Method for processing sensor data, method for processing output data, and signal processing unit
Patent Information
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-02-16
- Publication Date
- 2026-03-04
AI Technical Summary
Current radar sensor systems in vehicles, particularly FMCW radars, face challenges in accurately and efficiently identifying and classifying object targets in a vehicle environment, as they often rely on 2D-FFT for angle, distance, and speed information, which can be affected by noise and peak overlap, leading to less reliable and precise object detection.
A method that processes radar sensor data using FFT for signal filtering and peak detection, enriches output data with additional parameters like peak half-width and shape characteristics, and employs machine learning processes, such as neural networks, for improved object recognition and classification, enabling more precise and reliable identification of targets.
This approach enhances the reliability, precision, and speed of object detection and classification by providing enriched data for better peak characterization and using machine learning for improved recognition, allowing for more accurate identification of stationary and mobile objects in a vehicle environment.
Smart Images
Figure EP2024054004_31102024_PF_FP_ABST
Abstract
Description
[0001] Method for sensor data processing, method for output data processing and signal processing unit
[0002] The invention relates to a method for sensor data processing according to claim 1. Furthermore, the invention relates to a method for output data processing and a signal processing unit.
[0003] State of the art
[0004] In vehicles, radar sensors are preferably used to detect targets in the vehicle's surroundings. Such radar sensors are often designed as frequency-modulated continuous wave (FMCW) radars because they are inexpensive and offer high resolution. An FMCW radar transmits a finite sequence of separate linear frequency-modulated chirps in each coherent processing interval. At a receiver, the target echoes are mixed with the transmitted signal, resulting in a complex beat signal or intermediate frequency signal. The distance and speed information of the targets can be obtained from the frequencies of the intermediate frequency signal. Fast Fourier transforms (FFT) are typically used for this purpose.
[0005] To also obtain angular information of the targets, multiple transmit and receive antennas (MI MO radar) are used. Two-dimensional frequency estimation algorithms, particularly 2D FFT, are used to obtain the range, angle (angle of arrival), and velocity information of the received target echo signals. This results in signal spectra in the respective domains (angle, distance, velocity). Peaks in the signal spectra are identified through subsequent peak detection.
[0006] Disclosure of the invention
[0007] According to the present invention, a method with the features of claim 1 is proposed. This allows object targets in the vehicle's surroundings to be identified and marked more reliably, accurately, and quickly. The output data can be enhanced with additional information. The vehicle can be a motor vehicle, in particular a motor vehicle having at least three wheels, a two-wheeled vehicle, in particular a motorcycle or a bicycle. The vehicle can be a robot, for example, a robotic lawnmower.
[0008] The radar sensor can be an FMCW radar sensor. The radar sensor can be a MIMO radar sensor.
[0009] The targets can be objects or living beings. The targets can be stationary, especially signs, buildings, trees, or moving, especially other vehicles.
[0010] The sensor data can be processed or unprocessed measurement data from the radar sensor. The sensor data can be filtered and thus processed.
[0011] The signal processing can use an FFT (Fast Fourier Transformation).
[0012] A peak is the technical term derived from English for a locally significant signal peak value of a signal value distribution.
[0013] In the signal spectrum, the target objects detected by the radar sensor may each appear as peaks. The signal spectrum may contain distance information, angle information, and / or speed information about the detected targets in at least one dimension. A further dimension of the signal spectrum may include signal intensity.
[0014] The output data may include distance information, angle information and / or velocity information, and preferably information on the reflectivity of the object target associated with the peak.
[0015] In a preferred embodiment of the invention, it is advantageous if, in the information processing, the peak is assumed to be an idealized peak corresponding to a predetermined shape, and the additional output data includes the peak parameter of the idealized peak that at least approximates the peak. The idealized peak can be formed by a Gaussian function.
[0016] A preferred embodiment of the invention is advantageous in which the peak parameter is the half-width of the peak. The half-width refers to the width of the peak at half its height. It therefore indicates how wide a peak is at half its height.
[0017] In the case of superpositions of several peaks in the signal spectrum, suitable approximation can be carried out by, for example, the information processing determining the half-width depending on the peak progression in only one direction or in the direction of the next minimum in the processing data.
[0018] In a special embodiment of the invention, it is advantageous if the output data associated with the peak comprise exclusively the peak position and the peak parameter.
[0019] In a specific embodiment of the invention, it is advantageous if the additional output data includes a further peak parameter of the peak in addition to the peak parameter. This allows the shape of the peak to be described more precisely using the additional output data.
[0020] A preferred embodiment of the invention is advantageous in which the signal spectrum is two-dimensional and the peak parameter characterizes the shape of the peak in the signal spectrum. The two-dimensional signal spectrum can represent signal values versus distance, speed, or angle, in particular the elevation angle or the azimuth angle.
[0021] A preferred embodiment of the invention is advantageous in which the signal spectrum is three-dimensional, and the peak parameter characterizes the shape of the peak in a first dimension of the signal spectrum, and the further peak parameter characterizes the shape of the peak in the second dimension of the signal spectrum. The peak parameter can specify a half-width of the peak in the first dimension. The further peak parameter can specify a half-width of the peak in the second dimension.
[0022] The signal spectrum can also have more than three dimensions.
[0023] According to the present invention, a method with the features of claim 8 is further proposed. Object recognition can recognize the target objects. Object classification can classify the target objects.
[0024] In a preferred embodiment of the invention, the object recognition and / or object classification includes at least one machine learning process with input data formed by the output data, including the additional output data. The machine learning process can include the use of at least one neural network. The neural network can be trained using deep learning. The neural network can have multiple network layers. The object recognition and / or object classification can be performed by at least one object model comprising the neural network, with the output data as input data.
[0025] The additional output data can be used as input data for inference of the object model. The additional output data can be used to evaluate the object model.
[0026] According to the present invention, a signal processing unit with the features of claim 10 is further proposed. The signal processing unit can be a data processing unit, in particular a computer. The signal processing unit can be connected to the radar sensor for data transmission. The signal processing unit can be arranged in the vehicle.
[0027] Further advantages and advantageous embodiments of the invention emerge from the description of the figures and the illustrations.
[0028] Character description
[0029] The invention is described in detail below with reference to the figures. They show in detail:
[0030] Figure 1: A method for sensor data processing, a method for output data processing and a signal processing unit, each in a specific embodiment of the invention.
[0031] Figure 2: A signal spectrum during sensor data processing using the method in Figure 1.
[0032] Figure 1 shows a method for sensor data processing, a method for output data processing, and a signal processing unit, each in a specific embodiment of the invention. The method 10 for sensor data processing of sensor data 12 from a radar sensor 16 arranged in a vehicle 14 is preferably carried out during operation of the vehicle 14. The operation 18 of the radar sensor 16 detects object targets 20 in a vehicle environment 22 of the vehicle 14. First, signal processing 24 of the sensor data 12 is performed, followed by the provision 26 of processing data 32 obtained from the signal processing 24 and representable in at least one signal spectrum 28. The signal processing 24 can include noise suppression, signal filtering, and the like.
[0033] In the processing data 32, a peak detection 34 of at least one peak corresponding to one of the object targets 20 in the vehicle environment 22 takes place in the signal spectrum 28. After information processing 36 of the processing data 32 assigned to the detected peak, output data 40 generated by the information processing 36 is then outputted 38, which includes at least the peak position of the detected peak in the signal spectrum 28. The output data 40 comprises additional output data 42, which, in addition to the peak position, comprise at least one peak parameter of the peak. The additional output data 42 can be obtained by an additional processing operation 43. The additional processing operation 43 can be carried out with the information processing 36.
[0034] The signal processing 24 can in turn be carried out by feedback depending on the additional processing operation 43.
[0035] Also depicted is a method 44 for processing the output data 40, in which an object recognition 46 of at least one object target 20 in the vehicle environment 22 is performed based on the output data 40 identifying the object target 20 with the additional output data 42 as additional information. The object recognition 46 comprises a machine learning process 48 with input data 50 formed by the output data 40 and the additional output data 42.
[0036] The object recognition 46 can then output the recognized object target 52. The output data processing can alternatively or additionally perform an object classification.
[0037] The method 44 for output data processing and / or the method 10 for sensor data processing can be executed in a signal processing unit 54 in the vehicle 14. The signal processing unit 54 can be connected to the radar sensor 16 for data transmission.
[0038] Figure 2 shows a signal spectrum during sensor data processing using the method from Figure 1. Figure 2 a) shows a two-dimensional representation of the three-dimensional signal spectrum 28, which indicates the signal intensity of the sensor signal as a function of the distance r along the ordinate and the speed v along the abscissa. The signal spectrum 28 can be a range Doppler diagram. The two-dimensional representation in Figure 2 a) indicates the course of the signal intensity as a function of the distance r and the speed v via the contour lines 56.
[0039] A total of three peaks P can be seen, a first peak P1 and a second peak P2, which each correspond to object targets moving at the same speed v but at a different distance r from the radar sensor, and a third peak P3 in the signal spectrum 28, which corresponds to an object target that has a smaller distance r and a lower speed v than the other two object targets.
[0040] Figure 2 b) shows a section through the signal spectrum along the vertical line shown in Figure 2 a). The diagram plots the signal intensity as a function of distance r as the first dimension. The output data includes, for example, the peak position 58 of the first peak P1 in the distance dimension.
[0041] In information processing, the first peak P1 is assumed to be an idealized peak P' corresponding to a given shape, here the first idealized peak PT, which can be described by a Gaussian function, and a peak parameter 60, here a half-width B of the first idealized peak PT, which at least approximates the first peak P1, is determined in the distance dimension. The peak parameter 60 can characterize the shape of the first idealized peak PT and thus approximately the shape of the first peak P1, respectively, in the distance dimension. The output data assigned to the first peak P1 thus include the peak position 58 and the peak parameter 60, here the half-width Br.
[0042] Figure 2 c) shows a section through the signal spectrum along the horizontal line shown in Figure 2 a). The diagram plots the signal intensity as a function of velocity v as the second dimension. During the information processing, a further peak parameter 60', here a further half-width B' of the further first idealized peak P1" in the velocity dimension, is determined.
Claims
Patent claims 1. Method (10) for sensor data processing of sensor data (12) of at least one radar sensor (16) detecting object targets (20) in a vehicle environment (22) of a vehicle (14), comprising the following steps Signal processing (24) of the sensor data (12) and providing (26) processing data (32) obtained by the signal processing (24) and representable in at least one signal spectrum (28), peak detection (34) of at least one peak (P) in the signal spectrum (28) corresponding to one of the object targets (20), information processing (36) of the processing data (32) assigned to the peak (P), and output (38) of output data (40) generated from the information processing (36) and including at least the peak position (58) of the detected peak (P) in the signal spectrum (28), wherein the output data (40) has additional output data (42) which, in addition to the peak position (58), comprise at least one peak parameter (60) of the peak (P).
2. Method (10) according to claim 1, characterized in that in the information processing (36) the peak (P) is assumed to be an idealized peak (P') corresponding to a predetermined shape and the additional output data (42) comprise the peak parameter (60) of the idealized peak (P') which at least approximates the peak (P).
3. Method (10) according to claim 1 or 2, characterized in that the peak parameter (60) is a half-width (B) of the peak (P).
4. Method (10) according to one of the preceding claims, characterized in that the output data (40) assigned to the peak (P) comprise exclusively the peak position (58) and the peak parameter (60).
5. Method (10) according to one of the preceding claims, characterized in that the additional output data (42) comprise, in addition to the peak parameter (60), a further peak parameter (60') of the peak (P).
6. Method (10) according to one of claims 1 to 5, characterized in that the signal spectrum (28) is two-dimensional and the peak parameter (60) characterizes the shape of the peak (P) in the signal spectrum (28).
7. Method (10) according to one of claims 1 to 5, characterized in that the signal spectrum (28) is three-dimensional and the peak parameter (60) characterizes the shape of the peak (P) in a first dimension of the signal spectrum (28) and the further peak parameter (60') characterizes the shape of the peak (P) in the second dimension of the signal spectrum (28).
8. Method (44) for output data processing of output data (40) obtained by a method (10) according to one of the preceding claims, in which an object recognition (46) and / or object classification of at least one object target (20) in the vehicle environment (22) is carried out on the basis of the output data (40) characterizing the object target (20) with the additional output data (42) as additional information.
9. Method (44) according to claim 8, characterized in that the object recognition (46) and / or object classification includes at least one machine learning process (48) with input data (50) formed by the output data (40) including the additional output data (42).
10. Signal processing unit (54) for a vehicle (14), which is configured to carry out the method (10) for sensor data processing according to one of claims 1 to 7 and / or the method (44) for output data processing according to claim 8 or 9.