Electronic map intelligent construction and optimization method based on artificial intelligence

Abstract

The application discloses an intelligent construction and optimization method of an electronic map based on artificial intelligence, a navigation terminal collects user navigation path planning data, and blind areas in the navigation path are recognized according to a regional GPS error distribution map; if there are blind areas in the navigation path, feature markers corresponding to the blind areas are imported into a user electronic map from a feature marker information library, and secondary construction of the electronic map is completed; navigation is performed based on the secondary constructed electronic map, when the user enters a blind area, a camera device of the navigation terminal collects road scene images in real time, recognizes markers of the blind area, and matches the recognized markers with feature markers in the user electronic map to determine an estimated position of the user; whether the user deviates from the navigation path is judged according to the estimated position of the user, and path updating or error data recording is correspondingly performed; the application effectively solves the navigation precision problem of a weak GPS signal area, and improves the reliability of the electronic map in a complex environment.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electronic maps, and particularly relates to an intelligent construction and optimization method of an electronic map based on artificial intelligence. BACKGROUND

[0002] In GPS signal blocked areas such as high-rise dense areas, tunnels, underground parking lots and the like, the traditional GPS navigation system often has problems such as inaccurate positioning and signal loss, which seriously affects the user navigation experience. In the prior art, although auxiliary means such as an inertial navigation system (INS) and a Bluetooth beacon are used, such methods are either high in cost or insufficient in accuracy.

[0003] In recent years, computer vision technology has gradually attracted attention in the field of navigation. Some research attempts to assist positioning by recognizing visual features in the environment, but these methods often require the pre-construction of a complex three-dimensional scene model, which is large in calculation amount, poor in real-time performance, and limited in adaptability to environmental changes. In addition, most existing visual aided navigation systems adopt an independent working mode, lack effective data sharing and collaborative optimization mechanisms, and are difficult to form a large-scale precise positioning service. Furthermore, the existing navigation systems generally lack fine evaluation and prediction ability of GPS signal quality, and cannot identify potential positioning risk areas in advance, resulting in that users can only respond passively after encountering positioning problems. In a complex road network environment, such a lag response may cause the user to miss a key intersection or drive into the wrong road, increasing travel time and safety hazards. These problems seriously limit the reliability and user experience of the electronic map navigation system in complex environments. SUMMARY

[0004] The purpose of the present application is to provide an intelligent construction and optimization method of an electronic map based on artificial intelligence, which can predict GPS blind areas in a navigation path in advance by constructing a regional GPS error distribution map, and conduct secondary construction of an electronic map accordingly, thereby solving the problem that a lag response may cause the user to miss a key intersection or drive into the wrong road.

[0005] The specific technical solution is as follows: an intelligent construction and optimization method of an electronic map based on artificial intelligence, which comprises the following steps:

[0006] Step S1: a navigation terminal collects user navigation path planning data, and identifies blind areas in the navigation path according to a regional GPS error distribution map.

[0007] Step S2: if there is a blind area in the navigation path, feature markers corresponding to the blind area are imported from a feature marker information library into a user electronic map, and secondary construction of the electronic map is completed.

[0008] The feature markers at least include traffic signboards, road markings and building features.

[0009] Step S3, navigation based on the secondary constructed electronic map, when the user enters the blind area, the road scene image in the user driving process is collected in real time through the camera device of the navigation terminal, the markers of the blind area are recognized, and the recognized markers are matched with the feature markers in the user electronic map to determine the estimated position of the user.

[0010] Step S4, judging whether the user deviates from the navigation path according to the estimated position of the user;

[0011] When it is confirmed that the user does not deviate from the navigation path, the GPS positioning error data of the blind area is recorded and uploaded to the server to update the regional GPS error distribution map.

[0012] When it is confirmed that the user deviates from the navigation path, a new navigation path is recalculated based on the currently recognized marker position and returned to step S1.

[0013] Further, the method for identifying the blind area in the navigation path according to the regional GPS error distribution map specifically comprises:

[0014] The GPS positioning accuracy index P of each point in the navigation path is obtained according to the regional GPS error distribution map, and when the GPS positioning accuracy index P of a point in the navigation path is greater than a preset threshold P th , it is judged whether the point is a blind area in the navigation path, wherein P th is 8 meters.

[0015] The judgment method of the blind area specifically comprises: taking a point with a GPS signal quality lower than a predetermined threshold as the center, and marking a region with non-unique navigation paths within a range of 100 meters before and after the center as a blind area; the existence of non-unique navigation paths includes the existence of: intersections, interchange transition roads, and entrances and exits of main and auxiliary roads.

[0016] The GPS positioning accuracy index P is calculated by a real-time updating method:

[0017] Wherein, n is the GPS sampling number of a point in the navigation path, the latest 5 to 20 navigation positioning data of other users at the point are taken, (x i ,y i ) is the i-th GPS measurement coordinate, (x j ,y i ) is the reference coordinate, and the middle line coordinate of the road in the navigation path is taken.

[0018] Further, the marker information in the feature marker information library is collected in a crowd-sourcing manner, and when a marker is identified by not less than three different users and the position deviation is less than 5 meters, the marker information is confirmed to be valid and stored in the feature marker information library.

[0019] Each marker in the feature marker information library comprises the following entry information: marker type, marker GPS coordinate, marker image feature vector, and relative position of the marker in the road.

[0020] Further, the identified marker is matched with the feature marker in the user electronic map, and a feature point matching algorithm is used:

[0021] SIFT feature points are extracted from the image to be matched; and the cosine similarity between the feature point descriptors is calculated:

[0022] Wherein, A and B are the descriptor vectors of the identified marker picture and the feature marker picture respectively; when the similarity S(A, B) is greater than a preset threshold S th , it is determined that the matching is successful, wherein S th is 0.75.

[0023] Further, the method for judging whether the user deviates from the preset navigation path comprises:

[0024] Step S401, a probability graph model G of the marker position and the navigation path is established: G=(V, E, P); wherein V is a set of marker nodes, E is a set of connecting edges between nodes, and P is a node transition probability matrix.

[0025] Step S402, the current position of the user is estimated by a particle filter algorithm, the number of particles is 200 to 500, and the particle weight update formula is: Wherein, is the weight of the i-th particle at time t, is an observation likelihood function.

[0026] Step S403, the deviation distance D of the estimated position from the preset path is calculated:

[0027] Wherein, (x est ,y est ) is the estimated position coordinate of the user, (x j ,y j ) is the reference coordinate, and the middle line coordinate of the road in the navigation path is taken.

[0028] Step S405, when the deviation distance D is greater than a preset threshold D thWhen the distance D between the user and the i-th landmark is greater than 15 meters, it is determined that the user deviates from the preset navigation path. th The value of D is 15 meters.

[0029] Further, the method for determining the estimated position coordinates (x est ,y est ) of the user is as follows:

[0030] A user position estimation model is established based on the identified multiple landmarks: wherein P(x, y) is the probability of the user at coordinates (x, y), P i (x, y) is the probability of the user position estimated based on the i-th landmark, and m is the number of detected visible landmarks.

[0031] w i is the weight of the i-th landmark, and the weight calculation formula is: w i = V i · D i , V i is the visibility score of the landmark, with a value range of 0 to 1, D i is the distance influence factor of the landmark, and the calculation formula is: D i = exp(-d i / d0), d i is the estimated distance of the user to the landmark, and d0 is the characteristic distance decay coefficient, with a value of 100 meters.

[0032] An observation equation is established for each landmark: z i = h i (x, y) + v i ; wherein z i is the observation value of the landmark z i in the camera image, h i (x, y) is the theoretical observation value of the landmark i when the user is at coordinates (x, y), and v i is the observation noise, which is subject to Gaussian distribution σ i is related to the clarity and size of the landmark in the image.

[0033] Then the solving equation of the user position is:

[0034] Further, the calculation formula of the visibility score V i of the landmark is: V i = 0.4C i + 0.6S i ; wherein C i is the clarity score of the landmark, with a value range of 0 to 1, which is related to the contrast and blurring degree of the landmark in the image; S iThe size of the marker is scored, with a value ranging from 0 to 1, and is proportional to the pixel area occupied by the marker in the image.

[0035] Furthermore, when the number of detected visible markers m is greater than 3, 3 markers are randomly selected to calculate the initial position estimate.

[0036] Calculate the observation residuals r of other markers j =|z j -h j (x est ,y est )|;When the residual r j Less than the threshold r th When the threshold r is reached, the marker is added to the interior point set. th The value is 5% of the image width.

[0037] Repeat the above process 10 times, select the set of inliers with the largest number of inliers to recalculate the user's location, and use it as the user's final location estimate.

[0038] Furthermore, the image acquisition frequency f is dynamically adjusted based on vehicle speed and environmental complexity: f = f base ·(1+k1·v+k2·H); where, f base The base acquisition frequency is set to 0.5Hz, v is the vehicle speed in meters per second, k1 is the speed adjustment coefficient with a value of 0.02s / m, H is the environmental complexity, and k2 is the environmental complexity adjustment coefficient with a value of 0.5. p i Let K represent the proportion of the i-th type of region after image segmentation, and K be the number of segmentation categories.

[0039] Compared with the prior art, the beneficial effects of this invention are:

[0040] This invention constructs a regional GPS error distribution map, enabling early prediction of GPS blind spots in navigation paths and targeted secondary reconstruction of electronic maps. It collects and verifies feature markers through crowdsourcing, establishing a rich and reliable marker information database. Visual recognition technology is used to match markers in blind areas in real time, combined with a weighted estimation model to accurately calculate the user's location, effectively solving the positioning problem in environments with weak or unstable GPS signals. The system also implements closed-loop optimization, continuously updating the GPS error distribution map and marker information based on user navigation records, ensuring that navigation accuracy continuously improves with the increase in the number of users. This method requires no additional hardware and can be implemented using only existing smart terminals, making it highly practical and worthy of widespread application. Attached Figure Description

[0041] Figure 1 This is a flowchart of the AI-based intelligent construction and optimization method for electronic maps according to the present invention. DETAILED DESCRIPTION

[0042] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application are described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0043] As shown in Figure 1 , the present application is an intelligent construction and optimization method of an electronic map based on artificial intelligence, which comprises the following steps:

[0044] Step S1, a navigation terminal collects navigation path planning data of a user, and identifies a blind area in the navigation path according to a regional GPS error distribution map.

[0045] The method for identifying a blind area in the navigation path according to a regional GPS error distribution map is specifically as follows:

[0046] The GPS positioning accuracy index P of each point in the navigation path is obtained according to the regional GPS error distribution map, and when the GPS positioning accuracy index P of a certain point in the navigation path is greater than a preset threshold P th , it is determined whether the point is a blind area in the navigation path, wherein P th is 8 meters; in urban areas with high buildings, under bridges, in tunnels and in mountainous areas and gorges, the value of P often exceeds the preset threshold.

[0047] The navigation terminal collects navigation path planning data of a user, and this process is automatically performed after the user starts the navigation application and inputs the destination. For example, when the user drives from A to a certain shopping mall in B, the system will generate an optimal path based on road network data. At the same time, the system will query the regional GPS error distribution map database, which stores the GPS signal quality information of each region. On this path, the system may find that when passing under the East Third Ring Road interchange, the GPS positioning accuracy index P value reaches 12 meters, which is much greater than the threshold of 8 meters, and therefore marks this area as a blind area.

[0048] The method for determining a blind area is specifically as follows: taking a point with a GPS signal quality lower than a predetermined threshold as the center, and marking an area within a range of 100 meters before and after the center where the navigation path is not unique as a blind area; the existence of a navigation path not unique includes the existence of: an intersection, an interchange transition road, and a main and auxiliary road entrance. For example, when a certain intersection in a high building area is identified, the GPS signal quality P value is 9 meters, and within a range of 100 meters before and after the intersection, there are two optional driving routes (main road and auxiliary road), and the system will mark this area as a blind area.

[0049] The GPS positioning accuracy indicator P is calculated using a real-time updating method: where n is the number of GPS sampling times at a point on the navigation path, the latest 5-20 navigation positioning data of other users at the point are taken, (x i ,y i ) is the coordinate of the i-th GPS measurement, and (x j ,y j ) is the reference coordinate, which is the coordinate of the middle line of the road in the navigation path.

[0050] The calculation of the GPS positioning accuracy indicator P is based on historical navigation data. For example, in the aforementioned interchange blind area, the system extracts the GPS positioning data of 15 different users passing through this location in the last time. If the 15 measured GPS coordinate points deviate from the road centerline by distances of 10 meters, 9 meters, 12 meters, etc., the calculated P value will reach about 10.5 meters, which exceeds the threshold of 8 meters.

[0051] Step S2, if there is a blind area in the navigation path, the feature marker corresponding to the blind area is imported from the feature marker information library to the user electronic map, and the secondary construction of the electronic map is completed.

[0052] The feature marker at least includes traffic signs, pavement markings, and building features; the marker information in the feature marker information library is collected in a crowd-sourcing manner, and the marker information identified by a plurality of users in the same area is collected. When a marker is identified by no less than 3 different users and the position deviation is less than 5 meters, the marker information is confirmed to be valid and stored in the feature marker information library.

[0053] For example, in the above-mentioned intersection blind area, the system will download information including traffic signs indicating directions, pedestrian crosswalk lines, road name plates, and surrounding landmark buildings (such as the appearance of special stores). These information will be temporarily loaded into the user's electronic map to form a local augmented reality auxiliary navigation layer to provide a reference benchmark for subsequent visual auxiliary positioning. For another example, when a newly installed road sign is identified and uploaded by navigation equipment of at least 3 different users, and the position data uploaded by these users deviates by no more than 5 meters, the system will add the road sign as a valid marker to the database.

[0054] Each marker in the feature marker information library contains the following entry information: marker type, marker GPS coordinate, marker image feature vector, and relative position of the marker in the road.

[0055] Step S3, navigation based on the secondary constructed electronic map, when the user enters the blind area, through the camera of the navigation terminal, real-time collection of road scene images in the user driving process, recognition of the markers of the blind area, and matching of the recognized markers with the feature markers in the user electronic map to determine the estimated position of the user.

[0056] When the user drives into the blind area, the navigation system will automatically start the camera to collect the scene. For example, when the user's vehicle drives under the aforementioned overpass, the system will take pictures of the road scene in front at a frequency of 0.8 times per second, and detect whether there are pre-downloaded feature markers in the image. When the system recognizes the lane line markers and exit signs under the overpass, it will extract the SIFT feature points of these markers and match them with the reference images in the feature marker library. Through the matching result, the accurate position of the user is calculated, so that accurate navigation guidance can still be provided in the case of weak GPS signal.

[0057] Matching the recognized markers with the feature markers in the user electronic map uses a feature point matching algorithm:

[0058] Extracting SIFT feature points from the image to be matched; calculating the cosine similarity between the feature point descriptors:

[0059] Where A and B are the descriptor vectors of the recognized marker picture and the feature marker picture respectively; when the similarity S(A, B) is greater than the preset threshold S th , it is determined that the matching is successful, where S th takes a value of 0.75. For example: the system extracts SIFT feature points from the real-time collected intersection scene images to obtain descriptor vector A; then calculates the cosine similarity with the descriptor vector B of the speed limit sign reference image in the feature marker library in this area; when the calculation result S(A, B) reaches 0.83, exceeding the preset threshold of 0.75, the system confirms that the speed limit sign is successfully recognized, and it is used for subsequent position estimation calculation.

[0060] The method for determining the estimated position coordinates (x est ,y est ) of the user is:

[0061] Based on the recognized multiple markers, a user position estimation model is established: Where P(x, y) is the probability of the user at coordinates (x, y), P i (x, y) is the probability of the user's position estimated based on the i-th marker, and m is the number of detected visible markers.

[0062] w i is the weight of the i-th marker, and the weight calculation formula is: wi = V i · D i , V i is the visibility score of the landmark, ranging from 0 to 1, D i is the distance impact factor of the landmark, calculated as: D i = exp(-d i / d0), d i is the estimated distance from the user to the landmark, and d0 is the characteristic distance decay coefficient, taking the value of 100 meters.

[0063] The visibility score of a landmark directly affects its weight in position estimation. Taking a speed limit sign recognized in a navigation as an example, the sign occupies about 4% of the pixel area in the image (size score S = 0.65), the image has good clarity but slight light reflection (clarity score C = 0.7), according to the formula V = 0.4 x 0.7 + 0.6 x 0.65 = 0.67, the system gives the landmark a visibility score of 0.67. At the same time, since the landmark is about 30 meters away from the user, the calculated distance impact factor D = exp(-30 / 100) = 0.74, and finally the weight of the landmark in position estimation is 0.67 x 0.74 = 0.50.

[0064] An observation equation is established for each landmark: z i = h i (x, y) + v i ; where z i is the observed value of landmark z i in the camera image, h i (x, y) is the theoretical observation value of landmark i when the user is at coordinates (x, y), and v i is the observation noise, which follows a Gaussian distribution σ i is related to the clarity and size of the landmark in the image.

[0065] Then the solving equation for the user's position is:

[0066] For example, at a certain intersection, the system simultaneously recognizes three landmarks: a traffic light, a pedestrian crosswalk line, and a road name plate. According to the positions, sizes, and clarities of the three landmarks in the image, the system calculates their weights respectively: the traffic light weight is 0.42 (visibility score 0.85, distance 50 meters), the pedestrian crosswalk line weight is 0.36 (visibility score 0.7, distance 40 meters), and the road name plate weight is 0.28 (visibility score 0.6, distance 60 meters). The system weights and averages the position estimation results of the three landmarks according to the weights to finally determine the accurate position coordinates of the user.

[0067] Visibility score V of the marker i The calculation formula is: V i = 0.4C i + 0.6S i ; wherein, C i is the clarity score of the marker, the value range is 0 to 1, which is related to the contrast and blur degree of the marker in the image; S i is the marker size score, the value range is 0 to 1, which is proportional to the pixel area occupied by the marker in the image.

[0068] When the number of detected visible markers m is greater than 3, 3 markers are randomly selected to calculate the initial position estimate; the observation residual r j of other markers is calculated: r j = |z j -h est (x est ,y j )|; when the residual r th is less than the threshold r th , the marker is added to the inlier set, and the threshold r base is 5% of the image width; repeat the above process 10 times, and select the inlier set with the most inliers to recalculate the user position as the final position estimate of the user.

[0069] For example, at a crossroads in a certain business district, the system identifies 5 markers, but some of them may contain non-stable features such as temporary billboards. The system randomly selects 3 markers (such as traffic lights, street signs, and lane lines) to calculate the initial position, and then checks the observation residuals of other markers. If the observation residual of a temporary billboard reaches 7% of the image width, which exceeds the threshold of 5%, it is excluded. After 10 iterations, the system selects the largest inlier set containing 4 reliable markers and recalculates the user position based on the 4 markers to obtain a more accurate positioning result.

[0070] According to the vehicle speed and environmental complexity, the image acquisition frequency f is dynamically adjusted: f = f base ·(1+k1·v+k2·H); wherein, f base is the basic acquisition frequency, the value is 0.5 Hz, v is the vehicle speed, the unit is meter / second, k1 is the speed adjustment coefficient, the value is 0.02 s / m; H is the environmental complexity, k2 is the environmental complexity adjustment coefficient, the value is 0.5; the calculation formula of the environmental complexity is: p iPi is the proportion of the i-th region after image segmentation, and K is the number of segmentation categories. For example, when the user's vehicle is driving on an urban highway at a speed of 25 meters per second, the environmental complexity H is low (about 0.6), and the system calculates the acquisition frequency f = 0.5 x (1 + 0.02 x 25 + 0.5 x 0.6) = 1.15 Hz, that is, about 1.15 images are collected per second. When the vehicle slows down to 10 meters per second and enters a complex business district, the environmental complexity H increases to 1.2, and the calculated acquisition frequency f = 0.5 x (1 + 0.02 x 10 + 0.5 x 1.2) = 0.8 Hz.

[0071] Step S4, determining whether the user deviates from the navigation path according to the estimated position of the user.

[0072] The method for determining whether the user deviates from the preset navigation path comprises:

[0073] Step S401, establishing a probabilistic graph model G of the landmark position and the navigation path: G = (V, E, P); wherein V is a set of landmark nodes, E is a set of connecting edges between nodes, and P is a node transition probability matrix;

[0074] Step S402, estimating the current position of the user by a particle filtering algorithm, the number of particles is 200 to 500, and the particle weight update formula is: wherein, wi(t) is the weight of the i-th particle at time t, is an observation likelihood function; for example, at the entrance of an underground parking lot, the system distributes 300 particles to represent the possible user position. When the user enters the underground area and identifies the sign, the system updates the weight of each particle according to the observation result. After the weight is updated, the distribution of the particles gradually concentrates from a relatively dispersed state (indicating a higher position uncertainty) to the correct position near the sign.

[0075] Step S403, calculating the deviation distance D of the estimated position from the preset path:

[0076] wherein, (x est ,y est ) is the coordinate of the estimated position of the user, and (x j ,y j ) is the reference coordinate, which is the coordinate of the middle line of the road in the navigation path;

[0077] Step S405, when the deviation distance D is greater than a preset threshold D th , it is determined that the preset navigation path is deviated, wherein D th is 15 meters.

[0078] When it is confirmed that the user does not deviate from the navigation path, the GPS positioning error data of the blind area is recorded and uploaded to the server to update the regional GPS error distribution map. For example, when the user is driving under a viaduct, the user's position coordinates determined by visual recognition are (116.412345, 39.987654), and the coordinates of the nearest point on the preset navigation path are (116.412356, 39.987662). The calculated deviation distance D is about 1.5 meters, which is much smaller than the deviation threshold of 15 meters, so the system confirms that the user is still on the correct navigation path. At this time, the system records the deviation of the actual GPS coordinates of the current position from the road centerline and uploads this data to the server for updating the regional GPS error distribution map.

[0079] When it is confirmed that the user deviates from the navigation path, a new navigation path is recalculated based on the currently identified landmark position and the process returns to step S1.

[0080] When the system detects that the user deviates from the preset path, the re-planning process is immediately started. For example, the user originally planned to drive on the main road, but actually entered the parallel auxiliary road by mistake. The deviation distance calculated by visual positioning reaches 17 meters, which exceeds the threshold of 15 meters. At this time, the system recalculates the navigation path based on the current position and prompts the user, and then guides the user to return to the main road at an appropriate position or continue to advance to the destination along the auxiliary road.

[0081] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. An artificial intelligence-based electronic map intelligent construction and optimization method, characterized in that, The method comprises the following steps: Step S1, the navigation terminal collects the user's navigation path planning data, and identifies the blind area in the navigation path according to the regional GPS error distribution map; Step S2, if there is a blind area in the navigation path, the feature marker corresponding to the blind area is imported from the feature marker information library to the user's electronic map to complete the secondary construction of the electronic map; The feature marker at least includes a traffic sign, a road marking and a building feature; Step S3, based on the secondary constructed electronic map, navigation is carried out, when the user enters the blind area, the road scene image in the user's driving process is collected in real time through the camera device of the navigation terminal, the marker of the blind area is identified, and the identified marker is matched with the feature marker in the user's electronic map to determine the estimated position of the user; Step S4, whether the user deviates from the navigation path is judged according to the estimated position of the user; When it is confirmed that the user does not deviate from the navigation path, the GPS positioning error data of the blind area is recorded and uploaded to the server to update the regional GPS error distribution map; When it is confirmed that the user deviates from the navigation path, a new navigation path is recalculated based on the current identified marker position and returns to step S1.

2. The method of claim 1, wherein, The method for identifying the blind area in the navigation path according to the regional GPS error distribution map is specifically: According to the regional GPS error distribution map, a GPS positioning accuracy index P of each point in the navigation path is obtained, and when the GPS positioning accuracy index P of a point in the navigation path is greater than a preset threshold P th , it is judged whether the point is a blind area in the navigation path, wherein P th is 8 meters. The judgment method of the blind area is specifically: taking the point with the low GPS signal quality as the center, marking the area where the navigation path is not unique within the range of 100 meters before and after the center as the blind area; the existence of the navigation path not unique includes the existence of: intersection, interchange transition road, main and auxiliary road entrance; The GPS positioning accuracy index P is calculated by using a real-time updating method: Wherein, n is the GPS sampling times of a point in the navigation path, taking the nearest 5 to 20 times of navigation positioning data of other users at the point, (x i ,y i ) is the i-th GPS measurement coordinate, (x j ,y j ) is the reference coordinate, taking the middle line coordinate of the road in the navigation path.

3. The method of claim 2, wherein, The marker information in the feature marker information library is collected in a crowd sourcing manner, the marker information identified by a plurality of users in the same area is collected, when a marker is identified by not less than 3 different users and the position deviation is less than 5 meters, it is confirmed that the marker information is valid and stored in the feature marker information library; Each marker in the feature marker information library contains the following entry information: marker type, marker GPS coordinate, marker image feature vector and relative position of the marker in the road.

4. The method of claim 3, wherein, The identified marker is matched with the feature marker in the user's electronic map, and a feature point matching algorithm is adopted: SIFT feature points are extracted from the to-be-matched image; the cosine similarity between the feature point descriptors is calculated: wherein A and B are respectively the descriptor vectors of the identified marker picture and the characteristic marker picture; when the similarity S(A,B) is greater than a preset threshold S th , it is determined that the matching is successful, wherein S th is 0.

75.

5. The method of claim 4, wherein, The method for judging whether the user deviates from the preset navigation path comprises: Step S401, a probability graph model G of the marker position and the navigation path is established: G=(V, E, P); wherein V is a marker node set, E is a node connection edge set, and P is a node transition probability matrix; Step S402, estimate the current position of the user by a particle filtering algorithm, the number of particles is 200-500, and the particle weight updating formula is: wherein, wi(t) is the weight of the i-th particle at time t, is an observation likelihood function; Step S403, the deviation distance D of the estimated position and the preset path is calculated: wherein (x est ,y est ) is the estimated position coordinate of the user, and (x j ,y j ) is the reference coordinate, taking the middle line coordinate of the road in the navigation path. Step S405, when the deviation distance D is greater than a preset threshold D th , the deviation preset navigation path is determined, wherein D th is 15 meters.

6. The method of claim 5, wherein, determining an estimated position coordinate (x est ,y est ) of a user comprises: establish a user position estimation model based on the identified plurality of markers: where P(x, y) is the probability of the user being at coordinates (x, y), P i (x, y) is the probability of the user's position estimated based on the i-th marker, and m is the number of visible markers detected; w i is the weight of the ith landmark, and the weight calculation formula is: w i = V i · D i , V i is the visibility score of the landmark, and the value range is 0 to 1, D i is the distance influence factor of the landmark, and the calculation formula is: D i = exp(-d i / d0), d i is the estimated distance of the user to the landmark, and d0 is the feature distance attenuation coefficient, which is 100 meters; An observation equation is established for each marker: z i = h i (x, y) + v i ; where z i is the observed value of marker z i in the camera image, h i (x, y) is the theoretical observation of marker i when the user is at coordinates (x, y), and v i is the observation noise, which is Gaussian distributed σ i is related to the clarity and size of the marker in the image; The equation for solving the user position is:

7. The method of claim 6, wherein, The visibility score V of the marker i The calculation formula is: V i = 0.4C i + 0.6S i ; wherein C i is a clarity score of the marker, ranging from 0 to 1, related to the contrast and blurring of the marker in the image; S i is a marker size score, ranging from 0 to 1, proportional to the pixel area occupied by the marker in the image.

8. The method of claim 7, wherein, When the number of detected visible markers m is greater than 3, 3 markers are randomly selected to calculate the initial position estimation; Compute the observation residual r for other markers j = |z j - h j (x est , y est ); when the residual r j is less than a threshold r th , add the marker to the inlier set, the threshold r th is set to 5% of the image width; The above process is repeated 10 times, the inner point set with the largest number of inner points is selected to recalculate the user's position as the final position estimation of the user.

9. The method of claim 8, wherein, According to the vehicle driving speed and the environment complexity, the image acquisition frequency f is dynamically adjusted: f=f base ·(1+k1·v+k2·H); wherein, f base is the basic acquisition frequency, taking the value of 0.5 Hz, v is the vehicle speed, taking the unit of meter / second, k1 is the speed adjustment coefficient, taking the value of 0.02 s / m; H is the environment complexity, k2 is the environment complexity adjustment coefficient, taking the value of 0.

5.

10. The method of claim 9, wherein, The calculation formula of the environmental complexity is: p i is the proportion of the i-th region after image segmentation, and K is the number of segmentation categories.

Images