Vehicle position determination using radar sensor data

Abstract

The invention relates to a vehicle position determination (10) of a vehicle (46), comprising sensor data acquisition (12) by at least one radar sensor (14) of the vehicle (46) which detects the vehicle environment (40) and provides radar sensor data (R), sensor data processing (16) of the radar sensor data (R) and output as processed radar sensor data (Rp), position determination (20) of a vehicle position (42) of the vehicle (46) with respect to a map (44) by assigning the vehicle position (42) to map data (22) of the map (44), wherein the map data (22) are implemented as trained model parameters (27) of a computational model (26) and the computational model (26) processes the processed radar sensor data (Rp) as input data (24) during position determination (20) and outputs position data (P) for the vehicle position (42) as output data (28).

Description

[0001] The invention relates to a vehicle position determination according to the preamble of claim 1. State of the art

[0002] Vehicle positioning is used in driver assistance systems and automated driving systems to determine the vehicle's position within a vehicle environment. Various sensors and technologies are used to collect and process position information, including GPS, inertial measurement units (IMUs), LIDAR (light detection and ranging), radar sensors, and cameras. Using the sensor data provided by these sensors, the vehicle can calculate its position relative to known points in the vehicle environment, for example, through sensor matching. This involves comparing the sensor data with previously recorded sensor data to which a position is assigned.

[0003] The use of radar sensor data is also known for object detection. For example, DE 10 2020 206 639 A1 describes a method for detecting and / or locating objects based on radar sensor data that is compared with an object measurement model. Disclosure of the invention

[0004] According to the present invention, a vehicle position determination system with the features of claim 1 is proposed. This allows the map to which the determined vehicle position is related to require less memory to store the map data. Furthermore, the memory requirements of the map data can be more closely decoupled from the map size. The vehicle position determination can be performed more quickly and with greater energy efficiency. The vehicle position can be determined more accurately. The vehicle position determination can be performed reliably even under difficult environmental conditions, for example, in rain, fog, dirt, and / or darkness.

[0005] The latent map data can eliminate or simplify the following commonly used steps: Map generation by extracting landmarks and features, Saving the landmarks in compressed form, Comparison of the stored landmarks with the recorded sensor data, Fusion of potentially inaccurate position data with map-based localization.

[0006] The radar sensor data can also be used to detect objects in the vehicle's surroundings. The radar sensor data can contain the following information about the objects in the vehicle's surroundings: Distance and azimuth angle, alternatively a two-dimensional xy position, a speed, in particular a radial speed, preferably a radial speed compensated for the vehicle's own movement, a Radar Cross Section (RCS) of the radar reflections, an elevation angle of the radar reflections and / or an orientation, for example a yaw angle.

[0007] The vehicle can be a motor vehicle, a two-wheeled vehicle, or a mobile robot. The vehicle can use the determined vehicle position based on the output position data to control at least one vehicle function. The vehicle function can be assigned to a driver assistance system and / or an autonomous driving system of the vehicle.

[0008] The radar sensor can be an FMCW radar sensor. The radar sensor can be a long-range radar.

[0009] The radar sensor data can be displayed as a time-dependent, range-dependent, azimuth-dependent, and / or frequency-dependent representation. The radar sensor data can be displayed as a frequency diagram using a fast Fourier transform (FFT). The radar sensor data can include Doppler velocity information.

[0010] The sensor data processing can adjust the radar sensor data containing Doppler velocity information for the vehicle's speed.

[0011] The computational model can process temporally sequential radar sensor data. The computational model can comprise a recurrent neural network (RNN), in particular a long-short-term memory or gated recurrent unit, and / or a convolutional neural network (CNN). In the CNN, a kernel or filter can be applied to the input data as part of the convolutional operation.

[0012] The computational model can comprise a neural network with multiple network layers. Each network layer can contain a plurality of neurons.

[0013] The map can include features of a given geographical area, in particular the vehicle's surroundings and / or the Earth's surface. The geographical area can be a country or a sub-area of ​​a country. For example, the geographical area spanned by Germany can be divided into several, in particular 30, sub-areas. Each sub-area can be assigned its own map. Each map can be assigned its own calculation model.

[0014] When determining the vehicle position, the calculation model to be used to determine the vehicle position can be selected depending on the map, which in turn is selected depending on the sub-area in which the vehicle is currently located. The sub-areas can overlap geographically. If the vehicle is located in an overlapping area, the vehicle position can be determined by applying the calculation models assigned to the sub-areas representing the overlap. If multiple calculation models are used, the position data can be determined by subsequent data fusion, for example, averaging. The multiple calculation models can be applied serially or in parallel.

[0015] The map data can be specified as information in a latent space of the computational model. The map data can contain information about the position and / or dimensions of objects.

[0016] Objects can be other vehicles, pedestrians, cyclists, animals, buildings, devices, infrastructure elements, temporary obstacles such as construction sites or parked vehicles, or plants.

[0017] The calculation of the position data using the calculation model can be carried out in the vehicle or by a device outside the vehicle, for example a computing server.

[0018] The calculation model can be stored in the vehicle or in a storage device outside the vehicle, such as cloud storage. The calculation model can be updated, for example, during a vehicle inspection.

[0019] In a preferred embodiment of the invention, it is advantageous if the processed radar sensor data is processed as input data by forward propagation with the learned model parameters. During forward propagation, the input data can be present in the latent space. The input data can be processed in various network layers of the calculation model, each with the model parameters, in particular by matrix multiplication.

[0020] In a preferred embodiment of the invention, it is advantageous if the position data indicate a global and / or local vehicle position. The map can describe the immediate vehicle surroundings. The map can be a local map that at least partially includes the vehicle surroundings and / or a global map with global coordinates, for example, in the WGS84 coordinate system.

[0021] The local map can specify the vehicle position relative to the objects contained in the map. The local vehicle position is a vehicle position in the local map. The local map can have a fixed coordinate origin.

[0022] The global vehicle position can be specified in global coordinates, specifically longitude and latitude. The global vehicle position is a vehicle position on the global map.

[0023] A preferred embodiment of the invention is advantageous in which the calculation model has been trained using training data comprising radar sensor data and reference position data. The radar sensor data of the training data can be radar sensor data recorded from a fleet operation of several vehicles. The radar sensor data of the training data can be recorded during a vehicle training run.

[0024] The reference position data can be provided by a GNSS sensor, an accelerometer, and / or position data from other vehicle sensors. The reference position data indicates a verified and / or accurate vehicle position.

[0025] In a specific embodiment of the invention, it is advantageous if the reference position data comprises first reference position data from at least one first position sensor and position difference data. The position difference data can indicate the position difference between the first reference position data and the reference position data. The reference position data can be more precise than the first reference position data. The first reference position data can have a larger error or a larger deviation from the exact position than the reference position data.

[0026] In an advantageous embodiment of the invention, the sensor data acquisition includes position data from at least one position sensor. The position sensor can be a GNSS sensor, in particular a GPS sensor, for example a low-cost GNSS sensor. The sensor data acquisition can include linear, rotational, and / or acceleration data of a vehicle's acceleration.

[0027] In a specific embodiment of the invention, it is advantageous if the input data includes position data from the position sensor in addition to the processed radar sensor data. The input data may include acceleration data. The input data may include further information, in particular from further vehicle sensors, for example a yaw, pitch, and / or roll angle of the vehicle, map information, in particular lane information, aggregated environmental data, in particular a number of objects in the vehicle's surroundings and / or an altitude above sea level.

[0028] A preferred embodiment of the invention is advantageous in which the position sensor providing the position data of the input data is the first position sensor, and the position data of the input data is first position data. It is also possible for the input data to comprise only the radar sensor data, and for the calculation model to calculate position difference data based on this input data.

[0029] A preferred embodiment of the invention is advantageous in which the position data of the output data comprise position difference data, which is superimposed on the first position data in an output data processing step. The position difference data can be added to the first position difference data. The position difference data can be signed.

[0030] The output data may comprise exclusively the position difference data. The input data may comprise the radar sensor data, in particular with additional first position data.

[0031] A preferred embodiment of the invention is advantageous in which the number of learned model parameters is adapted to a map size and / or the number of map information items. The number of model parameters can increase with increasing map size and / or increasing map information or can remain constant.

[0032] The map information may include road layout, such as the direction and layout of roads, including curves, intersections, highways, one-way streets, and the like; terrain features, such as bridges, tunnels, buildings, parks, rivers, and other geographical features; traffic signs, such as speed limits, stop signs, traffic lights, and other traffic regulations; points of interest (POI), such as gas stations, restaurants, hospitals, schools, and other important locations.

[0033] The vehicle position determination can be carried out as a computer-implemented process.

[0034] Furthermore, the invention relates to a non-transitory, computer-readable medium containing instructions, the execution of which causes at least one processor to carry out the vehicle position determination with at least one previously described feature.

[0035] Further advantages and advantageous embodiments of the invention emerge from the description of the figures and the illustrations. Character description

[0036] The invention is described in detail below with reference to the figures. They show in detail: Fig. 1: A vehicle position determination in a special embodiment of the invention. Fig. 2: A radar sensor diagram of example radar sensor data. Fig. 3: A series of further radar sensor diagrams of example radar sensor data. Fig. 4: Another radar sensor diagram of example radar sensor data.

[0037] Fig. 1 shows a vehicle position determination in a specific embodiment of the invention. The vehicle position determination 10 determines a vehicle position of a vehicle by providing radar sensor data R via sensor data acquisition 12 by at least one radar sensor 14 of the vehicle that detects the vehicle's surroundings. This data is output as processed radar sensor data Rp via subsequent sensor data processing 16.

[0038] With a position determination 20 of a vehicle position with respect to a map by assigning the vehicle position to map data 22 of the map, the processed radar sensor data Rp are processed as input data 24 with a calculation model 26 that has the map data 22 as learned model parameters 27 and, with the calculation of the input data 24, outputs position data P relating to the vehicle position as output data 28. The processed radar sensor data Rp can be processed as input data 24 by forward propagation with the learned model parameters 27.

[0039] The position data P indicates, for example, a global vehicle position in a global map, for example a longitude and latitude of the vehicle on the earth's surface.

[0040] The calculation model 26 was trained, in particular, using training data 30 comprising radar sensor data R' and reference position data G. The reference position data G can be provided by a GNSS sensor, an acceleration sensor, and / or position data from other vehicle sensors. The reference position data G can include first reference position data G1 from at least one first position sensor 32 and position difference data dP.

[0041] The sensor data acquisition 12 can include first position data P1 from at least the first position sensor 32. The input data 24 can therefore include the first position data P1 in addition to the processed radar sensor data Rp. The calculation model 26 calculates the position data P, which includes the position difference data dP, based on this input data 24. In a subsequent output data processing 34, the first position data P1 is overlaid with the position difference data dP output by the calculation model 26 and output as processed position data Pp.

[0042] It is also possible for the input data 24 to contain only the radar sensor data R, and for the calculation model 26 to calculate the position difference data dP based on this input data 24. Accordingly, the calculation model 26 is trained with position difference data and radar sensor data as input data.

[0043] Fig. Figure 2 shows a radar sensor diagram of exemplary radar sensor data. The radar sensor diagram indicates the radar sensor data R, here in a two-dimensional representation with the spatial coordinates x and y. The radar sensor data R can be provided by a front radar sensor at the front of the vehicle. The coordinate origin O corresponds to the center of the vehicle front.

[0044] Using radar sensor data R, a building 36 and a roadway 38 in the vehicle's surroundings can be detected. This radar sensor data R is passed to the calculation model as input data, which then specifies the position data for the vehicle position on the local map 44 describing the immediate vehicle surroundings. This allows, for example, the vehicle position 42 to be specified in relation to the roadway 38 and the building 36 on the local map 44.

[0045] Fig. Figure 3 shows a series of additional radar sensor diagrams of exemplary radar sensor data. The radar sensor diagrams represent the radar sensor data R in long-range Doppler velocity recordings in the time dimension L, each after 2D FFT processing. The color intensity depends on radar reflections from objects detected in the vehicle's surroundings at the respective speed J and distance K.

[0046] Fig. Figure 4 shows another radar sensor diagram of exemplary radar sensor data. The range-azimuth diagram indicates the radar sensor data R as a function of a distance and an azimuth angle of the radar reflections to the radar sensor of the vehicle 46. For example, the two strongest intensities can represent a detected traffic sign 48 and a motorcycle 50. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] DE 10 2020 206 639 A1

[0003]

Claims

[1] Vehicle position determination (10) of a vehicle (46), comprising sensor data acquisition (12) by at least one radar sensor (14) of the vehicle (46) which detects the vehicle environment (40) and provides radar sensor data (R), Sensor data processing (16) of the radar sensor data (R) and output as processed radar sensor data (Rp), Determining (20) a vehicle position (42) of the vehicle (46) in relation to a map (44) by assigning the vehicle position (42) to map data (22) of the map (44), characterized by , that the map data (22) are implemented as learned model parameters (27) of a calculation model (26) and the calculation model (26) processes the processed radar sensor data (Rp) as input data (24) during the position determination (20) and outputs position data (P) relating to the vehicle position (42) as output data (28). [2] Vehicle position determination (10) according to claim 1, characterized bythat the processed radar sensor data (Rp) are processed as input data (24) by forward propagation with the learned model parameters (27). [3] Vehicle position determination (10) according to claim 1 or 2, characterized by that the position data (P) indicate a global and / or local vehicle position (42). [4] Vehicle position determination (10) according to one of the preceding claims, characterized by that the calculation model (26) was trained with training data (30) comprising radar sensor data (R') and reference position data (G). [5] Vehicle position determination (10) according to claim 4, characterized by that the reference position data (G) comprise first reference position data (G1) of at least one first position sensor (32) and position difference data (dP). [6] Vehicle position determination (10) according to one of the preceding claims, characterized bythat the sensor data acquisition (12) includes position data from at least one position sensor (32). [7] Vehicle position determination (10) according to claim 6, characterized by that the input data (24) comprise, in addition to the processed radar sensor data (Rp), position data of the position sensor (32). [8] Vehicle position determination (10) according to claims 5 and 7, characterized by that the position sensor (32) providing the position data of the input data (24) is the first position sensor (32) and the position data of the input data (24) are first position data (P1). [9] Vehicle position determination (10) according to claim 8, characterized by that the position data (P) of the output data (28) comprise position difference data (dP) which are superimposed on the first position data (P1) in an output data processing (34). [10] Vehicle position determination (10) according to one of the preceding claims, characterized bythat the number of learned model parameters (27) is adapted to a map size and / or number of map information of the map (44).

Images