RU2784310C1 - Method for mapping of area for autonomous vehicles - Google Patents

Abstract

FIELD: car industry.

SUBSTANCE: invention relates to methods for mapping of an area for autonomous vehicles (hereinafter – AV). A global area map is formed and saved, images received from AV cameras are processed with isolation of linear objects of a roadbed, a map of signs of isolated objects is formed, a local map is formed, initial assessments of the probability of finding signs of isolated objects in global and local map cells are set, signs of isolated objects are recorded in corresponding global and local map cells, based on the specified points, lines are formed, for each local map cell, the probability of finding a sign of an isolated object of the roadbed is assessed, the resulting section of the local map is applied to a corresponding section of the global map with summation of assessments of the probability of finding a sign of an isolated object, taking into account the previous assessment, the resulting map is obtained by binarization of the resulting map cells by a threshold value of the assessment of the probability of finding a sign of an isolated object of the roadbed.

EFFECT: increase in the accuracy of mapping for AV.

2 cl, 1 dwg

Description

The method of terrain mapping for autonomous vehicles (ATS) refers to the navigation aids of highly automated ATS and is designed to automatically map the area with visual linear features of the roadway (road boundaries, road markings, etc.) recognized on images from ATS cameras , and is created for the deployment of automatic telephone exchanges in a new territory.

Mapping is a set of methods and algorithms for creating maps containing information about the environment. In robotics, digital maps that reflect the working environment of a robot are used to determine their own position in space, plan a path, and avoid collisions. The quality of the map directly affects the accuracy of solving the listed problems, so mapping is one of the fundamental problems of robotics.

Since the advent of driver assistance systems and autonomous vehicles, digital road maps have become an integral part of intelligent transportation systems. Moreover, there are a number of studies demonstrating that high-accuracy digital road maps are a key factor in the development of SAE Class 3-5 autonomous vehicles.

Despite the fact that the problem of mapping has been widely studied, there is only a small number of works devoted to the creation of high-precision maps for autonomous vehicles. Existing mapping methods require the use of highly accurate positioning systems and sensors, such as the Global Positioning System (GPS) using Real Time Kinematic (RTK) methods and laser rangefinders (LIDAR).

Thus, the task of developing road infrastructure mapping algorithms that do not require the use of expensive sensor equipment and allows creating and updating high-precision digital road maps is an urgent task. Algorithms that use cameras as a source of sensory data meet the specified criteria. Cameras have a number of advantages over other sensors: they are relatively cheap, provide high spatial resolution, and are not susceptible to interference due to their passive nature.

The following technical solutions are known from the art.

A known system for determining the actual parameters of the carriageway (patent RU 2683618).

According to the method described in the materials of this patent, the determination of the parameters of the roadway includes the following steps:

1. Separation of image colors with a threshold for yellow (in HSV space) and white (in RGB space).

2. Applying anti-aliasing (blurring the image to remove noise).

3. Convert image to grayscale color mode.

4. Application of the Canny operator.

5. Prepare area of interest.

6. Applying the Hough transform to extract elements from an image.

7. Drawing translucent lines on the initial image (marking lines on the video image).

This method allows you to recognize the markings applied on the roadway, however, it does not describe the method for compiling and checking the correctness of compiling a map of the passed road markings.

There is a known method for localization and construction of navigation maps of a mobile service robot (patent RU 2740229) operated indoors based on the correction of the current coordinates of the robot and the trajectory traveled using stationary reflective beacons with an RFID tag installed on each of the beacons.

According to this method, a submap of obstacles is built at a specific position of the robot and the layers of the built submap are updated using an RGB-D camera. The submap of obstacles is tied to the coordinate system of the room according to the passed trajectory. When the robot moves to a given threshold value, a new submap of obstacles is built and the layers of the constructed submap are updated. The obstacle submap is combined into an obstacle map after each new obstacle submap is constructed. Then a global map is built using the SLAM map data, the static map, and taking into account the obstacle map data.

The construction of an obstacle submap in a specific position of the robot and updating the layers of the constructed obstacle submap according to this method is carried out using data from an RGB-D camera, lidar and ultrasonic rangefinders, which imposes increased requirements on the computing power of the system implementing the method, as well as the involvement of expensive sensor equipment leads to to increase the cost of such a system.

There is a known method for determining the boundaries of the road, the shape and position of objects on the road according to patent RU 2571871, including obtaining grayscale image frames from a camera installed on a vehicle, in which: the boundaries of objects are formed by converting a halftone image into a binary image characterized by the first and second binary colors, indicating the boundaries of the objects of the halftone image in the first binary color; allocate the image of the road surface by filling the first solid color inside the closed contour of the road, limited by the first binary color; determining the contour of the left, right and upper boundaries of the image of the road surface; detecting and recognizing three-dimensional objects, the images of which are in contact with the image contour of the road surface, by analyzing the shape of the contour, the left, right and upper boundaries of the image of the road surface, and then determine the position of these objects on the road; select images of flat objects on the road surface image that are below the upper boundary of the road surface image contour and are limited by the first binary color by filling these objects with a second solid color and determine the shape and position of these objects on the road.

The presented method is intended only for recognizing the markings of the road and its boundaries, as well as determining the shape and position of objects on the road, but does not allow you to generate a map of the area for the automatic telephone exchange, that is, to map the area.

There is a known method for simultaneously creating a map with multiple lanes and localizing a vehicle on the generated map (US2018053060), which includes the following steps:

- detecting a plurality of lane markings in each of the images of the far front, middle forward, right, left and rear camera fields of view;

- identifying each of the plurality of lane markings using the control lane markings stored in the database;

combining the identified lane markings to create a global lane map having a plurality of contiguous lanes for vehicles;

- localization of the main vehicle within the main lane on the global map of lanes.

A known method of mapping the carriageway (US2021180980), which includes the following steps:

- detection of image data using the camera unit;

- determination of road sections in the data of the detected image using the control unit;

- converting the detected image data, including certain areas of the roadway, into a top view and projecting the converted image data onto a plane that is at the level of the surface of the roadway;

- adding sections of the roadway to the grid map on the plane;

- aligning at least two grid maps from two different travel routes relative to each other and combining them;

- creation of the final grid map based on combined and combined grid maps.

It is also noted that the control unit is configured to determine a probability value for each determined area of the roadway to validate the mapping being performed and, if the probability value is above a predetermined threshold, to mark the corresponding area of the roadway as a roadway.

For these technical solutions, it is characteristic that they do not take into account or do not fully take into account errors in the detection of linear objects of the roadway (road markings and road boundaries) from images received from ATS cameras, which reduces the accuracy of building roadway maps.

The source materials ["Method of Mapping the Roadway", US2021180980] indicate that it is possible to determine the probability value for each specific area of the roadway in order to verify the validity of the mapping being performed, while if the probability value is higher than a predetermined threshold, mark the corresponding area of the roadway parts as a roadway. However, the materials of this source do not contain information on how this process is implemented.

The method known from the source US2021180980, the technical essence is the closest to the claimed invention and can act as a prototype.

The task to be solved by the claimed invention is to create a method for accurate and reliable mapping of the terrain for the vehicle by recognizing linear objects of the roadway (road markings and road boundaries) from images received from the cameras of the vehicle, followed by the formation of a map of the roadway, taking into account errors detection of signs of linear objects of the roadway, and the implementation of this method is performed without the involvement of expensive sensor equipment.

The technical result of the claimed invention is to improve the accuracy of terrain mapping for automatic telephone exchanges.

The claimed technical result is achieved through the implementation of the method of mapping the area for the PBX, according to which a global map of the area is formed, divided into cells, the specified map is recorded in the memory of the mapping module of the on-board computer of the PBX, from the camera located on the PBX, an image stream arrives at the input of the on-board mapping module after that, the obtained images are processed with the selection of linear objects of the roadway, a map of features of the selected linear objects of the roadway is formed, a local map divided into cells is formed to account for errors in the detection of linear objects of the roadway, initial estimates of the probability of finding the features of the selected linear objects are set of the roadway in the cells of the global and local maps, and the probability estimate in the cells of the global map is a priori, the signs of the selected linear objects of the roadway are recorded in the corresponding cells of the global and local maps, taking into account the pre-known camera calibration and the current location of the PBX, while the signs of the selected linear objects enter the cells of the local map in the form of break points of the selected broken lines, form line points based on the specified points, evaluate for each cell of the local map the probability of finding the sign of the selected linear object of the roadway, the resulting section of the local map is superimposed on the corresponding section of the global map with the summation of the estimates of the probability of finding the sign of the selected linear object of the roadway, taking into account the previous estimate, the resulting map of the roadway is obtained by binarizing the cells of the received map according to the threshold value of the probability estimate of finding the sign of the selected linear object roadbed object.

In a preferred embodiment of the method of mapping terrain for vehicles, the recognition of linear objects of the roadway from images received from the camera is carried out by the windowed Hough transform.

Figure 1 shows an illustration of the execution of the steps of the method for taking into account errors in the detection of signs of linear objects of the roadway, where 1 is the formation of a section of a local map with estimates of the probability of finding a sign of a selected linear object in each cell, 2 ‒ accounting for spatial errors in the detection of signs of linear objects of the roadway, 3 - obtaining the resulting roadway map by combining the local map section with the corresponding section of the global map.

The method of mapping the area for the automatic telephone exchange is carried out as follows.

At the first stage of the implementation of the method, a global map of the area is preliminarily formed, divided into cells, and the specified map is recorded in the memory of the mapping module of the ATS on-board computer.

During the movement of the vehicle, the input of the mapping module of the on-board computer of the vehicle receives a stream of images of the roadway from the cameras placed on the vehicle, after which the received images are processed with the recognition of linear objects of the roadway, preferably by the Hough window transform.

Road markings and roadway boundaries act as linear objects of the roadway, and as a result of image processing in this way, a plurality of broken lines are obtained, which are presumably road markings and roadway boundaries recognized in the image.

Then, a map of features of the selected linear objects of the roadway is formed.

To take into account errors in the detection of linear objects of the roadway (figure 1), a local map (1) is formed, divided into cells, and initial estimates of the probability of finding signs of selected linear objects of the roadway in the cells of the global and local maps are set, and the probability estimate in the cells of the global map is a priori.

So, the values of the cells of the local map at the time of creation contain zero values, and before the start of mapping for each cell of the global map, the presence or absence of linear roadway objects in it is considered equally probable, that is, the probability of this event is 0.5.

At the next stage, the signs of the selected linear objects of the roadway are recorded in the corresponding cells of the global and local maps, taking into account the previously known camera calibration and the current location of the vehicle, while transferring from the coordinate system associated with the camera to the coordinate system associated with the map.

The features of the selected linear objects enter the cells of the local map in the form of break points of the recognized broken lines, find intermediate values for building lines, preferably using the Bresenham algorithm. In addition, the width of the detected broken lines increases in the direction perpendicular to the direction of each segment, and taking into account the scale of the map. This is necessary to take into account the width of the road markings, regulated by the GOST R 51256-2018 standard. Technical means of organizing traffic. Road marking. Classification. Technical requirements.

In the cells of the local map corresponding to the broken lines expanded in this way, the values of the probability of finding the signs of the selected linear objects of the roadway are recorded in the form of a logarithmic value of the ratio of probabilities:

, (1)

, (one)

– вероятность истинно-положительного срабатывания детектора линейных признаков объектов дорожного полотна. where

- the probability of a true-positive operation of the detector of linear signs of roadbed objects.

Cells of the image, the values of which remained zero, are assigned a value equal to

, (2)

, (2)

– вероятность истинно-отрицательного срабатывания детектора линейных признаков объектов дорожного полотна.where

is the probability of a true-negative response of the detector of linear signs of roadbed objects.

и

предварительно оценивают для применяемого детектора на основе экспериментальных данных.

and

pre-estimated for the applied detector based on experimental data.

At the next step, spatial detection errors (2) are taken into account. It is assumed that the detection noise of each broken line segment can be modeled by a two-dimensional normal distribution.

и значением стандартного отклонения

. To take into account the spatial noise of detection, the local probability map obtained at the previous step is convolved with a Gaussian kernel of size

and the standard deviation value

.

Further, the obtained section of the local map is superimposed on the corresponding section of the global map with the summation of the estimates of the probability of finding the sign of the selected linear object of the roadway, taking into account the previous estimate (3) according to the following:

, (3)

, (3)

– результирующая оценка вероятности нахождения признака выделенного линейного объекта дорожного полотна в

ячейке локальной карты на k-ом шаге;where

is the resulting estimate of the probability of finding the sign of the selected linear object of the roadway in

cell of the local map at the kth step;

− оценка вероятности нахождения признака выделенного линейного объекта дорожного полотна в

ячейке локальной карты на k-ом шаге;

− estimation of the probability of finding a sign of a selected linear object of the roadway in

cell of the local map at the kth step;

− оценка вероятности нахождения признака выделенного линейного объекта дорожного полотна в

ячейке локальной карты на (k-1)-ом шаге;

− estimation of the probability of finding a sign of a selected linear object of the roadway in

cell of the local map at the (k-1)th step;

– априорная оценка вероятности нахождения признака выделенного линейного объекта дорожного полотна в

ячейке глобальной карты.

- a priori estimate of the probability of finding a sign of a selected linear object of the roadway in

cell of the global map.

Expression (3) is a rule for iterative updating of the estimates of the probabilities of finding the signs of the selected linear objects of the roadway in the cells of the local map. It should be noted that this update consists of only the summation operation, which makes the proposed method computationally efficient.

It should be noted that an a priori estimate of the probability of finding a sign of a selected linear object of the roadway, stored in the cells of the global map, is performed for each cell as follows:

(4)

(four)

– априорная вероятность нахождения признака выделенного линейного объекта дорожного полотна.where

– a priori probability of finding a sign of a selected linear object of the roadway.

,

соответственно. Данные ограничения необходимы для предотвращения неограниченного роста значений в ячейках карты, который возможен, например, когда АТС стоит на месте или долгое время в область видимости попадает один и тот же локальный участок карты. When summing the values of logarithms, the maximum and minimum allowable values of the sums in the cells are also entered

,

respectively. These restrictions are necessary to prevent an unlimited growth of values in the map cells, which is possible, for example, when the PBX stands still or the same local area of the map falls into the visibility area for a long time.

At the final stage, the resulting map of the roadway is obtained by binarizing the cells obtained by the specified combination of the map by the threshold value of the probability estimate of finding the feature of the selected linear object of the roadway, while considering that if the value of the cell is greater than the threshold value, then the space contains a linear feature of the roadway object , below, there is no linear feature.

Thus, the implementation of the method of mapping the terrain for the automatic telephone exchange provides the possibility of generating a high-precision roadway map that reflects information about the route of the automatic telephone exchange, which can be vectorized and used as a digital map for positioning a highly automated automatic telephone exchange, and periodic updating of the digital map will allow maintaining the current version, which, accordingly increase positioning accuracy.

Claims (2)

1. A method of terrain mapping for autonomous vehicles (ATS), characterized in that a global terrain map is formed, divided into cells, the specified map is written into the memory of the mapping module of the on-board computer of the ATS, an image stream is received from the camera located on the ATS to the input of the module mapping of the on-board computer of the automatic telephone exchange, after that the received images are processed with the selection of linear objects of the roadway, a map of features of the selected linear objects of the roadway is formed, a local map divided into cells is formed to account for errors in the detection of linear objects of the roadway, initial estimates of the probability of finding the features of the selected linear objects of the roadway in the cells of the global and local maps, and the probability estimate in the cells of the global map is a priori, the signs of the selected linear objects of the roadway are recorded in the corresponding cells of the global and local maps, taking into account ohm of a pre-known calibration of the camera and the current location of the vehicle, while the signs of the selected linear objects are received in the cells of the local map in the form of break points of the selected broken lines, based on the indicated points of the line, the probability of finding the sign of the selected linear object of the roadway is estimated for each cell of the local map , the resulting section of the local map is superimposed on the corresponding section of the global map with the summation of the estimates of the probability of finding the feature of the selected linear object of the roadway, taking into account the previous estimate, the resulting map of the roadway is obtained by binarizing the cells of the obtained map according to the threshold value of the probability estimate of finding the feature of the selected linear object of the roadway . 2. The method of terrain mapping for autonomous vehicles (ATS) according to claim 1, characterized in that the recognition of linear objects of the roadway from images received from the camera is carried out by the Hough window transform.

Images