CN105956542B - High-resolution remote sensing image road extraction method based on statistical matching of structural wire harnesses - Google Patents
High-resolution remote sensing image road extraction method based on statistical matching of structural wire harnesses Download PDFInfo
- Publication number
- CN105956542B CN105956542B CN201610272080.5A CN201610272080A CN105956542B CN 105956542 B CN105956542 B CN 105956542B CN 201610272080 A CN201610272080 A CN 201610272080A CN 105956542 B CN105956542 B CN 105956542B
- Authority
- CN
- China
- Prior art keywords
- road
- line
- section
- edge
- distance
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/182—Network patterns, e.g. roads or rivers
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a high-resolution remote sensing road extraction method for statistical matching of a structural wiring harness. The method comprises road baseline detection, wire harness profile feature statistics and road mode matching. The road baseline detection is to count the optimal marshalling connecting line object according to the edge segmentation marshalling result; the statistics of the profile features of the wire harness is to take a road base line as a horizontal line, calculate the distance between each object in the wire harness and the base line and form a one-dimensional feature vector as the structural features of the road profile; the road mode matching is to construct a Gaussian mixture model based on structural features of road side lines which are symmetrically distributed relative to a center line, and the model and the section statistical features are subjected to moving matching to detect the optimal road center line and side line positions; in the post-processing stage, the road extraction result needs to be verified and corrected according to the feature of the ground feature. The invention fully utilizes abundant line characteristics in the high-resolution remote sensing image, combines the artificial seed points as prior information, and has strong practicability.
Description
Technical Field
The invention relates to the technical field of remote sensing image application, in particular to a road extraction method of a high-resolution remote sensing image with statistical matching of structural wiring harnesses.
background
The road is used as a link for connecting regions and is the basis of national economic development. The road plays a significant role in urban land and economic activities, and the high-precision identification and extraction and real-time update of the road network information have important theoretical and practical significance for traffic management, urban planning, automatic vehicle navigation, emergency transaction processing and update of a geographic information system database. In recent years, with the promotion of the national urbanization process and the rapid development of economic construction, the demand of people for road transportation is continuously increased, thereby promoting the rapid development of road construction in China. According to the update summary report of the national basic geographic database, the update data volume of the road elements occupies a large proportion in the work of updating the terrain database, and is second to the elements of the residential areas. Therefore, the rapid acquisition and updating of the road element data become an important task for the construction of basic geographic information in China.
currently, the comprehensive judgment and adjustment of interior industry based on high-resolution remote sensing images is a main means for updating basic geographic elements including road networks. The advantages of large-area synchronous observation, high timeliness, economy and the like of the remote sensing technology enable the remote sensing image to be used for updating basic geographic information to have great advantages. Compared with field ground actual measurement data updating, the field interpretation mode based on the remote sensing image improves the collection efficiency of basic geographic element data, and is suitable for rapid updating of large-range roads.
in the process of extracting the road of the high-resolution remote sensing image, the semi-automatic road extraction method not only fully utilizes the cognitive, recognition and detection capabilities of people, but also exerts the calculation capability of a computer, and is considered as a better choice at present.
The urban arterial road has rich lane line objects, and the lane line objects are represented as rich line features in the high-resolution image, the line features are distributed in a bundle shape, and the overall trend and the rough width information of the road are identified explicitly. However, it is difficult to convert the visually perceived effects into computer-extractable results. The line segment grouping method commonly used in the medium-low resolution remote sensing image is difficult to be applied to the road extraction task of the high-resolution image: on one hand, the line features on the main road are not all straight line segments, and a certain radian exists in a bent road segment; on the other hand, the connection between parallel adjacent line objects has a great ambiguity due to the line characteristics being too rich. The task of extracting roads from the images with medium and low resolution focuses on extracting the road center line, and people have higher requirements on the road extraction method and can extract the road edge line under the support of the images with high resolution. Therefore, the trend of road semi-automatic extraction by fully utilizing rich line characteristics of high-resolution remote sensing images and combining manual operation is realized.
Disclosure of Invention
aiming at the problems in the background art, the invention provides a high-resolution remote sensing road extraction method for statistical matching of structural wire harnesses
the technical scheme of the invention is as follows:
a high-resolution remote sensing road extraction method for statistical matching of structural wire harnesses comprises the following steps:
step one, edge detection and pretreatment. Carrying out bilateral filtering on the remote sensing image, and then carrying out edge detection on the road image by using a Canny operator; detecting intersection points and inflection points of the edges and eliminating the intersection points and the inflection points; and filtering the edge section according to the connecting line direction of the buffer area and the seed point, and reserving effective road elements.
And step two, element fitting perception grouping. A method for element fitting perception grouping is provided, namely road elements are connected through analysis, selection and synthesis of features, and the road elements are expanded into road segments.
and step three, selecting a baseline. And taking the accumulated length of the edge primitives in the grouped object as a standard for selecting a base line, and taking the maximum accumulated length of the edge primitives as a road base line.
And step four, counting the structural features of the road section. The road section structure characteristics are provided to express the section structure of the road and help determine the side line and the middle line of the road.
and step five, matching and verifying the road mode. And providing a verification method based on pattern matching, verifying and detecting the road side line and the road middle line according to the priori knowledge, and correcting the obtained road line through verification feedback.
the specific process of the first step is as follows;
firstly, carrying out bilateral filtering on an image, filtering salt and pepper noise, simultaneously keeping edge characteristics, and then carrying out edge detection on a road image by using a Canny operator; the Canny edge detection result is the original edge characteristic of the road image, and in order to extract the road, effective road elements need to be obtained from the Canny edge detection result, namely independent edge segments capable of representing the road elements; the method comprises the steps of detecting intersection points and inflection points of edges and removing the intersection points and the inflection points to realize the decomposition of edge detection results, wherein edge objects after decomposition are straight line segments or arc segments with geometric characteristics; determining the image processing range of road extraction and the direction of the road by inputting the seed points, filtering the edge section according to the connecting line direction of the buffer area and the seed points, and reserving the effective road elements for supporting the subsequent marshalling processing.
The specific process of the second step is as follows;
The method comprises the steps of firstly respectively fitting road elements, and determining the optimal fitting order according to the position difference of a fitting curve and the elements, wherein the primary curve and the secondary curve are shown in formulas (1) to (4), and the formula (5) is an optimization condition for solving fitting parameters; then sorting according to the length of the element, and carrying out iterative marshalling processing between the edges according to the sequence;
y＝a0+a1x+a2x2 (1)
y＝a0+a1x (2)
(1) the fitting values represented by the formula (1) and the formula (2) are actual values, the formula (3) and the formula (4) are fitting values, the formula (1) to the formula (4) represents a fitting curve of a base line, the formula (i) represents the fitting curve x i and y i have subscripts, and the formula (5) represents the minimum error of the fitting curves from 1 to N.
the constraint conditions for grouping the fitted primitives comprise continuity f con, proximity f pro and similarity f sim;
From the above description, the primitive join probability function is constructed as shown in equation (6):
fconn＝c0fpro(gt,gl,l1,l2)+c1fcon(α,|θ1+θ2|)+c2fsim(Δp) (6)
c 0, c 1 and c 2 are weight coefficients of corresponding factors, l 1 and l 2 are two sections of primitives respectively, the variables related to the proximity are that the longitudinal distance g t is g t1 + g t2, the transverse distance g l is g l1 + g l2, g t1 and g l1 are the longitudinal distance and the transverse distance of the first section of primitives, g t2 and g l2 are the longitudinal distance and the transverse distance of the second section of primitives respectively, the variables related to the continuity are the included angle alpha and the collinearity theta 1 + theta 2 | of the line primitives, theta 1 and theta 2 are included angles between connecting lines of central points of the two primitives and the primitives, the similarity is that the characteristic difference delta p between the primitives is | p 1 -p 2 | and p 1, and p 2 are geometric and spectral characteristics corresponding to the line primitives;
if the fitted primitive is a quadratic curve, intercepting effective primitive segments from two ends of the primitive, and calculating connection probability according to the connection constraint condition; and when the deviation between the grouping result and the main direction of the road is large, performing rollback processing, and reselecting suboptimal primitives for grouping connection.
the concrete process of the third step is as follows;
The accumulated length of edge elements in the grouped object is used as a standard for selecting a base line, and the longer the accumulated length of the edge is, the more effective edges are supported by the grouped object; if a plurality of seed points are input to the current road section to be extracted, taking the area marked by every two adjacent seed points as a sub-road section, executing edge element accumulation length processing on each sub-road section, and respectively selecting corresponding base lines in each sub-road section; and translating the base lines corresponding to the sub-road sections at the positions of the public seed points of the two adjacent sub-road sections in a normal direction until the base lines in all the sub-road sections are connected end to form the base line of the whole road section.
The concrete process of the step four is as follows;
the section structure characteristics are formed by the normal distance from the element to the base line; calculating the distance between each element and the base line, wherein the method comprises the following steps: taking the normal distance from the middle point of the edge to the road base line as the section distance of the current edge object, wherein the section distance on the left side of the fitting road is a positive value and the section distance on the right side of the fitting road is a negative value according to the main direction of the fitting road;
After the edge section distance is obtained, all edge elements on the current road section are classified according to the section distance and are grouped and connected again, meanwhile, the length accumulated value of a newly grouped object is counted, and a one-dimensional line graph is generated according to the section distance and the length accumulated value, namely the section structure characteristic.
the concrete process of the step five is as follows;
According to the verification method based on pattern matching, a road is considered to have a symmetrical structure, namely the road is formed by two side lines, and the side lines are symmetrically distributed on two sides of a road center line or a virtual center line; dividing the road into a double-peak mode and a triple-peak mode according to whether the road has a central lane line or not, and respectively defining corresponding mixed Gaussian kernel functions as shown in formulas (7) and (8);
Wherein ω is road width; sigma is kernel function variance, and a span range of a road side line or a central line corresponding to a statistical peak value is identified; defining a road mode matching response function based on the kernel function, as shown in a formula (9);
wherein d min and d max are the minimum and maximum values of the section distance, l x is the section structure characteristic value corresponding to the x position of the section distance, namely the edge accumulated length value, x i is the kernel function offset, and v i is the response value of the current position mode matching;
At the moment, the detection problem of the road middle line and the side line is converted into the solution of the section position corresponding to the maximum value obtained by v i, omega in the kernel function is an unknown parameter, the possible road width is predicted according to the distance between the section positions corresponding to the peak values in the statistical characteristics, omega in the double peak mode is the section distance between two continuous peak values, omega in the three peak mode is the section distance between the head peak value and the tail peak value in three continuous peak values;
The process of detecting the edge accumulation length peak value in the section structure characteristic comprises two steps:
(a all local peaks in the feature are detected; assuming l (x) is a statistical feature length function, where x is an integer profile distance value (-d min ≦ x ≦ d max), the local peaks are detected under the following conditions:
(b) the smaller peaks are filtered: when the local peak value is far smaller than the global maximum peak value or smaller than the average edge statistical length, deleting the peak value from the peak value set, as shown in formula (11);
the method comprises the steps of obtaining a road base line, calculating a model matching response function, wherein epsilon 1 is a threshold value, l avg is an average edge statistical length, finally substituting all predicted omega into a kernel function p (x | omega, theta), calculating a response value v i corresponding to each offset x i position of the model matching response function, taking a position x i corresponding to the maximum value of the response value v i as a road center line position, and taking a statistical characteristic peak position corresponding to the road width omega in the kernel function as the left and right line positions of a current road, and translating the road base line according to the obtained position information in the normal direction to obtain a final road extraction result;
The verification feedback is the last step of the road extraction method, and two verification feedback conditions are provided;
the feedback condition one is as follows: the central line of the road extraction result needs to be located in a certain buffer range of the link of the road segment seed points, and a consistency index S (a, B) is defined as shown in formula 12:
when S (A, B) > T S, the current road extraction result is considered as a candidate road extraction result, and T s is a road length integrity threshold value;
the second feedback condition is that after pattern matching and sorting according to response values v i, a road extraction result set which meets the first feedback condition is selected, and the result that the difference between the pattern matching response value v i and the maximum response value v max of each extraction result is within a specified threshold value is used as a candidate result;
and when the effective road extraction result cannot be obtained under the constraint of the feedback condition, returning to the base line selection stage, reselecting the road base line, and performing section structure statistics and road extraction.
The invention fully utilizes abundant line characteristics in high-resolution remote sensing images, combines artificial seed points as prior information, has stronger practicability, introduces road structure characteristics as prior knowledge to carry out road mode matching, can verify and correct road extraction results according to surface feature characteristics, has higher accuracy and better effect, and is characterized in that:
(1) regarding edges in a road segment scene as a beam-shaped set, performing multi-level edge grouping based on fitting characteristics, and acquiring a grouping object with semantic characteristics from the multi-level edge grouping;
(2) The road structure characteristics are introduced as prior knowledge, the road boundary line is considered to be symmetrically distributed relative to the road center line, and a corresponding Gaussian mixture function is defined to express the structure characteristics; detecting final road center line and side line extraction results by traversing and matching road structure features and a Gaussian mixture function;
(3) And iteratively extracting the matching result of the road center line and the side line according to the position difference between the extraction result and the seed point connecting line section as feedback information, and selecting the optimal extraction result according to the iteration termination condition.
drawings
FIG. 1 is a flow processing diagram of a high-resolution remote sensing road extraction method for statistical matching of structural wiring harnesses.
Fig. 2 is a schematic diagram of line cell grouping.
FIG. 3 is a schematic diagram of a cross-sectional distance metric.
Detailed Description
The invention provides a high-resolution remote sensing road extraction method for statistical matching of a structural wiring harness. The road baseline detection is to count the optimal marshalling connecting line object according to the edge segmentation marshalling result; the statistics of the profile features of the wire harness is to take a road base line as a horizontal line, calculate the distance between each object in the wire harness and the base line and form a one-dimensional feature vector as the structural features of the road profile; the road mode matching is to construct a Gaussian mixture model based on structural features of road side lines which are symmetrically distributed relative to a center line, and the model and the section statistical features are subjected to moving matching to detect the optimal road center line and side line positions; in the post-processing stage, the road extraction result needs to be verified and corrected according to the feature of the ground feature.
the technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments, wherein a flow chart is shown in fig. 1, and the technical solution flow of the embodiments includes the following steps:
Step one, edge detection and pretreatment. Firstly, bilateral filtering is carried out on the image, the edge characteristics are kept while salt and pepper noise is filtered, and then Canny operators are used for carrying out edge detection on the road image. The Canny edge detection result is the original edge feature of the road image, and in order to extract the road, effective road elements, namely independent edge segments capable of representing the road elements, need to be obtained from the Canny edge detection result. From the practical perspective, the decomposition of the edge detection result is realized by detecting the intersection point and the inflection point of the edge and eliminating the intersection point and the inflection point, and the edge objects after decomposition are all straight line segments or arc segments with certain geometric characteristics. After the seed points are input, the image processing range of road extraction and the approximate direction of the road can be determined, the edge sections are filtered according to the connecting line direction of the buffer area and the seed points, and the retained road elements are effective road elements and can be used for supporting subsequent marshalling processing.
And step two, element fitting perception grouping. And combining and connecting the elements, namely connecting the road elements through analysis, selection and synthesis of features, and expanding the road elements into road segments. The method comprises the steps of firstly respectively fitting road elements, and determining the optimal fitting order according to the position difference of a fitting curve and the elements, wherein the primary curve and the secondary curve are shown in formulas (1) to (4), and the formula (5) is an optimization condition for solving fitting parameters; then sorting according to the length of the primitive and carrying out iterative grouping processing between the edges according to the sequence.
y＝a0+a1x+a2x2 (1)
y＝a0+a1x (2)
the main constraints for grouping the fitted primitives include continuity, proximity and similarity, (1) the continuity f con indicates that roads are generally continuous and that there is a general possibility of connection with gaps between primitives, (2) the proximity f pro indicates that the closer the distance between primitives, the more likely they are to be connected, (3) the similarity f sim indicates that primitives have the same direction, similar gray features and other collinear features, the greater the probability of connection between primitives.
From the above description, the primitive join probability function is constructed as shown in equation (6):
fconn＝c0fpro(gt,gl,l1,l2)+c1fcon(α,|θ1+θ2|)+c2fsim(Δp) (6)
The parameters are shown as a line element grouping schematic diagram in fig. 2, which shows the geometric relationship between two elements, wherein c 0, c 1 and c 2 are weight coefficients of corresponding factors, l 1 and l 2 are two elements respectively, the variables related to the proximity are longitudinal distance g t ═ g t1 + g t2, transverse distance g l ═ g l1 + g l2, g t1 and g l1 are the longitudinal distance and the transverse distance of the first element, g t2 and g l2 are the longitudinal distance and the transverse distance of the second element respectively, the variables related to the continuity are the line element included angle α and the collinearity θ ═ θ 1 + θ 2 |, θ 1 and θ 2 are the included angle between the central points of the two elements connected together, and the similarity is the characteristic difference Δ p ═ p 1 -p 2 |, p 1, p 2 is the geometric and spectral characteristic spectrum corresponding to the line elements.
If the fitted primitive is a quadratic curve, effective primitive segments need to be intercepted from the two ends of the primitive, and the connection probability is calculated according to the connection constraint conditions. And when the deviation of the grouping result and the road main direction is large, performing rollback processing, and reselecting suboptimal primitives for grouping connection, which is an iterative trial and error mechanism to ensure that a reasonable and optimal result is obtained in the grouping processing.
step three: and selecting a baseline. The accumulated length of the edge primitives in the grouped object is used as a standard for selecting a baseline, and the longer the accumulated length of the edge is, the more effective edges are supported by the grouped object. If a plurality of seed points (for the extraction task of a curve road section or a longer road) are input into the current road section to be extracted, the area marked by every two adjacent seed points is taken as a sub-road section, the processing is executed on each sub-road section, and the corresponding base line is respectively selected in each sub-road section. Because the end-to-end connection of the base line in each sub-road section cannot be ensured, in order to obtain a continuous base line in the whole road section, the base lines corresponding to the sub-road sections need to be translated at the positions of the public seed points of two adjacent sub-road sections in a normal direction until the base lines in all the sub-road sections are connected end-to-end to form the base line of the whole road section.
And step four, counting the structural features of the road section. The cross-sectional structural features are constituted by the normal distance of the elements from the base line. The method for calculating the distance between each element and the base line is shown in fig. 3, the normal distance from the middle point of the edge to the base line of the road is taken as the section distance of the current edge object, and the section distance on the left side of the fitted road is a positive value (d1) and the section distance on the right side is a negative value (d2) according to the main direction of the fitted road, so that the separability and the continuity of the statistical result in the section distance are ensured. Due to the difference of baseline selection, the calculated results of the section distances of the same edge object are different, so that the section distances in the statistical characteristic diagram of the subsequent section structure are uniformly standardized to be positive values, which is convenient for uniform expression and comparison.
and after the edge section distance is obtained, classifying all edge elements on the current road section according to the section distance, and regrouping and connecting. And meanwhile, counting the length accumulated value of the newly grouped object. And generating a one-dimensional line graph which is the section structure characteristic according to the section distance and the accumulated length value. According to experience, the edge characteristics corresponding to the section edge and the section position with the lane line or the isolation tape mark are rich. The cross-sectional distance locations corresponding to the building edges on both sides of the road also form local statistical peaks. It should be noted that, before counting the edge length, edge objects whose direction difference from the baseline exceeds a threshold value are filtered to ensure the validity of the statistical result.
Step five: and matching and verifying the road mode. A verification method based on pattern matching is provided, and the road is considered to have a symmetrical structure, namely the road is formed by bilateral lines, and the bilateral lines are symmetrically distributed on two sides of a road center line or a virtual center line (under the condition of no lane line). According to whether the road has a central lane line or not, the road is divided into a double-peak mode and a triple-peak mode, and corresponding mixed Gaussian kernel functions are respectively defined as shown in formulas (7) and (8).
wherein ω is road width; and sigma is kernel function variance, and the span range of the road side line or the central line corresponding to the statistical peak value is identified. And defining a road mode matching response function based on the kernel function, as shown in formula (9).
Wherein d min and d max are the minimum and maximum values of the section distance, l x is the section structure characteristic value corresponding to the x position of the section distance, namely the edge accumulated length value, x i is the kernel function offset, and v i is the response value of the current position pattern matching.
at this time, the detection problem of the road center line and the side line is converted into the method that omega in a section position kernel function corresponding to the maximum value obtained by v i is obtained as an unknown parameter, and the possible road width needs to be predicted according to the distance between the section positions corresponding to the peak values in the statistical characteristics.
The process of detecting the edge accumulation length peak value in the section structure characteristic comprises two steps:
(a) And detecting all local peaks in the features, and assuming l (x) is a statistical feature length function, wherein x is an integer section distance value (-d min ≦ x ≦ d max), the detection condition of the local peaks is as follows:
(b) the smaller peaks are filtered: when the local peak is much smaller than the global maximum peak or smaller than the average edge statistical length, the peak is removed from the set of peaks, as shown in equation (11).
And finally, substituting all predicted omega into a kernel function p (x | omega, theta), calculating a response value v i corresponding to each offset x i position of the pattern matching response function, taking a position x i corresponding to the maximum value of the response value v i as the road center line position, taking the statistical characteristic peak position corresponding to the road width omega in the kernel function as the left and right side line positions of the current road, and translating the road base line according to the obtained position information in the normal direction to obtain the final road extraction result.
the verification feedback is the last step of the road extraction method, and due to the complexity of the road scene, the road line obtained based on the processing process cannot be guaranteed to correspond to the actual road condition, and needs to be corrected through the verification feedback. And verifying that the feedback condition is two. Firstly, the central line of the road extraction result needs to be located in a certain buffer range of the link of the road segment seed points, and a consistency index S (a, B) is defined, as shown in formula 12:
When S (A, B) > T s, the current road extraction result is considered as a candidate road extraction result, and T s is a road length integrity threshold value.
In reality, a main road is usually provided with a plurality of remarkable line objects of non-road side lines such as a plurality of road lines or green isolation belts, and the like, influence of the remarkable line objects is considered to define a feedback condition II, after pattern matching and sorting according to a response value v i, a road extraction result set meeting the feedback condition I is selected, a result that the difference between the pattern matching response value v i and the maximum response value v max of each extraction result is within a certain range is taken as a candidate result, then a result corresponding to the maximum road width omega is selected as a final road extraction result, when an effective road extraction result cannot be obtained under the constraint of the feedback condition, a base line selection stage is returned, and a road base line is reselected to perform section structure statistics and road extraction.
Claims (4)
1. a high-resolution remote sensing road extraction method for statistical matching of structural wire harnesses is characterized by comprising the following steps: comprises the following steps;
Step one, edge detection and pretreatment;
carrying out bilateral filtering on the remote sensing image, and then carrying out edge detection on the road image by using a Canny operator; detecting intersection points and inflection points of the edges and eliminating the intersection points and the inflection points; filtering the edge section according to the connecting line direction of the buffer area and the seed point, and reserving effective road elements;
Secondly, element fitting perception grouping;
the specific process of the second step is as follows;
the method comprises the steps of firstly respectively fitting road elements, and determining the optimal fitting order according to the position difference of a fitting curve and the elements, wherein the primary curve and the secondary curve are shown in formulas (1) to (4), and the formula (5) is an optimization condition for solving fitting parameters; then sorting according to the length of the element, and carrying out iterative marshalling processing between the edges according to the sequence;
y＝a0+a1x+a2x2 (1)
y＝a0+a1x (2)
The constraint conditions for grouping the fitted primitives comprise continuity f con, proximity f pro and similarity f sim;
from the above description, the primitive join probability function is constructed as shown in equation (6):
fconn＝c0fpro(gt,gl,l1,l2)+c1fcon(α,|θ1+θ2|)+c2fsim(Δp) (6)
c 0, c 1 and c 2 are weight coefficients of corresponding factors, l 1 and l 2 are two sections of primitives respectively, the variables related to the proximity are that the longitudinal distance g t is g t1 + g t2, the transverse distance g l is g l1 + g l2, g t1 and g l1 are the longitudinal distance and the transverse distance of the first section of primitives, g t2 and g l2 are the longitudinal distance and the transverse distance of the second section of primitives respectively, the variables related to the continuity are the included angle alpha and the collinearity theta 1 + theta 2 | of the line primitives, theta 1 and theta 2 are included angles between connecting lines of central points of the two primitives and the primitives, the similarity is that the characteristic difference delta p between the primitives is | p 1 -p 2 | and p 1, and p 2 are geometric and spectral characteristics corresponding to the line primitives;
If the fitted primitive is a quadratic curve, intercepting effective primitive segments from two ends of the primitive, and calculating the connection probability according to a formula (6); when the deviation between the marshalling result and the main direction of the road is large, performing rollback processing, and reselecting suboptimal elements for marshalling connection;
Step three, selecting a base line;
the concrete process of the third step is as follows;
the accumulated length of edge elements in the grouped object is used as a standard for selecting a base line, and the longer the accumulated length of the edge is, the more effective edges are supported by the grouped object; if a plurality of seed points are input to the current road section to be extracted, taking the area marked by every two adjacent seed points as a sub-road section, executing edge element accumulation length processing on each sub-road section, and respectively selecting corresponding base lines in each sub-road section; translating the base lines corresponding to the sub-road sections at the positions of the public seed points of the two adjacent sub-road sections in a normal direction until the base lines in all the sub-road sections are connected end to form the base line of the whole road section;
Step four, statistics of road profile structural features;
The method comprises the steps of providing a road section structure characteristic to express a section structure of a road and help to determine a side line and a middle line of the road;
step five, matching and verifying the road mode;
and verifying and detecting the road side line and the road middle line according to the priori knowledge by using a verification method based on pattern matching, and correcting the obtained road line through verification feedback.
2. the method for extracting the high-resolution remote sensing road statistically matched with the structural wire harness according to claim 1, wherein the method comprises the following steps: the specific process of the first step is as follows;
Firstly, carrying out bilateral filtering on an image, filtering salt and pepper noise, simultaneously keeping edge characteristics, and then carrying out edge detection on a road image by using a Canny operator; the Canny edge detection result is the original edge characteristic of the road image, and in order to extract the road, effective road elements need to be obtained from the Canny edge detection result, namely independent edge segments capable of representing the road elements; the method comprises the steps of detecting intersection points and inflection points of edges and removing the intersection points and the inflection points to realize the decomposition of edge detection results, wherein edge objects after decomposition are straight line segments or arc segments with geometric characteristics; determining the image processing range of road extraction and the direction of the road by inputting the seed points, filtering the edge section according to the connecting line direction of the buffer area and the seed points, and reserving the effective road elements for supporting the subsequent marshalling processing.
3. the method for extracting the high-resolution remote sensing road statistically matched with the structural wire harness according to claim 2, wherein the method comprises the following steps: the concrete process of the step four is as follows;
the section structure characteristics are formed by the normal distance from the element to the base line; calculating the distance between each element and the base line, wherein the method comprises the following steps: taking the normal distance from the middle point of the edge to the road base line as the section distance of the current edge object, wherein the section distance on the left side of the fitting road is a positive value and the section distance on the right side of the fitting road is a negative value according to the main direction of the fitting road;
after the edge section distance is obtained, all edge elements on the current road section are classified according to the section distance and are grouped and connected again, meanwhile, the length accumulated value of a newly grouped object is counted, and a one-dimensional line graph is generated according to the section distance and the length accumulated value, namely the section structure characteristic.
4. the method for extracting the high-resolution remote sensing road statistically matched with the structural wire harness according to claim 3, wherein the method comprises the following steps: the concrete process of the step five is as follows;
according to the verification method based on pattern matching, a road is considered to have a symmetrical structure, namely the road is formed by two side lines, and the side lines are symmetrically distributed on two sides of a road center line or a virtual center line; dividing the road into a double-peak mode and a triple-peak mode according to whether the road has a central lane line or not, and respectively defining corresponding mixed Gaussian kernel functions as shown in formulas (7) and (8);
wherein ω is road width; sigma is kernel function variance, and a span range of a road side line or a central line corresponding to a statistical peak value is identified; defining a road mode matching response function based on the kernel function, as shown in a formula (9);
Wherein d min and d max are the minimum and maximum values of the section distance, l x is the section structure characteristic value corresponding to the x position of the section distance, namely the edge accumulated length value, x i is the kernel function offset, and v i is the response value of the current position mode matching;
at the moment, the detection problem of the road middle line and the side line is converted into the solution of the section position corresponding to the maximum value obtained by v i, omega in the kernel function is an unknown parameter, the possible road width is predicted according to the distance between the section positions corresponding to the peak values in the statistical characteristics, omega in the double peak mode is the section distance between two continuous peak values, omega in the three peak mode is the section distance between the head peak value and the tail peak value in three continuous peak values;
the process of detecting the edge accumulation length peak value in the section structure characteristic comprises two steps:
(a) Assuming l (x) is a statistical feature length function, wherein x is an integer section distance value, -d min ≦ x ≦ d max, then the local peak detection condition is:
(b) The smaller peaks are filtered: when the local peak value is far smaller than the global maximum peak value or smaller than the average edge statistical length, deleting the peak value from the peak value set, as shown in formula (11);
The method comprises the steps of obtaining a road base line, calculating a model matching response function, wherein epsilon 1 is a threshold value, l avg is an average edge statistical length, finally substituting all predicted omega into a kernel function p (x | omega, theta), calculating a response value v i corresponding to each offset x i position of the model matching response function, taking a position x i corresponding to the maximum value of the response value v i as a road center line position, and taking a statistical characteristic peak position corresponding to the road width omega in the kernel function as the left and right line positions of a current road, and translating the road base line according to the obtained position information in the normal direction to obtain a final road extraction result;
The verification feedback is the last step of the road extraction method, and two verification feedback conditions are provided;
the feedback condition one is as follows: the central line of the road extraction result needs to be located in a certain buffer range of the link of the road segment seed points, and consistency indexes S (A, B) are defined, as shown in formula (12):
When S (A, B) > T S, the current road extraction result is considered as a candidate road extraction result, and T s is a road length integrity threshold value;
The second feedback condition is that after pattern matching and sorting according to response values v i, a road extraction result set which meets the first feedback condition is selected, and the result that the difference between the pattern matching response value v i and the maximum response value v max of each extraction result is within a specified threshold value is used as a candidate result;
and when the effective road extraction result cannot be obtained under the constraint of the feedback condition, returning to the base line selection stage, reselecting the road base line, and performing section structure statistics and road extraction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610272080.5A CN105956542B (en) | 2016-04-28 | 2016-04-28 | High-resolution remote sensing image road extraction method based on statistical matching of structural wire harnesses |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610272080.5A CN105956542B (en) | 2016-04-28 | 2016-04-28 | High-resolution remote sensing image road extraction method based on statistical matching of structural wire harnesses |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105956542A CN105956542A (en) | 2016-09-21 |
| CN105956542B true CN105956542B (en) | 2019-12-10 |
Family
ID=56915882
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610272080.5A Active CN105956542B (en) | 2016-04-28 | 2016-04-28 | High-resolution remote sensing image road extraction method based on statistical matching of structural wire harnesses |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105956542B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108492276B (en) * | 2018-01-29 | 2021-03-19 | 中国人民解放军战略支援部队信息工程大学 | Similarity measurement-based vector road change detection method and device |
| CN108917591A (en) * | 2018-04-27 | 2018-11-30 | 中国铁道科学研究院集团有限公司基础设施检测研究所 | Rail profile autoegistration method and device under a kind of dynamic environment |
| CN108961286B (en) * | 2018-06-27 | 2021-11-16 | 武汉大学 | Unmanned aerial vehicle image segmentation method considering three-dimensional and edge shape characteristics of building |
| CN109887030A (en) * | 2019-01-23 | 2019-06-14 | 浙江大学 | Texture-free metal parts image position and posture detection method based on the sparse template of CAD |
| CN114708237A (en) * | 2022-04-14 | 2022-07-05 | 杭州小肤科技有限公司 | Detection algorithm for hair health condition |
| CN115641512B (en) * | 2022-12-26 | 2023-04-07 | 成都国星宇航科技股份有限公司 | Satellite remote sensing image road identification method, device, equipment and medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101364259B (en) * | 2008-04-09 | 2010-06-02 | 武汉大学 | Method for extracting road various information of multi-level knowledge driven panchromatic remote sensing image |
| CN101901343B (en) * | 2010-07-20 | 2012-05-23 | 同济大学 | Remote sensing image road extracting method based on stereo constraint |
| CN103258203A (en) * | 2013-05-20 | 2013-08-21 | 武汉大学 | Method for automatically extracting road centerline of remote-sensing image |
| CN104134080A (en) * | 2014-08-01 | 2014-11-05 | 重庆大学 | Method and system for automatically detecting roadbed collapse and side slope collapse of road |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8155391B1 (en) * | 2006-05-02 | 2012-04-10 | Geoeye Solutions, Inc. | Semi-automatic extraction of linear features from image data |
-
2016
- 2016-04-28 CN CN201610272080.5A patent/CN105956542B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101364259B (en) * | 2008-04-09 | 2010-06-02 | 武汉大学 | Method for extracting road various information of multi-level knowledge driven panchromatic remote sensing image |
| CN101901343B (en) * | 2010-07-20 | 2012-05-23 | 同济大学 | Remote sensing image road extracting method based on stereo constraint |
| CN103258203A (en) * | 2013-05-20 | 2013-08-21 | 武汉大学 | Method for automatically extracting road centerline of remote-sensing image |
| CN104134080A (en) * | 2014-08-01 | 2014-11-05 | 重庆大学 | Method and system for automatically detecting roadbed collapse and side slope collapse of road |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105956542A (en) | 2016-09-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105956542B (en) | High-resolution remote sensing image road extraction method based on statistical matching of structural wire harnesses | |
| CN109271928B (en) | Road network updating method based on vector road network fusion and remote sensing image verification | |
| CN108197583B (en) | Building change detection method based on graph cut optimization and image structure characteristics | |
| CN106780524B (en) | Automatic extraction method for three-dimensional point cloud road boundary | |
| CN108062517B (en) | Vehicle-mounted laser point cloud-based method for automatically extracting boundary line of unstructured road | |
| CN111079611B (en) | Automatic extraction method for road surface and marking line thereof | |
| Lee et al. | Fusion of lidar and imagery for reliable building extraction | |
| CN111047695B (en) | Method for extracting height spatial information and contour line of urban group | |
| CN103703490A (en) | Device for generating three-dimensional feature data, method for generating three-dimensional feature data, and recording medium on which program for generating three-dimensional feature data is recorded | |
| CN110619258B (en) | Road track checking method based on high-resolution remote sensing image | |
| CN104751478A (en) | Object-oriented building change detection method based on multi-feature fusion | |
| CN102147250A (en) | Digital line graph mapping method | |
| CN101901343A (en) | Remote sensing image road extracting method based on stereo constraint | |
| CN111652241B (en) | Building contour extraction method integrating image features and densely matched point cloud features | |
| CN106096497B (en) | A kind of house vectorization method for polynary remotely-sensed data | |
| CN103310199B (en) | Based on the vehicle model recognition methods of high-definition remote sensing data | |
| CN114332291A (en) | Oblique photography model building outer contour rule extraction method | |
| CN103578105B (en) | A kind of multi-source foreign peoples method for registering images based on provincial characteristics | |
| CN115690081A (en) | Tree counting method, system, storage medium, computer equipment and terminal | |
| CN106886988B (en) | Linear target detection method and system based on unmanned aerial vehicle remote sensing | |
| CN109187548A (en) | A kind of rock cranny recognition methods | |
| Zhang et al. | Knowledge-based image analysis for 3D edge extraction and road reconstruction | |
| CN111985507A (en) | Rock three-dimensional point cloud joint trace extraction method | |
| CN114972628A (en) | Building three-dimensional extraction method based on multispectral laser radar point cloud data | |
| CN116503705A (en) | Fusion method of digital city multi-source data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |