfying the conditions of formula (4-3), It is required that the ground identification feature points are pairs of points consisting of adjacent rising edge points and falling edge points, and that they meet a certain distance and satisfy formula (4-4). Δw=e p (x)-e v (x)>W (Official 4-4) e p (x) and e v (x) is the column pixel coordinates of the rising and falling points, Δw is the width of the rising and falling points, and W is the maximum number of pixels occupied by the lane line in the image; Define the lane feature points f(x) as pairs of e p (x) and e v The average value of (x), that is, formula (4-5), f(x)=(e p (x)+e v (x)) / 2 (Formula 4-5); The third step model fitting module is to establish a lane line model based on the structured road constraint hypothesis, and divide the lane line field of view into three parts: near field of view, middle field of view and far field of view. Among them, the near field of view and middle field of view area are matched with a straight line model, and the far field of view area adopts a weighted hyperbola fitting model. In the straight line fitting model, a straight line detection method combining the advantages of two algorithms is designed; first, the approximate range of the straight line is determined by using the Hough transform, and then the improved least squares method is used to obtain the precise straight line parameters for the feature point set in each area. Before curve fitting, a curve target pixel point set is established; each pixel point in the target pixel set should meet the following two requirements: Conditions: First, as many points as possible are on the lane line, and second, the number of points in the set is sufficient. Only sufficient data can reduce the interference caused by error pixels; based on the parameters of straight line detection, rely on the high accuracy of straight line detection to estimate the accuracy of lane line target pixel extraction in the far field of view; use the pixel point at the end point of the straight line as the starting seed point of the search, and start scanning and searching the image for gradient value correlation and gray value correlation from the starting seed point; first determine the direction of the lane line based on the gradient value correlation, define the search area, and then use the gray value correlation for local search; use a linear hyperbola model for fitting in the target pixel set, as shown in Formula 4-5, The parameter h in the formula represents the longitudinal position of the vanishing point of the road plane on the image plane, that is, the position of the vanishing line; k represents the curvature of the road marking; b l 、b r The values ​​of represent the distances of the left and right lane lines relative to the road markings. l =0 represents the lane marking on the right, when equation b r =0 represents the lane marking on the left; u H Indicates the distance from the lane line to the longitudinal axis; The fourth lane line extraction processing module is to extract the lane lines after the lane line model is fitted, select reliable lane lines, and remove false detection marking lines; perform two steps of processing, one is the vanishing point constraint relationship, and the other is the parallel relationship of multiple lane lines; In the vanishing point constraint relationship, the error tolerance area is specified in the horizontal and vertical directions respectively; assuming that the vanishing point position of the previous frame is p(x,y), the error tolerance area of ​​the vanishing point p'(x',y') is formula (4-6) T x and T y The range of movement in the horizontal and vertical directions respectively; In the parallel relationship of multiple lane lines, three control points are specified for each lane line, and these three control points are designated as the intersection points of the vision line and the lane line; according to the parallel relationship of the lane lines, the lane lines with the same control point sequence number should have similar distance, slope and angle; The fifth step tracking prediction module is in the actual acquisition system and most intelligent vehicle systems. The on-board camera directly obtains video stream information, and the vehicle movement is continuous in time and space. The Kalman filter is used to predict the movement state of each lane in real time and observe the match between the movement state and the current state. The sixth step of structuring the lane line detection output parameters is to perform unified structured processing on the lane line detection results.

2. The device for a lane line detection method based on vision according to claim 1, comprising: a front of a car, characterized in that The front of the automobile has a front windshield, a reinforcement plate is connected below the front windshield, the reinforcement plate is connected to the sheet metal of the front of the automobile and fixed with bolts, the bottom of the reinforcement plate is fixedly connected vertically to a support plate, the top of the reinforcement plate is connected to an adjustment cover plate through a built-in hinge, the left side of the reinforcement plate is fixedly connected to a left baffle, the right side of the reinforcement plate is fixedly connected to a right baffle, the support plate supports the bottom of the camera base, the camera base is connected to the reinforcement plate and fixed with screws, and the camera base is connected to a monocular camera.

3. The device for a vision-based lane line detection method according to claim 2, characterized in that :The camera base is inserted between the support plate, the left baffle plate, and the right baffle plate, the left baffle plate is provided with an arc-shaped left adjustment opening, the left adjustment rod is inserted into the arc-shaped left adjustment opening in a tight fit, the outer end of the left adjustment rod is connected to the left fixed limit plate, the left fixed limit plate is blocked on the outer side of the left baffle plate, the left adjustment rod is connected to the left pull rod, and the left pull rod is connected to the left side of the adjustment cover plate, the right baffle plate is provided with an arc-shaped right adjustment opening, the right adjustment rod is inserted into the arc-shaped right adjustment opening in a tight fit, the outer end of the right adjustment rod is connected to the right fixed limit plate, the right fixed limit plate is blocked on the outer side of the right baffle plate, the right adjustment rod is connected to the right pull rod, and the right pull rod is connected to the right side of the adjustment cover plate; the left adjustment rod and the right adjustment rod control the inclination angle of the adjustment cover plate.

Images