TW202115616A - Map data positioning system and method using roadside feature recognition having the advantages of low data volume, low computational complexity, high reliability - Google Patents

Abstract

The present invention provides a map data positioning system and method using roadside feature recognition, in which a road image map that includes a plurality of feature points is constructed by looking down a road for measurement, and a surrounding trip environment of a mobile vehicle in travel is detected to obtain a point cloud image, in which identification is made as to whether such feature points are included or not, and dynamic objects are filtered out in order to build a positioning map based on a road image map, remaining feature points in the point cloud image and characteristics of a plurality of roadside feature points. When a mobile vehicle is being driven, at least two roadside feature points ahead the vehicle are determined according to the positioning map to serve as reference points for calculation of a heading angle of the mobile vehicle in order to calculate the position of the mobile vehicle. The positioning map built by the present invention has the advantages of low data volume, low computational complexity, and high reliability, and can be used for automatic driving mobile vehicles to estimate high precision positions.

Description

Map data positioning system and method using roadside feature identification

The present invention relates to a positioning technology, in particular to a map data positioning system and method using roadside feature recognition.

Self-driving vehicles (referred to as self-driving vehicles) are also called unmanned vehicles. They can sense the surrounding environment and navigate without human operation, and can sense the surrounding environment with technologies such as radar, optical radar, satellite navigation, and computer vision. The advanced control system can convert the sensing data into appropriate navigation roads, obstacles and related signs.

Common positioning methods for self-driving cars include triangulation, Simultaneous Localization and Mapping (SLAM) positioning, Tag positioning and Fingerprint Based Map. Among them, the triangulation method needs to measure the distance between the target and three known position reference points, and find the intersection point of the circle with the three reference points as the center, but its disadvantage is that it requires more than three reference points, and There is no heading information, and the positioning accuracy is low; the synchronous positioning and map construction positioning method uses Lidar to scan the point cloud of the driving path, and then use the point cloud to compare the opponent method to estimate the position of the vehicle. However, the establishment of the point cloud is time-consuming and data Large amount of data, about 150MB of data per 1 km, and in an environment with few point cloud features, it will not be able to locate, and it is necessary to correct the absolute heading of the vehicle through the differential global positioning system (DGPS) and the vehicle steering dynamic model; tag positioning method Using the principle of trigonometric function, the optical radar scans the label of the known point, and then the vehicle position is reversed. For example, the coordinates of the bus stop plate are known as (x, y), the distance between the vehicle and the bus stop plate is d, and the angle is θ , The position of the vehicle is (x-dsinθ, y-dcosθ), but this technology also needs to correct the absolute heading of the vehicle through the differential global positioning system (DGPS) and the vehicle steering dynamic model, and the label deployment is not easy, and it is easily affected by road trees. , Pedestrians or other obstacles; the fingerprint-based map positioning method first scans the point cloud of the driving path with the optical radar of the first car, and compares the point cloud with the second car to estimate the position of the vehicle, but the establishment of the point cloud is time-consuming Although the amount of data is less than the synchronous positioning and map construction positioning method, the data must be coded and calculated one by one, which requires a large amount of calculation, and it also has the problem of inability to locate in an environment with few point cloud features.

Therefore, the present invention proposes a map data positioning system and method using roadside feature recognition to effectively solve the above-mentioned problems. The specific architecture and implementation methods will be described in detail below:

The main purpose of the present invention is to provide a map data positioning system and method using roadside feature recognition, which respectively obtain a road image map overlooking the road from the top of the road, and a point cloud image of the driving environment from a plane, using the technology of information space overlay , Quickly distinguish the road space and the roadside space from the road image map, and obtain the spatial information of specific object categories to filter out unwanted dynamic objects and retain static objects that can be used as roadside feature points to create high precision and Location map data with small amount of data.

Another object of the present invention is to provide a map data positioning system and method using roadside feature recognition, which uses an aerial camera to obtain a road image map. With a high-resolution camera, only low-cost aerial images can be obtained. High-precision road map.

Another object of the present invention is to provide a map data positioning system and method using roadside feature recognition, which further uses roadside feature points as reference points to calculate the heading angle of the mobile vehicle, and more accurately locate the position of the mobile vehicle .

To achieve the above object, the present invention provides a map data positioning method using roadside feature recognition, including: using at least one first detector to look down on the road for measurement, and building a road image map, the road image map including multiple features Point; install at least one second detector on at least one mobile vehicle, detect the surrounding driving environment when the mobile vehicle travels to obtain a point cloud image, identify whether it contains these feature points, and compare the feature points At least one dynamic object is filtered out, based on the road image map, the remaining feature points in the point cloud, and the feature attributes of multiple roadside feature points are set to build a positioning map; the positioning map is stored in In a mobile vehicle, when the mobile vehicle is traveling, one of the map data positioning systems in the mobile vehicle is used to scan the road ahead, and at least two roadside feature points ahead are determined based on the positioning map data, and used as The reference point calculates a heading angle of a mobile vehicle; and the heading angle of the mobile vehicle and the at least two reference points are used to calculate the position of the mobile vehicle.

According to the embodiment of the present invention, the method for constructing the location map data further includes the following steps: superimposing the road image map with the point cloud map to identify a road space and at least one side space; among the feature points The dynamic objects are filtered out, and a plurality of static objects are retained as the roadside feature points; the feature attributes of the roadside feature points are set; and based on the road image map and one of the superimposed maps of the point cloud, the The roadside feature points and these feature attributes are used to build the location map.

In addition, the roadside space divides the sidewalk, bicycle lane and/or arcade into the first and second roadside spaces from the inside out. The characteristic attributes include latitude and longitude coordinates, shape, size, height, and so on.

According to an embodiment of the present invention, the mobile vehicle captures a roadside image during driving, identifies at least one target object from it, and determines whether the target object is the roadside feature point by the feature attributes in the positioning map .

The present invention also provides a map data positioning system, which is installed in an on-vehicle system of a mobile vehicle, and uses the positioning map data built by the above method to perform mobile vehicle positioning. The map data positioning system includes: The database stores the location map data. The location map data includes multiple roadside feature points and the multiple feature attributes of the roadside feature points; the road side feature recognition module scans the road ahead according to the location map data and judges that it matches At least two of the characteristic attributes of the roadside feature points; a mobile vehicle heading angle estimation module that uses at least two of the roadside feature points as reference points to calculate the heading angle of one of the mobile vehicles; and The mobile vehicle position estimation module uses the heading angle of the mobile vehicle and the at least two reference points to calculate the position of the mobile vehicle.

The present invention provides a map data positioning system and method using roadside feature recognition, which captures a high-precision road image map from the air, and then overlays it on the point cloud map around the moving vehicle to quickly locate the road and the roadside area , And classify the dynamic objects and static objects in the image, delete the dynamic objects and keep only the static objects, not only can greatly reduce the data amount of the positioning map, but also only need two reference points to calculate the position, no need to use The triangulation method greatly reduces the computational complexity. When applied to the positioning of mobile vehicles for autonomous driving, the accuracy can reach centimeters. Compared with the accuracy of general satellite positioning, the accuracy is still within an acceptable range even if the error is 1 to 2 meters. The map data positioning method of the present invention can obviously ensure the accuracy and safety of the autonomous vehicle.

Please refer to Figure 1, which is a flowchart of the map data positioning method using roadside feature recognition of the present invention, which mainly includes four major steps. In step S10, the positioning map data is first constructed for use by mobile vehicles (such as autonomous vehicles); Step S12 is to identify the roadside feature points when the mobile vehicle is actually traveling; step S14 starts to correct the position of the mobile vehicle, estimate the heading angle of the mobile vehicle, and then calculate the position of the mobile vehicle in step S16. The detailed process is detailed below.

Figure 2 is a detailed flowchart of the establishment of the positioning map in the present invention. First, step S102 uses at least one first detector to look down the road from the top of the road for measurement to build a road image map. The first detector can be an aircraft equipped with an image capture device, such as an aerial camera. , Drones, remote-controlled aircraft, etc., the image capture device is a camera or video camera. As long as a high-resolution image capture device is installed on the aircraft, high-precision images can be captured. Therefore, the road image map includes multiple feature points. Such as dynamic objects such as vehicles and pedestrians and static objects such as traffic lights, stop signs, signs, buildings, and traffic signs. Among them, the judgment of dynamic objects is to presuppose that vehicles on the road space and pedestrians and moving objects on the roadside space are dynamic Objects, and delete vehicles and pedestrians in the road image map; step S104 install at least one second detector on at least one mobile vehicle, the second detector can be an optical radar (Lidar) or a camera, a camera Three-dimensional imaging technology can be used to generate three-dimensional images, and the moving vehicle can be a car. When the moving vehicle is moving, the second detector detects the driving environment around the moving vehicle and creates a point cloud on the scanned object surface. These point clouds are used to express the surface shape of the object. The higher the density of the point cloud, the more accurate the model can be established, and a deep, three-dimensional point cloud (point cloud) can be obtained. The information contained in it is the geometric information of the object, which can be identified. Whether these feature points are included; then, as described in step S106, the road image map and the point cloud map are superimposed using the information space to identify and classify the road space and the roadside space. The roadside space is defined widely and can be defined by In addition, the sidewalks, bicycle lanes and/or arcades are divided into first and second roadside spaces; in step S107, dynamic objects in the feature points are filtered out, and only static objects are retained as roadside feature points, and step S108 sets a plural number The feature attributes of the roadside feature points include the latitude and longitude coordinates, shape, size, height, etc. of the roadside feature points. Finally, in step S109, according to the superimposed map of the road image map and point cloud image, the remaining roadside feature points in the figure (static objects) ) And feature attributes of roadside feature points to build a location map.

When there is no static object in the roadside space, it means that there is no roadside feature point, and the roadside space can be deleted directly, leaving only road space, which will further reduce the amount of data for positioning maps.

Please refer to FIG. 3A to FIG. 3D, which are schematic diagrams of the process of establishing location map data in the present invention. Figure 3A is a high-precision road image map taken from above in the present invention. From the top, you can see which parts are roads and which parts are not roads (such as buildings, parks, parking lots, etc.). Figure 3B shows movement by reason The 3D point cloud image drawn by vehicle detection can identify roads, vehicles, pedestrians, buildings, traffic lights, stop signs, signboards, traffic signs and other characteristic points through the position of the mobile vehicle and high-precision road geometric spatial information; After the point cloud image of Figure 3B is overlaid on the road image map of Figure 3A, the overlay image of Figure 3C can be obtained, and the road space 10 and at least one side space 12, 14 can be classified on the overlay image, such as the first The roadside space 12 can be a bicycle lane, the second roadside space 14 is a sidewalk, or the first roadside space 12 is a sidewalk, and the second roadside space 14 is an arcade and a building. Dynamic objects such as pedestrians cannot be used as roadside feature points, so they are deleted. If there is no static object in the second roadside space 14, this roadside space can also be deleted. The final location map data is as shown in the 3D diagram. The roadside feature points in the road space include traffic lights, and the roadside feature points in the roadside space include landmarks such as buildings, electrical towers, and traffic signs, but this is only one implementation. For example, any characteristic object that can be used as a landmark or a feature point can be used as a roadside feature point, such as a signboard of a convenience store or fast food store, a signboard of a gas station, etc.

The feature attributes of the roadside feature points will be set according to different objects, such as the size, height and shape of the red street light, the size, height and shape of the bus stop sign, the size, height and shape of the store sign, etc., which are recorded in the location map one by one Zizhong.

When the positioning map data is created, it will be stored in a map data positioning system on the cloud platform or mobile vehicle. The map data positioning system can periodically update the latest information from the cloud. This map data positioning system can be installed in one of the mobile vehicles' on-board systems, and output the position information of the mobile vehicles after calculation. As shown in Figure 4, it is a structural diagram of the map data positioning system 22 of the present invention, which includes a database 222, a side feature identification module 224, a mobile vehicle heading angle estimation module 226, and a mobile vehicle The position estimation module 228, in which the database 222 stores location map data, the location map data includes multiple roadside feature points and multiple feature attributes of the roadside feature points; the environment sensing device 20 installed on the mobile vehicle scans the road ahead, The scan result is sent to the roadside feature identification module 224, and the roadside feature identification module 224 then determines whether there are feature points that meet the feature attributes in the scanned image according to the positioning map data. If there are at least two matching roadside feature points, then Use it as a reference point; the mobile vehicle heading angle estimation module 226 uses the reference point to calculate the heading angle of one of the mobile vehicles; the mobile vehicle position estimation module 228 then uses the mobile vehicle heading angle and reference point Calculate the position of the mobile vehicle.

In step S12 of Figure 1, the detailed flow chart of identifying roadside features when the mobile vehicle is traveling is shown in Figure 5. The environment sensing device 20 installed on the mobile vehicle can be a camera, a video camera or an optical radar. Capture roadside images while driving, use the processor of the on-board system to recognize at least one target from the image recognition processing technology, and determine whether the target is a roadside feature point by the feature attribute in the positioning map, which The judging method includes: in step S122, it is determined whether the target object meets the size of the roadside feature point, if so, then step S124 is followed to determine whether the target object meets the shape of the roadside feature point, and if so, then step S126 is followed to determine whether the target object meets the roadside feature point. The height of the feature point, if it is, then as described in step S128, the target object meets a certain roadside feature point, such as a traffic light; on the contrary, if any of the above judgments is negative, it means that the target object does not belong to any kind of roadside feature Click to end the judgment immediately, as in step S129.

其中，

，

。 同理，給予另一個路側特徵點的座標為(x₀ ,y₀ )，則移動載具的座標計算如下：

由於

且

，綜合上式後可描述為

，其中

，

，且

因此可得到

。When the direction of the head of the moving vehicle is not parallel to the road, the position of the moving vehicle in the next second will be far away from the straight forward position. Therefore, for precise positioning in the present invention, it is added as described in step S14 in Figure 1 The technology of calculating the heading angle of the mobile vehicle is described, and Figure 6 is a schematic diagram of calculating the heading angle and position of the mobile vehicle in the present invention. When the mobile vehicle is driving, the processor of the on-board system judges at least two road side feature points ahead according to the positioning map data, and uses it as a reference point to calculate a heading angle of the mobile vehicle. Assuming that the coordinates of the mobile vehicle are (x _v , y _v ) and the coordinates of the roadside feature points are (x ₁ , y ₁ ), then

among them,

,

. In the same way, the coordinates given to another roadside feature point are (x ₀ ,y ₀ ), then the coordinates of the mobile vehicle are calculated as follows:

due to

And

, Can be described as

,among them

,

And

So you can get

.

且

的等式後會先解出

，便可得到車頭與直線前進的夾角，即為移動載具航向角度。藉由此計算方法，只需二個路側特徵點做為參考點便可，無須如三角定位法需得到三個參考點才能計算移動載具位置。Above

And

Will be solved first after the equation of

, The angle between the front of the vehicle and the straight forward movement is obtained, which is the heading angle of the mobile vehicle. With this calculation method, only two roadside feature points are needed as reference points, and there is no need to obtain three reference points as in the triangulation method to calculate the position of the mobile vehicle.

，

Then proceed to the calculation of the position of the mobile vehicle as described in step S16 in Figure 1. According to at least two mobile vehicles estimated from the above at least two reference points, the position of the mobile vehicle is calculated as follows:

,

In summary, the map data positioning system and method using roadside feature recognition provided by the present invention are made by using low-cost aerial images to capture high-precision road image maps and the surrounding road environment captured during travel. The point cloud image is used to superimpose the information space to classify the road space and the roadside space, dynamic objects and static objects, and delete the dynamic objects and the empty roadside space, which greatly reduces the amount of data, and only requires two roadside features Point as a reference point can calculate the heading angle and position of the mobile vehicle. The calculation complexity is low but the reliability is high. The satellite positioning system is not used but the accuracy can reach centimeters. It is suitable for the navigation and positioning of autonomous vehicles.

Only the above are only preferred embodiments of the present invention, and are not used to limit the scope of implementation of the present invention. Therefore, all equivalent changes or modifications made in accordance with the characteristics and spirit of the application scope of the present invention shall be included in the patent application scope of the present invention.

10: Road space 12: First roadside space 14: Second roadside space 20: Environmental sensing device 22: Map data positioning system 222: database 224: Roadside feature recognition module 226: Mobile vehicle heading angle estimation module 228: Mobile Vehicle Position Estimation Module

Figure 1 is a flow chart of the map data positioning method using roadside feature recognition according to the present invention; Figure 2 is a detailed flow chart of the establishment of location map data in the present invention; Figures 3A to 3D are schematic diagrams of the process of establishing location map data; Figure 4 is a block diagram of the China Map Information Positioning System of the present invention; Figure 5 is a flowchart of identifying roadside feature points using location map data in the present invention; Figure 6 is a schematic diagram of calculating the heading angle and position of the mobile vehicle in the present invention.

Claims (15)

A map data positioning method using roadside feature recognition includes the following steps: Use at least one first detector to look down on the road for measurement, and build a road image map, the road image map including a plurality of feature points; Install at least one second detector on at least one mobile vehicle, detect the surrounding driving environment when the mobile vehicle travels to obtain a cloud image, identify whether it contains these feature points, and compare one of the feature points At least one dynamic object is filtered out, based on the road image map, the remaining feature points in the point cloud, and the feature attributes of the multiple roadside feature points are set to create a positioning map; Store the location map data in the mobile vehicle, and when the mobile vehicle is driving, use one of the map data positioning systems in the mobile vehicle to scan the road ahead, and determine at least two roadside features based on the location map data Point and use it as a reference point to calculate the heading angle of a mobile vehicle; and Use the heading angle of the mobile vehicle and the at least two reference points to calculate the position of the mobile vehicle. The map data positioning method using roadside feature recognition as described in claim 1, wherein the first detector is an aircraft equipped with an image capture device. The map data positioning method using roadside feature identification as described in claim 2, wherein the aircraft is an aerial camera, an unmanned aerial vehicle, a remote-controlled aircraft, etc. The map data positioning method using roadside feature recognition as described in claim 1, wherein the second detector is an optical radar (Lidar), a laser, a camera, or a sonar, etc. The map data location method using roadside feature recognition as described in claim 1, wherein the method for establishing the location map data further includes the following steps: Superimpose the road image map and the point cloud map to identify a road space and at least one side space; Filter out the dynamic objects among the feature points, and retain plural static objects as the roadside feature points; Set the characteristic attributes of the roadside characteristic points; and According to the road image map and a superimposed map of the point cloud map, the roadside feature points and the feature attributes, the positioning map data is constructed. The map data positioning method using roadside feature recognition as described in claim 5, wherein the feature points include dynamic objects such as vehicles and pedestrians, and static objects such as traffic lights, stop signs, signs, buildings, and traffic signs. The map data positioning method using roadside feature recognition as described in claim 5, wherein the roadside space divides the sidewalk, bicycle lane and/or arcade into the first and second roadside spaces from the inside to the outside. The map data positioning method using roadside feature recognition as described in claim 5, wherein the feature attributes include latitude and longitude coordinates, shape, size, height, etc. The map data positioning method using roadside feature recognition as described in claim 8, wherein the mobile vehicle captures roadside images during driving, identifies at least one target from it, and uses the features in the positioning map The attribute determines whether the target is the roadside feature point. The map data positioning method using roadside feature identification according to claim 1, wherein the map data is stored in a cloud platform or a map data positioning system of the mobile vehicle, and the map data positioning system can be set in the One of the mobile vehicles in the on-board system. A map data positioning system, which is installed in an on-vehicle system of a mobile vehicle, uses the positioning map data built in request item 1 to perform mobile vehicle positioning, and the map data positioning system includes: A database storing the location map data, the location map data including multiple roadside feature points and complex feature attributes of the roadside feature points; The road side feature identification module scans the front of the road based on the location map and determines at least two road side feature points that meet the feature attributes; A mobile vehicle heading angle estimation module, using at least two roadside feature points as reference points to calculate the heading angle of one of the mobile vehicles; and A mobile vehicle position estimation module uses the heading angle of the mobile vehicle and the at least two reference points to calculate the position of the mobile vehicle. The map data positioning system according to claim 11, wherein the roadside feature points include static objects such as traffic lights, stop signs, signboards, buildings, and traffic signs. The image data positioning system according to claim 11, wherein the characteristic attributes include latitude and longitude coordinates, shape, size, height, and so on. The image data positioning system according to claim 11, wherein the mobile vehicle includes an environment sensing device for scanning the image of the road ahead. The image data positioning system according to claim 14, wherein the environment sensing device is a camera, a video camera or an optical radar.

Images