Map information updating method and system

Abstract

The invention discloses a map information updating method and system, and relates to the technical field of map updating, and the method comprises the steps: according to each road sign object, using an editing distance and a cosine similarity to measure a first similarity between symbol elements in an element set of the road sign object and symbol elements in a map element set; measuring a second similarity between the color elements in the element set of the road sign object and the color elements in the map element set by using the Euclidean distance; measuring a third similarity between a central point position element in the element set of the road sign object and a central point position element in a map element set by using a Hausdorff distance; taking the intersection-union ratio of the contour frame elements in the element set of the road sign object and the contour frame elements in the map element set as fourth similarity; determining whether the four similarities are matched or not through the four similarities; if yes, updating the map. The map can be updated in time, and the real-time performance is good.

Description

Technical Field

[0001] The present invention relates to the technical field of map updating, and in particular to a method and system for updating map information. Background Art

[0002] High-precision maps can help cars perceive their precise location and understand the surrounding environment. As road scenes are constantly changing, high-precision maps need to be constantly updated to ensure the real-time and accuracy of their information. There are currently two main methods for updating high-precision maps: one is to use professional collection vehicles equipped with high-precision equipment to regularly collect data to update maps. This method has high update quality, but due to its high cost, it cannot be updated frequently and in a timely manner; the other is to install ordinary sensors on general vehicles and update maps in a crowdsourcing manner. This method is low-cost and fast to update, but the update quality is poor and the technology is difficult. This leads to a deviation between the map content and the actual road conditions, which may have a negative impact on the positioning and decision-making of driverless cars. Therefore, it is necessary to propose an effective method to achieve timely and accurate semantic information updates for high-precision maps.

[0003] In the prior art, patent CN111427904A discloses a method, device and electronic device for updating high-precision map data, which uses a collection vehicle to regularly recollect all road elements in the area that needs to be updated, and uses the obtained images and point cloud data to generate vectorized data of the elements. The collected data is compared with the historical data. If the comparison result is a change in the element data, the new vectorized data is used to directly batch replace the original high-precision map vectorized data of the area to complete the update. Patent CN116045964A discloses a method and device for updating high-precision maps, which obtains vehicle posture information, collects road images and navigation information, integrates observation features, optimizes vehicle posture, and uploads the updated map to the cloud when the confidence of the registered posture information is higher than the set threshold and the reprojection error exceeds the preset error threshold, and finally determines the updated content of the map elements through cloud statistical data.

[0004] However, when the above method determines the changes between the newly collected road data and the road data in the high-precision map, the road data is not further subdivided for comparison, so that unnecessary data is also updated when the road data in the high-precision map is updated, which consumes a lot of time. Summary of the invention

[0005] The embodiments of the present invention provide a map information updating method and system, which can solve the problem in the prior art that unnecessary data is updated when updating road data in a high-precision map, resulting in a large amount of time being wasted.

[0006] The embodiment of the present invention provides a method for updating map information, comprising the following steps: collecting road sign objects around the current position of a vehicle, taking multiple attributes of each road sign object as an element set of each road sign object; and obtaining a map around the current position of the vehicle, wherein the map comprises a map element set of each road sign object composed of multiple attributes of each road sign object; the road sign objects comprise: traffic signs, buildings and lane lines; the attributes comprise: symbols, colors, center point positions and contour boxes representing shapes; according to each road sign object, using edit distance and cosine similarity to measure the first similarity between the symbol elements in the element set of the road sign object and the symbol elements in the map element set; using Euclidean distance to measure the similarity between the road sign objects and the map element set; The method comprises the following steps: determining a second similarity between color elements in the element set of the object and color elements in the map element set; using Hausdorff distance to measure a third similarity between center point position elements in the element set of the landmark object and center point position elements in the map element set; taking the intersection-and-union ratio of outline frame elements in the element set of the landmark object and outline frame elements in the map element set as a fourth similarity; determining a matching relationship between the element set of each acquired landmark object and the map element set of each landmark object in the acquired map according to the first similarity, the second similarity, the third similarity and the fourth similarity, wherein the matching relationship includes: consistent elements, changed elements and newly added elements; and determining whether to update the map according to the matching relationship.

[0007] Furthermore, the step of using the edit distance and the cosine similarity to measure the first similarity between the symbol elements in the element set of the landmark object and the symbol elements in the map element set specifically includes: Obtain the symbol element from the element set of each landmark object collected, and record it as a string S_detect; obtain the symbol element from the map element set of each landmark object in the map, and record it as a string S_map; Initialize a two-dimensional matrix D of (|S_map|+1)×(|S_detect|+1), where |S_map| and |S_detect| both represent the length of the string; assign initial values ​​to the first row and first column of the two-dimensional matrix D so that D[i,0]=i, D[0,j]=j; For i=1to|S_map|, j=1to|S_detect|; when S_map[i-1]=S_detect[j-1], D[i,j]=D[i-1,j-1]; when S_map[i-1] is different from S_detect[j-1], D[i,j]=1+min(D[i-1,j], D[i,j-1], D[i-1,j-1]); The edit distance is obtained as L=D[|S_map|, |S_detect|]; Convert S_map and S_detect into one-hot encoding vectors v_map and v_detect respectively; Get the cosine similarity cos_sim of vectors v_map and v_detect: cos_sim= (v_map·v_detect) / (||v_map||*||v_detect||); Take the weighted average of the edit distance and cosine similarity to get the first similarity d1: d1=w5*(1-L / max(|S_map|, |S_detect|))+w6*cos_sim; Among them, w5 and w6 are weights.

[0008] Furthermore, the step of using the Euclidean distance to measure the second similarity between the color elements in the element set of the road sign object and the color elements in the map element set specifically includes: Obtain color elements from the element set of each landmark object collected, and record them as a color vector C_detect; obtain color elements from the map element set of each landmark object in the map, and record them as a color vector C_map; Get the Euclidean distance between the color vectors C_detect and C_map:

[0009] Normalize the Euclidean distance d to the interval [0, 1] to obtain the second similarity d2: d2=1-d / d_max; Wherein, d_max represents the Euclidean distance of the color space after the color vectors C_detect and C_map are converted into the same color space, and the color space is used to define and describe different colors in different scenes.

[0010] Furthermore, the method of using the Hausdorff distance to measure the third similarity between the center point position element in the element set of the landmark object and the center point position element in the map element set specifically comprises the following steps: Obtain outline box elements from the element set of each collected landmark object, and record them as set A={a1,a2,...an}; obtain outline box elements from the map element set of each landmark object in the map, and record them as set B={b1,b2,...bm}; Obtain the Hausdorff distance d3 between the contour box elements obtained in the element set of each collected landmark object and the contour box elements obtained in the map element set of each landmark object in the map: d3 = max(sup(inf d(a, b)), sup(inf d(b, a))); Where d(a,b) represents the Euclidean distance from point a to point b, a is each element in set A, b is each element in set B, sup represents the maximum value, and inf represents the minimum value; A Gaussian function is used for smooth mapping, and the Hausdorff distance d3 is normalized and mapped to the range of 0 to 1 to obtain the third similarity d3'.

[0011] Furthermore, the method of taking the intersection-and-union ratio of the outline frame elements in the element set of the road sign object and the outline frame elements in the map element set as the fourth similarity comprises the following specific steps: By using a computer graphics algorithm, a set of all pixel points within a contour frame obtained from an element set of each landmark object collected is determined, and a set of all pixel points within a contour frame obtained from a map element set of each landmark object in the map is determined, and the sets are merged into a pixel point set U; the number of pixel points in the pixel point set U is obtained as the area |U|; Determine the intersection of the set of all pixel points within the contour frame obtained in the element set of each collected landmark object and the set of all pixel points within the contour frame obtained in the map element set of each landmark object in the map, and record it as I; obtain the number of pixel points in the intersection I as the area |I|; Get the intersection-over-union (IOU) as the fourth similarity: IOU = |I| / |U|.

[0012] Furthermore, the step of determining the matching relationship between the element set of each collected landmark object and the map element set of each landmark object in the acquired map specifically includes: Weighting the first similarity, the second similarity, the third similarity, and the fourth similarity to obtain a matching metric; When the matching metric is greater than a preset threshold, it is determined that the difference between the matching metric and the preset threshold is greater than the preset difference, indicating that the element set of each acquired landmark object and the map element set of each landmark object in the acquired map are consistent elements; When the matching metric is greater than a preset threshold, it is determined that the difference between the matching metric and the preset threshold is less than the preset difference, indicating that the element set of each acquired landmark object and the map element set of each landmark object in the acquired map are changed elements; When the matching metric is less than a preset threshold, it means that the element set of each collected landmark object and the map element set of each landmark object in the acquired map are newly added elements.

[0013] Furthermore, the map is updated, and the specific steps include: Acquire multiple collection results of each attribute of each road sign object according to multiple vehicles collecting multiple collection results of each road sign object in the same area; According to the multiple collection results of each attribute of each road sign object, the mean μ and the standard deviation σ are obtained using Gaussian distribution; According to the set confidence level percentage, the corresponding z value is obtained using the standard normal distribution table of Gaussian distribution; Get the confidence interval of each attribute of each road sign object, the confidence interval formula is: [μ-z*σ,μ+z*σ]; According to each attribute of each newly collected road sign object, the distances d1 and d2 between each attribute and the two endpoints in the confidence interval are obtained respectively. The derivation formulas of the distances d1 and d2 are as follows: d1=|x'-(μ-z*σ)|; d2=|x'-(μ+z*σ)|; Among them, x' represents the attribute; Select the distance with the smaller distance value from d1 and d2, and use normalization to normalize the distance with the smaller distance value to the interval [0,1] as the confidence of each attribute of each landmark object; If the relationship between the element set of each newly collected landmark object and the map element set of each landmark object in the acquired map is a consistent element, then according to the confidence of each attribute of each landmark object newly collected multiple times, obtain the average increment of multiple confidences; add the average increment to the confidence of each landmark object in the map until the set highest confidence threshold is reached; If it is a changing element, then according to the confidence of each attribute of each landmark object newly collected multiple times, obtain the average reduction of multiple confidences, and subtract the average reduction from the confidence of each landmark object in the map until the set minimum confidence threshold is reached; If it is a newly added element, each newly collected landmark object will be inserted into the map and the initial confidence will be set.

[0014] A map information updating system, comprising: A road sign object acquisition module is used to collect road sign objects around the current position of the vehicle, and use multiple attributes of each road sign object as an element set of each road sign object; and obtain a map around the current position of the vehicle, the map containing a map element set of each road sign object composed of multiple attributes of each road sign object; the road sign objects include: traffic signs, buildings and lane lines; the attributes include: symbol, color, center point position and outline box representing shape; A similarity analysis module is used to measure, according to each road sign object, a first similarity between symbol elements in an element set of the road sign object and symbol elements in a map element set using edit distance and cosine similarity; to measure a second similarity between color elements in an element set of the road sign object and color elements in a map element set using Euclidean distance; to measure a third similarity between center point position elements in an element set of the road sign object and center point position elements in a map element set using Hausdorff distance; and to use an intersection-over-union ratio of outline frame elements in an element set of the road sign object and outline frame elements in a map element set as a fourth similarity; A map update module is used to determine the matching relationship between the element set of each collected landmark object and the map element set of each landmark object in the acquired map according to the first similarity, the second similarity, the third similarity and the fourth similarity, wherein the matching relationship includes: consistent elements, changed elements and newly added elements; and determine whether to update the map according to the matching relationship.

[0015] The embodiment of the present invention provides a method and system for updating map information. Compared with the prior art, the beneficial effects thereof are as follows: The road sign objects around the current position of the vehicle are collected, and the multiple attributes of each road sign object are used as the element set of each road sign object; then the similarity of each attribute of each collected road sign object and each attribute of each road sign object in the map is measured, and then the matching relationship between each attribute of each collected road sign object and each attribute of each road sign object in the map is determined through the similarity. The matching relationship includes: consistent elements, changed elements and newly added elements. According to each matching relationship, it is determined whether the map needs to be updated, thereby avoiding updating of unnecessary data in the map, so that the map can be updated more timely. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart provided for an embodiment of the present invention; Figure 2 A high-precision map delivery scenario diagram provided in an embodiment of the present invention, wherein (a) is V2I transmission, (b) is joint transmission, and (c) is V2V transmission; Figure 3 A road sign object graph provided by an embodiment of the present invention; Figure 4 A map update flow chart provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0018] See also Figure 1 , an embodiment of the present invention provides a method for updating map information, comprising the following steps: Step 1: Collect the road sign objects around the current position of the vehicle, and use the multiple attributes of each road sign object as the element set of each road sign object; and obtain the map around the current position of the vehicle, the map contains the map element set of each road sign object composed of the multiple attributes of each road sign object; road sign objects include: traffic signs, buildings and lane lines; attributes include: symbols, colors, center point positions and outlines representing shapes. Map types include but are not limited to high-precision maps and ordinary-precision maps.

[0019] Step 2: For each landmark object, use edit distance and cosine similarity to measure the first similarity between the symbol elements in the landmark object's element set and the symbol elements in the map element set; use Euclidean distance to measure the second similarity between the color elements in the landmark object's element set and the color elements in the map element set; use Hausdorff distance to measure the third similarity between the center point position elements in the landmark object's element set and the center point position elements in the map element set; and use the intersection-and-union ratio of the outline box elements in the landmark object's element set and the outline box elements in the map element set as the fourth similarity.

[0020] Step 3: Determine the matching relationship between the element set of each collected landmark object and the map element set of each landmark object in the acquired map based on the first similarity, the second similarity, the third similarity and the fourth similarity, wherein the matching relationship includes: consistent elements, changed elements and newly added elements; determine whether to update the map based on the matching relationship.

[0021] The specific implementation method is as follows: (1) Distribution of high-precision maps.

[0022] The target vehicle obtains its own location information through GPS during driving, and sends its own location information to the surrounding road test units (RSU) and other vehicles through V2I (Vehicle-to-Infrastructure) and V2V (Vehicle-to-Vehicle) communications. RSU can store a certain amount of high-precision map data, and the two cars driving in opposite directions also contain the high-precision map information needed by each other. Therefore, RSU and other vehicles can transmit high-precision maps to the target vehicle. There are three specific transmission scenarios: like Figure 2 As shown, two cars are driving in opposite directions on a city road. Figure 2 Vehicles located below the inner road are considered as target vehicles required by the HD map. Figure 2 The vehicle above the inner road stores the required high-definition map information in the direction of the target vehicle. Since the communication distance of V2V communication is shorter than that of V2I, it can reduce path loss, transmission delay and power consumption. Based on this, V2V transmission is given priority. V Represents the distance between two vehicles traveling in opposite directions, d VM Indicates the maximum communication distance of V2V.

[0023] exist Figure 2 In (a), when d V >d VM When the vehicle is in motion, the RSU transmits high-precision map information to the target vehicle through V2I.

[0024] exist Figure 2 In (b), when d V <d VM When the vehicle is in motion, both the RSU and the vehicle transmit high-precision map information to the target vehicle.

[0025] exist Figure 2 (c), d V The distance between the two vehicles becomes very small, that is, the two vehicles are very close to each other, and the vehicles transmit high-precision maps through V2V communication.

[0026] (2) After the target vehicle obtains the high-precision map through the above method, it is necessary to obtain the surrounding feature information of the vehicle based on the target detection method and match it with the high-precision map. The following uses traffic signs as an example to illustrate this process.

[0027] (2.1) The vehicle uses sensors such as cameras and lidar to collect road sign objects around the current location, such as traffic signs, buildings, lane lines, etc. The on-board unit (OBU) uses the currently mature feature recognition algorithm to extract the attributes of each object, such as traffic sign type, color, center point location coordinates, outline box (here "outline box" refers to a rectangular or polygonal area surrounding the target object, used to represent the geometric shape of the object), and other attributes to obtain a traffic sign element set {O}.

[0028] (2.2) The vehicle queries the high-precision map around its own position and obtains a set of map elements {G} of traffic signs, where each map element Gj also contains corresponding attributes such as type, color, position, and shape.

[0029] (2.3) The vehicle uploads the traffic sign element set {O} and the map element set {G} obtained above to the edge server. The edge server calculates the similarity between each detected object Oi and all map element sets {G}.<Oi, Gj> : (2.3.1) Calculate the similarity d1 between the two traffic signs.

[0030] First, the symbol information of the corresponding traffic sign is obtained from the high-precision map, recorded as the string S_map, as a reference standard. Then, the detected sign symbol is recognized using OCR (Optical Character Recognition) technology, recorded as the string S_detect.

[0031] The similarity between the detected symbol and the symbol in the high-precision map is calculated by using the edit distance (edit distance is a classic indicator for measuring the similarity between two strings, which represents the minimum number of editing operations required to convert one string into another) and the cosine similarity, denoted as d1.

[0032] Edit distance calculation: 1. Initialize a two-dimensional matrix D of (|S_map|+1)x(|S_detect|+1), where |S| represents the length of string S.

[0033] 2. Assign initial values ​​to the first row and first column: D[i,0]=i, D[0,j]=j. For i=1to|S_map|, j=1to|S_detect|: if S_map[i-1]==S_detect[j-1], then D[i,j]=D[i-1,j-1], otherwise D[i,j]=1+min(D[i-1,j], D[i,j-1], D[i-1,j-1]), and the final edit distance = D[|S_map|, |S_detect|].

[0034] Cosine similarity calculation: Convert S_map and S_detect to one-hot encoding vectors v_map and v_detect respectively.

[0035] 2. Calculate the cosine similarity of two vectors: cos_sim = (v_map·v_detect) / (||v_map||*||v_detect||) The larger the cosine similarity, the more similar the two traffic signs are.

[0036] Similarity score calculation: The edit distance and cosine similarity can be weighted averaged to obtain the final symbol content similarity score d1: d1=w5*(1-edit distance / max(|S_map|, |S_detect|))+w6*cos_sim. The weights w5 and w6 can be adjusted according to actual conditions.

[0037] (2.3.2) Calculate the color similarity d2 of the two traffic signs.

[0038] First, the color information of the corresponding sign element is obtained from the high-precision map, which is recorded as the color vector C_map. This color vector can be a value in a color space such as RGB, HSV, or Lab. Then the color of the detected sign symbol is measured and converted to the same color space as C_map, which is recorded as the color vector C_detect.

[0039] Calculate color distance: Calculate the Euclidean distance between C_map and C_detect:

[0040] Similarity calculation: Normalize the calculated color distance d to the interval [0, 1] as the color similarity score d2: d2 = 1-d / d_max, where d_max is the maximum distance in the color space, which is used to normalize d.

[0041] (2.3.3) Calculate the similarity d3 between the two traffic sign positions.

[0042] The position similarity mainly calculates the Hausdorff distance d3 between the two positions.

[0043] First, the contour of the detected object Oi is represented as a set A={a1,a2,...an}, and the contour of the matching candidate object Gj obtained from the high-precision map is represented as a set B={b1,b2,...bm}.

[0044] The Hausdorff distance is defined as: d3 = max(sup(inf d(a, b)), sup(inf d(b, a))), where d(a, b) represents the Euclidean distance from point a to point b, sup represents the maximum value, and inf represents the minimum value. Use the Gaussian function for smooth mapping, normalize the position distance d3, map it to the range of 0 to 1, and set it as d3'.

[0045] (2.3.4) Calculate the shape similarity of the two traffic signs, that is, the intersection over union (IOU) of their shapes.

[0046] Use computer graphics algorithms to determine the set of all pixels within the contour box range, calculate the union of the two contour boxes A={a1,a2,...an} and B={b1,b2,...bm}, that is, the set of all pixels contained in A and B, set to U, and calculate the number of pixels in the union U as the area |U|. Calculate the intersection of the two contour boxes, that is, the set of pixels contained in both A and B, set to I, and calculate the number of pixels in the intersection I as the area |I|. IOU = intersection area / union area, that is: IOU = |I| / |U|.

[0047] (2.4) Finally, the matching metric D is used to determine whether it is a consistent element, a changed element, or a newly added element.

[0048] The d1, d2, d3' and IOU calculated above are weighted to generate a new matching metric D: D=w1*d1+w2*d2+w3*(1-d3')+w4*IOU. The weights w1, w2, w3 and w4 can be set to 0.25 at first, and then adjusted according to the actual effect. If D is greater than the threshold τ, it indicates a preliminary match. If D is less than the threshold τ, it indicates a mismatch. The threshold τ can be initially set to 0.7, which means that a match is only considered when the position and shape are relatively close. τ can also be learned through training data. The following is the specific situation of judging whether it is a consistent element, a changed element, or a newly added element based on the matching metric D.

[0049] Match consistent elements: For each detected object Oi, calculate its matching metric D with the set of all map elements {G}. If D is greater than a given threshold τ (such as 0.7), Oi is considered to match the corresponding map element Gj.<Oi, Gj> A pair is an element that "matches the same".

[0050] There are elements with changes: for those elements with matching metric D>τ, but the matching D value is low<Oi,Gj> Yes, it means that there is a certain discrepancy between the detection result and the original map. This may be caused by changes in the location, shape and other attributes of the road elements. By statistically analyzing the observation results of multiple vehicles, if the change exceeds a certain confidence threshold, it can be determined as an element with "changes" and the element attributes need to be updated.

[0051] Newly added elements: For all detected Oi, if the matching metric D with all Gj is lower than the threshold τ, it means that the Oi has no corresponding element in the original high-precision map and is judged as a newly added element. By statistically analyzing the consistent observation results of multiple vehicles, if the confidence of the newly added element exceeds the set threshold, it will be added to the high-precision map.

[0052] (2.5) For matching<Oi, Gj> For Oi, a matching record is generated, including the ID, time, location and other information of the matching element. For unmatched Oi, an unmatched record is generated. Both matching and unmatched records are sent to the edge server to update the confidence of the map.

[0053] (2.6) For other point or line elements, similarity judgments based on other features can be added. For example, for point elements, the calculation of texture similarity can be added; for line elements, the calculation of length and curvature similarity can be added. Similarly, the weighted sum of the above similarity indicators is performed to obtain the final similarity score, which is used to determine whether the point or line elements match. In this way, the similarity of different types of road sign elements can be more comprehensively evaluated.

[0054] (3) When updating a high-precision map based on information from a single vehicle’s single observation, obstacles may block the data, resulting in unreliable update data. The present invention uses a vehicle crowdsourcing method based on vehicle-road collaboration to update the high-precision map.

[0055] (3.1) Data integration: The edge server organizes the map matching data of the same local HD map uploaded by multiple vehicles within a certain time interval. The uploaded data is sorted and classified in chronological order and grouped into corresponding map areas. For multiple detection results of the same object, data alignment is performed, and then the detection results of different vehicles are integrated into a unified HD map coordinate system.

[0056] (3.2) Statistical analysis: Gaussian distribution statistics are performed on multiple detection results of the same object, including data such as location and shape, and statistical parameters such as the mean and variance of different attributes are calculated. Then, the confidence interval of the detection results is estimated by fitting the statistical model. The specific calculation process is as follows: (3.2.1) Data collection: Collect multiple detection results of the same map element in different vehicles and at different times, including the detected location coordinates (x, y, z), shape outline (rectangle or polygon) and other attribute values.

[0057] (3.2.2) Gaussian distribution fitting: For each attribute (such as position x, position y, etc.), taking position x as an example, we use Gaussian distribution to fit its multiple observations: take the average of all the observations of the attribute and calculate the mean μ; calculate the standard deviation σ to measure the discreteness of the attribute value. The Gaussian distribution probability density function obtained by fitting is: p(x) = 1 / (sqrt(2π)σ) * exp(-(x-μ)^2 / (2σ^2)).

[0058] (3.2.3) Confidence interval estimation: According to the required confidence level (usually 95% or 99%), consult the standard normal distribution table of Gaussian distribution, obtain the corresponding z score, and calculate μ±z*σ as the confidence interval [a, b] of the attribute. For example, the 95% confidence level corresponds to z=1.96, then the confidence interval is [μ-1.96σ, μ+1.96σ]. In this way, a confidence interval [a, b] can be obtained for each attribute.

[0059] (3.2.4) Confidence calculation: Calculate the distance d1=|x'-a|, d2=|x'-b| between the new observation x' and the interval endpoints a and b, and normalize the smaller distance to the interval [0,1] as the confidence of the attribute: conf = min(d1,d2) / max(|a-μ|,|b-μ|).

[0060] (3.2.5) Comprehensive confidence: According to the minimum method, here we take the position confidence and shape confidence as examples, and take the smaller value as the final comprehensive confidence. Comprehensive confidence = min(confidence_position, confidence_shape), the value range of comprehensive confidence is [0, 1], 1 represents high confidence, and 0 represents low confidence.

[0061] (3.3) Noise filtering and map matching: Detect and exclude outliers and abnormal detection results to improve the robustness of statistical analysis. Match the filtered statistical results with the existing map and mark the elements that match consistently, have changed, and have been newly added.

[0062] (3.4) Map update: (3.4.1) For consistent elements, calculate the average increment through the confidence values ​​of multiple matching records to maintain or improve its confidence. At the same time, set an upper limit for the confidence increment, for example, it cannot exceed 0.95.

[0063] The purpose of setting an upper limit on the confidence increment is to: 1) Avoid overconfidence.

[0064] Even if the confidence of an element is high after multiple matches, some margin must still be left to prevent potential errors or abnormalities. For example, factors such as sensor detection accuracy, environmental changes, and occlusion may lead to misjudgment or deviation. The confidence limit ensures that the system will not draw overconfident conclusions due to some short-term detection results.

[0065] 2) Leave room for future changes.

[0066] Road scenes are dynamic, and traffic signs, road conditions, etc. may change over time. If the confidence is set to an absolute 100%, it means that the system will believe that the current map elements are completely correct and will not change.

[0067] (3.4.2) For changed elements, update the attributes of the changed elements based on the original high-precision map, calculate the average reduction through the confidence values ​​of multiple matching records, and reduce the confidence. At the same time, set the lower limit of the confidence reduction, for example, it cannot be lower than 0.1.

[0068] The purpose of setting a lower limit for confidence reduction is to: 1) Prevent data from being easily denied.

[0069] The lower limit of confidence reduction ensures that the system does not drastically reduce the confidence of a map element due to a few inaccurate detections. Without the lower limit, the confidence may be excessively weakened by a small amount of false information, causing the system to prematurely consider these map elements invalid or need to be removed.

[0070] 2) Maintain the robustness of existing map information.

[0071] Map elements, such as traffic signs and lane lines, are usually long-term stable and do not change frequently. Therefore, a design that does not reduce the confidence level too low can help maintain the robustness of existing maps and avoid excessive modification of map elements when there is not enough evidence. By retaining a confidence level of at least 0.1, important map information can be prevented from being deleted or significantly modified due to noise or short-term misjudgment.

[0072] (3.4.3) For the newly added elements, insert them into the original high-precision map and initialize the confidence level, for example, 0.5, and then increase the confidence level based on the number of detections.

[0073] (3.4.4) Set confidence change thresholds for key elements. For elements related to traffic rules and determining autonomous driving behavior, such as traffic signs, traffic lights, lane lines, etc., more stringent thresholds are used. For example, the confidence change threshold for conventional elements is 0.1, and the confidence change threshold for key elements can be set to 0.05. At the same time, different levels of key elements can have different change thresholds.

[0074] (3.5) Change review and map release: Additional manual review is performed on key change elements to avoid errors caused by automatic updates. All updated maps are uploaded to the cloud for integration and then sent to vehicles for use, while collecting map feedback from vehicles.

[0075] The present invention includes multiple key technologies, which are as follows: (1) High-precision map distribution mechanism: The present invention utilizes V2I and V2V communication technologies to dynamically select the optimal distribution method according to the distance between vehicles, and can flexibly and efficiently transmit high-precision map information to the target vehicle, thereby improving the efficiency of map distribution.

[0076] (2) Vehicle-mounted feature information acquisition and matching method: This method uses vehicle-mounted sensors to collect surrounding information, such as traffic signs, buildings and other attributes, and matches and correlates this information with the original high-precision map. It can accurately determine whether there are consistency, changes or additions to map elements.

[0077] (3) Map information update method based on vehicle-road collaboration: This solution uses data collected by multiple vehicles to perform statistical analysis and noise filtering on the edge server, calculate the confidence of map elements, and maintain consistent elements, update changed elements, and add new elements. At the same time, a stricter confidence change threshold is set for key elements, and manual review is performed.

[0078] (4) Resource integration and efficiency improvement: This method makes full use of the perception, communication, and computing capabilities of the vehicle-road cooperative system and distributes computing tasks between the on-board unit and the edge server, thereby improving the quality and efficiency of updates and reducing the computing load on the cloud.

[0079] The advantages of the present invention are: (1) In terms of map distribution mechanism, the invention makes full use of V2V and V2I communication technologies and can select different distribution methods according to the distance between vehicles, thereby improving the flexibility and efficiency of distribution.

[0080] (2) In terms of feature matching, this invention uses edge servers for feature matching, which greatly improves computing efficiency and reduces the computing load on the cloud.

[0081] (3) It makes full use of vehicle-road collaboration, using both on-board perception and roadside infrastructure to achieve effective resource integration. At the same time, it uses vehicle crowdsourcing based on vehicle-road collaboration to update high-precision maps, thereby improving the quality of map updates.

[0082] A map information updating system, comprising: The road sign object acquisition module is used to collect road sign objects around the current position of the vehicle, and use the multiple attributes of each road sign object as the element set of each road sign object; and obtain a map around the current position of the vehicle, the map contains a map element set of each road sign object composed of the multiple attributes of each road sign object; the road sign objects include: traffic signs, buildings and lane lines; the attributes include: symbols, colors, center point positions and contour boxes representing shapes. The similarity analysis module is used to measure the first similarity between the symbol elements in the element set of the road sign object and the symbol elements in the map element set using the edit distance and cosine similarity according to each road sign object; use the Euclidean distance to measure the second similarity between the color elements in the element set of the road sign object and the color elements in the map element set; use the Hausdorff distance to measure the third similarity between the center point position elements in the element set of the road sign object and the center point position elements in the map element set; use the intersection and union ratio of the contour box elements in the element set of the road sign object and the contour box elements in the map element set as the fourth similarity. A map update module is used to determine the matching relationship between the element set of each collected landmark object and the map element set of each landmark object in the acquired map according to the first similarity, the second similarity, the third similarity and the fourth similarity, wherein the matching relationship includes: consistent elements, changed elements and newly added elements; and determine whether to update the map according to the matching relationship.

[0083] A specific embodiment is as follows: 1. The target vehicle obtains a high-precision map.

[0084] 2. Obtain feature information about the vehicle’s surroundings based on target detection methods and match it with high-precision maps.

[0085] (1) Use sensors such as cameras and lidar to collect road signs around the current location, such as traffic signs, buildings, lane lines, etc. Figure 3 shown.

[0086] (2) The vehicle queries the high-precision map around its own location and obtains a set of map elements {G}, where each map element contains attributes such as location, shape, and type.

[0087] (3) The vehicle uploads the semantic element set {O} and the map element set {G} obtained above to the edge server. The edge server calculates the similarity between each detected object and all map element sets {G}. For each pair <,>.

[0088] 3. Update high-precision maps using vehicle crowdsourcing based on vehicle-road collaboration.

[0089] (1) The edge server organizes the map matching data of the same local HD map uploaded by multiple vehicles within a certain time interval.

[0090] (2) Perform Gaussian distribution statistics on multiple detection results of the same object, including data such as position and shape, and calculate statistical parameters such as the mean and variance of different attributes.

[0091] (3) Detect and exclude outliers and abnormal test results to improve the robustness of statistical analysis.

[0092] (4) Update the map, such as Figure 4 shown.

[0093] (5) Conduct additional manual review for key changed elements to avoid errors caused by automatic updates.

[0094] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for updating map information, characterized in that: The following steps are involved: Collecting landmark objects around the current position of the vehicle, using multiple attributes of each landmark object as an element set of each landmark object; and obtaining a map around the current position of the vehicle, the map comprising a map element set of each landmark object composed of the multiple attributes of each landmark object; The road sign objects include: traffic signs, buildings and lane lines; the attributes include: symbol, color, center point position and contour box representing the shape; According to each road sign object, the first similarity between the symbol elements in the element set of the road sign object and the symbol elements in the map element set is measured using the edit distance and the cosine similarity; the second similarity between the color elements in the element set of the road sign object and the color elements in the map element set is measured using the Euclidean distance; the third similarity between the center point position elements in the element set of the road sign object and the center point position elements in the map element set is measured using the Hausdorff distance; the intersection and union ratio of the outline frame elements in the element set of the road sign object and the outline frame elements in the map element set is taken as the fourth similarity; A matching relationship between an element set of each collected landmark object and a map element set of each landmark object in the acquired map is determined according to the first similarity, the second similarity, the third similarity and the fourth similarity, wherein the matching relationship includes: consistent elements, changed elements and newly added elements; and whether to update the map is determined according to the matching relationship.

2. A method for updating map information as claimed in claim 1, characterized in that: The method of using the edit distance and the cosine similarity to measure the first similarity between the symbol elements in the element set of the landmark object and the symbol elements in the map element set comprises the following specific steps: Obtain the symbol element from the element set of each landmark object collected, and record it as a string S_detect; obtain the symbol element from the map element set of each landmark object in the map, and record it as a string S_map; Initialize a two-dimensional matrix D of (|S_map|+1)×(|S_detect|+1), where |S_map| and |S_detect| both represent the length of the string; assign initial values ​​to the first row and first column of the two-dimensional matrix D so that D[i,0]=i, D[0,j]=j; For i=1to|S_map|, j=1to|S_detect|; when S_map[i-1]=S_detect[j-1], D[i,j]=D[i-1,j-1]; when S_map[i-1] is different from S_detect[j-1], D[i,j]=1+min(D[i-1,j], D[i,j-1], D[i-1,j-1]); The edit distance is obtained as L=D[|S_map|, |S_detect|]; Convert S_map and S_detect into one-hot encoding vectors v_map and v_detect respectively; Get the cosine similarity cos_sim of vectors v_map and v_detect: cos_sim= (v_map·v_detect) / (||v_map||*||v_detect||); Take the weighted average of the edit distance and cosine similarity to get the first similarity d1: d1=w5*(1-L / max(|S_map|, |S_detect|))+w6*cos_sim; Among them, w5 and w6 are weights.

3. A method for updating map information as claimed in claim 1, characterized in that: The method of using the Euclidean distance to measure the second similarity between the color elements in the element set of the road sign object and the color elements in the map element set specifically comprises the following steps: Obtain color elements from the element set of each landmark object collected, and record them as a color vector C_detect; obtain color elements from the map element set of each landmark object in the map, and record them as a color vector C_map; Get the Euclidean distance between the color vectors C_detect and C_map: ； Normalize the Euclidean distance d to the interval [0, 1] to obtain the second similarity d2: d2=1-d / d_max; Wherein, d_max represents the Euclidean distance of the color space after the color vectors C_detect and C_map are converted into the same color space, and the color space is used to define and describe different colors in different scenes.

4. A method for updating map information as claimed in claim 1, characterized in that: The method of using the Hausdorff distance to measure the third similarity between the center point position element in the element set of the landmark object and the center point position element in the map element set specifically comprises the following steps: Obtain outline box elements from the element set of each collected landmark object, and record them as set A={a1,a2,...an}; obtain outline box elements from the map element set of each landmark object in the map, and record them as set B={b1,b2,...bm}; Obtain the Hausdorff distance d3 between the contour box elements obtained in the element set of each collected landmark object and the contour box elements obtained in the map element set of each landmark object in the map: d3 = max(sup(inf d(a, b)), sup(inf d(b, a))); Where d(a,b) represents the Euclidean distance from point a to point b, a is each element in set A, b is each element in set B, sup represents the maximum value, and inf represents the minimum value; A Gaussian function is used for smooth mapping, and the Hausdorff distance d3 is normalized and mapped to the range of 0 to 1 to obtain the third similarity d3'.

5. A method for updating map information as claimed in claim 1, characterized in that: The specific steps of taking the intersection-and-union ratio of the outline frame elements in the element set of the road sign object and the outline frame elements in the map element set as the fourth similarity include: By using a computer graphics algorithm, a set of all pixel points within a contour frame obtained from an element set of each landmark object collected is determined, and a set of all pixel points within a contour frame obtained from a map element set of each landmark object in the map is determined, and the sets are merged into a pixel point set U; the number of pixel points in the pixel point set U is obtained as the area |U|; Determine the intersection of the set of all pixel points within the contour frame obtained in the element set of each collected landmark object and the set of all pixel points within the contour frame obtained in the map element set of each landmark object in the map, and record it as I; obtain the number of pixel points in the intersection I as the area |I|; Get the intersection-over-union (IOU) as the fourth similarity: IOU = |I| / |U|.

6. A method for updating map information as claimed in claim 1, characterized in that: The specific steps of determining the matching relationship between the element set of each collected landmark object and the map element set of each landmark object in the acquired map include: Weighting the first similarity, the second similarity, the third similarity, and the fourth similarity to obtain a matching metric; When the matching metric is greater than a preset threshold, it is determined that the difference between the matching metric and the preset threshold is greater than the preset difference, indicating that the element set of each acquired landmark object and the map element set of each landmark object in the acquired map are consistent elements; When the matching metric is greater than a preset threshold, it is determined that the difference between the matching metric and the preset threshold is less than the preset difference, indicating that the element set of each acquired landmark object and the map element set of each landmark object in the acquired map are changed elements; When the matching metric is less than a preset threshold, it means that the element set of each collected landmark object and the map element set of each landmark object in the acquired map are newly added elements.

7. A method for updating map information as claimed in claim 1, characterized in that: The specific steps of updating the map include: Acquire multiple collection results of each attribute of each road sign object according to multiple vehicles collecting multiple collection results of each road sign object in the same area; According to the multiple collection results of each attribute of each road sign object, the mean μ and the standard deviation σ are obtained using Gaussian distribution; According to the set confidence level percentage, the corresponding z value is obtained using the standard normal distribution table of Gaussian distribution; Get the confidence interval of each attribute of each road sign object, the confidence interval formula is: [μ-z*σ,μ+z*σ]; According to each attribute of each newly collected road sign object, the distances d1 and d2 between each attribute and the two endpoints in the confidence interval are obtained respectively. The derivation formulas of the distances d1 and d2 are as follows: d1=|x'-(μ-z*σ)|; d2=|x'-(μ+z*σ)|; Among them, x' represents the attribute; Select the distance with the smaller distance value from d1 and d2, and use normalization to normalize the distance with the smaller distance value to the interval [0,1] as the confidence of each attribute of each landmark object; If the relationship between the element set of each newly collected landmark object and the map element set of each landmark object in the acquired map is a consistent element, then according to the confidence of each attribute of each landmark object newly collected multiple times, obtain the average increment of multiple confidences; add the average increment to the confidence of each landmark object in the map until the set highest confidence threshold is reached; If it is a changing element, then according to the confidence of each attribute of each landmark object newly collected multiple times, obtain the average reduction of multiple confidences, and subtract the average reduction from the confidence of each landmark object in the map until the set minimum confidence threshold is reached; If it is a newly added element, each newly collected landmark object will be inserted into the map and the initial confidence will be set.

8. A map information updating system, characterized in that: include: A road sign object acquisition module is used to collect road sign objects around the current position of the vehicle and use multiple attributes of each road sign object as an element set of each road sign object; and obtaining a map of the surroundings of the current position of the vehicle, the map comprising a set of map elements for each road sign object composed of a plurality of attributes of each road sign object; The road sign objects include: traffic signs, buildings and lane lines; the attributes include: symbol, color, center point position and contour box representing the shape; A similarity analysis module is used to measure, according to each road sign object, a first similarity between symbol elements in an element set of the road sign object and symbol elements in a map element set using edit distance and cosine similarity; to measure a second similarity between color elements in an element set of the road sign object and color elements in a map element set using Euclidean distance; to measure a third similarity between center point position elements in an element set of the road sign object and center point position elements in a map element set using Hausdorff distance; and to use an intersection-over-union ratio of outline frame elements in an element set of the road sign object and outline frame elements in a map element set as a fourth similarity; A map update module is used to determine the matching relationship between the element set of each collected landmark object and the map element set of each landmark object in the acquired map according to the first similarity, the second similarity, the third similarity and the fourth similarity, wherein the matching relationship includes: consistent elements, changed elements and newly added elements; and determine whether to update the map according to the matching relationship.

Images