CN110008872B - Road network extraction method combining vehicle track and remote sensing image - Google Patents

Road network extraction method combining vehicle track and remote sensing image Download PDF

Info

Publication number
CN110008872B
CN110008872B CN201910229817.9A CN201910229817A CN110008872B CN 110008872 B CN110008872 B CN 110008872B CN 201910229817 A CN201910229817 A CN 201910229817A CN 110008872 B CN110008872 B CN 110008872B
Authority
CN
China
Prior art keywords
road network
points
connecting line
remote sensing
representative
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910229817.9A
Other languages
Chinese (zh)
Other versions
CN110008872A (en
Inventor
解利军
倪忠义
郑耀
陈建军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201910229817.9A priority Critical patent/CN110008872B/en
Publication of CN110008872A publication Critical patent/CN110008872A/en
Application granted granted Critical
Publication of CN110008872B publication Critical patent/CN110008872B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/182—Network patterns, e.g. roads or rivers

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a road network extraction method combining a vehicle track and a remote sensing image. The road network generated by the invention has the characteristics of wide coverage, high precision, high updating speed and the like.

Description

Road network extraction method combining vehicle track and remote sensing image
Technical Field
The invention relates to the field of data mining, image recognition and information visualization, in particular to a method for generating a road network based on vehicle track data and a high-definition remote sensing image.
Background
The street map and the traffic network are the basis for constructing the smart city, and the perfect and accurate road network map has great social and application values. Common algorithms for generating and updating road networks are mainly classified into three categories: firstly, field measurement is carried out based on professional GPS equipment and a surface measurement technology, the method depends on professional road measurement vehicles and data acquisition personnel, and the defects of long working period, unstable measurement precision, high cost and maintenance cost and the like exist; secondly, the road network map is extracted from the high-definition remote sensing image based on the image processing technology, but the method is limited by the existing image processing technology, and the method is difficult to realize automatic operation, so that the extraction efficiency is low. And thirdly, constructing a road network through spontaneous Geographic Information (abbreviated as VGI), wherein the VGI refers to a map created, edited, managed and maintained by a user in an online network cooperation manner, such as a wiki satellite map, an open street map and the like. The latest and detailed road information can be collected by using the method, but the map quality updated in the method depends on the skill level of the user and the accuracy of the data, so the method is not suitable for the situation that certain requirements are made on the accuracy of the road network.
Disclosure of Invention
The invention aims to provide a road network extraction method combining a vehicle track and a remote sensing image aiming at the defects of the prior art. The invention has low cost and high efficiency.
The purpose of the invention is realized by the following technical scheme: a road network extraction method combining vehicle tracks and remote sensing images comprises the following steps:
(1) acquiring vehicle track data and a high-definition remote sensing image;
(2) generating a primary road network using vehicle trajectory data, the step comprising the sub-steps of:
(2.1) data preprocessing: judging and processing the missing of vehicle track data, judging and preprocessing a noise point and judging and processing a stop point;
(2.2) representative point extraction: extracting representative points from the vehicle track data, wherein the representative points store position information in the vehicle track data, and each representative point represents the average position of a nearby sampling point;
(2.3) connecting line segment extraction: extracting a connecting line segment from the vehicle track data, wherein the connecting line segment stores time sequence information in the vehicle track data and is used for establishing a connection relation between the representative points;
(2.4) road network generation: and interpolating the connecting line segments by utilizing a Delaunay triangulation network and a Dijkstra algorithm to construct a primary road network.
(3) Identifying a road region in the high-definition remote sensing image by using a convolutional neural network, wherein the step comprises the following substeps:
and (3.1) selecting road network segments from the primary road network constructed in the step 2.4, and combining with a high-definition remote sensing image to construct a training data set. Expanding a training data set through horizontal overturning, random deduction and rotation operation;
(3.2) constructing a convolutional neural network for identifying a road area in the high-definition remote sensing image, and training the convolutional neural network by using the training data set expanded in the step 3.1 to obtain a convolutional neural network model with the area characteristic;
and (3.3) identifying the road area in the high-definition remote sensing image by using the convolutional neural network model obtained in the step 3.2 and combining a sliding window algorithm.
(4) And (3) drawing the primary road network constructed in the step (2.4) into an image, superposing the image and the road area identified in the step (3.3), and finally extracting a skeleton in the superposed image by using a skeleton extraction algorithm to serve as the road network.
Further, the representative points in step 2.2 are used to represent n position sampling points { p ] within a circle with radius r meters1,p2,…,pnMean position of the points represented by
Figure BDA0002005400040000021
Where r is the representative point representing the radius, n is the number of position sampling points represented by the representative point, p1,p2,…,pnFor the n position sampling points, the position of the sample,
Figure BDA0002005400040000022
for the position of the representative point, t is the time of the last sampling point for updating the position of the representative point, i.e. t equals max (t)1,t2,…,tn)，t1,t2,…,tnIs the sampling time of n position sample points.
Further, the connecting line segment in step 2.3 refers to a line segment connecting two representative points a and b, and is represented by L ═ a, b, n, where a and b represent the two representative points connected by the connecting line segment, and n represents the number of times the connecting line segment appears.
Further, in the step 2.4, the Delaunay triangulation network and the Dijkstra algorithm are used to interpolate the connecting line segment to construct the primary road network, the representative points in the step 2.2 are used to construct the Delaunay triangulation network, and then the Dijkstra algorithm is used to search the shortest path between the representative points at the two ends of the connecting line segment in the step 2.3, where the shortest path is the interpolation result of the connecting line segment, and the distance measurement formula between the two representative points is the α -th power of the euclidean distance between the two representative points.
Further, the size of the identification area of the convolutional neural network in the step 3.2 is set to be 30 × 30 pixels, the size of the image identified by the incoming convolutional neural network is 250 × 250 pixels, the identification area is in the right center of the incoming image, and the image around the identification area is used as the context feature for assisting in identification.
The method has the advantages that the primary road network is extracted by using vehicle track data, then a training set is constructed by using the primary road network and the high-definition remote sensing image and is used for training the convolutional neural network, the step of manually marking a training sample is omitted, finally the road areas identified by the primary road network and the convolutional neural network are overlapped, and a skeleton diagram is extracted to serve as the road network. The road network generation method provided by the invention can generate a road network with high precision, high coverage rate and high real-time property with lower cost.
Drawings
FIG. 1 is a road network generation flow diagram;
FIG. 2 is a schematic illustration of a partial vehicle trajectory missing;
FIG. 3 is a schematic illustration of noise points in a trace;
FIG. 4 is a schematic illustration of a dwell point in a trace;
FIG. 5 is a schematic diagram of a trajectory representing a point representation;
FIG. 6 is a schematic diagram of the construction of a Delaunay triangulation network using representative points;
FIG. 7 is a diagram illustrating interpolation results with side length as weight;
FIG. 8 is a schematic diagram of a trajectory-generating road network result;
FIG. 9 is a tag data diagram;
FIG. 10 is a sample data diagram corresponding to the tag data of FIG. 9;
FIG. 11 is a schematic diagram of a convolutional neural network architecture;
FIG. 12 is a schematic diagram of a sliding window algorithm;
FIG. 13 is a schematic diagram of a recognized high-definition remote sensing image;
fig. 14 is a diagram of a result of identifying the high-definition remote sensing image in fig. 13 using a convolutional neural network.
Detailed Description
The technical solution of the present invention will be further described with reference to the following embodiments.
The invention provides a road network extraction method combining vehicle tracks and remote sensing images, which mainly comprises the following four steps as shown in figure 1: 1. acquiring vehicle track data and a high-definition remote sensing image; 2. generating a primary road network using the vehicle trajectory data; 3. identifying a road area in the high-definition remote sensing image by using a convolutional neural network; 4. and drawing the primary road network into an image, overlapping the image with the identified road area, and finally extracting a skeleton in the overlapped graph by using a skeleton extraction algorithm to serve as the road network.
1. The method comprises the steps of obtaining vehicle track data and high-definition remote sensing images, obtaining the vehicle track data from a taxi company or a logistics company, and downloading the high-definition remote sensing images from map websites such as a hundred-degree satellite map.
2. Generating a primary road network using vehicle trajectory data, the step consisting essentially of the sub-steps of:
2.1 vehicle trajectory data preprocessing
Defining position sampling points: the position sampling point refers to a coordinate point of a vehicle position measured by vehicle-mounted positioning equipment in the driving process of a vehicle. Where x and y correspond to the longitude and latitude of the location of the vehicle, respectively, and are measured by the positioning device, and t represents the time when the vehicle is at the location.
Let pi，pi+1Is two adjacent position sampling points in a vehicle track, where pi＝{xi,yi,ti}，pi+1＝{xi+1,yi+1,ti+1}. The vehicle trajectory may be pre-processed when the following three conditions occur:
judging and processing missing of track data: the vehicle locating device signal is disturbed or interrupted, which results in the interruption of the vehicle trajectory as shown in fig. 2, and if p3 and p4 are directly connected, an error trajectory is formed, which results in the problem of subsequent road network extraction. The invention judges whether the input track has deficiency (| t) according to the time interval between two adjacent position sampling pointsi+1-ti|＞tmaxWherein t ismaxIndicating 5 times the sampling interval of the positioning device), for absenceThe more serious trace (the time interval between two position sampling points is more than 10 times of the set sampling time interval of the positioning equipment) is directly abandoned.
Judging and processing noise points: the interference of the positioning device signal can also cause the positioning positions of part of the position sampling points to deviate from the actual vehicle driving route, so as to form noise points, thereby greatly influencing the road network extraction result, as shown in fig. 3. The invention can calculate the average speed between two position sampling points according to the distance and the time interval between two adjacent position sampling points, then searches the noise point in the track according to the average speed and eliminates the noise point (the average speed between the two position sampling points)
Figure BDA0002005400040000041
Judging and processing a stop point: some positioning apparatus vehicles still maintain the working state when stopping, so that a large number of sampling points with the same (close) positions may appear in the track data within a period of time, and a stopping point is formed, as shown in fig. 4. A stagnation point in the vehicle trajectory may create a large amount of data redundancy, resulting in a significant reduction in program operating efficiency. For the dwell point in the track, the main characteristic is that the positions of a plurality of continuous position sampling points are not changed (or are changed less), the invention finds the dwell point in the track and deletes redundant points through the characteristic (the invention)
Figure BDA0002005400040000042
Wherein epsilondIndicating a positioning error of the positioning device).
2.2 representative Point extraction
Defining a representative point: for representing n position sampling points { p) within a circle of radius r meters1,p2,…,pnMean position of the points represented by
Figure BDA0002005400040000043
Where r is the representative point representing the radius, n is the number of position sampling points represented by the representative point, p1,p2，…,pnFor the n position sampling points, the position of the sample,
Figure BDA0002005400040000044
for the position of the representative point, t is the time of the last sampling point for updating the position of the representative point, i.e. t equals max (t)1,t2,…,tn)，t1,t2,…,tnIs the sampling time of n position sample points.
Process of extracting and updating representative points from vehicle trajectory data: first, let the representative point set rps be empty, for each position sampling point in the input track, { x, y, t }, searching for a representative point closest to the point and not more than r meters away from the point from the representative point set (the present invention implements searching for the nearest point through a K-d tree):
1) and if no representative point meeting the requirement exists, adding a representative point to the representative point set, wherein the added representative point is expressed as { x, y,1, t }.
2) If a representative point rp which is the closest to the point and less than r meters away from the point exists, the representative point rp is updated to be
Figure BDA0002005400040000051
And synchronizes the change in rp to the set of representative points.
2.3 connecting line segment extraction
Defining a connecting line segment: the connecting line segment refers to a line segment connecting two representative points a and b, and is represented by L ═ a, b, n, where a and b represent the two representative points connected by the connecting line segment, and n represents the number of times the connecting line segment appears.
A process of extracting a connecting line segment from the vehicle trajectory data: firstly making a connecting line segment set lines empty, and then representing the input track from an original position sampling point by { p } according to the result of the representative point extraction1,p2,…,pnConverting into a representative point representation { rp }1,rp2,…,rpnWhere rp1Is p1Is a representative point of, rp2Is p2… …, rpnIs pnRepresentative point of (a); and finally, according to the sequence of the representative points in the track, grouping every two points into one group, and sequentially adding the groups into a line set of connecting line segments, namely, the lines is [ (rp)1,rp2),(rp2,rp3),……,(rpn-1,rpn)]。
2.4 Primary road network Generation
The invention provides an interpolation algorithm based on Dijkstra algorithm and Delaunay triangulation network, which is used for realizing interpolation of connecting line segments by inserting representative points between longer connecting line segments and constructing a primary road network.
A distance measure between two representative points is defined: the distance between two representative points is a power of the euclidean distance between them.
It is assumed that a graph obtained by converting an input original trajectory into representative points is shown in fig. 5, where line segments in the graph are connecting line segments and points are representative points. A Delaunay triangulation network is constructed for all representative points in the graph, as shown in fig. 6.
For any connecting line segment, the representative points at the two ends of the connecting line segment are taken as a starting point and an end point, the Dijkstra algorithm is used for searching the shortest path between the starting point and the end point of all the connecting line segments in the graph 5 according to the defined distance measurement, and the shortest path is taken as the interpolation result of the connecting line segment. When alpha is larger than 1, the path selected by the interpolation algorithm is biased to the path containing more short line segments, and the trend is more obvious when alpha is larger. Note that when interpolating a connected line segment and dividing it into a plurality of shorter connected line segments, if these shorter connected line segments occur repeatedly, the number of repeated occurrences of the connected line segment needs to be recorded, and a connected line segment with a larger number of repeated occurrences indicates that the reliability of the line segment is higher. The final interpolation result is shown in fig. 7, where the representative points in the graph are sequentially connected along the direction of the road, and all the connected line segments correspond to one Delaunay edge, and the obtained result is highly similar to the shape of the road.
And (3) generating a road network by using the truck track of a certain urban partial area according to the method, dividing the road network into 4 grades of 1-3 times, 4-6 times, 7-9 times and more than 9 times according to the repeated occurrence times of the connecting line segments, and drawing the road network on the same picture from thin to thick. The results are shown in FIG. 8.
3. Identifying a road region in the high-definition remote sensing image by using a convolutional neural network, wherein the step comprises the following substeps:
3.1 training dataset construction
Selecting a road network map formed by connecting line segments with the repetition times higher than 9 times from the road network generated by the vehicle track data as a label data set (part of data is shown in figure 9), downloading a high-definition remote sensing image of a corresponding area from a hundred-degree satellite map website as a sample data set (part of data is shown in figure 10), and constructing a training data set for training the convolutional neural network by combining the label data set and the sample data set. Because the number of the training data sets obtained in the steps is limited, the training data sets need to be expanded, and the training data sets are expanded to be several times of the original data by performing horizontal overturning, random deduction and rotation operations on the training data sets.
3.2 construction and training of convolutional neural networks
In the process of road identification, a lot of interference may exist in input data, for example, trees growing on both sides of a road block part of the road, which may cause the accuracy of a convolutional neural network to be reduced; furthermore, areas such as roofs, parking lots, and railways are very similar to roads, and even human identification can be erroneous. It is necessary to classify images in conjunction with the contextual characteristics of the classified images when classifying data input. Thus, the present invention sets the identification area to a size of 30 x 30 pixels, while the size of the image identified by the incoming convolutional neural network is 250 x 250 pixels. The identification area is positioned in the center of the input image, and images around the identification area are used as context features for assisting in identification.
The convolutional neural network structure is constructed as shown in fig. 11, and then a convolutional neural network model with regional characteristics for the city is trained by using the training data set.
3.3 road region identification
The input of the convolutional neural network designed by the invention is an image with the size of 250 multiplied by 250 pixels, but in general, the size of a high-definition remote sensing image to be recognized far exceeds the value, so that a sliding window method is needed to traverse the whole input image, recognize the image area corresponding to each window, and then combine the image into a complete image. A schematic diagram of the sliding window algorithm is shown in fig. 12. The three smaller squares in the figure are 30 x 30 pixels in size, representing the identification area size of the convolutional neural network, while the three larger squares are 250 x 250 pixels in size, representing the image size input to the convolutional neural network, with a step size of one identification area width per window sliding.
The remote sensing image shown in fig. 13 is identified by using a sliding window algorithm, and the identification result is shown in fig. 14, where the darker area in the image is a road area identified by the convolutional neural network, and the lighter area is a non-road area identified by the convolutional neural network.
4. Road network synthesis
Although the coverage of the road area extracted from the high-definition remote sensing image by using the convolutional neural network is wide, a part of the road is not correctly identified, or some areas are wrongly identified as the road area, and a series of intermittent small areas exist. Although the coverage area of the primary road network extracted by using the vehicle track data is smaller than the road area identified by the convolutional neural network, a part of the coverage area is not correctly identified by the convolutional neural network, so that a road network with a wider coverage area can be obtained by overlapping the road areas identified by the primary road network and the convolutional neural network.
Firstly, a primary road network extracted from vehicle track data is drawn into an image, and the thickness of each line segment is increased to make the line segment have the same width as the thinnest road in an actual road network. This image is then superimposed with the road region image identified by the convolutional neural network. At this time, there are still some small isolated regions in the superimposed image, and these regions are generally caused by the recognition error of the convolutional neural network. Normal roads should be connected with each other, so that these erroneously recognized regions can be removed by judging the size of the area (an area smaller than 1000 square meters is regarded as an erroneously recognized region). And finally, extracting the skeleton of the superposed image by using a skeleton extraction algorithm to form a more complete and accurate road network map.

Claims (4)

1.一种基于车辆轨迹数据和高清遥感图像的道路网络生成方法，其特征在于，包括如下步骤：1. a road network generation method based on vehicle trajectory data and high-definition remote sensing images, is characterized in that, comprises the steps: (1)获取车辆轨迹数据和高清遥感图像；(1) Obtain vehicle trajectory data and high-definition remote sensing images; (2)利用车辆轨迹数据生成初级道路网络，该步骤包括以下子步骤：(2) Using vehicle trajectory data to generate a primary road network, this step includes the following sub-steps: (2.1)数据预处理：包括车辆轨迹数据缺失的判断与处理、噪音点的判断预处理以及停留点的判断与处理；(2.1) Data preprocessing: including the judgment and processing of missing vehicle trajectory data, the judgment and preprocessing of noise points, and the judgment and processing of stop points; (2.2)代表点提取：从车辆轨迹数据中提取代表点，代表点保存了车辆轨迹数据中的位置信息，每个代表点表示附近采样点的平均位置；(2.2) Representative point extraction: extract representative points from the vehicle trajectory data, the representative points save the position information in the vehicle trajectory data, and each representative point represents the average position of the nearby sampling points; (2.3)连接线段提取：从车辆轨迹数据中提取连接线段，连接线段保存了车辆轨迹数据中的时序信息，连接线段用于建立代表点之间的连接关系；(2.3) Extraction of connecting line segments: extracting connecting line segments from the vehicle trajectory data, the connecting line segments save the time sequence information in the vehicle trajectory data, and the connecting line segments are used to establish the connection relationship between representative points; (2.4)道路网络生成：利用Delaunay三角网和Dijkstra算法对连接线段进行插值，构建初级道路网络；(2.4) Road network generation: use Delaunay triangulation and Dijkstra algorithm to interpolate connecting line segments to construct a primary road network; (3)利用卷积神经网络识别高清遥感图像中的道路区域，该步骤包括以下子步骤：(3) Identify road areas in high-definition remote sensing images using convolutional neural networks, and this step includes the following sub-steps: (3.1)从步骤(2.4)构建的初级道路网络中挑选道路网络片段，结合高清遥感图像，构造训练数据集；通过水平翻转、随机抽取以及旋转操作扩充训练数据集；(3.1) Select road network segments from the primary road network constructed in step (2.4), combine high-definition remote sensing images, and construct a training data set; expand the training data set through horizontal flipping, random extraction, and rotation operations; (3.2)构建一个用于识别高清遥感图像中的道路区域的卷积神经网络，并使用步骤(3.1)扩充的训练数据集训练该卷积神经网络，得到一个具有区域特色的卷积神经网络模型；(3.2) Construct a convolutional neural network for identifying road areas in high-definition remote sensing images, and train the convolutional neural network using the training dataset expanded in step (3.1) to obtain a convolutional neural network model with regional characteristics ; (3.3)使用步骤(3.2)获得的卷积神经网络模型，结合滑动窗口算法对高清遥感图像中的道路区域进行识别；(3.3) using the convolutional neural network model obtained in step (3.2), combined with the sliding window algorithm to identify the road area in the high-definition remote sensing image; (4)将步骤(2.4)构建的初级道路网络绘制成图像，再将该图像与步骤(3.3)识别出的道路区域叠加，最后利用骨架提取算法提取出叠加图形中的骨架作为道路网络；(4) Draw the primary road network constructed in step (2.4) into an image, then superimpose the image with the road area identified in step (3.3), and finally use the skeleton extraction algorithm to extract the skeleton in the superimposed graph as the road network; 所述步骤(2.4)中利用Delaunay三角网和Dijkstra算法对连接线段进行插值，构建初级道路网络，首先使用步骤(2.2)中的代表点构建Delaunay三角网，然后使用Dijkstra算法搜索步骤(2.3)中的连接线段两端的代表点之间的最短路径，该最短路径就是连接线段的插值结果，此时两个代表点之间的距离为两个代表点之间的欧式距离的α次方。In the step (2.4), the Delaunay triangulation network and the Dijkstra algorithm are used to interpolate the connecting line segments to construct the primary road network. First, the Delaunay triangulation network is constructed using the representative points in the step (2.2), and then the Dijkstra algorithm is used to search the step (2.3). The shortest path between the representative points at both ends of the connecting line segment, the shortest path is the interpolation result of the connecting line segment, and the distance between the two representative points is the α power of the Euclidean distance between the two representative points. 2.根据权利要求1所述的基于车辆轨迹数据和高清遥感图像的道路网络生成方法，其特征在于：所述步骤(2.2)中代表点用于表示半径为r米的圆内的n个位置采样点{p1，p2，…，pn}的平均位置，该代表点表示为
Figure FDA0002858577200000011
其中r为代表点表示半径，n为代表点表示的位置采样点的数量，p1，p2，…，pn为n个位置采样点，
Figure FDA0002858577200000012
Figure FDA0002858577200000013
为该代表点的位置，t是最后一个更新该代表点位置的位置采样点的时间，即t＝max(t1，t2，…，tn)，t1，t2，…，tn是n个位置采样点的采样时间。2. the road network generation method based on vehicle trajectory data and high-definition remote sensing images according to claim 1, is characterized in that: in described step (2.2), the representative point is used to represent the n position in the circle that radius is r meters The average position of the sampling points {p 1 , p 2 , ..., p n }, the representative point is expressed as
Figure FDA0002858577200000011
where r is the radius of the representative point, n is the number of position sampling points represented by the representative point, p 1 , p 2 , ..., p n is the n position sampling points,
Figure FDA0002858577200000012
Figure FDA0002858577200000013
is the position of the representative point, t is the time of the last position sampling point to update the position of the representative point, that is, t=max(t 1 , t 2 ,..., t n ), t 1 , t 2 ,..., t n is the sampling time of n location sampling points. 3.根据权利要求1所述的基于车辆轨迹数据和高清遥感图像的道路网络生成方法，其特征在于：所述步骤(2.3)中连接线段是指连接两个代表点a，b之间的线段，表示为L＝{a，b，n}，其中a，b表示被连接线段连接的两个代表点，n表示该连接线段出现的次数。3. the road network generation method based on vehicle trajectory data and high-definition remote sensing images according to claim 1, is characterized in that: in described step (2.3), connecting line segment refers to connecting two representative points a, the line segment between b , expressed as L={a, b, n}, where a, b represent two representative points connected by a connecting line segment, and n represents the number of occurrences of the connecting line segment. 4.根据权利要求1所述的基于车辆轨迹数据和高清遥感图像的道路网络生成方法，其特征在于：所述步骤(3.2)中卷积神经网络的识别区域大小设置为30×30像素，而传入卷积神经网络识别的图像大小为250×250像素，识别区域在传入图像的正中心，识别区域周围的图像作为上下文特征辅助识别。4. The road network generation method based on vehicle trajectory data and high-definition remote sensing images according to claim 1, characterized in that: in the step (3.2), the size of the recognition area of the convolutional neural network is set to 30 × 30 pixels, and The size of the image recognized by the incoming convolutional neural network is 250 × 250 pixels, the recognition area is in the center of the incoming image, and the images around the recognized area are used as contextual features to assist in the recognition.
CN201910229817.9A 2019-03-25 2019-03-25 Road network extraction method combining vehicle track and remote sensing image Active CN110008872B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910229817.9A CN110008872B (en) 2019-03-25 2019-03-25 Road network extraction method combining vehicle track and remote sensing image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910229817.9A CN110008872B (en) 2019-03-25 2019-03-25 Road network extraction method combining vehicle track and remote sensing image

Publications (2)

Publication Number Publication Date
CN110008872A CN110008872A (en) 2019-07-12
CN110008872B true CN110008872B (en) 2021-04-06

Family

ID=67168060

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910229817.9A Active CN110008872B (en) 2019-03-25 2019-03-25 Road network extraction method combining vehicle track and remote sensing image

Country Status (1)

Country Link
CN (1) CN110008872B (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110852342B (en) * 2019-09-26 2020-11-24 京东城市(北京)数字科技有限公司 Road network data acquisition method, device, equipment and computer storage medium
CN110990661B (en) * 2019-10-23 2023-06-30 南通大学 A Method of Extracting Entrances and Exits of Regions of Interest Based on Road Network Constraints and Density Clustering
CN110686678A (en) * 2019-10-23 2020-01-14 众虎物联网(广州)有限公司 A method and device for generating road network based on electromagnetic fingerprint collection path
CN111881370B (en) * 2020-05-21 2024-12-31 滴图(北京)科技有限公司 A method and system for delineating the outline of an area of interest
CN113379754A (en) * 2020-12-02 2021-09-10 哈尔滨理工大学 Road center line extraction method based on vehicle-mounted GPS data and neural network
CN113190538A (en) * 2021-03-31 2021-07-30 北京中交兴路信息科技有限公司 Road construction method and device based on track data, storage medium and terminal
CN114383600B (en) * 2022-03-23 2022-07-19 北京百度网讯科技有限公司 Processing method, device, electronic device and storage medium for map
CN114943842A (en) * 2022-04-18 2022-08-26 北京纵目安驰智能科技有限公司 Road network extraction method, system, medium and terminal based on vehicle track data
CN114911887B (en) * 2022-04-22 2023-06-23 阿波罗智联(北京)科技有限公司 Data processing method, device, equipment and storage medium
CN114972988B (en) 2022-05-16 2024-03-26 北京百度网讯科技有限公司 Road network extraction method, device, equipment and storage medium
CN115984279B (en) * 2023-03-20 2023-11-17 银河航天(北京)网络技术有限公司 Path determination method, system, electronic device and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104554258A (en) * 2013-10-28 2015-04-29 通用汽车环球科技运作有限责任公司 Path planning for evasive steering manuever employing virtual potential field technique
CN106197419A (en) * 2016-06-23 2016-12-07 北京建筑大学 Take barrier and the indoor navigation method of subspace subdivision and system into account
CN108920481A (en) * 2018-04-20 2018-11-30 中国地质大学（武汉） A kind of road network method for reconstructing and system based on mobile phone location data

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090140887A1 (en) * 2007-11-29 2009-06-04 Breed David S Mapping Techniques Using Probe Vehicles
EP3319568B1 (en) * 2015-07-08 2022-02-23 AMO Development, LLC Laser surgical systems with laser scan location verification
US9767565B2 (en) * 2015-08-26 2017-09-19 Digitalglobe, Inc. Synthesizing training data for broad area geospatial object detection
CN105787937B (en) * 2016-02-25 2019-03-01 武汉大学 A kind of high resolution remote sensing image road change detecting method based on OSM
CN106525057A (en) * 2016-10-26 2017-03-22 陈曦 Generation system for high-precision road map
CN108615452B (en) * 2018-04-03 2019-08-20 华南理工大学 An Unknown Road Extraction Method Based on Multi-resolution Processing of Trajectory Points of People and Vehicles
CN108830209B (en) * 2018-06-08 2021-12-17 西安电子科技大学 Remote sensing image road extraction method based on generation countermeasure network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104554258A (en) * 2013-10-28 2015-04-29 通用汽车环球科技运作有限责任公司 Path planning for evasive steering manuever employing virtual potential field technique
CN106197419A (en) * 2016-06-23 2016-12-07 北京建筑大学 Take barrier and the indoor navigation method of subspace subdivision and system into account
CN108920481A (en) * 2018-04-20 2018-11-30 中国地质大学（武汉） A kind of road network method for reconstructing and system based on mobile phone location data

Also Published As

Publication number Publication date
CN110008872A (en) 2019-07-12

Similar Documents

Publication Publication Date Title
CN110008872B (en) Road network extraction method combining vehicle track and remote sensing image
CN107766808B (en) Method and system for clustering of moving trajectories of vehicle objects in road network space
CN113450596B (en) Spatio-temporal DP method based on feature point extraction of ship trajectory
CN108802776B (en) Bus GPS (global positioning system) deviation rectifying method based on abnormal point elimination and track compression algorithm
CN109410586B (en) Traffic state detection method based on multi-metadata fusion
CN108286979B (en) A method and device for acquiring high-precision navigation path data, as well as a navigation system and a driving system
CN105957342B (en) Track grade road plotting method and system based on crowdsourcing space-time big data
Chen et al. Roads digital map generation with multi-track GPS data
CN106023589B (en) A kind of track of vehicle reconstructing method based on bayonet socket data
CN109215347A (en) A kind of traffic data quality control method based on crowdsourcing track data
CN106323301B (en) A method and device for obtaining road information
Li et al. Integrated use of spatial and semantic relationships for extracting road networks from floating car data
CN101290725A (en) Digital road map generation method and map generation system
CN112530158B (en) Road network supplementing method based on historical track
CN113392169A (en) High-precision map updating method and device and server
CN115481207B (en) Trajectory reconstruction method based on floating vehicle satellite positioning data
CN108253976A (en) It is a kind of fully by the three stage Online Map matching algorithms in vehicle course
CN104422451A (en) Road recognition method and road recognition apparatus
JP4584344B1 (en) Traveling link determination device and link travel time calculation device
CN114037948B (en) Vehicle-mounted road point cloud element vectorization method and device based on transfer active learning
CN114707616A (en) Method, device and equipment for identifying incidental relationship between tracks
CN116935065A (en) Lane line instance detection method and system based on fusing and fusion
CN114462609A (en) A Trajectory Restoration Method for Floating Car Data Based on Hidden Markov Model
CN108615452B (en) An Unknown Road Extraction Method Based on Multi-resolution Processing of Trajectory Points of People and Vehicles
CN117741717A (en) Detection and reduction of GPS bias

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant