An interactive map display method based on map image recognition

Abstract

The application discloses an interactive map display method based on map image recognition and relates to the technical field of map display. In view of the problem that paper maps have limited expression contents in the prior art, the application combines augmented reality technology with traditional paper maps, and to some extent, the defects of the two are made up. For example, once the contents of a paper map are published, the contents are fixed, and the map contents are greatly limited by factors such as a map surface, a drawing scale, a resolution and the like. The augmented reality technology is used for enhanced expression of the paper map, the map contents and other geographic information can be expanded on a mobile terminal, for example, superimposed enhanced symbols, topography, a three-dimensional model, text, pictures, animation, video and other data, virtual information is supplemented and sublimated to real and tangible maps, and the richness and comprehensiveness of map expression contents are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of map display, in particular to an interactive map display method based on map image recognition. BACKGROUND

[0002] Map service is an important measure to protect spatial rights and interests and contact real life. The previous map service mainly refers to paper maps. Traditional paper maps, as an important carrier for expressing, transmitting and researching geographic information, have been widely used in various fields such as tourism, surveying and mapping, and ecology due to their advantages of convenient reading, clear information expression, high collection value, etc. However, paper maps also have some defects such as lack of personalization, limited expression content, monotonous form of expression, one-way transmission mode, and long update cycle. SUMMARY

[0003] The purpose of the present application is to solve the problem of limited expression content of paper maps in the prior art, and to provide an interactive map display method based on map image recognition.

[0004] The technical scheme adopted by the present application to solve the above technical problem is:

[0005] An interactive map display method based on map image recognition, comprising the following steps:

[0006] Step 1: Obtain an electronic base map and a paper map image to be recognized;

[0007] Step 2: Preprocess the electronic base map and the paper map image to be recognized, respectively, and the preprocessing specifically comprises:

[0008] Step 2-1: Filter the image;

[0009] Step 2-2: Perform histogram equalization on the filtered image;

[0010] Step 2-3: Perform edge enhancement on the histogram equalized image;

[0011] Step 3: For the edge enhanced electronic base map and the paper map image to be recognized, use the SIFT algorithm to match the paper map image with the electronic base map, wherein the electronic base map is associated with display data, and the display data includes thematic vector data, image data, multimedia information, statistical analysis chart data and three-dimensional model data;

[0012] Step 4: Display the display data corresponding to the paper map image.

[0013] Further, the filtering processing is performed by a bilateral filter, the bilateral filter generates a distance template using a two-dimensional Gaussian function and generates a value domain template using a one-dimensional Gaussian function, and a generation formula of the distance template coefficient is as follows:

[0014]

[0015] wherein, is a central coordinate of the template window, is a coordinate of other coefficients of the template window, is a standard deviation of the Gaussian function;

[0016] a generation formula of the value domain template coefficient is as follows:

[0017]

[0018] wherein, represents a pixel value of the image at a point ; is a central coordinate of the template window; is a coordinate of other coefficients of the template window; is a standard deviation of the Gaussian function;

[0019] a template of the bilateral filter is represented as:

[0020] .

[0021] Further, the histogram equalization is represented as:

[0022]

[0023] wherein, is a minimum value of a cumulative distribution function, M and N respectively represent a pixel number of a length and a width of the image, L is a gray level number, and v is a pixel value in the original image.

[0024] Further, the edge enhancement is an image edge enhancement based on a Laplacian operator.

[0025] Further, the SIFT algorithm is an improved SIFT algorithm, and the improved SIFT algorithm specifically performs the following steps:

[0026] Step 1: respectively extracting feature points from a paper map image and an electronic base map, and the feature point extraction specifically includes the following steps:

[0027] Step 11: performing Canny operator edge detection on the image by using an OpenCV library;

[0028] Step 12: constructing a scale space according to the image after the edge detection;

[0029] Step 13: extreme point detection in scale space, to get local extreme points;

[0030] Step 14: screening of local extreme points, deleting unstable contrast and false extreme points, then positioning the remaining extreme points in the original image to get key points;

[0031] Step 15: first fitting of key points with quadratic function, then extreme value of the fitted function, derivative is zero, then it is a real extreme point, i.e. real key point, real key point is represented as:

[0032]

[0033] Where X is the distance relative to the key point, D is the key point value, Q is the weight set according to the position relationship of the key point and the map boundary, extreme value, extreme point offset;

[0034] Step 16: extraction of feature points of real key points, and finally determination of the direction of feature points;

[0035] The specific steps of step 16 are:

[0036] First, calculate the gradient direction of all pixel points in the field centered on the real key point, then normalize the gradient direction to 8 directions, each direction represents a 45 degree range, then calculate the number of real key points and neighborhood points in each direction domain, and generate a gradient direction histogram according to the calculation result, in the gradient direction histogram, the direction with the most number of real key points and neighborhood points in the direction domain is the main direction of the current real key point;

[0037] Step 17: generating feature point descriptor according to real key point and direction of real key point, and then getting feature point;

[0038] Step 2: matching of paper map image and feature points of electronic base map.

[0039] Further, the scale space is constructed by DOG pyramid.

[0040] Further, the specific steps of extreme point detection in step 13 are:

[0041] In two-dimensional image space, each pixel point is taken as a center point, and the 8 pixel points in the 3*3 neighborhood of the center point are compared with the center point;

[0042] In the scale space of the same group, each pixel point is taken as a center point, and the 2*9 pixel points of the center point and the two layers of images above and below the center point are compared to select the maximum value and the minimum value in the range.

[0043] Further, the gradient direction of the pixel point is represented as:

[0044]

[0045] Wherein, The position of the pixel point.

[0046] Further, the specific steps of the step 2 are:

[0047] Taking the feature point as the center, a 16*16 pixel size area window is first established, then the gradient amplitude and direction of each pixel point are calculated, then the area window is divided into 4*4 total 16 sub-regions, each sub-region has 4*4 pixel points, the histogram information of 8 gradient directions is counted in each sub-region, and finally a 4*4*8 total 128-dimensional feature vector is obtained, and then the Euclidean distance of the feature vectors of the key points in the two images is calculated to judge the similarity,

[0048] The specific steps of calculating the Euclidean distance of the feature vectors of the key points in the two images to judge the similarity are:

[0049] A feature point in the paper map image is taken out, and then the two feature points with the closest Euclidean distance to the feature point in the electronic base map are found, among the two feature points, if the nearest distance divided by the second nearest distance is less than a threshold value, then the pair of matching points is accepted, the threshold value is set to 0.7, and if the nearest distance divided by the second nearest distance is not less than the threshold value, then the pair of matching points is not accepted.

[0050] The beneficial effects of the present application are:

[0051] The present application combines the augmented reality technology with the traditional paper map, to some extent, makes up for the defects of the two, and is influenced by factors such as the size, the mapping scale, the resolution and the like, and the map content has great limitations, through the augmented reality technology, the paper map is enhanced to express, the map content and other geographic information can be expanded on the mobile terminal, such as superimposed enhanced symbols, topography, three-dimensional models, texts, pictures, animations, videos and the like, let the virtual information become the supplement and sublimation of the real perceptible map, and the richness and comprehensiveness of the map expression content are improved; in addition, the paper map lacks user interaction function, and by using the augmented reality technology, an interactive platform can be provided for the user, such as click, gesture and the like, and the user experience is improved.

[0052] The application can not only extend the information expressed by paper maps, but also enrich the forms of expression of maps. The limitations of the expression content of paper maps can be improved by the augmented reality technology, and in the digital era of rapid development of the Internet, the traditional paper map culture can be promoted and popularized more deeply. In the technical scheme of the application, the map interaction and experience form has the practicability of maps and the convenience of augmented reality technology, provides different interactive visual experiences for different users, is beneficial to expand the breadth and depth of map services, and better plays the basic role of "two supports and one improvement" of surveying and mapping work in the new period. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is a feature extraction and matching flowchart;

[0054] Figure 2 is a feature point descriptor example Figure 1 ;

[0055] Figure 3 is a feature point descriptor example Figure 2 ;

[0056] Figure 4 is a feature point matching threshold diagram. DETAILED DESCRIPTION

[0057] It should be particularly noted that the various embodiments disclosed in the application can be combined with each other without conflict.

[0058] Specific implementation one: refer to Figure 1 To specifically explain this embodiment, the interactive map display method based on map image recognition in this embodiment comprises the following steps:

[0059] Step one: obtaining an electronic base map and a paper map image to be recognized;

[0060] Step two: respectively pre-processing the electronic base map and the paper map image to be recognized, and the pre-processing specifically comprises:

[0061] Step two one: filtering the image;

[0062] Step two two: histogram equalization of the image after filtering;

[0063] Step two three: edge enhancement of the image after histogram equalization;

[0064] Step three: using SIFT algorithm to match the paper map image with the electronic base map, the electronic base map is associated with the display data, the display data includes thematic vector data, image data, multimedia information, statistical analysis chart data and three-dimensional model data.

[0065] Step four: displaying the display data corresponding to the paper map image.

[0066] In the past, the recognition prototype used in the research based on augmented reality technology is generally an image, which generally has the characteristics of clear color contrast, rich content, high definition and high quality, and as a sample, a large number of evenly distributed feature point sequences can be extracted in the training, so that the ideal image recognition goal can be achieved. However, compared with general images, map data has dense lines, low definition, and not enough clear color contrast. It is found that when the same feature point extraction method as general images is used, the extracted feature points are less and unevenly distributed, resulting in poor map image recognition effect, low recognition rate and high error rate. In view of this situation, a series of image processing methods are performed on the map data before the training of the map base map sample to highlight the main content and weaken the secondary information, so as to improve the feature point extraction and matching ability of the map data and achieve good map recognition effect.

[0067] Image filtering

[0068] Image filtering, that is, suppressing the noise existing in the map data while trying to preserve the details of the map. This noise is different from the salt and pepper noise (black and white intensity value), impulse noise (pure white intensity value or black intensity value) and Gaussian noise (randomly distributed noise of any amplitude) in general images. The noise targeted by the application refers to the information that affects the extraction of feature points in map data, such as low-level, small-range, and complex boundary data with trivial lines, such as community, school, and hospital boundaries, and high-repetition place name and address data. When the map data is scaled to the size of an ordinary image, these noise data have low definition but redundant content, which can cause low contrast and unobvious features in the map image.

[0069] This noise belongs to Gaussian distribution noise, that is, each point has noise, and the amplitude of the noise is randomly distributed. This noise is suitable for filtering map data using a high-speed filter. The Gaussian filter is a linear filter that can effectively suppress noise, but it will smooth the image to some extent, which is somewhat inconsistent with our goal of improving contrast. Therefore, we use a bilateral filter, which is a nonlinear filter that can achieve the effect of preserving edges and reducing noise smoothing.

[0070] The bilateral filter uses a two-dimensional Gaussian function to generate a distance template and a one-dimensional Gaussian function to generate a value domain template. The generation formula of the distance template coefficient is as follows:

[0071]

[0072] wherein, is the center coordinate of the template window; is the coordinate of other coefficients of the template window; is the standard deviation of the Gaussian function. The filter template generated by using the formula and the template used by the Gaussian filter are not different.

[0073] The generation formula of the value domain template coefficient is as follows:

[0074]

[0075] wherein, the function represents an image to be processed, represents the pixel value of the image at point ; is the center coordinate of the template window; is the coordinate of other coefficients of the template window; is the standard deviation of the Gaussian function.

[0076] The multiplication of the above two templates obtains the template of the bilateral filter:

[0077]

[0078] The application uses the function bilateralFilter in the OpenCV class library to implement the corresponding bilateral filter, that is:

[0079] public static void bilateralFilter(Mat src, Mat dst, int d, doublesigmaColor, double sigmaSpace), wherein src and dst are input image and output image respectively, d represents the diameter range of each pixel neighborhood in the filtering process, sigmaColor represents the sigma value of the color space filter, the greater the value of the parameter, the wider the color of the pixel neighborhood that is smoothed to the same value domain, and sigmaSpace represents the sigma value of the filter in the coordinate space. If the value is large, the pixels of similar color in a larger region will be affected.

[0080] Histogram equalization

[0081] The final purpose of the map image processing of the present application is to enhance the contrast of the image. There are two methods of enhancing the contrast in the digital image processing, which are direct enhancement and indirect enhancement. The histogram equalization is a common method of indirectly enhancing the contrast, which is a method of adjusting the histogram of the image to adjust the contrast of the image. The histogram equalization is to stretch the image nonlinearly and redistribute the pixel values of the image so that the number of pixels in a certain gray scale range is approximately the same.

[0082] The histogram equalization can reduce the gray levels of the image, thereby blurring some details. In addition, for some histograms with high peaks, the processed image can be greatly enhanced. Especially for the map image, the color of the internal lines is similar, and the contrast is close. Through the histogram equalization, the gray values can be better distributed on the histogram, and the local contrast can be enhanced without affecting the overall contrast. The histogram equalization is widely used in X-ray images, and the bone structure can be better displayed after the processing. The map data of the present application is similar to the X-ray image.

[0083] The calculation formula of the histogram equalization is as follows, is the minimum value of the cumulative distribution function, M and N represent the number of pixels of the length and width of the image respectively, L is the number of gray levels (for example, if the image is 8-bit depth, the number of gray levels is 2^8=256, which is the most common number of gray levels), and v is the pixel value of the original image:

[0084]

[0085] The present application uses the function equalizeHist in the OpenCV library to realize the corresponding histogram equalization, that is,

[0086] public void equalizeHist (InputArray src, OutputArray dst), wherein src and dst are the input image and the output image respectively.

[0087] Edge enhancement

[0088] The recognition of the map data of the present application is largely the recognition of the map boundary data, such as the provincial boundary and the county boundary. These boundaries are often located at the edge of the image in the map, and these edges are at the image information mutation point in the image. However, these boundary data are generally smooth arc-shaped geometries. Therefore, in order to improve the accuracy of edge detection, the present application needs to enhance and sharpen the image edges.

[0089] Image edge enhancement is a processing method that highlights certain information in an image while weakening other information. It is a technique that highlights edges in an image (or video) where the brightness values (or color tones) of adjacent pixels (or regions) differ greatly. The enhanced image can more clearly show the boundaries of different types of objects or phenomena, or the traces of linear images, facilitating the identification of different types of objects and the delineation of their distribution range.

[0090] The present application uses the OpenCV class library to implement image edge enhancement based on the Laplacian operator.

[0091] Map local processing

[0092] As a carrier of geographic information data, maps have a rich form of information expression. The points, lines, and surfaces presented by maps, as well as their derived symbols, shapes, and colors, all contain rich information. This study refers to all of these as map information. Map information is presented in various ways in maps. Data information such as government agencies, enterprises, and public transportation routes in urban built-up areas, as well as information such as vegetation species distribution, crop rotation and fallow, and river seasonal trends in forest, grassland, and farmland regions, all belong to the category of map information.

[0093] Map information is not evenly distributed. In densely populated urban areas, map information is concentrated and abundant, while in natural regions far from human activity, map information is sparse and scattered. Therefore, given the uneven distribution of map information, how to deeply mine and extract map information is a key and difficult point of this study.

[0094] For the richness of the information covered by map data, this study extracts and analyzes this information by setting a certain map information threshold, discarding map information greater than or less than the threshold, and ultimately obtaining map information with a uniform distribution of data volume.

[0095] For example, a regular prefecture administrative district information map, because its scope covers all the terrain data within the prefecture jurisdiction, the information covered by its map data is not evenly distributed,

[0096] Specifically, the map information of urban and township built-up areas is relatively abundant and dense, while the map information of mountainous, farmland, and river areas is relatively simple and sparse. This study discards some simple mountainous and forested areas and retains map data from areas with relatively sharp lines and abundant information based on the characteristics of this map data.

[0097] Feature extraction and matching method

[0098] The application uses a feature point extraction and matching method improved based on SIFT algorithm. The Scale Invariant Feature Transform (SIFT) algorithm is proposed by Professor David G Lowe of Columbia University in 1999 based on the feature detection method based on invariant technology. SIFT is a scale space-based image local feature description operator that remains invariant to image scaling, rotation, and even affine transformation. Professor Lowe improved the SIFT algorithm more systematically in 2004.

[0099] The SIFT algorithm extracts image local features by finding extreme points in different scale spaces and extracting the size, position, direction, scale, and other information of the extreme points, and describing the feature points using the information. The SIFT operator remains invariant to rotation, scale, and brightness changes, and has a certain degree of stability to changes in viewing angle and noise. The feature points searched by the SIFT algorithm are mainly some prominent pigment points that remain stable and do not change due to factors such as light, affine change, and noise, such as corner points, edge points, dark points in bright areas, and bright points in dark areas. For fast and accurate extraction and matching of massive feature data in map data, very good results can be obtained. In addition, for the color simplicity and unobvious features of map data compared to general images, the SIFT operator can exhibit its multi-quantity, i.e., a small number of objects can generate a large number of SIFT feature vectors.

[0100] The SIFT algorithm is improved for the characteristics of map data. The interactive map system referred to in the application mainly identifies and matches county boundary data in map data, such as identifying several county boundaries of Hegang City and displaying corresponding Hegang City thematic data. For map data such as place names and addresses, the same county boundary data in different maps is considered to belong to the same Hegang City map and can be identified. Therefore, in the feature point extraction of this study, we first extract the outer county boundary using the improved edge detection method, then set the corresponding weight according to the position relationship between the key points searched by the SIFT algorithm and the county boundary, calculate the total value of each key point based on the weight and the operator, and continue to extract feature points from key points with higher values, and ignore key points with lower values. The specific algorithm flow is shown in Figure 1 .

[0101] Edge detection based on Canny operator

[0102] The target of edge detection in the present application is to express more peripheral boundary data in the map while keeping the original map attributes, significantly reducing the data size and expression content of the map. The Canny edge detection operator is a multi-stage edge detection algorithm. On the basis of first-order differentiation, it increases the non-maximum suppression and double threshold detection. Its target is to find the optimal edge detection algorithm, i.e. to identify as many actual edges in the image as possible. The detected edge points should correspond to the actual edge points one by one. The steps of the algorithm mainly include: Gaussian filter smoothing processing, calculating the gradient intensity and angle direction of the image, applying the non-maximum suppression method to eliminate false detection, and using the double threshold method to select potential boundaries.

[0103] The present application also uses the OpenCV class library to perform Canny operator edge detection:

[0104] public void Canny (InputArray image, OutputArray edges, double threshold1, double threshold2, int apertureSize = 3, bool L2gradient = false);

[0105] Wherein image and edges are input image and output image respectively, threshold1 represents the first hysteresis threshold, threshold2 represents the second hysteresis threshold, apertureSize represents the aperture size of the applied Sobel operator, and L2gradient represents an identifier for calculating the image gradient amplitude.

[0106] SIFT algorithm scale space construction

[0107] The scale space of the image solves the problem of how to describe the image at all scales. Scale space construction is to perform scale transformation on the original base image to obtain the image representation sequence in the multi-scale space, extract the main contour in the scale space, and take it as a feature vector to realize key point extraction. The basis of scale space construction is the DOG pyramid (difference of Gaussians pyramid). The image pyramid is a structure for explaining the image in multiple resolutions. The Gaussian pyramid is a collection of images blurred by different Gaussian kernels under certain size conditions after the original image is filtered by a Gaussian low-pass filter. The DOG pyramid is constructed on the basis of the Gaussian pyramid. Through experiments, different blur effect images of the original image at each scale can be obtained, which can simulate the formation process of the target image on the retina when the human eye is close to the target from near to far.

[0108] Extreme point detection in scale space

[0109] After the scale space is constructed, the extreme value point detection can be performed in the space. We compare each pixel point with all the adjacent points in the range, and select the maximum and minimum values in the range. In the two-dimensional image space, the center point is compared with the eight points in its 3*3 neighborhood. In the same group of scale space, the center point is compared with the 2*9 points in the upper and lower adjacent layers of images, to ensure that the detected key points are local extreme points in the scale space and the two-dimensional image space.

[0110] Positioning key points

[0111] Since the DOG value is sensitive to noise and edges, many extreme points can be detected in the above scale space. We need to screen them out and discard the unstable and false extreme points. In addition, we need to determine the exact position of these detected local extreme points in the original image, i.e., positioning.

[0112] Screening key points according to edge weight

[0113] The large number of local extreme points obtained in the previous step need to be further screened to remove unstable and false extreme points. We fit a quadratic function to the key points, and then take the extreme value of the function to perform Taylor's quadratic expansion on each key point function D(x, y):

[0114]

[0115] The key points obtained by the above method are extreme points in discrete space, not real extreme points. We need to obtain real extreme points in continuous space through two conversions. First, we fit a quadratic function to the key points, i.e., perform Taylor's quadratic expansion on each key point function . This expansion simulates a quadratic curve around the key point. Then we take the extreme value of the fitted function, and the derivative is zero, which is the extreme value, to obtain the extreme value:

[0116]

[0117] where X is the distance relative to the key point, and D is the key point value. Q is the weight set according to the position relationship between the key point and the map boundary. The greater the weight, the greater the key point extreme value. When the extreme value <0.03, we consider that the key point extreme value does not meet the requirements, and it is a false extreme point, which may be an unstable key point or a noise point. It will be eliminated. The screened key points and the accurate positions of the key points obtained by this process are the sign of the partial derivative, the offset of the extreme point.

[0118] Positioning image feature points and determining feature direction

[0119] After the above steps we get a more stable extreme point, the next need to get the key point direction information, this key point direction information need to the image angle and rotation has not invariable deformation, to ensure the accuracy and efficiency of feature point matching. Direction information is achieved by calculating the gradient of each feature point, the gradient amplitude of any feature point is shown as follows:

[0120]

[0121] Gradient direction is:

[0122]

[0123] Wherein the scale of L is the scale of each feature point.

[0124] Determination of feature point direction is not simply according to the gradient direction of key points, but according to the histogram formula, first calculate the gradient direction of all points in the field centering on the key point, then normalize these gradient directions to 36 directions, each direction represents a 10 degree range, then calculate the number of key points falling in each direction domain respectively, thus generating a gradient direction histogram, the direction with the maximum y value in the gradient direction histogram is the main direction of the current key point.

[0125] Generating feature point descriptor

[0126] After extracting the feature points and determining the direction, we describe the key points, that is, to define the key points in a mathematical form, the feature point descriptor contains the key points and the neighborhood points. For example, we take 8*8 64 small windows with the key point as the center, each small window represents a pixel in the scale space of the key point and the neighborhood points, the gradient amplitude and direction of each small window are calculated according to the above formula, the arrow direction is the gradient direction and the arrow length is the gradient amplitude, then the Gaussian weighted calculation is used. Then divide the 8*8 grid into 4 4*4 sub-regions, calculate the gradient direction histogram of 8 directions in each sub-region to form a seed point, there are 4 seed points in the test area, each seed point has 8 direction vector information. This 2*2*8 latitude vector is the descriptor of the center feature point, as shown in Figure 2 and Figure 3 .

[0127] Matching feature points

[0128] The experimental results of Professor Lowed show that the descriptor of 4*4*8, i.e. 128-dimensional vector, is the best for describing each key point. That is, a 16*16 pixel size area window is established with the key point as the center, the gradient amplitude and direction of each pixel block are calculated, then the pixel block is divided into 4*4, i.e. 16 sub-regions, each sub-region has 4*4 pixels, the histogram information of 8 gradient directions is counted in each sub-region, and finally a 4*4*8, i.e. 128-dimensional feature vector is obtained.

[0129] The premise of feature point matching is to obtain the feature point descriptor of the original image and the matching image, and then the Euclidean distance of the key point feature vectors in the two images is calculated to determine the similarity, such as taking a feature point in image 1, then finding the two feature points with the closest Euclidean distance in image 2, and in the two key points, if the nearest distance divided by the second nearest distance is less than the threshold value, then the pair of matching points is accepted. Professor Lowe recommends a threshold value of 0.8. In this study, a large number of pictures are matched, and the results are shown in the following figure. The threshold value is set to 0.5-0.7, which is the best. When the threshold value is low, the matching rate is low, and when the threshold value is high, the matching rate is high, but the accuracy is low, and it is easy to identify errors, so the threshold value used in this application is 0.7. The feature point matching threshold value is shown in Figure 4 .

[0130] Elimination of false matching

[0131] In the feature point matching process, due to reasons such as light and shielding, false matching may occur. While optimizing the method of Euclidean distance threshold value of feature vector, RANSAC (Random Sample Consensus method) is also used to eliminate false matching points. The RANSAC algorithm is a method of obtaining the optimal parameter model by iterative training from a set of data containing false matching points. The points that do not conform to the optimal parameter model are false matching points. In this study, the RANSAC algorithm is realized by using the OpenCV library. First, the feature points are matched according to the feature vector, then the findFundamentalMat method is used to obtain the vector basis matrix RansacStatus, which describes the corresponding relationship between the pixel points in the three-dimensional scene. According to the result of the basis matrix, the false matching points are deleted, i.e. the points with RansacStatus[0]=0.

[0132] It should be noted that the specific embodiments are only an explanation and description of the technical solutions of the present application, and cannot limit the scope of protection. Any partial change made according to the claims and description of the present application shall still fall within the protection scope of the present application.

Claims

1. An interactive map presentation method based on map image recognition, characterized by The method comprises the following steps: Step 1: obtaining an electronic base map and a paper map image to be recognized; Step 2: respectively pre-processing the electronic base map and the paper map image to be recognized, and the pre-processing specifically comprises: Step 211: performing filter processing on the image; Step 212: performing histogram equalization on the image after the filter processing; Step 213: performing edge enhancement on the image after the histogram equalization; Step 3: matching the paper map image with the electronic base map by using a SIFT algorithm, the electronic base map being associated with display data, and the display data comprising thematic vector data, image data, multimedia information, statistical analysis chart data and three-dimensional model data; Step 4: displaying the display data corresponding to the paper map image.

2. The map image recognition-based interactive map presentation method of claim 1, wherein The filter processing is performed by using a bilateral filter, the bilateral filter using a two-dimensional Gaussian function to generate a distance template and using a one-dimensional Gaussian function to generate a value domain template, the generation formula of the distance template coefficient being as follows: wherein, is the center coordinate of the template window, is the coordinate of the other coefficients of the template window, is the standard deviation of the Gaussian function; The generation formula of the value domain template coefficient is as follows: wherein, represents a pixel value of the image at point ; is a center coordinate of the template window; is a coordinate of other coefficients of the template window; is a standard deviation of the Gaussian function; The template of the bilateral filter is represented as follows: 。 3. The map image recognition-based interactive map presentation method of claim 2, wherein The histogram equalization is represented as follows: where, is the minimum value of the cumulative distribution function, M and N represent the number of pixels in the length and width of the image, respectively, L is the number of gray levels, and v is the pixel value in the original image.

4. The map image recognition-based interactive map presentation method of claim 3, wherein The edge enhancement is edge enhancement of the image based on a Laplacian operator.

5. The map image recognition-based interactive map presentation method of claim 4, wherein The SIFT algorithm is an improved SIFT algorithm, and the improved SIFT algorithm specifically performs the following steps: Step 1: respectively extracting feature points from the paper map image and the electronic base map, and the specific steps of the feature point extraction are as follows: Step 11: performing Canny operator edge detection on the image by using an OpenCV library; Step 12: constructing a scale space according to the image after the edge detection; Step 13: detecting extreme points in the scale space to obtain local extreme points; Step 14: screening the local extreme points to delete extreme points with unstable contrast and errors, and then positioning the remaining extreme points in the original image to obtain key points; Step 15: first fitting the key points with a quadratic function, then finding extreme values of the fitted function, and the derivative being zero is a real extreme point, i.e., a real key point, and the real key point is represented as follows: Wherein, X is the distance relative to the key point, D is the key point value, Q is the weight set according to the position relationship of the key point and the map boundary, is the extreme value, is the offset of the extreme point; Step 16: extracting feature points of the real key point, and finally determining the direction of the feature point; The specific steps of step 16 are as follows: First, calculating the gradient direction of all pixel points in the field centered on the real key point, then normalizing the gradient direction into 8 directions, each direction representing a 45-degree range, then respectively calculating the number of real key points and neighborhood points falling in each direction domain, and generating a gradient direction histogram according to the calculation result, in the gradient direction histogram, the direction with the largest number of real key points and neighborhood points in the direction domain is the main direction of the current real key point; Step 17: generating a feature point descriptor according to the real key point and the direction of the real key point, and then obtaining the feature point; Step 2: matching the feature points of the paper map image with the electronic base map.

6. The map image recognition-based interactive map presentation method of claim 5, wherein The scale space is constructed by using a DOG pyramid.

7. The map image recognition-based interactive map presentation method of claim 6, wherein The specific steps of the extreme point detection in step 13 are as follows: In the two-dimensional image space, each pixel point is taken as a center point, 8 pixel points in the 3*3 neighborhood of the center point are compared, and the maximum and minimum values in the range are selected; In the scale space within the same group, each pixel point is taken as a center point, 2*9 pixel points of the two adjacent image layers above and below the center point are compared, and the maximum and minimum values in the range are selected.

8. The map image recognition-based interactive map presentation method of claim 5, wherein The gradient direction of the pixel point is represented as: wherein, is the position of the pixel point.

9. The map image recognition based interactive map presentation method of claim 8, wherein The specific steps of step 2 are: Taking the feature point as the center, a 16*16 pixel size region window is first established, then the gradient amplitude and direction of each pixel point are calculated, then the region window is divided into 4*4 sub-regions, each sub-region has 4*4 pixel points, the histogram information of 8 gradient directions is counted in each sub-region, finally a 4*4*8 total of 128-dimensional feature vector is obtained, and then the Euclidean distance of the feature vectors of the key points in the two images is calculated to judge the similarity, The specific steps of calculating the Euclidean distance of the feature vectors of the key points in the two images to judge the similarity are: A feature point in the paper map image is taken out, then the two feature points with the closest Euclidean distance to the feature point in the electronic base map are found, in the two feature points, if the nearest distance divided by the second nearest distance is less than the threshold value, the pair of matching points is accepted, the threshold value is set to 0.7, if the nearest distance divided by the second nearest distance is not less than the threshold value, the pair of matching points is not accepted.

Images