A road texture feature detection method, device and automatic driving system
Through the road texture feature detection method, dense parallax map and Gabor filter are used to extract texture features, which solves the problem of missed target detection in the autonomous driving system, and improves road perception accuracy and driving safety.
Patent Information
- Application Number
- CN202110906132.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-08-09
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2041-08-09
AI Technical Summary
In the prior art, there are problems of missed target detection, making it difficult to detect the road environment ahead in real time, affecting the comfort and safety of autonomous driving.
The road texture feature detection method is adopted to obtain dense parallax maps, calculate point cloud information, perform homography transformation and Gabor filter to extract texture features, tracking and detection of road texture features is achieved.
提高了道路感知功能的多帧融合准确性,增强了自动驾驶系统对道路环境的实时检测能力,提高了驾驶的舒适性和安全性。
Smart Images

Figure CN113838111B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent driving technology, and in particular to a road texture feature detection method, device and automatic driving system. Background Art
[0002] At present, the research in the field of automobile assisted driving is constantly improving with the development of image processing technology. The research on this technology realizes functions such as detection of imaging objects in front through the sensor's ability to perceive the environment and understand the scene.
[0003] Monitoring the road ahead and assessing its dangerous conditions are key links in the development of autonomous driving technology. If the road environment ahead can be detected in real time, driving comfort and safety can be improved.
[0004] In view of this, the present invention is proposed. Summary of the invention
[0005] The main purpose of the present invention is to disclose a road texture feature detection method, device and automatic driving system, which are used to solve the problems of false omissions in target detection in the prior art.
[0006] To achieve the above object, according to one aspect of the present invention, a road texture feature detection method is disclosed, and the following technical solution is adopted:
[0007] A road texture feature detection method includes: obtaining a dense disparity map of a road scene based on a left view and a right view of the same road scene; presetting a detection area in a world coordinate system, and intercepting a corresponding detection area in the dense disparity map based on the preset detection area; calculating point cloud information of the detection area based on image information and disparity information of the detection area; performing homography transformation using two-dimensional image information and point cloud information to convert the image of the detection area to a projection plane; extracting lateral texture features of the projection plane using a Gabor filter, and tracking and detecting road texture features through multi-frame texture feature matching.
[0008] Furthermore, the preset detection area in the world coordinate system and the interception of the corresponding detection area in the dense disparity map based on the preset detection area include: calculating the preset detection area in the world coordinate system according to the detection depth requirement and the detection width requirement, and then converting the preset detection area into the detection area in the image coordinate system by means of affine transformation.
[0009] Furthermore, the calculation of the detection area point cloud information based on the image information and parallax information of the detection area includes: segmenting the image information of the detection area, processing the image information, completing the conversion of the image coordinate system to the world coordinate system according to the binocular stereo vision system imaging model and the pinhole imaging model, and obtaining the point cloud information of the detection area. The specific conversion formula is as follows:
[0010] Z=b*F / disp
[0011] X=(Img x -cx)*b / disp
[0012] Y=(Img y -cy)*b / disp
[0013] Where b is the distance between the left and right cameras of the binocular stereo vision imaging system model; F is the focal length of the camera; cx and cy are the image coordinates of the camera principal point, Img x and Img y is the image coordinate point within the detection area; disp is the image point coordinate (Img x , Img y ) where X is the horizontal distance from the camera, Y is the vertical distance from the camera, and Z is the depth from the camera.
[0014] Furthermore, the use of the two-dimensional image information and the point cloud information to perform homography transformation to convert the image of the detection area to the projection plane includes: calculating the homography transformation matrix through the two-dimensional image information and the point cloud information, and the transformation is expressed as:
[0015]
[0016] Project x =X*scale x
[0017] Project y =Z*scale z
[0018] Where X is the lateral distance from the camera and Z is the depth from the camera; Img x and Img y is the image coordinate point within the detection area; Proj x and Proj y is the projection surface coordinate point in the detection area; H is the homography transformation matrix.
[0019] Furthermore, the extraction of the transverse texture features of the projection plane by using a Gabor filter and the tracking and detection of the road texture features by matching the multi-frame texture features include:
[0020] According to the requirements, a specific Gabor function is selected and the Gabor filter kernel is designed to realize the multi-scale and multi-directional horizontal texture feature extraction of the image. The extraction formula is as follows:
[0021]
[0022] in,
[0023] x′=x cosθ+y sinθ
[0024] y′=-x sinθ+y cosθ
[0025] λ is the wavelength, and its size directly affects the filtering scale of the filter; θ is the direction of the filter, and the filter can detect textures in different directions by changing θ; ψ is the phase shift; γ is the spatial aspect ratio, which determines the shape of the filter; σ is the variance of the Gaussian filter; extract the real part of the Gabor function, and convolve the real part to achieve texture feature detection. The real part is:
[0026]
[0027] λ is the wavelength, and its size directly affects the filtering scale of the filter; θ is the direction of the filter, and the filter can detect textures in different directions by changing θ; ψ is the phase shift; γ is the spatial aspect ratio, which determines the shape of the filter; σ is the variance of the Gaussian filter.
[0028] Furthermore, the Gabor filter is used to extract the transverse texture features of the projection plane, and the road texture feature tracking detection is realized by matching the texture features of multiple frames, including: tracking the texture information of the previous and next frames: using the normalized correlation coefficient matching method to match the texture information of the previous frame with the texture information of the next frame, that is, using the correlation coefficient formula to calculate the correlation coefficient to obtain the correlation coefficient matrix, and the correlation coefficient matrix is defined as:
[0029]
[0030] in, is the grayscale mean of the texture area in the previous frame, is the grayscale mean of the area to be matched in the next frame. The correlation coefficient indicates the approximation between the two. The closer the correlation coefficient is to 1, the more linear the relationship between the two is. The peak point in the correlation coefficient matrix NCC(i,j) is taken as the best matching point P. The corrected displacements in the X and Z directions are |C1_x-P_x| and |C1_y-P_y|, respectively. The average of the differences between the coordinates of multiple texture centroids and the best matching point is taken and converted into the actual distance, namely ΔX and ΔZ. ΔX and ΔZ are the lateral and longitudinal movement distances between the two frames of image scenes.
[0031] According to another aspect of the present invention, a road texture feature detection device is provided, and the following technical solution is adopted:
[0032] A road texture feature detection device comprises: an acquisition module, which is used to acquire a dense disparity map of a road scene based on a left view and a right view of the same road scene; a capture module, which is used to preset a detection area in a world coordinate system, and capture the corresponding detection area in the dense disparity map based on the preset detection area; a calculation module, which is used to calculate point cloud information of the detection area based on the image information and disparity information of the detection area; a transformation module, which is used to perform homography transformation using two-dimensional image information and point cloud information, and convert the image of the detection area to a projection plane; and an extraction module, which is used to extract the lateral texture features of the projection plane using a Gabor filter, and realize the tracking and detection of road texture features through multi-frame texture feature matching.
[0033] Furthermore, the interception module includes: an affine transformation module, which is used to calculate a preset detection area in the world coordinate system according to the detection depth requirement and the detection width requirement, and then convert the preset detection area into a detection area in the image coordinate system by means of affine transformation.
[0034] Furthermore, the calculation module includes: a segmentation module, which is used to segment the image information of the detection area and process the image information, complete the conversion of the image coordinate system to the world coordinate system according to the binocular stereo vision system imaging model and the pinhole imaging model, and obtain the point cloud information of the detection area. The specific conversion formula is as follows:
[0035] Z=b*F / disp
[0036] X=(Img x -cx)*b / disp
[0037] Y=(Img y -cy)*b / disp
[0038] Where b is the distance between the left and right cameras of the binocular stereo vision imaging system model; F is the focal length of the camera; cx and cy are the image coordinates of the camera principal point, Img x and Img y is the image coordinate point within the detection area; disp is the image point coordinate (Img x , Img y ) where X is the horizontal distance from the camera, Y is the vertical distance from the camera, and Z is the depth from the camera.
[0039] According to another aspect of the present invention, an automatic driving system is provided, and the following technical solution is adopted:
[0040] The automatic driving system includes the above-mentioned vehicle pseudo road texture feature detection device.
[0041] The present invention performs a series of processing on images of the same road scene, calculates point cloud information of the detection area using image information and parallax information of the detection area, and performs homography transformation using two-dimensional image information and point cloud information, thereby converting the image of the detection area to a projection plane; extracts the lateral texture features of the projection plane using a Gabor filter, and realizes tracking and detection of road texture features through multi-frame texture feature matching, thereby helping the road perception function of assisted driving to improve the accuracy of multi-frame fusion. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0043] Figure 1 A flow chart of a vehicle road texture feature detection method according to an embodiment of the present invention;
[0044] Figure 2 The binocular camera grayscale image and the corresponding disparity image described in the embodiment of the present invention;
[0045] Figure 3 A schematic diagram of a world coordinate system established based on a left camera according to an embodiment of the present invention;
[0046] Figure 4 A specific flow chart of the vehicle road texture feature detection method according to an embodiment of the present invention;
[0047] Figure 5 Schematic diagram of a vehicle road texture feature detection device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0048] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.
[0049] Figure 1 The present invention is a flowchart of a method for detecting vehicle road texture features according to an embodiment of the present invention.
[0050] See also Figure 1 As shown, a road texture feature detection method includes:
[0051] S101: Acquire a dense disparity map of the road scene based on a left view and a right view of the same road scene;
[0052] S103: Preset a detection area in the world coordinate system, and intercept a corresponding detection area in the dense disparity map based on the preset detection area;
[0053] S105: Calculating point cloud information of the detection area based on the image information and disparity information of the detection area;
[0054] S107: Performing homography transformation using the two-dimensional image information and the point cloud information to convert the image of the detection area to a projection plane;
[0055] S109: Using a Gabor filter to extract the transverse texture features of the projection plane, and achieving road texture feature tracking and detection through multi-frame texture feature matching.
[0056] Specifically, in step S101, a dense disparity map of the road scene is obtained based on the left view and the right view of the same road scene. First, a binocular camera imaging mathematical model is established, and the disparity is calculated point by point using the original grayscale image of the binocular camera (only for the overlapped and valid image area of the left and right cameras), and a disparity map corresponding to the original image is obtained, such as Figure 2 As shown, based on the disparity map, a template window of a fixed size is selected in the left view, and a sliding window of the same size is traversed and selected in the right view. The maximum sliding distance corresponds to the maximum disparity Disp max , calculate the matching cost values of the two windows one by one, when the sliding distance u and the distance z measured by the laser point cloud satisfy u=bf / z, set the matching cost value corresponding to the current sliding distance to 0, and set the matching cost values corresponding to other sliding distances to the maximum value, repeat the above steps for each pixel in the left view, generate the initial matching cost atlas, then construct the energy function, set the penalty parameter, use a smaller penalty parameter P1 for small parallax difference to adapt to the slope or curved surface, and use a larger penalty parameter P2 for large parallax difference to prevent depth (parallax) discontinuity, and aggregate the generated path cost data items. When the disparity of pixel p is disp, it corresponds to r path cost values, and the sum of multiple path cost values is recorded as the aggregate cost value. Select the disparity corresponding to the minimum aggregate cost value of each pixel as the final disparity value of the pixel, repeat the above steps for each pixel in the left view, and generate the dense disparity map. In step S103, a detection area is preset in the world coordinate system, and the detection area corresponding to the dense disparity map is intercepted based on the preset detection area; the binocular camera is installed in the middle position in front of the vehicle, and the world coordinate system established based on the left camera is as follows: Figure 3 shown.
[0057] X-axis: When facing forward, points to the right side of the vehicle;
[0058] Y axis: perpendicular to the ground and pointing upwards;
[0059] Z axis: points to the vehicle's forward direction;
[0060] Some parameters of the two cameras are known: the base distances B of the two optical centers, the focal length f, and the row and column coordinates of the optical center of the left camera (crow, ccol). If the row and column coordinates of a pixel are (row, col) and the parallax is d, the three-dimensional coordinate calculation formula is as follows:
[0061]
[0062]
[0063]
[0064] Then, in the world coordinate system, the corresponding detection area in the dense disparity map is intercepted based on the preset detection area. First, the preset detection area in the world coordinate system is calculated according to the detection depth requirement and the detection width requirement, and then the preset detection area is converted into the detection area in the image coordinate system through affine transformation.
[0065] In step S105, the point cloud information of the detection area is calculated based on the image information and parallax information of the detection area, specifically including: segmenting the image information of the detection area, processing the image information, completing the conversion of the image coordinate system to the world coordinate system according to the binocular stereo vision system imaging model and the pinhole imaging model, and obtaining the point cloud information of the detection area. The specific conversion formula is as follows:
[0066] Z=b*F / disp
[0067] X=(Img x -cx)*b / disp
[0068] Y=(Img y -cy)*b / disp
[0069] Where b is the distance between the left and right cameras of the binocular stereo vision imaging system model; F is the focal length of the camera; cx and cy are the image coordinates of the camera principal point, Img x and Img y is the image coordinate point within the detection area; disp is the image point coordinate (Img x , Img y ) where X is the horizontal distance from the camera, Y is the vertical distance from the camera, and Z is the depth from the camera.
[0070] In step S107, a homography transformation is performed using the two-dimensional image information and the point cloud information to convert the image of the detection area to the projection plane, specifically including: calculating the homography transformation matrix using the two-dimensional image information and the point cloud information, and the transformation is expressed as:
[0071]
[0072] Project x =X*scale x
[0073] Project y =Z*scale z
[0074] Where X is the lateral distance from the camera and Z is the depth from the camera; Img x and Img y is the image coordinate point within the detection area; Proj x and Proj y is the projection surface coordinate point in the detection area; H is the homography transformation matrix.
[0075] Furthermore, the use of Gabor filter to extract the transverse texture features of the projection plane and to achieve road texture feature tracking and detection through multi-frame texture feature matching includes: selecting a specific Gabor function according to the requirements and designing the Gabor filter kernel to achieve multi-scale and multi-directional transverse texture feature extraction of the image. The extraction formula is as follows:
[0076]
[0077] in,
[0078] x′=x cosθ+y sinθ
[0079] y′=-x sinθ+y cosθ
[0080] λ is the wavelength, and its size directly affects the filtering scale of the filter; θ is the direction of the filter, and the filter can detect textures in different directions by changing θ; ψ is the phase shift; γ is the spatial aspect ratio, which determines the shape of the filter; σ is the variance of the Gaussian filter;
[0081] Extract the real part of the Gabor function and perform convolution on the real part to achieve texture feature detection. The real part is:
[0082]
[0083] λ is the wavelength, and its size directly affects the filtering scale of the filter; θ is the direction of the filter, and the filter can detect textures in different directions by changing θ; ψ is the phase shift; γ is the spatial aspect ratio, which determines the shape of the filter; σ is the variance of the Gaussian filter; the texture image is filtered using a Gabor filter, and each Gabor filter only allows the texture corresponding to its frequency to pass, so that other textures that do not meet the parameter conditions are suppressed. The horizontal texture of the filtered projection image will be enhanced, while the textures in other directions will be weakened. The filtering results are binarized and connected domains are detected, and connected domains with a length greater than the preset threshold are retained and stored.
[0084] In step S109, the transverse texture features of the projection plane are extracted by using a Gabor filter, and the road texture feature tracking detection is realized by matching multiple frames of texture features, which specifically includes: the transverse texture features of the projection plane are extracted by using a Gabor filter, and the road texture feature tracking detection is realized by matching multiple frames of texture features, including: tracking the texture information of the previous and next frames: using the normalized correlation coefficient matching method to match the texture information of the previous frame with the texture information of the next frame, that is, using the correlation coefficient formula to calculate the correlation coefficient to obtain the correlation coefficient matrix, and the correlation coefficient matrix is defined as:
[0085]
[0086] in, is the grayscale mean of the texture area in the previous frame, is the grayscale mean of the area to be matched in the next frame. The correlation coefficient indicates the approximation between the two. The closer the correlation coefficient is to 1, the more linear the relationship between the two is. The peak point in the correlation coefficient matrix NCC(i,j) is taken as the best matching point P. The corrected displacements in the X and Z directions are |C1_x-P_x| and |C1_y-P_y|, respectively. The average of the differences between the coordinates of multiple texture centroids and the best matching point is taken and converted into the actual distance, namely ΔX and ΔZ. ΔX and ΔZ are the lateral and longitudinal movement distances between the two frames of image scenes.
[0087] Figure 4 This is a specific flow chart of the vehicle road texture feature detection method described in an embodiment of the present invention.
[0088] See also Figure 4 As shown, as a preferred implementation, the vehicle road texture feature detection method specifically includes:
[0089] Step 41: Calculate a dense disparity map based on the acquired left image and right image;
[0090] Step 42: Detection area calculation;
[0091] Step 43: Convert the image coordinate system to the world coordinate system;
[0092] Step 44: Calculate the homography transformation matrix and generate the projection surface;
[0093] Step 45: road surface texture detection;
[0094] Step 46: Road surface texture matching;
[0095] Step 47: Calculate the moving distance between adjacent frames.
[0096] Therefore, the present invention processes images of the same road scene, calculates point cloud information of the detection area using image information and parallax information of the detection area, and performs homography transformation using two-dimensional image information and point cloud information, thereby converting the image of the detection area to a projection plane; extracts the lateral texture features of the projection plane using a Gabor filter, and realizes tracking and detection of road texture features through multi-frame texture feature matching, thereby helping the road perception function of assisted driving to improve the accuracy of multi-frame fusion.
[0097] Figure 5 Schematic diagram of a vehicle road texture feature detection device according to an embodiment of the present invention.
[0098] See also Figure 5 As shown, the vehicle road texture feature detection device specifically includes: an acquisition module 50, which is used to acquire a dense disparity map of the road scene based on the left view and the right view of the same road scene; a capture module 52, which is used to preset a detection area in the world coordinate system, and capture the corresponding detection area in the dense disparity map based on the preset detection area; a calculation module 54, which is used to calculate the point cloud information of the detection area based on the image information and disparity information of the detection area; a transformation module 56, which is used to perform a homography transformation using the two-dimensional image information and the point cloud information, and convert the image of the detection area to the projection plane; an extraction module 58, which is used to extract the lateral texture features of the projection plane using a Gabor filter, and realize the tracking and detection of road texture features through multi-frame texture feature matching.
[0099] Optionally, the interception module 52 includes: an affine transformation module (not shown in the figure), which is used to calculate a preset detection area in the world coordinate system according to the detection depth requirement and the detection width requirement, and then convert the preset detection area into a detection area in the image coordinate system through an affine transformation.
[0100] An automatic driving system provided by the present invention includes the above-mentioned vehicle pseudo-road texture feature detection device.
[0101] The present invention performs a series of processing on images of the same road scene, calculates point cloud information of the detection area using image information and parallax information of the detection area, and performs homography transformation using two-dimensional image information and point cloud information, thereby converting the image of the detection area to a projection plane; extracts the lateral texture features of the projection plane using a Gabor filter, and realizes tracking and detection of road texture features through multi-frame texture feature matching, thereby helping the road perception function of assisted driving to improve the accuracy of multi-frame fusion.
[0102] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.
Claims
1. A road texture feature detection method, It is characterized in that include: Based on the left view and right view of the same road scene, a dense disparity map of the road scene is obtained. First, a mathematical model of binocular camera imaging is established. The original grayscale image of the binocular camera is used to calculate the disparity point by point to obtain a disparity map corresponding to the original image. On the basis of the disparity map, a template window of a fixed size is selected in the left view, and a sliding window of the same size is traversed and selected in the right view. The maximum sliding distance corresponds to the maximum disparity, and the matching cost values of the two windows are calculated one by one. When the sliding distance u and the distance z measured by the laser point cloud satisfy u=bf / z, the matching cost value corresponding to the current sliding distance is set to 0. And let the matching cost values corresponding to other sliding distances be the maximum value, repeat the above steps for each pixel point in the left view, generate an initial matching cost atlas, then construct an energy function, set the penalty parameter, use the penalty parameter P1 for small disparity difference, and use the penalty parameter P2 for large disparity difference. When the disparity of pixel p is disp, it corresponds to r path cost values, and the sum of multiple path cost values is recorded as the aggregate cost value; select the disparity corresponding to the minimum aggregate cost value of each pixel as the final disparity value of the pixel, repeat the above steps for each pixel point in the left view, and generate the dense disparity map; A detection area is preset in the world coordinate system, and a corresponding detection area in the dense disparity map is intercepted based on the preset detection area; Calculate the point cloud information of the detection area based on the image information and disparity information of the detection area; Using the two-dimensional image information and point cloud information to perform homography transformation, the image of the detection area is converted to the projection plane; The lateral texture features of the projection plane are extracted using Gabor filters, and the road texture feature tracking and detection is achieved through multi-frame texture feature matching; The method of extracting the transverse texture features of the projection plane by using a Gabor filter and tracking and detecting the road texture features by matching multiple frames of texture features includes: Select a Gabor function according to the requirements and design the Gabor filter kernel to realize the multi-scale and multi-directional horizontal texture feature extraction of the image. The extraction formula is as follows: in, x'=x cosθ+y sinθ y'=-x sinθ+y cosθ x, y represent the horizontal and vertical coordinates of the projection surface coordinate point in the detection area; x', y' represent the coordinates of x, y in the frequency domain after the projection surface coordinate point in the spatial domain is converted to the frequency domain; λ is the wavelength, and its size directly affects the filtering scale of the filter; θ is the direction of the filter, and the filter can detect textures in different directions by changing θ; Ψ is the phase shift; γ is the spatial aspect ratio, which determines the shape of the filter; σ is the variance of the Gaussian filter; Extract the real part of the Gabor function and perform convolution on the real part to achieve texture feature detection. The real part is: λ is the wavelength, whose size directly affects the filtering scale of the filter; θ is the direction of the filter, and the filter can detect textures in different directions by changing θ; Ψ is the phase shift; γ is the spatial aspect ratio, which determines the shape of the filter; σ is the variance of the Gaussian filter.
2. The road texture feature detection method according to claim 1, It is characterized in that The presetting of the detection area in the world coordinate system and intercepting the corresponding detection area in the dense disparity map based on the presetting detection area include: The preset detection area in the world coordinate system is calculated according to the detection depth requirement and the detection width requirement, and then the preset detection area is converted into the detection area in the image coordinate system by affine transformation.
3. The road texture feature detection method according to claim 1, It is characterized in that The calculation of the detection area point cloud information based on the image information and the parallax information of the detection area comprises: The image information of the detection area is segmented and processed. According to the binocular stereo vision system imaging model and the pinhole imaging model, the image coordinate system is transformed into the world coordinate system to obtain the point cloud information of the detection area. The specific conversion formula is as follows: Z=b*F / disp X=(Img x -cx))*b / disp Y=(Img y -cy)*b / disp Where b is the distance between the left and right cameras of the binocular stereo vision imaging system model; F is the focal length of the camera; cx and cy are the image coordinates of the camera principal point, Img x and Img y are the horizontal and vertical coordinates of the image coordinate point in the detection area; disp is the disparity value of the image coordinate point in the detection area; X is the horizontal distance from the camera, Y is the vertical distance from the camera, and Z is the depth from the camera.
4. The road texture feature detection method according to claim 1, It is characterized in that The method of using the two-dimensional image information and the point cloud information to perform homography transformation to convert the image of the detection area to the projection plane includes: The homography transformation matrix is calculated through the two-dimensional image information and point cloud information. The transformation is expressed as: Proj x =X*scale x Project y =Z*scale z Where X is the lateral distance from the camera and Z is the depth from the camera; Img x and Img y are the horizontal and vertical coordinates of the image coordinate points within the detection area; Project x and Proj y are the horizontal and vertical coordinates of the projection surface coordinate points within the detection area; H is the homography transformation matrix.
5. A road texture feature detection device, used to implement the method according to any one of claims 1 to 4, It is characterized in that include: An acquisition module, used for acquiring a dense disparity map of the road scene based on a left view and a right view of the same road scene; A capture module, used to preset a detection area in a world coordinate system, and capture a corresponding detection area in a dense disparity map based on the preset detection area; A calculation module, used for calculating the point cloud information of the detection area based on the image information and parallax information of the detection area; A transformation module, used to perform homography transformation using two-dimensional image information and point cloud information to transform the image of the detection area to a projection plane; The extraction module is used to extract the lateral texture features of the projection plane using a Gabor filter, and to track and detect road texture features through multi-frame texture feature matching.
6. The road texture feature detection device according to claim 5, It is characterized in that The interception module comprises: The affine transformation module is used to calculate the preset detection area in the world coordinate system according to the detection depth requirement and the detection width requirement, and then convert the preset detection area into the detection area in the image coordinate system by means of affine transformation.
7. The road texture feature detection device according to claim 5, It is characterized in that The calculation module comprises: The segmentation module is used to segment the image information of the detection area and process the image information. According to the binocular stereo vision system imaging model and the pinhole imaging model, the image coordinate system is converted to the world coordinate system to obtain the point cloud information of the detection area. The specific conversion formula is as follows: Z=b*F / disp X=(Img x -cx)*b / disp Y=(Img y -cy)*b / disp Where b is the distance between the left and right cameras of the binocular stereo vision imaging system model; F is the focal length of the camera; cx and cy are the image coordinates of the camera principal point, Img x and Img y are the horizontal and vertical coordinates of the image coordinate point in the detection area; disp is the disparity value of the image coordinate point in the detection area; X is the horizontal distance from the camera, Y is the vertical distance from the camera, and Z is the depth from the camera.
8. An autonomous driving system, It is characterized in that The invention comprises the road texture feature detection device as described in any one of claims 5 to 7.
Citation Information
Cited By
A multi-functional comprehensive detection device and system for road based on automatic driving
CN122901418A
