Object detection in clutter radar data

Abstract

The invention discloses object detection in clutter radar data. Specifically, object detection in clutter-affected radar data is disclosed. A method (100) of detecting an object in radar data, comprising: obtaining (110) radar data from radar measurements of a scene, where the radar data comprises intensities relating to a plurality of points in a range-Doppler plane or a range-angle plane; parameterizing (112) a one-dimensional subset of the range-Doppler plane or the range-angle plane as appropriate by means of a parameter; extracting (113) an intensity below a pre-configured clutter threshold from the radar data relating to the subset; regularizing (114) the extracted intensity as a function of a parameter to obtain a signal mask; deriving (115) a detection threshold from the signal mask; and performing object detection (116) in which the detection threshold is applied to the radar data relating to the subset.

Description

Technical Field

[0001] Within the general technical field of radar technology, this disclosure relates to methods and apparatus for detecting objects in radar data that includes signal content representing clutter. In particular, it proposes methods and apparatus for object detection based on range-Doppler data affected by radar clutter. Background Technology

[0002] In radar technology, clutter (or radar clutter) is generally understood to include targets that a radar device captures and displays, but which are undesirable because they reduce the probability of detecting the desired, actual target. What specific target represents clutter varies between different radar devices and different usage scenarios. For example, dense clouds are of interest to weather radar, but for radar monitoring aircraft flying through or behind clouds, dense clouds act as clutter.

[0003] Clutter targets can be modeled as point scatterers or extended scatterers. In the acquired radar data, clutter targets in the second category may have non-zero ranges in the range, angle, and / or Doppler dimensions. Such clutter targets may be confined to a surface (such as surface clutter like swaying grass, puddles, etc.) or they may be space-filling (such as volumetric clutter like free water droplets, dust particles, etc.).

[0004] Threshold-based object detection, a widely adopted approach, involves comparing the radar intensity of different data units (e.g., range-Doppler units) with a detection threshold. To perform object detection in cluttered radar images, false alarms can be limited by setting the detection threshold high enough (e.g., above the 90th percentile intensity). However, this also limits the ability to detect weak objects in radar images.

[0005] Object detectors specifically designed for cluttered environments have been proposed in the literature and implemented in high-end devices. These detectors typically involve highly complex algorithms that perform well but consume significant computational resources. Furthermore, such dedicated object detectors are often difficult to implement in power-constrained devices, such as battery-powered portable devices. There is a desire to propose alternative, computationally more efficient solutions for object detection in cluttered radar data. Summary of the Invention

[0006] One object of this disclosure is to provide a method and apparatus for object detection in radar data affected by radar clutter. A particular object is that computationally simplified methods and apparatus for this purpose are available. A further object is to provide such methods and apparatus suitable for battery-powered applications. A further object is to provide object detection methods and apparatus that selectively apply consistent or inconsistent detection thresholds depending on the characteristics of the scene. A further object is to provide object detection techniques universally applicable to radar data in various formats, including at least range-Doppler data and range-angle data. A further object is to provide object detection methods and apparatus particularly suitable for certain known types of clutter. A further object is to achieve object detection by selectively applying the novel techniques proposed herein if significant clutter is present, and by applying existing methods in other cases.

[0007] At least some of these objectives are achieved by means of the invention as defined in the independent claims. The dependent claims relate to advantageous embodiments of the invention.

[0008] In a first aspect of this disclosure, a method for detecting objects in radar data is provided. The method includes: obtaining radar data from radar measurements of a scene, wherein the radar data includes intensities associated with multiple points in a range-Doppler plane or a range-angle plane; parameterizing a one-dimensional subset of the range-Doppler plane or range-angle plane by a parameter, depending on the situation; and extracting clutter thresholds below a pre-configured threshold from the radar data associated with the subset. The intensity is determined; the extracted intensity is regularized as a function of parameters to obtain a signal mask; a detection threshold is derived from the signal mask; and object detection is performed, wherein the detection threshold is applied to radar data relevant to a subset.

[0009] It should be understood that the act of “parameterizing” a subset does not require active mathematical operations, but can be equivalent to noting down one of the basic variables (range, Doppler velocity, and angle) that is suitable as a parameter. For example, the range variable can be a suitable parameter for a constant Doppler subset of range-Doppler data, and the Doppler variable can be a suitable parameter for a constant range subset. It should be further understood that the “clutter threshold” has been pre-configured by, for example, the user or system owner, and the clutter threshold is typically a globally applicable value intended to be applied to various scenarios and / or for any time and date used to perform the method. The clutter threshold can be based on an expected or estimated upper bound of the typical clutter intensity that will be encountered in one or more scenarios to be measured. Furthermore, the “detection threshold” represents a reference value in threshold-based object detection, i.e., the minimum intensity value that will be identified as a possible object in radar data.

[0010] In other words, the method according to the first aspect includes dividing the range-Doppler plane or range-angle plane (as the case may be) into one or more one-dimensional subsets, and applying the regularized signal mask to a signal below the clutter threshold. A novel combination of fitting radar data and using a signal mask as the basis for a detection threshold. As the inventors have recognized, this achieves one or more of the following advantages:

[0011] The detection threshold is adapted to actual radar data, that is, to actual measurements in the actual scenario, and specifically, to radar data for each subset. Variations in the detection threshold are allowed between subsets.

[0012] The method allows for the use of non-uniform detection thresholds with different heights for different parameter values.

[0013] Because the signal mask is based only on signals below the clutter threshold. This method uses radar data, so it suppresses the influence of stronger radar data that usually originates from non-clutter targets.

[0014] • Below the clutter threshold Regularization of radar data tends to eliminate noise and short-range artifacts that are usually irrelevant to considerations within object detection.

[0015] In a second aspect of this disclosure, a signal processing apparatus having processing circuitry is provided, the processing circuitry being configured to detect an object in radar data by performing the methods described above.

[0016] In a third aspect, this disclosure provides a computer program containing instructions for causing a computer or, in particular, a signal processing apparatus to perform the methods described above. The computer program may be stored or distributed on a data carrier. As used herein, "data carrier" can be a temporary data carrier, such as a modulated electromagnetic wave or light wave, or a non-temporary data carrier. Non-temporary data carriers include volatile and non-volatile memories, such as permanent and non-permanent storage media of the magnetic, optical, or solid-state types. Still within the scope of "data carrier," such memory may be permanently mounted or portable.

[0017] The second and third aspects of this disclosure generally share the effects and advantages of the first aspect, and they can be implemented with corresponding degrees of technical changes.

[0018] In some embodiments of object detection methods and / or signal processing devices, the detection threshold is determined by offset. Add to the signal mask to derive from the signal mask, where the offset is constant relative to the parameters. If processing multiple subsets, pre-configured offset values ​​can be used. Apply to all subsets. Alternatively, different offset values ​​can be used. Furthermore, depending on the spatial characteristics of the subset, different offset values ​​can be used. For example, if the subset corresponds to an increase in a constant range (in the range-Doppler plane or the range-angle plane), a reduced offset value can be used to partially compensate for the fact that radar reflections at a greater distance are attenuated. Another option is to use an offset calculated based on radar data from the subset under consideration.

[0019] Below the clutter threshold Regularization of the extracted intensities can be performed using any suitable method. In some embodiments, a smoothing function of parameters parameterized over a subset is fitted to the extracted intensities. For example, the smoothing function can be a polynomial such as a quadratic polynomial (less than second order), a cubic polynomial (less than third order), or more generally a combination of piecewise quadratic polynomials. The smoothing function can further be a combination of piecewise cubic polynomials (spline fitting). In other embodiments, regularization is applied to the extracted intensities; this type of regularization can include a smoothing operation such as convolution with a convolution kernel or convolution matrix with non-zero backing.

[0020] In some embodiments, object detection includes estimating a noise floor based on radar data and using conventional methods. The noise floor is then compared to a signal mask. If the noise floor substantially matches the signal mask, it is inferred that the radar data is substantially free of clutter, allowing the application of simpler object detection methods. For example, a detection threshold can be calculated based on the noise floor, and the signal mask can be disregarded thereafter. This can ultimately benefit the quality of object detection due to the availability of highly accurate noise floor estimation methods (noise statistics) in the literature. Accordingly, these embodiments selectively apply the novel object detection technique proposed herein if significant clutter is present, otherwise applying existing methods.

[0021] In some embodiments, object detection is performed for multiple one-dimensional subsets of the distance-Doppler plane or the distance-angle plane, using appropriate detection thresholds as appropriate. By combining the outputs of two or more subsets, a more complete understanding of the objects present in the scene can be obtained. Alternatively or additionally, the reliability of object detection can be further improved by combining the outputs of adjacent, intersecting, or overlapping subsets. For example, if two or more overlapping or neighboring detections are found in different subsets, this indicates a higher probability of the object's presence. Further alternatively, global object detection includes using a global detection threshold obtained by combining detection thresholds from two or more subsets (e.g., by fitting a smoothing function to the detection thresholds from all subsets). Object detection is then performed as a joint operation in which global object detection uses the global detection threshold.

[0022] Referring to radar data in the range-Doppler plane, some embodiments involve a one-dimensional subset of range slices (constant range) parameterized by Doppler velocities. In this case, the signal mask can be determined by operations that favor even functions (e.g., attempting to fit the extracted intensity to a smoothed even function (to be used as the signal mask), or if multiple smoothed functions are equally consistent with the extracted intensity, there is a higher probability of choosing an even function as the signal mask). To achieve this preference, the signal mask can, for example, be determined based on an even-fitted model such as a cosine series or a sum of a finite number of cosines. In particular, it is preferable to determine the signal mask using an even function centered at zero or near-zero Doppler velocity, which the inventors have observed is generally correct for clutter reflections. In the specific case of swaying grass or other vegetation, the instantaneous velocity obviously has an expected value of zero. Except in cases of strong (radial) winds, raindrops tend to move at low horizontal velocities. Stationary clutter targets, such as reflected puddles or suspended reflective particles in still air, are of course inherently stationary.

[0023] Referring to radar data in the range-angle plane, some embodiments involve a one-dimensional subset of range slices (constant range) parameterized by the angle of arrival. When determining the signal mask with reference to this parameterization, it is easy to consider exploratory methods applicable to known clutter sources. For example, raindrops can be expected to produce omnidirectional (isotropic) clutter, a patch of windblown grass may occupy the lower half of the field of view, and tree leaves are typically confined to bounded angular intervals.

[0024] Generally, unless expressly defined herein, all terms used in the claims should be interpreted according to their ordinary meaning in the art. Unless expressly stated otherwise, all references to “a / the element, device, component, apparatus, step, etc.” should be openly interpreted as referring to at least one instance of that element, device, component, apparatus, step, etc. Unless expressly stated otherwise, the steps of any method disclosed herein need not be performed in the exact order presented. Attached Figure Description

[0025] Aspects and embodiments will now be described by way of example with reference to the accompanying drawings, in which:

[0026] Figure 1 This is a flowchart of a method for detecting objects in radar data according to embodiments herein;

[0027] Figure 2 Depicts a scene monitored by radar equipment and a signal processing device configured to perform object detection in the received radar data;

[0028] Figure 3 The diagram shows what has been drawn as Doppler velocity. and distance The function is based on the clutter detection threshold of intensity;

[0029] Figure 4 The figure shows the Doppler velocity plotted at a constant distance. The signal mask and detection threshold of the function;

[0030] Figure 5A , Figure 5B and Figure 5C The diagram illustrates distance-Doppler ( Plane or distance-angle ( Different combinations of one-dimensional subsets of a plane;

[0031] Figure 6 The diagram shows the angle of arrival. and distance The function is based on the clutter detection threshold of intensity; and

[0032] Figure 7 The diagram shows the angle of arrival plotted as a constant distance. The signal mask and detection threshold of the function. Detailed Implementation

[0033] In the following description, aspects of this disclosure will be described more fully with reference to the accompanying drawings, on which certain embodiments of the invention are illustrated. However, these aspects may be implemented in many different forms and should not be construed as limiting; rather, these embodiments are provided by way of example so that this disclosure will be comprehensive and complete, and will fully convey the scope of all aspects of the invention to those skilled in the art. Throughout the specification, the same reference numerals refer to the same elements.

[0034] Detecting objects in distance-Doppler data

[0035] To detect objects in cluttered radar data, the inventors have conceived a method 100 universally applicable to radar data in both the range-Doppler and range-angle planes. To convey the full generality of the inventors' contributions while providing a detailed description for putting the object detection method into practice, a novel object detection method will be described once for the range-Doppler case, and once for the range-angle case.

[0036] Reference Figure 1The flowchart in the diagram describes some embodiments of method 100. Method 100 takes range-Doppler radar data derived from radar measurements as input, wherein the radar measurements have been performed by the same entity as the entity executing method 100 or by a different entity. The entity executing method 100 does not require radar equipment to its control, but radar data is sufficient; therefore, the entity executing method 100 can simply be a general-purpose processor with general data input and data output capabilities.

[0037] One possible application of method 100 is target detection or target tracking, especially in situations where radar clutter is expected. For illustration, Figure 2 An example scenario 290 and a radar device 230 arranged to monitor scenario 290 are illustrated. Scenario 290 is located in a built environment containing a vehicle 291, plants 292, a building 293, and a garden sprinkler 294. It is assumed that vehicle 291 and building 293 constitute expected radar targets of interest to the owner or operator of radar device 230, while plants 292 and water droplets from sprinkler 294 cause unwanted radar reflections that would obscure radar data. (In some use cases, static objects like building 293 can also be considered unintended radar targets.) The object detection method 100 to be described aims to avoid reporting reflections from plants and water droplets as positive results, which would be false alarms in this case.

[0038] As those skilled in the art will know, radar device 230 includes a radar transmitter 231 and a radar receiver 232. The radar transmitter 231 is configured to transmit an outgoing radio frequency (RF) beam toward scene 290, and the radar receiver 232 is configured to receive an incident RF beam as a reflection (backscattering) of the outgoing RF beam departing from an object in scene 290. Radar device 230 may be a frequency modulated continuous wave (FMCW) radar device. Although not explicitly stated... Figure 2 As shown, but it should be understood, radar device 230 may further include drive circuitry and control circuitry. The teachings of this disclosure are not limited to single-transmitter, single-receiver radar devices, but can also be applied to radar devices comprising multiple physical transmitters and / or multiple physical receivers. Such radar devices can operate according to a repetitive transmission sequence or a transmission schedule with another configuration. In particular, the radar device can operate according to a MIMO method, such as a time-division multiplexing (TDM) multiple-input multiple-output (MIMO) method.

[0039] exist Figure 2In the setup depicted, radar device 230 shares its data with signal processing apparatus 240 configured to perform object detection via a wired or wireless data connection 250. For implementation of this invention, only a one-way data connection 250 (towards signal processing apparatus 240) is required. Signal processing apparatus 240 may include processing circuitry 241, a memory 242 suitable for storing a computer program 243, a data interface to radar device 230, and internal communication lines (data bus, etc.). Processing circuitry 241 may be, for example, a general-purpose (programmable) circuit, application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or system-on-a-chip with one or more processing cores. Recall that a radar signal processing chain may include a sequence of functional levels starting from the antenna side: mixing, analog-to-digital conversion, RF front-end processing based on intermediate frequency (IF) signals, and possibly digital beamforming. Different processing chains may integrate these levels to varying degrees. Therefore, signal processing apparatus 240 performing method 100 may be adapted to be deployed as a general-purpose radar baseband processor, a combined front-end and beamforming apparatus, or a dedicated digital beamforming apparatus.

[0040] In the first of method 100 Step 110 In this process, radar data from radar measurements in scenario 290 is obtained. The radar data includes intensities associated with multiple points in the range-Doppler plane. Each intensity value may correspond to a combination of range and Doppler cells (collectively referred to as a range-Doppler cell), or it may correspond to a range-Doppler value representing the center of the observation that contributes to that intensity value. Intensity refers to the amplitude of the incident RF propagating from the target in scenario 290 to radar receiver 232.

[0041] For the purpose of mathematical description of step 110, and for simplicity, it will be assumed that radar device 230 is a single-input single-output device, wherein radar transmitter 231 transmits six pulses. Application pulse length and pulse repetition time The pulse is reflected by the target in scene 290 and measured by radar receiver 232 as occurring within a time interval. discrete points in The sampled intensity value. (In practical implementations, discretization can be more refined, and calculations can be based on data from a larger number of pulses. Using common knowledge and / or consulting textbooks and other references, those skilled in the art can derive the following mathematical expressions to suit practical applications, and they can also derive corresponding expressions applicable to multi-input and / or multi-output radars.) The sampled IF intensity value can be written in matrix form as follows:

[0042]

[0043] Each row corresponds to one of the pulses, and each item can be interpreted as a time sample of that pulse. Distance information can be obtained by applying a discrete harmonic transform, such as DFT or FFT, to each row of the IF signal. If FFT is used, this will produce the following distance spectrum (“Distance FFT”):

[0044]

[0045] The row dimensions of this matrix now correspond to the distance, where, This can be interpreted as a distance unit, that is, an interval in the radial distance to the reflecting object. The column dimension remains the same as the six pulses. Correspondingly, all information in the matrix has been derived from measurement data read from a virtual array element. This is achieved through... Each column is further subjected to an FFT to obtain the distance-Doppler spectrum (or "Doppler FFT"):

[0046]

[0047] Matrix that is usually complex Each term in the equation can be understood as an element in a discrete representation of the range-Doppler spectrum. For example... The superscript should be understood as referring to the first... Velocity (or Doppler) unit and the first Distance cell, or simply the first Range-Doppler cells. It is important to note that velocity is a signed quantity; in this sense, the range-Doppler spectrum allows for the distinction between radial motion toward the radar and radial motion away from the radar.

[0048] In fact, there is dedicated hardware (e.g., chipsets, optionally integrated in TDM MIMO radar devices) for calculating the range-Doppler spectrum, which enables the acquisition of radar data in range-Doppler format without having to perform the above calculations. Since such hardware and the corresponding algorithm library are available to the implementer of method 100, step 110 of obtaining radar data in method 100 should be considered complete once the data according to equation (3) is available.

[0049] In the next step of method 100 Step 112 In this context, a one-dimensional subset of the range-Doppler plane is parameterized by a single parameter. Example subset 510 represents Doppler velocities on the horizontal axis. And the vertical axis represents distance. of Figure 5A , Figure 5B and Figure 5C The image in the middle shows...

[0050] In clearly indicating discrete distance-Doppler points Figure 5A In this context, subset 510 corresponds to a constant distance value: , , And so on. Each subset can be accessed via speed. Speed ​​of rescaling (in, Or more generally, by speed Parameterize any non-decreasing continuous function.

[0051] exist Figure 5B In this context, the subset corresponds to a line in the distance-Doppler plane, and is of the form: ,in And it is a constant and Different individual subsets have different values. Each such subset can be ordered according to... or Or any equivalent form via real parameters Parameterize it. Although in Figure 5B While not explicitly stated, range-Doppler data can be discrete because radar data only has integer pairs of values. (or some other discretization of the range-Doppler plane) intensity values. In this case, radar data associated with a subset can be effectively correlated with that subset and the integer points where it intersects.

[0052] exist Figure 5C In the range-Doppler plane shown, there exist two one-dimensional subsets 510.1 and 510.2 corresponding to constant velocity values, for example... and . Figure 5C It further shows that it can be represented by, for example The parameterization of a one-dimensional circular subset is performed using the angle parameters.

[0053] Generally, parameterizing a subset involves determining appropriate parameters so that the points in the subset can be traversed by changing the parameters. Then, each point corresponds to a parameter value. Parameter It can be a scalar parameter that takes a value from a real number or a subrange thereof. As shown in the example above, a given parameterization is not unique, as it can be replaced by a shifted or rescaled version.

[0054] In the following of method 100 Step 113 In this process, radar data with a threshold below a pre-configured clutter threshold is extracted from radar data associated with a subset of the data. The intensity of the clutter is determined. The extracted intensity values ​​will be further processed in subsequent steps. Therefore, the configuration of the clutter threshold is not part of this method 100. Instead, the clutter threshold may have been pre-configured, for example, by the user or system owner. Clutter Threshold Typically, it is a globally applicable value intended to be applied to various scenarios and / or for any time and date used to execute methods. Clutter threshold It can be based on the expectation or estimation of the upper bound of the typical clutter intensity that will be encountered in one or more scenarios to be measured.

[0055] Figure 3 The distance-Doppler plane 300 is shown by The specified example clutter threshold is 310. Figure 4 The location of the example clutter threshold 310 in a subset of the distance-Doppler plane is shown.

[0056] Next, in Step 114 In this process, the extracted intensities are processed as samples of a (unknown) function of the parameters of a subset. A signal mask is obtained by applying a regularization operation. The regularization operation may include fitting a smoothing function of the parameters to the extracted intensities 114.1. In this sense, the smoothing function may be a continuous function of the parameters, particularly a function of piecewise continuously differentiable parameters. For example, the smoothing function may be a polynomial such as a quadratic polynomial (less than second order), a cubic polynomial (less than third order), or more generally a piecewise combination of quadratic or cubic polynomials. In other embodiments, as an alternative to the function fitting operation 114.1, the regularization operation acts directly on the extracted intensity values ​​and transforms them into intensity values ​​that are more regular with respect to the parameters. The direct regularization operation may include smoothing by convolving with a convolution kernel with non-zero support (e.g., a Gaussian kernel or a kernel with compact support) or by performing discrete convolution with the convolution vector.

[0057] In some embodiments, where a one-dimensional subset of the range-Doppler plane is a range slice (constant range) parameterized by Doppler velocity, the signal mask 420 can be determined by a regularization operation 114, preferably an even function, such as a function symmetric about the origin or about another input value. For example, the signal mask 420 can be determined based on an even pseudo-model (e.g., a cosine series over an interval or a sum of a finite number of cosines over an interval), wherein the pseudo-model is fitted 114.1 to the extracted intensity value 410 by determining one or more shape parameters (e.g., cosine coefficients) of the pseudo-model. Specifically, given that clutter reflections are often so centrally located, the signal mask 420 can preferably be determined by an even function centered at or near zero Doppler velocity. In specific examples such as outdoor vegetation swaying in the wind and wave crests on water, the instantaneous velocity has an expected value of zero. A stationary clutter target is obviously stationary relative to a stationary radar receiver 232. Raindrops tend to move at low horizontal speeds unless they experience strong winds. Optionally, if the (radial) wind speed is known, the signal mask 420 can be determined based on a hypothetical model centered on the Doppler velocity corresponding to that wind speed. The method 100 according to these embodiments is particularly well suited for clutter targets of the types described above.

[0058] The horizontal axis represents velocity. And the vertical axis represents strength. of Figure 4 The figure illustrates an example of what the radar data looks like after the extraction operation in step 113. (Not in...) Figure 4 The image shows unextracted radar data. The subset considered represents a constant range, and it can be determined by velocity. Parameterization was performed. None of the extracted intensity values ​​(410) exceeded the clutter threshold. The signal mask 420 has been obtained by regularizing the extracted intensity value 410, and therefore can be said to be a smooth approximation of the extracted intensity value 410. The signal mask 420 can be spliced ​​together, for example, at two junctions between the flat tail and the raised center portion, according to a quadratic or cubic polynomial, where the first derivative of the signal mask 420 is discontinuous.

[0059] Continuing with the description of method 100, the execution flow proceeds to deriving the detection threshold 430 from the signal mask 420. step Step 115The detection threshold 430 will be used to detect objects in the range-Doppler data associated with subset 510. More precisely, the intensity of the range-Doppler cells is compared to the detection threshold, and if the intensity exceeds the detection threshold, a target is considered to be potentially present in scene 290 at that distance and moving at that Doppler velocity. The detection threshold 430 is derived from the signal mask 420 in the sense that it inherits some characteristics of the signal mask 420, such as its waveform or other characteristics that may represent clutter regions of the subset. (Compared to the clutter threshold...) Unlike other parameters, the detection threshold 430 is typically not constant; that is, it is inconsistent with the parameters of a subset. When no clutter is present, the signal mask 420 can be approximately correlated with a noise substrate that is substantially consistent with the parameters.

[0060] Examples representing the characteristics of clutter regions are: Figure 4 The elevated central portion of the signal mask 420 seen in the image; a copy of this elevated portion (possibly after translation and / or rescaling) should also preferably be present in the detection threshold 430. Portions outside the clutter characteristics (tails) are not necessarily inherited by the detection threshold 430, but they can be replaced by a pre-configured detection threshold.

[0061] To achieve these objectives while deriving the detection threshold 430 from the signal mask 420, such as Figure 4 As shown, the offset can be Added to signal mask 420, where the offset is relative to the parameter It is constant. If, as will be discussed in more detail below, multiple subsets 510 are processed within the execution of method 100, the pre-configured offset values ​​can be used. Apply to all subsets. Alternatively, different offset values ​​can be used. Furthermore, depending on the spatial characteristics of the 510 subset, particularly different offset values ​​can be used. For example, if the subset increases with a constant distance... Correspondingly (each subset is parameterized by Doppler velocity), a reduced offset value can be used. This partially compensates for the fact that radar reflections at greater distances attenuate to a greater extent due to the longer propagation distance. Another option is to use an offset calculated based on the noise level of the radar data considered from a subset of the data. Preferably, the offset is positively correlated with the noise level of the radar data.

[0062] Use a constant clutter threshold Together with its value An offset that varies with the spatial characteristics of subset 510 will produce a detection threshold that varies relative to the same spatial characteristics. A detection threshold with equivalent variation can be achieved by using a constant offset. Together with its value This is achieved by a clutter threshold that varies with the same spatial characteristics of subset 510. Similarly, the desired dependence on noise can be introduced via a clutter threshold or via an offset.

[0063] Then, the execution flow of method 100 can proceed to object detection. Step 116 Object detection is threshold-based, where the detection threshold 430 derived in step 115 is applied to radar data relating to the considered subset 510. As described above, due to the specific data-adaptive nature of the detection threshold 430, object detection 116 has a low probability of reporting clutter data as objects.

[0064] The detection threshold 430 can be applied to all radar data (i.e., all intensity values) related to subset 510. Alternatively, to reduce the number of comparisons and memory operations, some radar data is excluded from object detection based on some heuristic or shortcut. For example, if the detection threshold 430 is calculated by adding a constant positive offset to the signal mask 420, all the intensities extracted in step 113 are significantly smaller than the detection threshold 430, and it can be considered that any of these intensities is unlikely to lead to effective object detection. Therefore, object detection can be conveniently limited to a supplement of the extracted intensities.

[0065] In some embodiments, the execution of method 100 may terminate after object detection 116 is completed. The output of object detection 116 may be presented through a user interface.

[0066] In other embodiments, such as by Decision point 117 It is suggested that, for at least one additional one-dimensional subset 510 of the distance-Doppler plane, the steps of parameterization 112, extraction 113, regularization 114, thresholding 115 and object detection 116 be repeated (from the "no" branch of decision point 117).

[0067] In global object detection Step 118 In this process, the outputs of two or more subsets can be combined to achieve a more complete understanding of the objects present in scene 290.

[0068] Alternatively or additionally, the reliability of object detection can be further improved by combining the outputs of adjacent, intersecting, or overlapping subsets. For example, if two or more overlapping or neighboring detections are found in different subsets, this indicates that the probability of an object being present is higher than if only a single detection were found. Accordingly, the output of object detection can optionally be presented along with an indication of the relative probability (reliability) of each detected object.

[0069] Further, alternatively, or additionally, global object detection 118 includes using a global detection threshold obtained by combining detection thresholds 430 from two or more subsets 510. The global detection threshold can be obtained by combining two variables (distance)... and Doppler The smoothing function is determined by fitting a detection threshold 430 from two or more subsets 510. (This can be equivalently achieved by using distance...) and Doppler The smoothing function is fitted to a signal mask 420 from two or more subsets and then a global detection threshold is derived from the smoothing function. Then, the object detection step 116 for the two or more subsets 510 is performed jointly and using the global detection threshold, that is, the global detection threshold is applied to the intensity values ​​of the entire range-Doppler plane, rather than to its subsets.

[0070] In some embodiments, regardless of whether one or more subsets 510 are processed, method 100 may further include estimating the noise basis based on radar data. Step 111 Highly accurate noise base estimation methods are known in the art. When a noise base is available, the derivation 115 of the detection threshold 430 includes a preliminary sub-step evaluating whether the noise base substantially corresponds to the signal mask 420; if this is true, the detection threshold 430 is calculated based on the noise base rather than the signal mask 420. As mentioned above, when there is no clutter, the signal mask 420 can approximately correspond to the noise base according to conventional techniques. The use of a flat detection threshold 430 simplifies object detection 116. In embodiments, the noise base can be set to, for example, equal to the median intensity or the second-highest intensity. The traditional method of using intensity percentage points is used for estimation 111, where, .

[0071] Detecting objects in distance-angle data

[0072] The description will now include the distance-angle plane (where the coordinate axes are distances). and Angle of Arrival (AoA) An embodiment of method 100 is to take radar data relating to the intensity of multiple points in the radar as input.

[0073] It is assumed that radar device 230 is configured, specifically by including a radar array, to resolve characteristics regarding the AoA. The radar array can consist of a single physical transmitter and multiple physical receivers, or it can have multiple physical transmitters and a single physical receiver. The effective number of elements in a physical radar array equals the number of physical receivers. MIMO (Multiple-Input Multiple-Output) radar arrays have multiple physical receivers and... A physical transmitter, and this produces a number of physical transmitters. A virtual radar array with elements, wherein... This refers to the number of physical receivers. Physical transmitters in a MIMO radar array can use multi-carrier signals for synchronized feedback, or alternate between physical transmitters (TDM MIMO). Although this embodiment of method 100 is not limited to any of the radar configurations outlined, the mathematical description will involve radar transmitter 231 having a single physical transmitter ( Furthermore, the radar receiver 232 has eight physical receivers RX1, RX2, ..., RX8. A simple example.

[0074] In step 110, radar data in intensity form related to multiple points in the range-angle plane 600 is obtained. The basic calculations can follow one of the following methods. The quantities discussed in step 110 above... In this configuration, each physical receiver has one copy. This includes eight sets of sampled IF intensity values:

[0075] ,

[0076] Eight distance spectrums:

[0077] ,

[0078] And eight distance-Doppler spectra:

[0079] .

[0080] To illustrate, the distance-Doppler spectrum of the first physical receiver RX1 can have this appearance:

[0081]

[0082] According to the first method, in order to calculate the radial velocity Distance-angle spectrum of a moving object, from all spectra Collection of the first Distance-Doppler units are used to form an array signal:

[0083] .

[0084] The phase shift between elements is given as the sum of the phase shift caused by velocity and the phase shift caused by AoA. Due to the path difference between physical receivers, the phase shift caused by AoA can be observed when AoA is non-zero in the plane of the receiver array.

[0085] In the preparation of AoA estimation, the phase shift caused by velocity is eliminated by a phase compensation method, thereby obtaining a compensated array signal:

[0086]

[0087] The phase compensation method can be any known phase compensation method from the literature. For example, the phase compensation method described in patent publication US10627483B2 can be applied. The Doppler phase used in this method... According to the RF beam emitted towards scene 290 Pulse repetition time and carrier frequency To calculate. Compensated array signal The phase difference between two elements and the phase shift caused by AoA Correspondingly, this phase shift AoA is related to the following formula:

[0088]

[0089] in, It refers to the spatial spacing of the elements. This applies to the compensated array signal. Applying FFT operations to generate as a target The strength of the function of the angle. And further distance values. Correspondingly, for array signals These calculations are repeated to produce a complete range-angle image. The intensities that make up the range-angle image can be called range-angle units. This is the data received in step 110 of method 100.

[0090] According to the second method, the range-angle spectrum of an object traveling at any radial velocity is calculated based on the range FFT. This is done according to the range spectrum of different receivers. The following array vectors are based on the same pulse (here: ) and the same distance (here: The corresponding elements are used to form:

[0091] .

[0092] Applying FFT operations to array vectors Generate as a target The strength of the function of the angle. And further distance values. Correspondingly, for array signals Repeat these calculations to produce a complete distance-angle image.

[0093] In step 112 of method 100, a one-dimensional subset of the distance-angle plane 600 is parameterized by a parameter. Based on the description above, which need not be repeated here, this step can be performed exactly as in the distance-Doppler plane. Also relevant is... Figure 5A , Figure 5B and Figure 5C Example subset 510 is illustrated in the figure, where the horizontal axis represents the Doppler velocity. And the vertical axis now represents AoA .

[0094] In the next Step 113 In this process, radar data with a threshold below a pre-configured clutter threshold is extracted from radar data associated with a subset of the data. Its strength is 410. Figure 6 The distance-angle plane 600 is shown as... The specified example clutter threshold is 310. Figure 7 The location of an example clutter threshold 310 is shown in a subset of the distance-angle plane 600.

[0095] In the next Step 114 In this process, the extracted intensity 410 is processed into samples of a function of the parameters of a subset. A signal mask 420 is obtained by applying regularization. This process is performed by... Figure 7 The figure shows that the horizontal axis represents AoA. And the vertical axis represents strength. The subset under consideration represents a constant distance, and it can be accessed via AoA. The parameters are parameterized as multiples of the extracted intensity. As explained above, the regularization operation can include fitting a smoothing function of the parameters to the extracted intensity [114.1]. Example smoothing functions have already been mentioned above and need not be repeated here.

[0096] In this embodiment, the one-dimensional subset of the distance-angle plane 600 can be obtained through AoA. Perform parameterized range slicing (constant distance). This allows for consideration of various exploratory methods or hypotheses regarding the sources of known clutter, as desired by the implementer. Known clutter sources can include omnidirectional sources (raindrops, suspended dust, etc.) and local sources occupying bounded angular intervals (e.g., localized vegetation). It is conceivable... Figure 7 The figure illustrates the presence of clutter in the form of reflections from the leaves of a tree located to the right of the main incident direction on radar receiver 232. Signal mask 420 follows this inconsistent clutter contribution and therefore ignores it during the subsequent object detection step 116.

[0097] In a further step 115, a detection threshold 430 is derived from the signal mask 420. The detection threshold 430 will be used to detect objects in the range-angle data associated with subset 510. More precisely, if the intensity of a range-angle cell exceeds the detection threshold 430, a target located at that range and AoA is considered likely to exist in scene 290. Similar to the case of range-Doppler data, the detection threshold 430 is generally not constant, but it only rises above the noise floor when local clutter sources are present.

[0098] Examples representing the characteristics of clutter angular regions are: Figure 7 The signal mask 420 in the middle has a smooth, rounded peak on the right side; a copy of this peak shape can be expected to reappear in the detection threshold 430. Parts other than those representing clutter features are not necessarily inherited by the detection threshold 430, but they can be replaced by pre-configured detection thresholds.

[0099] To achieve these objectives while deriving the detection threshold 430 from the signal mask 420, such as Figure 7 As shown, the offset can be Add to signal mask 420. As explained above, offset. The parameter relative to the subset (here: AoA) is constant. It may optionally have a dependency on the spatial characteristics (e.g., distance) of the subset 510, or it may be consistent across all subsets 510. Considerations related to these choices of implementation have been discussed above and are equally applicable to the case of distance-angle data.

[0100] Then, the execution flow of method 100 can continue to threshold-based object detection. 116 In this step, the detection threshold 430 derived in step 115 is applied to the radar data relating to the considered subset 510. Because the detection threshold 430 is derived in a manner adapted to the radar data, object detection 116 has a low probability of reporting clutter data as an object.

[0101] Further developments of the method 100 described above in the context of range-Doppler data are also applicable to embodiments that process range-angle data.

[0102] The aspects of this disclosure have been described above with reference to several embodiments. However, as will be readily understood by those skilled in the art, other embodiments besides those disclosed above are also possible within the scope of the invention as defined by the appended claims.

Claims

1. A method for detecting objects in radar data, the method comprising: The radar data is obtained from the radar measurements by performing radar measurements of the scene or by receiving radar data from the radar device via a data interface, wherein the radar data includes the intensity of multiple points in the range-Doppler plane or the range-angle plane. The distance-Doppler plane or a one-dimensional subset of the distance-angle plane is parameterized by a single parameter. Extract the intensity of clutter below a pre-configured intensity threshold from the radar data associated with the subset; The extracted intensity is regularized as a function of the parameters to obtain a signal mask; Derive the detection threshold from the signal mask; and Perform object detection, wherein the detection threshold is applied to the radar data relating to the subset.

2. The method according to claim 1, wherein, Deriving the detection threshold involves adding an offset to the signal mask, wherein the offset is constant relative to the parameter.

3. The method according to claim 2, wherein, The offset has a pre-configured value.

4. The method according to claim 2, wherein, The offset is calculated based on the radar data associated with the subset.

5. The method according to any one of the preceding claims, wherein, The intensity extracted by regularization includes fitting a smoothing function of the parameters to the extracted intensity.

6. The method according to claim 5, wherein, The smoothing function is defined by a quadratic polynomial piecewise or a cubic polynomial piecewise.

7. The method of claim 1, further comprising: The noise floor is estimated based on the radar data. The derivation of the detection threshold includes evaluating whether the noise basis substantially matches the signal mask, and if so, calculating the detection threshold based on the noise basis.

8. The method of claim 1, further comprising: The parameterization, extraction, regularization, thresholding, and object detection are repeated for at least one additional one-dimensional subset of the distance-Doppler plane or the distance-angle plane.

9. The method of claim 8, further comprising: Performing global object detection includes combining the output of object detection from two or more subsets and / or using a global detection threshold obtained by combining detection thresholds from two or more subsets.

10. The method according to claim 1, wherein, In the case of the range-Doppler plane, the one-dimensional subset corresponds to a range slice parameterized by Doppler velocity.

11. The method according to claim 10, wherein, The regularization is preferably performed using an even function for the signal mask.

12. The method according to claim 11, wherein, The even function is centered at zero Doppler velocity.

13. The method according to claim 1, wherein, In the case of the distance-angle plane, the one-dimensional subset corresponds to a distance slice parameterized by the angle of arrival.

14. A signal processing apparatus comprising processing circuitry configured to detect an object in radar data by performing the method of claim 1.

15. A computer program comprising instructions that, when executed by a computer, cause the computer to perform the method according to claim 1.

Images