CN110807771A - Defect detection method for road deceleration strip - Google Patents
Defect detection method for road deceleration strip Download PDFInfo
- Publication number
- CN110807771A CN110807771A CN201911049884.9A CN201911049884A CN110807771A CN 110807771 A CN110807771 A CN 110807771A CN 201911049884 A CN201911049884 A CN 201911049884A CN 110807771 A CN110807771 A CN 110807771A
- Authority
- CN
- China
- Prior art keywords
- image
- deceleration strip
- color block
- road
- defective
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
- G06T2207/30252—Vehicle exterior; Vicinity of vehicle
- G06T2207/30256—Lane; Road marking
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a defect detection method of a road deceleration strip, which comprises the steps of firstly obtaining a road image of deceleration strip information, separating a deceleration strip outline image from the road image, obtaining a deceleration strip outline image with gray scale and a gray scale road image without the deceleration strip, then obtaining a gradient map of the deceleration strip outline image with gray scale, further obtaining edge lines between color blocks, and obtaining that the deceleration strip corresponding to the gradient map is a perfect deceleration strip when the edge lines are between yellow color blocks and black color blocks; when the edge lines are the edge lines among the yellow color blocks, the defective color blocks and the black blocks, yellow color block images, defective color block images and black block images can be obtained and are marked in sequence, so that the gray value comparison in the next step is facilitated, finally the total defective number, the defective yellow block number and the defective black block number can be finally obtained through the color block colors at the two ends of the defective color blocks, the operation efficiency is high, and the potential harm to vehicles and personnel caused by the defect of the road deceleration strip is solved.
Description
Technical Field
The invention relates to the field of road detection, in particular to a defect detection method for a road deceleration strip.
Background
As an important transportation facility in a road, a deceleration strip can slightly arch a road surface to achieve the purpose of vehicle deceleration, and the deceleration strip is most commonly a deceleration strip with black blocks and yellow blocks arranged at intervals. The operating condition of deceleration strip is comparatively abominable, often damages because of reasons such as vibration, impact, insolate and ageing, and the nail that can make when installing the deceleration strip exposes outside like this to cause latent harm to vehicle and personnel, this becomes the device that seriously influences traffic safety on the contrary.
At present, the existing technical scheme mainly adopts a machine learning method for deceleration strips with black color blocks and yellow color blocks arranged alternately, based on a convolutional neural network, judges whether the deceleration strips are contained in the road or not and whether the deceleration strips are defective or not by training positive and negative sample sets, but the calculated amount and the storage space of the deceleration strips are extremely large in the operation process, so that the road deceleration strip defect real-time detection cannot be normally applied in engineering, and therefore the detection of the deceleration strips needs to be improved.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method for detecting the defect of a road deceleration strip, which can be used for identifying the deceleration strip in the road in real time and further detecting the defect deceleration strip to obtain the number and the color of the defect deceleration strip.
The invention is realized by the following technical scheme:
a method for detecting defects of a road deceleration strip comprises the following steps,
step 2, solving a gradient map of the profile image of the deceleration strip with gray scale, solving edge lines among color blocks of the deceleration strip in the gradient map by applying a watershed algorithm,
when the edge line is found to be the edge line between the yellow color block and the black color block, the deceleration strip corresponding to the gradient map is obtained to be a complete deceleration strip;
when the edge line is the edge line between the yellow color block, the defective color block and the black block, separating out a yellow color block image, a defective color block image and a black block image, marking according to the sequence of the yellow color block image, the defective color block image and the black color block image, and then calculating the average gray value of the yellow color block image, the defective color block image and the black block image according to the sequence of the marks; calculating the average gray value of the gray road image without the deceleration strip, and recording the average gray value as theta;
and 4, obtaining the total defect number, the defect yellow block number and the defect black block number according to the color block colors at the two ends of the defect color block.
Preferably, in step 1, a road image containing deceleration strip information is acquired as follows;
step a1The method comprises the following steps that a binocular camera is installed at the front upper part of a detection vehicle, a vibration acceleration sensor is installed on a rear axle of the detection vehicle, and the binocular camera continuously photographs the road surface on the detection vehicle in a motion state;
then, sequentially carrying out edge cutting, graying and perspective change on the road image acquired by the binocular camera to obtain a road image after perspective change, wherein the ratio of the width of the left edge and the right edge to the total width of the image during edge cutting is 8-10%, and the ratio of the height of the upper edge of the image to the total height of the image is 10-15%;
step a2Sequentially screening and detecting Hough lines of the road image with perspective change according to a template matching mode to obtain image coordinate data of two end points of a plurality of groups of lines, and calculating the coordinate data of two end points of the lines stored in the same array to obtain the distance of each group of lines;
step a3Setting standard parameters of a standard deceleration strip in an image coordinate system as a threshold, discarding a straight line with a y coordinate difference lower than the threshold and a straight line with an abscissa coordinate difference larger than the threshold, keeping the longest straight line and the next longest straight line, recording coordinates of four corner points of the two straight lines, and selecting the four corner points by using a rectangular frame, wherein the rectangular length-width ratio is smaller than that of the standard deceleration strip;
step a4Carrying out inverse perspective change on the road image framed and selected by the rectangular frame, then calculating the time of the detection vehicle 1 reaching the deceleration strip 3, finally backtracking the image of the moment corresponding to the pulse signal in the time range, and acquiring the road image containing deceleration strip information;
the pulse signal is obtained by the vibration acceleration sensor and reaches a set threshold value after being subjected to gain variation and arithmetic mean filtering in sequence.
Further, step a4And when the Hamming distance between two images is larger than a threshold value, discarding the previous image, retaining the current image, and comparing the Hamming distance with the Hamming distance of the next image, so that the same deceleration belt only has one image.
Preferably, in step 1, histogram equalization is performed on the obtained deceleration strip profile image with gray scale, and then a gradient map of the deceleration strip profile image after histogram equalization is obtained.
Preferably, in step 2, the following process is adopted to calculate the average gray value of the gray road image without the speed bump;
traversing each pixel of the gray-scale road image without the deceleration strip in three cycles, taking the corner point as the starting point or the ending point of each cycle according to the recorded contour corner point of the deceleration strip, accumulating and summing the gray-scale values of each pixel of the rest images, calculating the total number of the pixels in the images according to the segmented cycle mode, and finally obtaining the average gray-scale value of the gray-scale road image without the deceleration strip.
Preferably, in step 3, black color blocks, yellow color blocks and defective color blocks are determined in the following manner;
the average gray value of the yellow patch image, the gray patch image and the black patch image is αiWhen i is 1 to n, when αi<θ1The color block is black block, if αi>θ2The color block is yellow when theta1<αi<θ2And the color block is a defective color block.
Preferably, in step 4, the total number of defects is determined as follows;
when a defective color block is sandwiched between two identical color blocks, it is a case of missing a single color block, and when a defective color block is located between two different color blocks, it is a case of continuously missing two color blocks, and the total number of defects is equal to the number of single color block defects × 1+ the number of continuous color block defects × 2.
Preferably, the number of defective yellow blocks and the number of defective black blocks are determined in the following manner in step 4;
when a defective color block is between two yellow blocks, it is a defective black color block, when a defective color block is between two black blocks, it is a defective yellow color block, when a defective color block is between two different color blocks, it is a yellow color block and a black color block, respectively, the number of defective yellow blocks is equal to the number of yellow blocks defective by a single color block × 1+ the number of continuous color block defective × 1, and the number of defective black blocks is equal to the number of black blocks defective by a single color block × 1+ the number of continuous color block defective × 1.
Compared with the prior art, the invention has the following beneficial technical effects:
the invention discloses a defect detection method of a road deceleration strip, which is different from a deep learning scheme, and comprises the steps of firstly obtaining a road image containing deceleration strip information with black blocks and yellow blocks arranged at intervals, separating a deceleration strip outline image from the road image, obtaining a gray deceleration strip outline image and a gray road image without a deceleration strip after gray scale, then obtaining a gradient map of the gray deceleration strip outline image, further obtaining edge lines between the color blocks of the deceleration strip, and obtaining the deceleration strip corresponding to the gradient map as a complete deceleration strip when the obtained edge lines are the edge lines between the yellow blocks and the black blocks; when the edge lines are the edge lines among the yellow color blocks, the defective color blocks and the black blocks, yellow color block images, defective color block images and black block images can be obtained and are marked in sequence, so that the gray value comparison of the next step is facilitated, and finally the total defective number, the defective yellow block number and the defective black block number can be finally obtained through the color block colors at the two ends of the defective color blocks; the method can accurately identify the total missing number and the color of the color blocks of the current deceleration strip in real time, has high operation efficiency, has higher defect detection rate compared with a deep learning scheme, and solves the potential harm to vehicles and personnel caused by the defect of the road deceleration strip.
Furthermore, similarity detection is carried out on the images at the moment corresponding to the pulse signals through a perceptual hash algorithm, so that the same deceleration strip only has one image, the calculation speed can be increased, the defect detection rate is higher, and the resource waste caused by repeated defect detection of the same deceleration strip on a road is avoided; because the difference of regions and climates can cause the difference of illumination and humidity, the method based on the convolutional neural network machine learning cannot ensure higher detection rate, and therefore, the method has greater advantages.
Further, in consideration of the fact that in the actual operation process, the image is too dark or overexposed due to factors such as weather and tree shadow, histogram equalization is performed, the gray value is enlarged to improve the image contrast, the brightness difference between the yellow block and the black block is increased, and the accuracy is improved.
Drawings
Fig. 1 is a schematic diagram of the defect detection of the road deceleration strip.
Fig. 2 is a schematic process diagram of the road deceleration strip defect detection.
FIG. 3 is a schematic diagram of the vibration deceleration sensor signal processing and gain and filtering of the present invention.
Fig. 4a is a diagram of pulse signals generated by a vibration acceleration sensor when a detection vehicle passes through a deceleration strip.
Fig. 4b is a graph of the variable gain processing of the pulse signal of fig. 4 a.
FIG. 4c is a graph of an arithmetic mean filter of the pulse signal of FIG. 4 a.
Fig. 5 is a schematic flow chart of a binocular camera image acquisition preprocessing algorithm of the invention.
Fig. 6 is a schematic flow chart of a deceleration strip multi-stage screening method of the invention.
Fig. 7 is a schematic flow chart of a road deceleration strip defect detection method of the invention.
FIG. 8 is a schematic diagram of a defect situation of a monochrome block of the deceleration strip according to the present invention.
FIG. 9 is a schematic diagram illustrating a defective dual-color block of the deceleration strip according to the present invention.
In the figure: the device comprises a detection vehicle 1, a binocular camera 2, a deceleration strip 3 and a vibration acceleration sensor 4.
Detailed Description
The present invention will now be described in further detail with reference to specific examples, which are intended to be illustrative, but not limiting, of the invention.
The invention discloses a defect detection method of a road deceleration strip, as shown in figure 1, a binocular camera 2 is arranged at the front upper part of a detection vehicle 1 and used for acquiring road images in front of the detection vehicle 1 and images of a deceleration strip 3 in real time, a vibration acceleration sensor 4 is arranged on a magnetic mounting base, the magnetic mounting base is adsorbed on a rear axle of the detection vehicle 1 and used for acquiring vibration signals in the operation process of the detection vehicle 1 and further determining whether the detection vehicle 1 passes through the deceleration strip 3, a rotary encoder is arranged at one side of a rear wheel of the detection vehicle 1 and can acquire stable rotation speed information, and a Beidou satellite positioning instrument and an industrial personal computer are arranged in a room of the detection vehicle.
The binocular cameras 2 are two cameras with the same model, the optical axes of the two cameras are parallel, the binocular cameras 2 collect road images in front of the detection vehicle 1 and images of the deceleration strip 3 in real time and transmit the road images and the images back to the industrial personal computer through data lines for rapid image preprocessing, and the method sequentially comprises the steps of cutting image edges, graying, perspective change, template matching, Hough line detection, deceleration strip boundary line screening and deceleration strip rectangular frame framing as shown in fig. 5; the vibration acceleration sensor 4 acquires vibration conditions of the detection vehicle 1 in the normal running process of a road and when the detection vehicle passes through the deceleration strip 3, and an obtained vibration waveform signal is transmitted back to the industrial personal computer for analysis; the square wave signal that rotary encoder will gather returns the industrial computer, and the industrial computer can carry out real-time location to detecting car 1 according to the tire diameter of setting for and this square wave automatic calculation current speed of a motor vehicle, big dipper satellite positioning appearance to record the geographical position of deceleration strip 3 when detecting car 1 and passing through deceleration strip 3, pinpoint when making things convenient for later stage maintenance deceleration strip 3, these processes are as shown in fig. 2. In the present embodiment, the rotary encoder employs a photoelectric encoder.
In combination with the system arrangement form, the industrial personal computer directly controls the binocular camera 2 and the vibration acceleration sensor 4 in real time according to the square waves output by the rotary encoder, specifically, the binocular camera 2 collects one frame every M square waves, and the vibration acceleration sensor 4 samples vibration signals every N square waves. Every M square wave binocular cameras collect a frame and can guarantee that the image quantity is the same in equal displacement to too much collection leads to the operand increase when avoiding the low-speed, and the frequency of gathering is low excessively and omits deceleration strip information when the high-speed.
In order to improve the identification accuracy of the deceleration strip and avoid the interference of complex information in a road to identification, a three-level screening method automatically performed by an industrial personal computer is designed by combining signals of a vibration acceleration sensor in an image preprocessing stage. Specifically, as shown in fig. 6, the first-stage screening method adopts a deceleration strip template matching mode, searches and matches the acquired image with the existing deceleration strip image stored in the industrial personal computer, if the correlation coefficient of the deceleration strip image is greater than a set threshold value, the original image may contain a deceleration strip, and performs hough line detection of the next stage, otherwise, stops preprocessing and discards the current image; the second-stage screening method is Hough straight line detection, a plurality of groups of straight lines obtained after the Hough straight line detection are screened, the length standard of the deceleration strip under an image coordinate system is specifically obtained through a limited number of tests and is set as a threshold, the straight lines are compared with the threshold, short straight lines, straight lines with the difference between two y coordinates lower than the threshold and straight lines with the difference between horizontal coordinates larger than the threshold are omitted, two longest horizontal lines and two next longest horizontal lines are reserved, four coordinates of the two straight lines are obtained and are the coordinates of the corner points of the deceleration strip, the outline of a suspected deceleration strip is extracted, frames which cannot normally extract the outline are discarded and are not stored, and the extracted image of the suspected deceleration strip outline is temporarily stored so as to be screened at the third stage; the third-level screening method comprises three steps, wherein the first step is that the distance between a detection vehicle and a deceleration strip is measured according to an image acquired by a binocular camera, the second step is that the running speed v of the detection vehicle is calculated according to a photoelectric encoder, the running speed v is (hr)/c, wherein h is a capture value of the rotary encoder, namely the pulse number output by the rotary encoder within one second, r is the rolling radius of a detection wheel, c is the line number of the rotary encoder, namely the pulse number output by the rotary encoder at one rotation speed, the third step is that the time information of the detection vehicle reaching the deceleration strip is calculated in a reverse mode, if a signal acquired by a vibration acceleration sensor is subjected to variable gain and arithmetic average filtering processing, a pulse signal exists in the time range and the amplitude of the pulse signal reaches a set threshold value, the situation that a profile of the deceleration strip is contained in a previous road image can be judged, and at the moment, a Beidou satellite, if there is no pulse signal in this time range, it can be determined that the suspected image contains no deceleration strip, and then it is discarded.
A group of images corresponding to the pulse signal can be traced back according to square wave information of the photoelectric encoder, and the specific tracing method is that a frame is collected by every M square waves of a camera, the time of a detection vehicle reaching a deceleration strip under the pulse signal is t, the group of images is started to be (ht)/M, the hash value of each image deceleration strip part of the group of images is calculated, the similarity of the hash value is calculated, a unique road image of the same deceleration strip is obtained by screening, the deceleration strip image is subjected to deceleration strip defect detection, and the total number and the color of the deceleration strip defect color blocks are calculated.
In particular, the method for screening the only one road image of the same deceleration strip comprises the following steps,
step 1a, printing a checkerboard, and pasting the checkerboard on a plane as a calibration object;
step 1b, shooting some photos in different directions for the calibration object by adjusting the direction of the calibration object or the binocular camera;
step 1c, extracting characteristic points, such as angular points, from the picture;
step 1d, estimating five internal parameters and all external parameters in the feature points under the ideal and distortion-free condition;
step 1e, estimating distortion coefficients of the characteristic points under the actual radial distortion by using a least square method;
step 1f, optimizing estimation by using a maximum likelihood method, and improving the estimation precision of a distortion coefficient;
step 1g, respectively obtaining a rotation matrix and a translation matrix of two cameras relative to the same coordinate system according to the parallax of the chessboard angular points in the two cameras, and finally obtaining five internal parameters, three external parameters, two distortion coefficients and a rotation translation matrix between the two cameras with high estimation precision;
step 2, edge cutting is carried out on the road image original image collected by the binocular camera, the ratio of the width of the left edge and the right edge of the cut road image to the total width of the image is 8% -10%, the ratio of the height of the upper edge of the cut image to the total height of the image is 10% -15%, so that unnecessary sidewalk information is well removed, deceleration strip information is reserved, and unnecessary operation is reduced to improve the calculation efficiency;
4, perspective change is carried out, a change matrix is obtained through calculation according to four points of the left lane line and the right lane line, so that perspective transformation is carried out on the gray-scale map, and deceleration strip distortion generated under overlooking shooting is eliminated;
when the detection vehicle runs to a lane to be detected, starting a binocular camera to start acquisition, calibrating by using a lane image without a deceleration strip to respectively acquire a left camera perspective transformation matrix and a right camera perspective transformation matrix, solving 4 groups of corresponding points of the transformation matrices, and selecting four groups of points including front points and rear points of lane lines on the left side and the right side of the road image, wherein the conversion formula is as follows:
wherein (u, v) is the original image coordinate, and (x ', y', w ') is the transformed three-dimensional coordinate, and the transformed coordinate is converted from two-dimension to three-dimension and divided by w', that is, the original image coordinate is the original image coordinate, and the transformed three-dimensional coordinate is the original image coordinate
The transformed image pixel coordinates.
Obtaining by solution:
based on the four groups of points and the corresponding coordinates of the quadrilateral points, i.e., (0, 0) ->(x0，y0),(1，0)->(x1，y1),(1，1)->(x2，y2),(0，1)->(x4，y4) In order to simplify the operation and ensure that the image of the deceleration strip does not generate distortion, a plane after transformation can be selected to be parallel to the original plane, and the parameters of the transformation matrix are obtained as follows:
a11＝x1-x0，a21＝x2-x1，a31＝x0，a12＝y1-y0，a22＝y2-y1，a32＝y0，a13＝0，a23＝0，a33＝0；
step 5, template matching, namely performing deceleration strip template matching on the image subjected to perspective change to improve the identification accuracy of the deceleration strip, calling a deceleration strip template image stored in an industrial personal computer to perform sliding matching on the current image by adopting a normalized square difference matching method, calculating to obtain the maximum normalized matching coefficient of the image, comparing the maximum normalized matching coefficient with a set threshold value, and primarily screening the image;
specifically, let the deceleration strip template be T (m, n), perform a translational sliding search on the image S (W, H) for the template T (m, n), and the area currently covered by the template be a sub-image SijWherein i, j is the coordinate of the lower left corner of the subgraph, and the search range is as follows: i is more than or equal to 1 and less than or equal to W-n, and j is more than or equal to 1 and less than or equal to H-m. Calculating the similarity coefficient of each sub-graph, normalizing the similarity coefficient to obtain a template matching correlation coefficient, wherein the formula is as follows:
find the maximum value Rmax(im，jm) For the most probable deceleration strip region in the current frame, compare RmaxAnd setting a threshold value gamma, if Rmax>Gamma, determining that the current frame is suspected to contain a deceleration strip area, and otherwise, discarding the current image;
step 6, Hough line detection, namely firstly carrying out contour detection on the road gray level image to improve the detection effect, and then adopting progressive probability Hough line detection to improve the line identification rate;
the outline detection adopts a Sobel operator to carry out edge detection, the size of a template is 3 multiplied by 3, the template can adopt the following steps:
the gradient amplitude G (x, y) ═ G | of the pixel point can be obtained by convolution operation of the template and the original imagex|+|Gy| then choose the threshold τ if G (x, y)>Tau, then the point is an edge point, otherwise, the point is a non-edge point; the Sobel operator is selected to be easily realized in space, so that a good edge detection effect can be generated, and meanwhile, the influence of noise is small due to the introduction of local averaging; and (3) carrying out progressive probability Hough line detection on the contour of the road and the object on the road obtained by edge detection, screening out possible curves and irregular lines, and keeping the lower line.
The progressive probability type Hough line detection is an improvement of the standard Hough line detection, only n points in the outline of each small region are processed randomly, and therefore compared with the full processing of the standard Hough line detection, the operation amount and the occupied memory can be reduced. Specifically, the progressive probabilistic Hough line detection is performed as follows,
step 6.1, equally dividing the contour image into a plurality of small spaces, specifically, dividing the contour image into a left part, a middle part and a right part, wherein the left part, the middle part and the right part respectively occupy 1/3, each part is divided into an upper small space, a middle small space and a lower small space, each small space corresponds to an accumulator acc (rho, theta), and the initial value of the accumulator acc (rho, theta) is 0; putting all detected edge points into an edge point set to be processed to form an edge point set to be processed;
step 6.2, detecting whether the edge point set to be processed is empty, and finishing the algorithm if the edge point set to be processed is empty; otherwise, randomly taking out a group of pixel points from the edge point set to be processed, projecting the pixel points to a polar coordinate system, calculating corresponding theta values under all rho values, and adding one to the corresponding accumulator acc (rho, theta);
6.3, deleting the taken points from the edge point set to be processed;
step 6.4, judging whether the updated accumulator value is larger than a threshold thr, entering step 6.5 if the updated accumulator value is larger than the threshold thr, and otherwise, returning to step 6.2;
step 6.5, determining a straight line by the parameter corresponding to the accumulator value which is larger than the threshold thr and obtained in the step 6.4, deleting the point which is to be processed and is centrally positioned on the straight line, and clearing the accumulator;
and 6.6, checking whether all the edge point sets are empty, and returning to the step 6.2 to continue processing if unprocessed point sets still exist. After contour detection and progressive probabilistic Hough line detection, some line segments which cannot be contours of the deceleration strip can be primarily screened out, and finally a plurality of groups of line end point data are reserved;
step 7, hough straight line screening, obtaining image coordinate data of two end points of a plurality of straight lines after hough straight line detection, screening the plurality of hough straight lines to obtain a deceleration strip boundary contour, and as an improvement, the screening method comprises the following steps: calculating the coordinate data of two straight points stored in the same array to obtain the straight line distance, and neglecting the pixel coordinate difference below a threshold value; data with the difference of x-axis coordinates exceeding a set value are discarded; keeping the longest and the next longest of the remaining Hough lines, recording coordinates of four corner points, and selecting by using a rectangular frame, wherein the aspect ratio of the rectangle is smaller than the aspect ratio of the deceleration strip, and the aspect ratio of the general shorter deceleration strip is about 10: 1;
step 8, selecting a rectangular frame, respectively obtaining end point coordinates of an upper straight line and a lower straight line of the deceleration strip after the left image and the right image are subjected to Hough straight line detection and screening, and selecting the outline of the deceleration strip according to the four coordinates;
step 9, inverse perspective change is carried out, coordinates of a certain corresponding corner point in the left image and the right image are multiplied by an inverse matrix of a perspective change matrix, and then the pixel coordinate position of the deceleration strip boundary corner point in the original image can be inversely calculated;
step 10, calculating the distance between the deceleration strip and the vehicle body under a world coordinate system, and obtaining the coordinates of an inner parameter matrix, an outer parameter matrix and a distortion matrix obtained by calibrating a binocular camera and the coordinates of corner points corresponding to the contours of the deceleration strip in the left image and the right image under an image coordinate system according to a similar triangle principle by the following calculation formula:
wherein, TxIs the base line distance between the binocular cameras, f is the focal length of the cameras, d is a certain angular pointDisparity, μ, in left and right imagesR、μlThe angle points are respectively a certain angle point of the deceleration strip, the angle points can be selected at will, preferably two angle points on the bottom edge and x pixel coordinate points, mu, on the left image and the right image are selectedoThe pixel coordinate origin of the images collected by the right-side camera and the left-side camera is the upper left angular point, and s is the resolution of the binocular camera used by the invention;
and step 11, calculating the hash similarity,
in order to improve the calculation speed and avoid the resource waste caused by repeated defect detection of the same deceleration strip on a road, before the defect detection of the image containing the deceleration strip, the similarity detection is carried out on the image within the range of the pulse signal threshold value, the adopted method is a perception hash algorithm, the hash value of the deceleration strip in each image is calculated, a corresponding characteristic character string is generated, the difference of 0 and 1 digit number between the two images is compared to obtain the similarity of the images, when the similarity is greater than the threshold value gamma, the previous image is deleted, the current image is reserved, the hash value of the next image is calculated for comparison, the similarity between the two images is circularly calculated, if the similarity is greater than the set threshold value, the deceleration strip appearing in the two images is considered to be the same deceleration strip, the previous image is discarded, and if the similarity is less than the set threshold value, the two images are simultaneously reserved, to be subsequently processed;
the method comprises the steps of sequentially carrying out Hash similarity calculation on a plurality of deceleration strip images acquired in a pulse signal, which mainly comprises the following steps,
step 11a, extracting a deceleration strip image in a rectangular frame;
step 11b, scaling the image of the deceleration strip to 32 × 32, and calculating the DCT transformation of the image to obtain a 32 × 32 DCT coefficient matrix;
step 11c, reserving 8 x 8 matrixes at the upper left corner of the matrix;
step 11d, calculating the gray average value of all 1024 pixels;
step 11e, comparing the gray scale of each pixel with the average value, wherein the gray scale is greater than or equal to the average value and is marked as 1; less than the average value, and is marked as 0;
step 11f, calculating the hash value of the image, and arranging the pixel numbers in sequence to form a 64-bit integer, namely the fingerprint of the picture;
step 11g, as an improvement, comparing hash values of the current image and the previous image, calculating Hamming distances of the two images, if the Hamming distances reach a set threshold value, determining the two images as the same deceleration strip, and discarding the previous image; the hash value of the next deceleration strip image is calculated by the method, and compared with the current deceleration strip image, the same deceleration strip image only has one image through cyclic calculation.
In order to improve the sensitivity of the vibration acceleration sensor for acquiring pulse signals generated when the vibration acceleration sensor passes through the deceleration strip, the analog signals of the vibration acceleration sensor are amplified and subjected to variable gain processing, and smooth filtering processing is performed on digital signals subjected to A/D sampling, so that the amplitude of the pulse signals is larger, the amplitude of random vibration generated on a road surface is reduced, a higher threshold value can be set to ensure the accuracy, the random interference of the road surface is eliminated, and the road surface noise is suppressed. As shown in fig. 3, 4a, 4b, and 4c, the process mainly includes analog signal gain-changing processing and digital signal arithmetic mean filtering.
The main process of the variable gain processing comprises the step of carrying out gain processing on the voltage signal passing through the charge amplifier according to the amplitude of the voltage signal, particularly the amplitude is larger than a set upper limit threshold theta1Or less than a set lower threshold value-theta2The gain coefficient of the part (2) is greater than 0, and the larger the absolute value of the amplitude is, the larger the gain coefficient is, the amplification processing is carried out; the gain signal of the signal with the amplitude between the upper threshold and the lower threshold is smaller than 0, and the smaller the absolute value of the amplitude is, the smaller the gain coefficient is, the reduction processing is performed, and finally the pulse signal when the vehicle passes through the deceleration strip is more remarkable in the random vibration of the road surface, so that a higher threshold epsilon can be set to distinguish the pulse signal. FIG. 4a is a diagram of a pulse signal generated by a vibration acceleration sensor when a detected vehicle passes through a deceleration strip, wherein the pulse signal contains pulses with certain amplitude and vibration noise mixed therein, after the pulse signal is subjected to variable gain processing, as shown in FIG. 4b, the pulse amplitude is improved, the random vibration amplitude of the road surface is suppressed between an upper threshold and a lower threshold, and after the random vibration amplitude is subjected to arithmetic average filtering, as shown in FIG. 4c, the signal is subjected to arithmetic average filteringThe waveform is smoother, and a higher threshold value can be set to screen out the waveform when the deceleration strip passes. Specifically, as shown in fig. 3, a vibration signal sensed by the vibration acceleration sensor is output as a charge q, a voltage output with a fixed gain is realized through a charge amplifier of a first stage, a primary amplification of the signal is realized, then a variable gain output is realized through a variable gain operational amplifier of a second stage, the a/D converter further performs smoothing filtering processing, the model of a control unit is STM32F103, the control unit is a positive feedback control circuit formed by a Cortex-M3 CPU with an ARM32 bit inner core, and the control unit controls the variable gain operational amplifier through MCU voltage feedback to amplify a low-frequency pulse signal and suppress random vibration generated on a road surface. The gain calculation of the variable gain amplifier adopts the existing method, which specifically comprises the following steps:
Gain(dB)＝40VG+G0
wherein gain (dB) is amplification factor, VGIs a differential input gain control voltage in the range-0.5V, and V is set when the voltage difference across the control unit exceeds a threshold valueG>0, the gain control voltage is correspondingly increased if the voltage difference is larger, G0Is the starting point of the gain.
The main process of arithmetic average filtering comprises continuous sampling of the vibration acceleration sensor for multiple times, summing of sampling values, and further dividing by the sampling times to obtain the average value of the point so as to reduce the influence of random vibration of the road surface, namely:
specifically, after sampling, software filtering is used, and a conventional arithmetic mean filtering algorithm can be adopted to well generate transmitted random vibration from the road surface, that is, to find a Y value, so that the sum of squares of errors between the Y value and each sampling value x (k) (k is 1 to N) is minimum, that is:
wherein
The square of the error between each sample value and the mean value, and N is the number of samples per time.
The limit principle is obtained by a unitary function:
the selection of N is not too large, and can be generally 10-15, so as to improve the sensitivity and reduce the real-time running time. And when the amplitude of the processed vibration acceleration signal exceeds a set threshold epsilon, the detected vehicle is considered to pass through a deceleration strip at the moment.
The invention discloses a method for detecting defects of a road deceleration strip, which is applied to road images containing deceleration strip information after the road images are subjected to early-stage image rapid preprocessing and screened by a vibration acceleration sensor, as shown in figure 7, the main flow is as follows,
step 12, extracting a region of interest (namely ROI) of the deceleration strip, separating the profile of the deceleration strip from an original image into a single new image according to coordinates of four corner points of a rectangular frame obtained in preprocessing, reserving the original image, and carrying out gray level on the new image and the original image;
step 13, histogram equalization is carried out on the deceleration strip with the gray level, and the gray level value is enlarged to improve the image contrast, increase the brightness difference of yellow blocks and black blocks and improve the accuracy by considering that the image is too dark or overexposed due to factors such as weather and tree shadow in the actual running process of the detection vehicle and before edge segmentation;
the method comprises the specific steps of carrying out,
step 13.1, listing the gray levels of the original image and the transformed image, wherein the transformed image is the new image in the step 12, and L is the number of the gray levels;
step 13.2, counting the number of pixels of each gray level in all the images;
step 13.3, calculating histogram probability value P (i) ═ Ni/N and histogram probability cumulative value P (j) ═ P (1) + P (2) + P (3) + … + P (i) of the original image;
step 13.4, calculating the transformed gray value by using a gray value transformation function, and rounding: j ═ INT [ (L-1) Pj +0.5 ];
step 13.5, determining a pixel mapping relation, performing gray mapping, namely determining a gray conversion relation i → j, and correcting the gray value f (m, n) ═ i of the original image to g (m, n) ═ j according to the gray value i → j;
step 13.6, gray mapping: operating the original image, and mapping each pixel into a new pixel;
and step 14, dividing the ROI edge of the deceleration strip,
a watershed algorithm is adopted, the main steps are that a gradient map of the gray level image obtained in the step 13 is obtained, the watershed algorithm is applied on the basis of the gradient map, local maximum values and local minimum values are found out, edge lines among color blocks of the deceleration strip are obtained, and then the ROI of the deceleration strip can be segmented;
when the divided edge line is the edge line between the yellow color block and the black color block, obtaining that the deceleration strip corresponding to the gradient map is a complete deceleration strip; when the divided edge lines are the edge lines among the yellow color blocks, the defective color blocks and the black color blocks, separating yellow color block images, defective color block images and black color block images to obtain a plurality of deceleration strip yellow blocks, defective color blocks and black color blocks, separating new images which are independent of each other, and labeling according to the sequence from left to right in the whole deceleration strip;
considering a complete deceleration strip, the specific principle is that a gray image of the deceleration strip is regarded as a topological plane, the gray value of a yellow block is high and is regarded as a bulge of the topological plane, and the gray value of a black block is lowest and is regarded as a concave part of the plane; firstly, a gradient image is constructed, the detection of gray level change is completed by gradient, a gradient operator is defined as,
the ROI edge segmentation of the deceleration strip is only interested in gradient change in the horizontal direction, namely a one-dimensional template [ -11 ] is selected for filtering f (x, y), the operation direction of the deceleration strip is specified to be from right to left, a gradient map and a tiny region in the gradient map are obtained through calculation, the tiny region is a region with unobvious gradient change, the overflow process is supposed to be increased by single gray, when the gray is increased to the tiny value region, the region forms a connected region, gray values are continuously increased to form different connected regions, a boundary line is formed among the regions, namely the boundary of a yellow block and a black block, each color block is segmented and extracted from right to left, and a plurality of color block minimaps are obtained and are numbered;
step 15, calculating the average gray value and the sequence of each color block,
traversing the color blocks at the first positions of the deceleration strip, namely, all pixels of the first color blocks from the right, keeping the consistency with the previous step, summing and accumulating the gray values of all the pixels, calculating the total number of the pixels in the image, obtaining the average gray value of the image, and circularly executing the operation on the color blocks at the positions in sequence according to the method until the last image is obtained, wherein the average gray values of all the color blocks are sequenced according to the original serial number position;
step 16, calculating the average gray value of the residual road surface image of the original image,
traversing each pixel of the image in three cycles, accumulating and summing gray values of each pixel point of the residual image by taking the corner point as a starting point or an ending point of each cycle according to the recorded contour corner point of the deceleration strip, calculating the total number of the pixel points in the image according to the segmented cycle mode, and solving the average gray value of the image;
and step 17, judging the color of the color block,
setting a numerical range for the obtained gray value of the remaining road surface according to the average gray value of the remaining road surface image, recording the average gray value of the image obtained in the step 5 as theta, and recording the minimum gray value as theta1The maximum gray value is theta2In order to avoid the confusion between the gray value of the black block and the gray value of the road surface and to better distinguish the gray value of the yellow block from the gray value of the road surface, the theta-theta should be used1<θ2-theta, e.g. theta is 50, theta2May be 80, theta1May be 40; mixing the aboveAverage gray α and theta for each color patch1、θ2For comparison, if α<θ1If the color block is a black block, if θ is1<α<θ2If α, the color block is a defective color block>θ2If the color block is yellow, the color block is yellow; sequencing the colors of the color blocks obtained by calculation according to the color block numbers in sequence;
step 18, calculating the total number of the defective color blocks and judging the color,
if the defective color block is clamped between two identical color blocks, the condition of a defective single color block is adopted, and when the defective color block is positioned between two yellow color blocks, the defective black color block is adopted, and when the defective color block is positioned between two black color blocks, the defective yellow color block is adopted; if the defective color block is located between two different color blocks, the two color blocks are continuously defective, and the defective color blocks are yellow and black, respectively, as shown in fig. 8 and 9. Finally, the total color block defect number is calculated in an accumulation way,
the total number of defects is equal to the number of single defect of color block multiplied by 1+ the number of continuous defect of color block multiplied by 2, and the color of the color block lacking is recorded:
the number of the defective yellow blocks is equal to the number of the defective yellow blocks of a single color block multiplied by 1+ the number of the defective continuous color blocks multiplied by 1;
the number of defective black blocks is equal to the number of defective black blocks of a single color block multiplied by 1+ the number of defective continuous color blocks multiplied by 1;
in summary, the invention provides a method for detecting the defect of a road deceleration strip based on Hough line detection, which comprises the steps of firstly obtaining a road image by a binocular camera at a certain frequency, preprocessing the road image to obtain an image frame suspected to contain the deceleration strip, not storing the image not containing the deceleration strip, screening the previous suspected image by a vibration acceleration sensor, deleting the image not passing the screening, detecting the defect of the screened image of the deceleration strip, finally judging whether the current deceleration strip is the defect deceleration strip or not, and judging the number and the color of the defect color blocks of the defect deceleration strip.
The above description is only one embodiment of the present invention, and is only for the purpose of assisting understanding of the system components and the core method and algorithm of the present invention, and the protection scope of the present invention is not considered to be limited thereto, and it should be noted that any person skilled in the art may make any changes or substitutions without any creative thinking work, and the purpose and form are the same, and the embodiments of the present invention should be included.
Claims (8)
1. A defect detection method of a road deceleration strip is characterized by comprising the following steps,
step 1, acquiring a road image containing deceleration strip information, wherein the deceleration strip is a deceleration strip with black blocks and yellow blocks arranged alternately, separating a deceleration strip outline image from the road image, and obtaining a deceleration strip outline image with gray scale and a gray scale road image without the deceleration strip after the deceleration strip outline image and the road image with the deceleration strip outline separated are subjected to gray scale;
step 2, solving a gradient map of the profile image of the deceleration strip with gray scale, solving edge lines among color blocks of the deceleration strip in the gradient map by applying a watershed algorithm,
when the edge line is found to be the edge line between the yellow color block and the black color block, the deceleration strip corresponding to the gradient map is obtained to be a complete deceleration strip;
when the edge line is the edge line between the yellow color block, the defective color block and the black block, separating out a yellow color block image, a defective color block image and a black block image, marking according to the sequence of the yellow color block image, the defective color block image and the black color block image, and then calculating the average gray value of the yellow color block image, the defective color block image and the black block image according to the sequence of the marks; calculating the average gray value of the gray road image without the deceleration strip, and recording the average gray value as theta;
step 3, setting a range of the gray value of the gray road image without the deceleration strip according to theta, and enabling theta to be equal to theta1<θ2- θ, where θ1Is the minimum gray value, theta, of the gray-scale road image without the deceleration strip2Respectively comparing the average gray values of the yellow color block image, the defective color block image and the black color block image calculated in the step 2 with theta for the maximum gray value of the gray road image without the deceleration strip1、θ2The comparison was carried out to determine the color blocks black, yellow anddefective color blocks;
and 4, obtaining the total defect number, the defect yellow block number and the defect black block number according to the color block colors at the two ends of the defect color block.
2. The method for detecting the defect of the road deceleration strip according to claim 1, wherein in the step 1, a road image containing deceleration strip information is obtained according to the following steps;
step a1The method comprises the following steps that a binocular camera is installed at the front upper part of a detection vehicle, a vibration acceleration sensor is installed on a rear axle of the detection vehicle, and the binocular camera continuously photographs the road surface on the detection vehicle in a motion state;
then, sequentially carrying out edge cutting, graying and perspective change on the road image acquired by the binocular camera to obtain a road image after perspective change, wherein the ratio of the width of the left edge and the right edge to the total width of the image during edge cutting is 8-10%, and the ratio of the height of the upper edge of the image to the total height of the image is 10-15%;
step a2Sequentially screening and detecting Hough lines of the road image with perspective change according to a template matching mode to obtain image coordinate data of two end points of a plurality of groups of lines, and calculating the coordinate data of two end points of the lines stored in the same array to obtain the distance of each group of lines;
step a3Setting standard parameters of a standard deceleration strip in an image coordinate system as a threshold, discarding a straight line with a y coordinate difference lower than the threshold and a straight line with an abscissa coordinate difference larger than the threshold, keeping the longest straight line and the next longest straight line, recording coordinates of four corner points of the two straight lines, and selecting the four corner points by using a rectangular frame, wherein the rectangular length-width ratio is smaller than that of the standard deceleration strip;
step a4Carrying out inverse perspective change on the road image framed and selected by the rectangular frame, then calculating the time of the detection vehicle 1 reaching the deceleration strip 3, finally backtracking the image of the moment corresponding to the pulse signal in the time range, and acquiring the road image containing deceleration strip information;
the pulse signal is obtained by the vibration acceleration sensor and reaches a set threshold value after being subjected to gain variation and arithmetic mean filtering in sequence.
3. The method for detecting the defect of the road deceleration strip according to claim 2, wherein the step a is carried out in a step b4And when the Hamming distance between two images is larger than a threshold value, discarding the previous image, retaining the current image, and comparing the Hamming distance with the Hamming distance of the next image, so that the same deceleration belt only has one image.
4. The method for detecting the defect of the road deceleration strip according to claim 1, wherein in step 1, the obtained deceleration strip profile image with the gray scale is subjected to histogram equalization, and then a gradient map of the deceleration strip profile image subjected to histogram equalization is obtained.
5. The method for detecting the defect of the road deceleration strip according to claim 1, wherein in the step 2, the average gray value of the gray road image without the deceleration strip is calculated by adopting the following process;
traversing each pixel of the gray-scale road image without the deceleration strip in three cycles, taking the corner point as the starting point or the ending point of each cycle according to the recorded contour corner point of the deceleration strip, accumulating and summing the gray-scale values of each pixel of the rest images, calculating the total number of the pixels in the images according to the segmented cycle mode, and finally obtaining the average gray-scale value of the gray-scale road image without the deceleration strip.
6. The method for detecting the defects of the road deceleration strip according to claim 1, wherein in step 3, black blocks, yellow blocks and defect blocks are determined in the following manner;
the average gray value of the yellow patch image, the gray patch image and the black patch image is αiWhen i is 1 to n, when αi<θ1The color block is black block, if αi>θ2The color block is yellow when theta1<αi<θ2And the color block is a defective color block.
7. The method for detecting the defects of the road deceleration strip according to claim 1, wherein in the step 4, the total number of defects is determined in the following manner;
when a defective color block is sandwiched between two identical color blocks, it is a case of missing a single color block, and when a defective color block is located between two different color blocks, it is a case of continuously missing two color blocks, and the total number of defects is equal to the number of single color block defects × 1+ the number of continuous color block defects × 2.
8. The method for detecting the defects of the road deceleration strip according to claim 1, wherein the number of the yellow defect blocks and the number of the black defect blocks are determined in the following manner in the step 4;
when a defective color block is between two yellow blocks, it is a defective black color block, when a defective color block is between two black blocks, it is a defective yellow color block, when a defective color block is between two different color blocks, it is a yellow color block and a black color block, respectively, the number of defective yellow blocks is equal to the number of yellow blocks defective by a single color block × 1+ the number of continuous color block defective × 1, and the number of defective black blocks is equal to the number of black blocks defective by a single color block × 1+ the number of continuous color block defective × 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911049884.9A CN110807771B (en) | 2019-10-31 | 2019-10-31 | Defect detection method for road deceleration strip |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911049884.9A CN110807771B (en) | 2019-10-31 | 2019-10-31 | Defect detection method for road deceleration strip |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110807771A true CN110807771A (en) | 2020-02-18 |
| CN110807771B CN110807771B (en) | 2022-03-22 |
Family
ID=69489711
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911049884.9A Active CN110807771B (en) | 2019-10-31 | 2019-10-31 | Defect detection method for road deceleration strip |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110807771B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111401341A (en) * | 2020-06-02 | 2020-07-10 | 弗徕威智能机器人科技（上海）有限公司 | Deceleration strip detection method and device based on vision and storage medium thereof |
| CN111862131A (en) * | 2020-07-31 | 2020-10-30 | 易思维（杭州）科技有限公司 | Adhesive tape edge detection method and application thereof |
| CN113012427A (en) * | 2021-02-09 | 2021-06-22 | 上海同陆云交通科技有限公司 | Improved vehicle-mounted lightweight inspection system and method |
| CN113836964A (en) * | 2020-06-08 | 2021-12-24 | 北京图森未来科技有限公司 | Method and device for detecting lane line corner |
| CN113989510A (en) * | 2021-12-28 | 2022-01-28 | 深圳市万物云科技有限公司 | River drainage outlet overflow detection method and device and related equipment |
Citations (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6233364B1 (en) * | 1998-09-18 | 2001-05-15 | Dainippon Screen Engineering Of America Incorporated | Method and system for detecting and tagging dust and scratches in a digital image |
| US7088862B1 (en) * | 1997-11-26 | 2006-08-08 | Cognex Corporation | Fast high-accuracy multi-dimensional pattern inspection |
| US20070195993A1 (en) * | 2006-02-22 | 2007-08-23 | Chao-Ho Chen | Method for video object segmentation |
| CN101620732A (en) * | 2009-07-17 | 2010-01-06 | 南京航空航天大学 | Visual detection method of road driving line |
| WO2011025452A1 (en) * | 2009-08-24 | 2011-03-03 | Singapore Health Services Pte Ltd | A method and system for detecting disc haemorrhages |
| CN102184378A (en) * | 2011-04-27 | 2011-09-14 | 茂名职业技术学院 | Method for cutting portable data file (PDF) 417 standard two-dimensional bar code image |
| CN102324102A (en) * | 2011-10-08 | 2012-01-18 | 北京航空航天大学 | Method for automatically filling structure information and texture information of hole area of image scene |
| CN102436644A (en) * | 2011-11-02 | 2012-05-02 | 南京物联网研究院发展有限公司 | Unstructured road detection method based on adaptive edge registration |
| CN102665030A (en) * | 2012-05-14 | 2012-09-12 | 浙江大学 | Improved bilinear Bayer format color interpolation method |
| CN102682292A (en) * | 2012-05-10 | 2012-09-19 | 清华大学 | Method based on monocular vision for detecting and roughly positioning edge of road |
| CN103512888A (en) * | 2013-06-05 | 2014-01-15 | 北京化工大学 | Cigarette packet seal defect detecting system based on image identification technology |
| CN104200210A (en) * | 2014-08-12 | 2014-12-10 | 合肥工业大学 | License plate character segmentation method based on parts |
| CN104268860A (en) * | 2014-09-17 | 2015-01-07 | 电子科技大学 | Lane line detection method |
| CN104599270A (en) * | 2015-01-18 | 2015-05-06 | 北京工业大学 | Breast neoplasms ultrasonic image segmentation method based on improved level set algorithm |
| CN105404868A (en) * | 2015-11-19 | 2016-03-16 | 电子科技大学 | Interaction platform based method for rapidly detecting text in complex background |
| CN105427291A (en) * | 2015-11-12 | 2016-03-23 | 北京建筑大学 | Method for detecting vector edges of multispectral remote sensing images |
| CN106226834A (en) * | 2016-07-14 | 2016-12-14 | 昆山市交通工程试验检测中心有限公司 | A kind of vehicular road surface manhole disappearance detection method |
| CN107705301A (en) * | 2017-09-29 | 2018-02-16 | 南京中设航空科技发展有限公司 | A kind of highway graticule damage testing method based on unmanned plane highway map picture |
| CN108573234A (en) * | 2018-04-19 | 2018-09-25 | 孙磊 | City manhole cover miss status identifying system |
| CN108830165A (en) * | 2018-05-22 | 2018-11-16 | 南通职业大学 | A kind of method for detecting lane lines considering front truck interference |
| CN109255776A (en) * | 2018-07-23 | 2019-01-22 | 中国电力科学研究院有限公司 | A kind of transmission line of electricity split pin defect automatic identifying method |
| CN110210451A (en) * | 2019-06-13 | 2019-09-06 | 重庆邮电大学 | A kind of zebra line detecting method |
-
2019
- 2019-10-31 CN CN201911049884.9A patent/CN110807771B/en active Active
Patent Citations (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7088862B1 (en) * | 1997-11-26 | 2006-08-08 | Cognex Corporation | Fast high-accuracy multi-dimensional pattern inspection |
| US6233364B1 (en) * | 1998-09-18 | 2001-05-15 | Dainippon Screen Engineering Of America Incorporated | Method and system for detecting and tagging dust and scratches in a digital image |
| US20070195993A1 (en) * | 2006-02-22 | 2007-08-23 | Chao-Ho Chen | Method for video object segmentation |
| CN101620732A (en) * | 2009-07-17 | 2010-01-06 | 南京航空航天大学 | Visual detection method of road driving line |
| WO2011025452A1 (en) * | 2009-08-24 | 2011-03-03 | Singapore Health Services Pte Ltd | A method and system for detecting disc haemorrhages |
| CN102184378A (en) * | 2011-04-27 | 2011-09-14 | 茂名职业技术学院 | Method for cutting portable data file (PDF) 417 standard two-dimensional bar code image |
| CN102324102A (en) * | 2011-10-08 | 2012-01-18 | 北京航空航天大学 | Method for automatically filling structure information and texture information of hole area of image scene |
| CN102436644A (en) * | 2011-11-02 | 2012-05-02 | 南京物联网研究院发展有限公司 | Unstructured road detection method based on adaptive edge registration |
| CN102682292A (en) * | 2012-05-10 | 2012-09-19 | 清华大学 | Method based on monocular vision for detecting and roughly positioning edge of road |
| CN102665030A (en) * | 2012-05-14 | 2012-09-12 | 浙江大学 | Improved bilinear Bayer format color interpolation method |
| CN103512888A (en) * | 2013-06-05 | 2014-01-15 | 北京化工大学 | Cigarette packet seal defect detecting system based on image identification technology |
| CN104200210A (en) * | 2014-08-12 | 2014-12-10 | 合肥工业大学 | License plate character segmentation method based on parts |
| CN104268860A (en) * | 2014-09-17 | 2015-01-07 | 电子科技大学 | Lane line detection method |
| CN104599270A (en) * | 2015-01-18 | 2015-05-06 | 北京工业大学 | Breast neoplasms ultrasonic image segmentation method based on improved level set algorithm |
| CN105427291A (en) * | 2015-11-12 | 2016-03-23 | 北京建筑大学 | Method for detecting vector edges of multispectral remote sensing images |
| CN105404868A (en) * | 2015-11-19 | 2016-03-16 | 电子科技大学 | Interaction platform based method for rapidly detecting text in complex background |
| CN106226834A (en) * | 2016-07-14 | 2016-12-14 | 昆山市交通工程试验检测中心有限公司 | A kind of vehicular road surface manhole disappearance detection method |
| CN107705301A (en) * | 2017-09-29 | 2018-02-16 | 南京中设航空科技发展有限公司 | A kind of highway graticule damage testing method based on unmanned plane highway map picture |
| CN108573234A (en) * | 2018-04-19 | 2018-09-25 | 孙磊 | City manhole cover miss status identifying system |
| CN108830165A (en) * | 2018-05-22 | 2018-11-16 | 南通职业大学 | A kind of method for detecting lane lines considering front truck interference |
| CN109255776A (en) * | 2018-07-23 | 2019-01-22 | 中国电力科学研究院有限公司 | A kind of transmission line of electricity split pin defect automatic identifying method |
| CN110210451A (en) * | 2019-06-13 | 2019-09-06 | 重庆邮电大学 | A kind of zebra line detecting method |
Non-Patent Citations (2)
| Title |
|---|
| LIN, Q等: "Watershed Segmentation Based on Histogram Threshold and Lowest Mean Edge Value", 《APPLIED MECHANICS & MATERIALS》 * |
| 屠昕等: "一种基于分水岭分割的快速图像修复算法", 《复旦学报（自然科学版）》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111401341A (en) * | 2020-06-02 | 2020-07-10 | 弗徕威智能机器人科技（上海）有限公司 | Deceleration strip detection method and device based on vision and storage medium thereof |
| CN111401341B (en) * | 2020-06-02 | 2020-09-08 | 弗徕威智能机器人科技（上海）有限公司 | Deceleration strip detection method and device based on vision and storage medium thereof |
| CN113836964A (en) * | 2020-06-08 | 2021-12-24 | 北京图森未来科技有限公司 | Method and device for detecting lane line corner |
| CN113836964B (en) * | 2020-06-08 | 2023-11-10 | 北京图森未来科技有限公司 | Method and device for detecting corner points of lane lines |
| CN111862131A (en) * | 2020-07-31 | 2020-10-30 | 易思维（杭州）科技有限公司 | Adhesive tape edge detection method and application thereof |
| CN111862131B (en) * | 2020-07-31 | 2021-03-19 | 易思维（杭州）科技有限公司 | Adhesive tape edge detection method and application thereof |
| CN113012427A (en) * | 2021-02-09 | 2021-06-22 | 上海同陆云交通科技有限公司 | Improved vehicle-mounted lightweight inspection system and method |
| CN113012427B (en) * | 2021-02-09 | 2022-05-24 | 上海同陆云交通科技有限公司 | Improved vehicle-mounted lightweight inspection system and method |
| CN113989510A (en) * | 2021-12-28 | 2022-01-28 | 深圳市万物云科技有限公司 | River drainage outlet overflow detection method and device and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110807771B (en) | 2022-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110807771B (en) | Defect detection method for road deceleration strip | |
| US11551341B2 (en) | Method and device for automatically drawing structural cracks and precisely measuring widths thereof | |
| CN108182383B (en) | Vehicle window detection method and device | |
| CN112598922B (en) | Parking space detection method, device, equipment and storage medium | |
| CN112488046B (en) | Lane line extraction method based on high-resolution images of unmanned aerial vehicle | |
| CN113781537B (en) | Rail elastic strip fastener defect identification method and device and computer equipment | |
| CN109708658B (en) | Visual odometer method based on convolutional neural network | |
| CN109559324A (en) | A kind of objective contour detection method in linear array images | |
| CN116612103B (en) | Intelligent detection method and system for building structure cracks based on machine vision | |
| CN116862910B (en) | Visual detection method based on automatic cutting production | |
| CN111753749A (en) | Lane line detection method based on feature matching | |
| CN107506753B (en) | Multi-vehicle tracking method for dynamic video monitoring | |
| CN114511832B (en) | Lane line analysis method and device, electronic device and storage medium | |
| CN115841633A (en) | Power tower and power line associated correction power tower and power line detection method | |
| CN112598743B (en) | Pose estimation method and related device for monocular vision image | |
| CN111444916A (en) | License plate positioning and identifying method and system under unconstrained condition | |
| KR102260556B1 (en) | Deep learning-based parking slot detection method and apparatus integrating global and local information | |
| CN113053164A (en) | Parking space identification method using look-around image | |
| CN115830514B (en) | Whole river reach surface flow velocity calculation method and system suitable for curved river channel | |
| CN106886609B (en) | Block type rural residential area remote sensing quick labeling method | |
| CN116385477A (en) | Tower image registration method based on image segmentation | |
| CN115331193A (en) | Parking space identification method, parking space identification system, electronic equipment and storage medium | |
| CN111539279B (en) | Road height limit detection method, device, equipment and storage medium | |
| CN115239623A (en) | Method for detecting contact force of bow net | |
| CN113963230A (en) | Parking space detection method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |