EP3724811A1

EP3724811A1 - Method for creating a feature-based localization map for a vehicle, taking into consideration characteristic structures of objects

Info

Publication number: EP3724811A1
Application number: EP18814502.3A
Authority: EP
Inventors: Peter Christian Abeling; Daniel Zaum
Original assignee: Robert Bosch GmbH
Current assignee: Robert Bosch GmbH
Priority date: 2017-12-14
Filing date: 2018-11-21
Publication date: 2020-10-21
Also published as: CN111480165A; US20200348685A1; US11442462B2; WO2019115192A1; JP2021507376A; DE102017222810A1; JP7427587B2; KR20200097772A

Abstract

The invention relates to a method for creating a feature-based localization map for a vehicle, comprising the following steps: determining data of at least one object in the environment of the vehicle; identifying characteristic structures of the at least one object; combining the characteristic portions to form a simplification structure of the object; and inserting the simplification structure into the feature-based localization map.

Description

description

title

Method for creating a feature-based localization map for a vehicle taking into account characteristic structures of objects

The invention relates to a method for creating a feature-based localization map for a vehicle. The invention further relates to a

Device for creating a feature-based location map for a

Vehicle. The invention further relates to a computer program product.

State of the art

Techniques from the field of simultaneous localization and mapping (SLAM) are known in order to create environmental maps for individual measuring vehicles or large vehicle fleets by means of sensor data. Corresponding applications are found above all in the fields of robotics, logistics, vehicle technology, aviation, consumer goods, etc.

In order to create accurate environmental maps from vehicle sensor observations (e.g., in the form of radar, video, lidar, etc.), graph-based SLAM techniques have become particularly popular.

In SLAM without local reference, an automated measurement vehicle (e.g., robot, not shown) travels along an unknown path with defined measurement positions, the measurement vehicle moving through

Estimate motion estimates (e.g., by inertial sensors,

Wheel rotation sensors, etc.). From each position, the measuring vehicle determines its surroundings by means of a series of measurements by targeting identified observations of landmarks. Each of the landmarks may be observed multiple times from each observation point, so there are more landmark measurements than current landmarks.

The goal of Graph-SLAM or Full-SLAM is to determine from the measurements of the measuring vehicle the true path of the measuring vehicle which the measuring vehicle has taken through the surroundings or the real positions of the surroundings. This is done by comparing the landmarks to each Measuring point were determined performed. In this process, the true map of the environment is to be determined. This is done by corresponding measurements of the same landmark from different measurement points, which are used to perform a measurement vehicle path estimation and environmental detection simultaneously.

Disclosure of the invention

An object of the present invention is to provide an improved method for creating a feature-based location map for a vehicle.

The object is achieved according to a first aspect with a method for creating a feature-based localization map for a vehicle, comprising the steps:

Determining data of at least one object in the environment of the vehicle; Identifying characteristic structures of the at least one object;

Combining the characteristic structures into one

Simplification structure of the object; and

Insert the simplification structure into the feature-based

Location Map.

In this way, semantic analysis from the determined data, e.g. performed in the form of raw sensor data, with the purpose of the complex

Structures to simplify structures together. In this way, a data volume of the digital feature-based location map can be reduced and a quality of localization significantly improved.

According to a second aspect, the object is achieved with a device for creating a feature-based localization map for a vehicle, comprising:

a determination device for determining data of at least one object in the environment of the vehicle;

an identifying device for identifying characteristic structures of the at least one object; a summarizing means for summarizing the characteristic structures into a simplification structure of the object; and

an inserter for inserting the simplification structure into the feature-based location map.

Advantageous developments of the method are the subject of dependent claims.

An advantageous development of the method provides that at least one of: radar sensor, ultrasound sensor, lidar sensor, camera is used to determine the data of the at least one object in the environment of the vehicle. In this way, the method can be realized with a multiplicity of different sensor concepts. The mentioned

Detection facilities separately and / or in a combined

Operation record the data of at least one object in the environment of the vehicle.

A further advantageous development of the method provides that prior knowledge is used to identify the characteristic structures of the at least one object. In this way, the identification of the characteristic structures of the at least one object can be performed more efficiently and faster. For example, for this purpose

Catalogs with product types or types of structures are used that are kept up to date.

A further advantageous development of the method provides that the card is updated at defined intervals, preferably after a defined number of days, more preferably after a defined number of hours. In this way, a high utility of such a trained feature-based location map is supported for many users.

A further advantageous development of the method provides that data of objects in the environment of vehicles are provided by means of a crowdsourcing-based approach. Advantageously, in this way, non-dedicated investigation vehicles must capture the data of the objects, but it can ordinary road users to create the proposed feature-based localization map. A determination quality is advantageously increased in this way.

The invention will be described in more detail below with further features and advantages with reference to several figures.

Disclosed method features are analogous to corresponding disclosed device features and vice versa. this means

In particular, features, technical advantages and embodiments relating to the method analogously result from corresponding embodiments, features and technical advantages relating to the device for creating a feature-based localization map for a vehicle and vice versa.

In the figures shows:

Fig. 1 is a schematic representation of an effect of

conventional SLAM-based map generation;

Fig. 2 principle representations of an effect of

conventional and inventive SLAM-based map generation;

FIG. 3 shows basic illustrations of effects of conventional and inventive SLAM-based map generation; FIG.

Fig. 4 is a block diagram of a proposed device for

Creating a feature-based location map for a vehicle; and

5 shows a basic sequence of a proposed method for creating a feature-based localization map for a vehicle.

Description of embodiments In the following, an automated motor vehicle can also be understood synonymously as a partially automated motor vehicle, autonomous motor vehicle and partially autonomous motor vehicle.

Methods and systems for mapping busy roads using sensors installed in vehicles (e.g., cameras, radar sensors,

Ultrasonic sensors, etc.) are known. In addition to the sensors mentioned, these systems generally also have a radio interface (for example realized via a connectivity unit) for transmitting the measured sensor data to a server. In this way, entire vehicle fleets can map their collective environment using the vehicle sensors by providing their sensor data, e.g. transferred to a server. The transmission of such so-called "fleet mapping data" is known.

The sensor data is collected on the server and a digital map for the relevant road section is generated from the data of several trips and / or vehicles. The digital maps determined in this way (also called HAD maps, AD maps or HD maps) are used, inter alia, for locating automatically moving vehicles in the digital map (for example for determining trajectories). Here, so-called landmarks are used, which are listed in the digital map with their exact geographical position.

Typical landmarks are eg lane markings, street signs, guardrails, etc. If an automatically driving vehicle recognizes one or more landmarks with the aid of the vehicle sensor system and can clearly find these landmarks in the digital map, this can result in a very accurate relative position of the vehicle relative to the landmark of the digital map be derived. Density and quality of the landmarks thus significantly affect a quality of localization with respect to the accuracy of the detected position. In reality, there are sections of track that have many and well-usable landmarks, as well as sections that have poor coverage of landmarks, which may result in poor localization quality. The mentioned graph-based SLAM algorithms are used for the purpose of mapping and can in principle be divided into two basic steps:

1st SLAM frontend

This step envisages identifying identical features of multiple runs of the same area by comparing recognized landmarks.

The identified relationships between measurement positions of different journeys are expressed as edges of a graph representation.

2nd SLAM backend

In this case, an optimization of the graph results of the aforementioned front-end step with the purpose of finding an optimal solution, which all

Conditions fulfilled.

As stated above, one of the key aspects of SLAM-generated maps is that the data from multiple trips along the route be included in the map. Therefore, most Real World objects are represented multiple times in the map after the optimization step, after being observed during each trip that contributes to the map.

Therefore, optimized data from multiple trips is condensed in a final clustering step called feature map creation with the goal of reducing map size and optimizing map content for localization feasibility. All the above steps are already known in the art.

In the field of automated driving, SLAM techniques are used to generate maps from crowdsourced data that allow for precise feature-based localization. With the announced products "Radar Road Signature" and "Video Road Signature" Robert Bosch is very active in this technical field. The feature-based localization provides that a mapped representation of observed objects in the environment of the vehicle is compared with data of a current environmental sensation of the vehicle and compared with them. This results in an estimate of the current one

Vehicle position in relation to the localization map.

A clustering step in SLAM-based map generation, as previously explained, generally works well with punctiform targets using well-known algorithms, e.g. Masts for traffic signs.

1 shows in three views a), b), c) a principle of a determination of a balanced radar localization map and a corresponding idealized compression of this map using known methods for clustering punctiform objects. It can be seen in Fig. 1a by means of a radar sensor detected guard rail attachment points of the card. In Fig. 1c is indicated that the detected guard rail attachment points are sufficiently represented by individual points in the compressed representation.

Nonetheless, some structures are under representation

Use of punctiform representations of guardrail attachment points not sufficiently described. This can disadvantageously diminish the efficiency of a localization map that provides only punctiform objects for alignment in the course of localization.

The invention proposes an improved process with the following advantages:

A densification of radar-based localization maps using not only punctiform objects but also objects having arbitrary structures and shapes represented in the form of "simplification structures"

- Resulting reduced data volume of resulting compressed maps - Improved localization results due to using the

Simplification structures achieved improved object representations

A key idea of the present invention is to provide representations of feature-based location maps not only using point-like representations of objects, but to provide for describing complex structures in a more complex manner to thereby facilitate alignment of data with the object

Improve localization map and localization quality. After a semantic description of all kinds of complex shapes, which can detect a radar sensor, is not feasible, it is proposed to complex structures with a larger, but compared to a

semantic description still represent a small number of points.

A principle of this procedure is shown in three representations a), b) and c) of FIG. 2. FIG. 2 a shows an exemplary complex object in the form of a symbolized electric power pole, which is represented in FIG. 2 b in a conventional manner by a single point-shaped object.

According to the invention, it is proposed to represent the electric power pole by means of a structure with several points as in FIG. 2c, which realizes a semisemantic representation of the object "power pole". This semisemantic representation abstracts data from a current radar measurement in a more accurate manner and, in the above-explained calibration process of the acquired data, allows a higher precision, as is the case in principle in two

Representations a) and b) of Fig. 3 is shown.

Fig. 3a shows probability ranges A1, A2 of a localization associated with a specific representation of the object. Recognizable are a relatively large region A1 of high localization quality and a relatively large region A2 of low localization quality, based on a punctiform representation e.g. an object "power pole" in the feature-based

Location Map.

In contrast, Fig. 3b, which is shown in the same scale as Fig. 3a, that the two said probability ranges A1, A2 by a Representation of the object "power pole" in the above-mentioned semi-semantic manner with several points are advantageously substantially reduced. As a result, it is achieved that a more accurate location for the vehicle can be achieved in comparison with the ratios of FIG. 3a by means of the feature-based localization map and detected radar sensor data.

Advantageously supported by the fact that an autonomous or automated driving function can be maintained longer for an automated vehicle. In the event that the localization by means of the detected sensor data and the feature-based location map is no longer possible, for example, the autonomous or automated driving function is switched off, whereupon the driver at least temporarily manual control of the vehicle

must take over.

Although the method explained by way of example with a radar sensor

It goes without saying that the proposed method can also be used with other sensors, e.g. a lidar, ultrasonic sensor, or a camera can be performed.

Preferably, the creation of such a feature-based

A location-based map with a crowd-sourcing-based approach such that a large number of vehicles that do not act as dedicated discovery vehicles transmit the captured data to a central location (not shown) that updates the feature-based location map, for example, at a defined high update frequency of a few days or hours.

The vehicles providing the feature-based data can thus both be data providers for the feature-based localization map and at the same time users of the map currently being created.

4 shows, in principle, a block diagram of a device 100 for creating a feature-based localization map.

One recognizes a determination device 110 for determining data by means of an object in the environment of the vehicle. The determination device 110 is functionally connected to an identification device 120, which is used for Identifying characteristic structures of the at least one object is provided. The identification device 120 is operatively connected to a summarizer 130, which is provided for summarizing the characteristic structures into a simplification structure of the object. The summarizer 130 is operatively connected to an inserter 100 for inserting the

Simplification structure is provided in the feature-based localization map.

Advantageously, the proposed method can be implemented as a software with program code means for running on an electronic device 100, whereby an easy changeability and adaptability of the method is supported.

FIG. 5 shows a basic sequence of an embodiment of the invention

inventive method.

In a step 200, a determination of data of at least one object in the environment of the vehicle is performed.

In a step 210, identification of characteristic structures of the at least one object is performed.

In a step 220, a summary of the characteristic structures to a simplification structure of the object is performed.

In step 230, an insertion of the simplification structure into the feature-based location map is performed.

One skilled in the art will suitably modify and / or combine the features of the invention without departing from the gist of the invention.

Claims

claims

A method of creating a feature-based location map for a vehicle, comprising the steps of:

Combining the characteristic structures into one

Simplification structure of the object; and

Insert the simplification structure into the feature-based

Location Map.

2. The method of claim 1, wherein for determining the data of the at least one object in the environment of the vehicle at least one of:

Radar sensor, ultrasonic sensor, lidar sensor, camera is used.

3. The method according to claim 1 or 2, wherein for identifying the

characteristic sections of the at least one object prior knowledge is used.

4. The method according to any one of the preceding claims, wherein the card at defined intervals, preferably after a defined number of days, even more preferably after a defined number of hours, is updated.

5. The method according to any one of the preceding claims, wherein data provided by objects in the environment of vehicles by means of a crowdsourcing based approach.

6. A device (100) for creating a feature-based location map for a vehicle, comprising:

a determination device (110) for determining data of at least one object in the environment of the vehicle; an identification device (120) for identifying

characteristic structures of the at least one object;

a summarizing means (130) for combining the characteristic portions into a simplification structure of the object; and

an insertion device (140) for inserting the

Simplification structure in the feature-based localization map.

7. Computer program product with program code means for carrying out the method according to one of claims 1 to 5, when it is on a

Device (200) for creating a feature-based location map for a vehicle is running or stored on a computer-readable medium.