Method for detecting an object in a radar signal of automotive grade

Abstract

The method includes acts of collecting radar signals reflected from objects within a field of view. Range-angle-Doppler bins are formed that represent three-dimensional objects within the field of view. Background noise within the range-angle-Doppler bins is removed using a local median operation over a selected dimension of the range-angle-Doppler bins. Low energy peak regions are masked by removing radial velocity values ​​in the selected dimension to form a sparse two-dimensional range-angle grid. The radar signals reflected from objects within the field of view are processed to extract reflection point detections. The reflection point detections are tracked according to a short-term filter rule to form tracked reflection point detections. The tracked reflection point detections are formed into clusters. The clusters are processed using a long-term filter rule.

Description

Technical Field

[0001] Cross - Reference to Related Applications This application claims priority to U.S. Provisional Patent Application No. 63 / 143,154, filed on January 29, 2021, the content of which is incorporated herein by reference. This application is related to U.S. Provisional Patent Application No. 63 / 123,403, entitled "METHOD, APPARATUS AND RADAR SYSTEMS FOR TRACKING OBJECTS", filed on December 9, 2020, the entire content of which is incorporated herein by reference.

[0002] The present disclosure relates to techniques for detecting up - sampled objects in automotive - grade radar signals. More specifically, the present disclosure describes background estimation and peak region verification in a radar environment, as well as subsequent filtering techniques.

Background Art

[0003] An autonomous vehicle (AV) is a vehicle configured to navigate a road based on sensor signals output by sensors of the AV, and the AV navigates the road without human input. The AV is configured to identify and track objects (such as vehicles, pedestrians, bicycles, stationary objects, etc.) based on sensor signals output by sensors of the AV, and to perform driving operations (such as acceleration, deceleration, turning, stopping, etc.) based on the identified and tracked objects.

[0004] As a result of advancements in sensing technologies (e.g., object detection and position tracking), control algorithms, and data infrastructure, the use of automation in the operation of road vehicles such as passenger cars and trucks is increasing. By combining adaptive cruise control (ACC), lane keeping assistance (LKA), electronic power assist steering (EPAS), adaptive front steering, parking assistance, antilock braking (ABS), traction control, electronic stability control (ESC), blind spot detection, GPS and map databases, vehicle-to-vehicle communication, and various other enabling technologies, it is possible for a vehicle to operate autonomously (i.e., with little or no driver intervention).

[0005] In the field of autonomous or semi-autonomous operation of vehicles such as aircraft, ships, and land vehicles, particularly manned or unmanned vehicles, sensing the surroundings of the vehicle and tracking objects in the vicinity of the vehicle can be considered important for advanced functions. These functions can range from driver assistance systems at various levels of autonomy to full autonomous operation of the vehicle.

[0006] In certain environments, multiple different types of sensors are used to sense the surroundings of a vehicle, such as monocular or stereo cameras, light detection and ranging (LiDAR) sensors, and radio detection and ranging (radar) sensors. Different sensor types have different characteristics that can be utilized for different tasks.

[0007] Embodiments of the present disclosure relate to ways of processing measurement data of a radar system, thereby reducing the fusion of computationally intensive sensor data (e.g., range, angle, and velocity). This is particularly useful when a certain parameter array needs to be embedded before processing another parameter array such as range or velocity.

[0008] A radar system typically provides measurement data, particularly range, Doppler, and / or angle measurements (azimuth and / or elevation angle) with high precision in the radial direction. This enables accurate measurement of the (radial) distance and (radial) velocity between different reflection points and the (respective) antennas of the radar system within the field of view of the radar system.

[0009] The radar system transmits (radiates) a radar signal within the field of view of the radar system, and the radar signal is reflected by an object present within the field of view of the radar system and received by the radar system. The transmitted signal is, for example, a frequency modulated continuous wave (FMCW) signal. The radial distance can be measured by utilizing the travel time of the radar signal, and the radial velocity is measured by utilizing the frequency shift caused by the Doppler effect.

[0010] By repeating the transmission and reception of the radar signal, the radar system can observe the field of view of the radar system over time by providing measurement data comprising a plurality of, particularly consecutive, radar frames.

[0011] Each individual radar frame may be, for example, a range azimuth frame or a range Doppler azimuth frame. If elevation angle data is available, a range Doppler azimuth elevation frame is also conceivable.

[0012] Furthermore, a radar system with a chirp sequence having a frequency offset and / or a time offset between chirps can be used. In that case, the resulting radar frame has a fast time axis, a slow time axis, and an azimuth / elevation angle. In the case of an equidistant chirp sequence where the start frequency for each chirp is the same, the range axis corresponds to the fast time axis and the Doppler axis corresponds to the slow time axis. Subsequently, range / Doppler notation is used, but this technique functions similarly for fast time notation and slow time notation as well.

[0013] In each of a plurality of radar frames, a plurality of reflection points that can form a cloud of reflection points can be detected. However, each reflection point or cluster of points within a radar frame does not, by itself, contain a semantic meaning. Therefore, semantic segmentation of the radar frame is necessary to evaluate ("understand") the situation around the vehicle.

[0014] Segmentation of the radar frame means that a meaning is assigned to individual reflection points within a single radar frame. For example, a reflection point can be assigned to a background of the scene, a foreground of the scene, a building, a wall, a stationary object such as a parked vehicle or a part of a road, and / or a moving object such as another vehicle, a bicycle, and / or a pedestrian in the scene.

[0015] Generally, a radar system observes the specular reflection of the transmitted signal radiated from the radar system because the object being sensed tends to have a smoother reflection characteristic than the (modulated) wavelength of the transmitted signal. Therefore, the acquired radar frame does not contain a continuous area representing a single object, but rather contains single prominent reflection points (such as the edge of a bumper) dispersed throughout the area of the radar frame.

[0016] Radar data forms a three-dimensional complex-valued array (also known as a radar cube) with dimensions corresponding to azimuth angle (in degrees), radial velocity (Doppler), and radial distance (range). When the magnitude of each angle-Doppler-range bin is obtained, it reveals the amount of energy that the radar sensor recognizes as coming from that point (angle and range) in space for that radial velocity.

[0017] The problem in the art arises from the vast amount of energy data embedded in the cube. As a result, processing the data in a real-time environment becomes impossible. One solution currently found in the art involves processing one dimension (i.e., parameter) at a time. However, this is not useful for certain applications such as 3D cube processing. SUMMARY OF THE INVENTION PROBLEM TO BE SOLVED BY THE INVENTION

[0018] In the art, there is a demonstrated need to perform background estimation and peak estimation while substantially preserving the three-dimensional radar cube. MEANS FOR SOLVING THE PROBLEM

[0019] A method for compressing data within a radar cube while preserving significant information is disclosed. Peak verification and peak region masking are used to enhance and separate range-angle-Doppler bins corresponding to objects within a scene. The noise background is estimated using an efficient (approximate) local median operation, which is used to mask bins not belonging to objects within the scene. Peak region verification can then be performed on peaks located within a thresholded cube. Low energy peak regions are masked. This is accomplished by removing one dimension (preferably, radial velocity) of the cube using a median operation to avoid the need to compute computationally expensive background statistics in 3D. As a result, an efficient 2D local median is applied to a sparse range-angle grid, which is then upsampled with bilinear interpolation.

[0020] According to one aspect, the present disclosure applies a first median operator to one dimension of the radar cube, thereby reducing further computational processing.

[0021] According to another aspect, a second median operator is applied to the remaining two or more dimensions.

[0022] According to one or more aspects, the result is upsampled.

[0023] According to one or more aspects, the upsampling is performed using bilinear interpolation.

[0024] According to some aspects, the first median operator is applied to radial velocity.

[0025] According to some aspects, the second median operator is applied to a sparse range angle grid.

[0026] According to one aspect, the method can include peak region verification, or be used for peak region verification, or used in conjunction with peak region verification.

[0027] According to one aspect, the method can include threshold masking, or be used for threshold masking, or used in conjunction with threshold masking.

[0028] According to one aspect, the method can include background estimation, or be used for background estimation, or used in conjunction with background estimation.

[0029] The present disclosure is best understood by reading the following detailed description in conjunction with the accompanying drawings. It is emphasized that, in accordance with standard practice in the industry, various features are not necessarily drawn to scale and are used for illustrative purposes only. Where scale is explicitly or implicitly shown, it provides only one exemplary example. In other embodiments, for clarity of discussion, the dimensions of various features may be arbitrarily enlarged or reduced. Similarly, for clarity and brevity, not all components are labeled in all drawings.

[0030] To more fully understand the nature and advantages of the present invention, reference should be made to the following detailed description of the preferred embodiments, in conjunction with the accompanying drawings.

Brief Description of the Drawings

[0031]

Figure 1A

Figure 1B

Figure 2

Figure 3A

Figure 3B

Figure 4

Figure 5A

Figure 5B

Figure 6

Figure 7

Figure 8A

Figure 8B

Figure 9

Figure 10

Figure 11

DETAILED DESCRIPTION OF THE INVENTION

[0032] The present disclosure relates to techniques for two-dimensional detection of objects in automotive grade radar signals. More specifically, the present disclosure describes background estimation and peak region verification in a radar environment. During radar signal processing, background subtraction tends to be performed overly aggressively, resulting in a large amount of computational effort. One method disclosed herein performs a median operation on one dimension of the radar cube.

[0033] This results in the removal of one dimension from the calculations when computing background statistics, such as from 3D to 2D, or from 4D to 3D, such that it can be used in multi-dimensional processing. In the example of going from 3D to 2D, the median operation can be used to remove the Doppler velocity. Subsequently, the local median 2D can be applied to a sparse range-angle grid. The result can then be upsampled using bilinear interpolation. Further, false peaks can be removed by a verification step.

[0034] The following description and drawings set forth particular exemplary implementations of the present disclosure in detail and illustrate several exemplary ways in which the various principles of the present disclosure may be carried out. However, the exemplary embodiments do not cover many possible implementations of the present disclosure. Other objects, advantages, and novel features of the present disclosure will be described hereinafter with reference to the drawings where applicable.

[0035] The present disclosure generally relates to millimeter-wave sensing, but other wavelengths and applications are not beyond the scope of the invention. Specifically, the method relates to a sensing technique called frequency-modulated continuous-wave (FMCW) radar, which is very popular in the automotive and industrial fields.

[0036] FMCW radar measures the range, velocity, and angle of arrival of objects in front. At the heart of FMCW radar is a signal called a chirp. FIGS. 1A and 1B show exemplary radar chirps as a function of time known in the art.

[0037] A chirp is a sine curve or sine wave whose frequency increases linearly with time. FIG. 1A shows this as an amplitude-versus-time, i.e., A-t plot. Referring to FIG. 1B, the chirp starts as a sine wave at a frequency fc, the frequency gradually increases, and finally reaches a value where B is added to fc, where B is the bandwidth of the chirp. The frequency of the chirp increases linearly with time, and the word linear is an important term. Thus, in an f-t plot, the chirp becomes a straight line with a slope S.

[0038] Therefore, a chirp is a continuous wave whose frequency is linearly modulated. Therefore, the term frequency-modulated continuous wave, or more simply FMCW, is used.

[0039] FIG. 2 shows an exemplary auto-grade radar system according to some embodiments. This is represented as a simplified block diagram of an FMCW radar with a single TX antenna and a single RX antenna. In one or more embodiments, the radar operates as follows. The synthesizer generates a chirp. This chirp is transmitted by the TX antenna. The chirp is then reflected by an object such as a car. The reflected chirp is received at the RX antenna. The RX signal and the TX signal are mixed in a mixer.

[0040] The resulting signal is called an intermediate (IF) signal. The IF signal is prepared for signal processing by low-pass (LP) filtering and sampled using an analog-to-digital converter (ADC). Next, the importance of the mixer will be described in more detail.

[0041] FIGS. 3A and 3B show the frequency difference in exemplary transmitted and received radar chirps according to some embodiments. In one or more embodiments, this difference is estimated using a mixer. As is known in the art, a mixer has two inputs and one output. When two sine waves are input to the two input ports of the mixer, the output of the mixer also becomes a sine wave as shown below.

[0042] The instantaneous frequency of the output is equal to the difference between the instantaneous frequencies of the two input sine curves. Therefore, the frequency of the output at any given time is equal to the difference between the input frequencies of the two time-varying sine curves at that time. Tau τ represents the time round-trip delay from the radar to the object and back. It can also be expressed as twice the distance to the object divided by the speed of light. A single object in front of the radar generates an IF signal with a constant frequency given by S2d / c.

[0043] Figure 4 shows two exemplary range matrices embedded by a radar frame according to some embodiments. The radar frame (left) has a time TF and includes a plurality of chirps 1 to N, each separated in time by Tc.

[0044] Each row corresponds to one chirp. That is, for each chirp, there is one row for the chirp index, i.e., N rows for N chirps. Each box in a particular row represents one ADC sample. Therefore, if each chirp is sampled M times, there are M columns in the matrix. Next, the conversion of the data matrix in the range matrix and the velocity matrix will be described.

[0045] Figure 5A shows the creation of a chirp range matrix from a previous data matrix according to some embodiments. As described above, each row corresponds to samples from a particular chirp. To determine the range, a range FFT is performed for each row. The fast Fourier transform (FFT) is an algorithm that computes the discrete Fourier transform (DFT) of a sequence or its inverse transform (IDFT). Fourier analysis transforms a signal from the original domain (often time or space) to a representation in the frequency domain and vice versa.

[0046] When range-FFT is applied, the object is decomposed in the range. As would be understood by one of ordinary skill in the art, the x-axis is actually the frequency corresponding to the range FFT bin. However, since the range is proportional to the IF frequency, it can be directly plotted as the range axis. Thus, FIG. 5A is a matrix of chirps where each chirp has an array of frequency bins. According to the above description, these bins directly correspond to the range via the IF.

[0047] FIG. 5B shows the creation of a velocity range matrix from a previous chirp index range matrix according to some embodiments. Doppler-FFT is performed along the columns of these range-FFT results shown in FIG. 5A. Thereby, the object is decomposed in the velocity dimension.

[0048] As can be understood, FIG. 5B shows two objects moving at two different velocities within the third range bin. Similarly, there are three objects moving at three different velocities within the eighth range bin. Note that these are accurate for a fixed range angle. Next, angle determination will be described in more detail.

[0049] FIG. 6 shows an exemplary antenna array used to calculate the angle according to some embodiments. At least two receiver (RX) antennas are required for angle estimation. To estimate the distance, the difference in the distance of the object to each of these antennas is utilized. Thus, the transmitting (TX) antenna transmits a signal that is a chirp. It is reflected from the object, and one ray goes from the object to the first RX antenna and another ray goes from the object to the second RX antenna.

[0050] In this example shown in FIG. 6, the ray to the second RX antenna needs to travel slightly longer. That is, an additional distance of delta d. This additional distance results in an additional phase of omega equal to 2pi delta d divided by lambda. Thus, this is the phase difference between the signal at this antenna and the signal at this antenna.

[0051] Figure 7 shows an exemplary range-angle-velocity radar cube according to some embodiments. As would be understood by one of ordinary skill in the art, when assembling the matrix, a 3D radar cube with range-angle-velocity axes is obtained. The methods disclosed herein describe techniques for processing and interpreting radar data extracted from one (or more) 77-GHz DigiMMIC (FMCW) radar sensors mounted on a moving vehicle, although other frequencies and applications are not beyond the scope of the present disclosure.

[0052] Radar cube data is in the form of a three-dimensional complex-valued array having dimensions corresponding to azimuth angle (angle), radial velocity (Doppler), and radial distance (range). The magnitude at each angle-Doppler-range bin is obtained to represent the amount of energy that the radar sensor considers to come from that point (angle and range) in space for that radial velocity. For demonstration purposes, a linear antenna array oriented in a direction parallel to the ground is assumed. Pillars with peak energy can be selected for peak verification, which will be described later in the present disclosure.

[0053] Figures 8A and 8B show scenes before and after image processing according to some embodiments. The purpose of the present disclosure is foreground extraction / background subtraction. Figure 8A is a raw image of a vehicle within a radar cube. From the relatively noisy image shown in Figure 8A, distinct peak energy regions can be extracted. The high-level idea is to separate reflections from objects within the vehicle scene by combining background noise estimation and masking of the peak regions. The former finds regions in the radar signal that stand out from the background, and the latter verifies and separates them for further processing.

[0054] The results are shown in Figure 8B. Figure 8B is a processed image of a vehicle within a radar cube having detected peaks (dots) and peak regions (non-black pixels). Next, the process will be described in more detail.

[0055] Foreground extraction can be performed by various methods for suppressing noise and artifacts and extracting prominent regions of the radar cube. In most, if not all, cases, it is necessary to estimate a background model and somehow remove it, or create a mask that emphasizes the desired bins and suppresses other bins through element-wise multiplication.

[0056] CFAR (Constant False Alarm Rate) thresholding is perhaps the most well-known and well-studied technique and involves the estimation of a background model by local averaging. Constant False Alarm Rate (CFAR) detection refers to a general form of an adaptive algorithm used in radar systems to detect the reflections of targets against the background of noise, clutter, and interference.

[0057] The fundamental idea is that the noise statistics may be non-uniform across the array. CA-CFAR (Cell Averaging) calculates a moving average while excluding the region (guard cell) at the center of the averaging window to avoid including the desired object in the background estimation. OS-CFAR (Order Statistics) performs the same calculation but uses a percentile operation instead of an average. Considering the background model (the estimated value of the background value at each bin) b ijk the foreground can be estimated as follows for some factor α that controls the amount of background suppression.

[0058]

Equation

[0059] One or more objectives of the present disclosure are to efficiently separate objects in an automotive scene from noise. The motivation is that automotive radar signals are composed of reflections from objects (e.g., cars, bicycles, buildings), and conventional methods may not be able to simultaneously achieve accuracy, efficiency, and usability in a machine learning pipeline.

[0060] Previous solutions have attempted only 2D or full 3D background estimation. Other solutions do not include peak verification or retention of peak regions.

[0061] Solutions to similar problems in related and other fields cannot be retained in the application to 3D radar data cubes. Also, barriers to computational efficiency arise.

[0062] FIG. 9 shows an exemplary method for detecting objects in automotive-grade radar according to some embodiments. The method includes medmed background modeling. Medmed refers to median-median, although other averaging techniques are not outside the scope of the present disclosure. The median is the value in an ordered set of upper and lower values with an equal number of values, or the arithmetic mean of the two central values if there is no single central numerical value.

[0063] The OS-CFAR described above is costly and lacks some flexibility as it operates on the entire radar cube. This implementation can be improved without significantly affecting the results as follows. For the purpose of explanation, assume that the background model is a function of only range-angle and not Doppler. By doing so, a full 3D moving median can be approximated by a median on Doppler followed by a 2D moving median on range-angle.

[0064] This allows for more flexible handling in the selection of the (2D) window size and overlap. The moving median on the sparse grid can be calculated and quickly upsampled to the original range-angle grid using widely used and optimized image resampling techniques. Two median operations give rise to the name of this method.

[0065] Referring to FIG. 9, the radar cube contains a large amount of data including both background clusters and energy clusters. These energy clusters or clouds can be the target areas identified during peak detection. However, before that is possible, it is necessary to remove (subtract) part of the large background to obtain a more efficiently processable and manageable data cube.

[0066] Typically in the art, positive thresholding is applied to the radar cube to compress the data by effectively discarding a large amount of information, i.e., data below the threshold. Image thresholding is a simple and effective way to divide an image into foreground and background. This image analysis technique is a type of image segmentation that separates objects by converting a grayscale image into a binary image. Image thresholding is most effective in high-contrast images.

[0067] By not performing threshold processing too aggressively, a larger area of the surrounding region can be maintained. This is useful for intelligent perceptual analysis. Configurational perception or intelligent perception is a theory of perception in which the perceiver uses sensory information and other sources of information to construct a cognitive understanding of a stimulus. In contrast to this top-down approach, there is a bottom-up approach of direct recognition. Perception is more of a hypothesis, and the evidence supporting this is that "perception enables appropriate actions in general with respect to the characteristics of objects that are not being sensed", which means that we react to obvious things such as doors even though we are only "seeing a long, thin rectangle as a half-open door".

[0068] In this embodiment, through intelligent perception, the radar system can not only identify the position, speed, and direction of an object (e.g., a car), but also positively associate it with a previously identified car. That is, the radar system can more easily temporally associate objects between scenes, thereby reducing unexpected situations.

[0069] In one or more embodiments, the fourth dimension includes tangential velocity. In state-of-the-art radar systems, Doppler can only resolve velocities closer to the observer. That is, approaching or moving away from the radar antenna. In some embodiments, tangential velocity can resolve velocities that are substantially perpendicular to the plane of the observer. More simply put, this includes left-right or up-down movement and velocity (since it is a vector, actually velocity). The 4D radar cube is called a tesseract.

[0070] Returning to FIG. 9, an absolute value operation is performed on the radar cube. As is known, the absolute value returns the magnitude of a complex number, which is simply the distance from the origin in the complex plane. In one or more embodiments, a median value is performed to achieve some threshold criterion. A moving median is desirable as a more robust statistic for averaging because the mean value can be lost by objects with large weights. That is, the median value can ignore high-energy objects in the scene. This is important in the background subtraction step.

[0071] However, since the median value requires sorting, the computational complexity increases. Assuming that the background does not depend on the radial velocity direction, one dimension of the radar cube is truncated as follows. A moving median in the radial velocity direction is performed. Then, this median value is used to reduce the 3D radar cube to a mere range - angle by replacing the velocity median in three dimensions. Its importance will be explained in more detail in the description of FIG. 10.

[0072] Once the background estimation is complete, masking can be performed. Masking removes background noise while leaving the target peak regions intact. As a result, a threshold cube that is computationally much easier to hold is obtained. This enables peak detection and filtering, which will also be described later in this disclosure.

[0073] Aspects of radar cube processing include using short - term track hypotheses (referred to as "tracklets") to model detections. Tracklets are clustered into groups to form long - term track hypotheses (referred to as "groups"). Tracklets and groups are subject to birth - death rules and various filters, as will be described below.

[0074] Next, the peak list can be processed using various state-of-the-art filtering and detection methods. These methods do not account for small variations in the reflection points for each measurement. For example, the reflection points on the curved surface of an object such as a car bonnet may change their position depending on the orientation and position of the car. Subsequently, this movement of the reflection points adds an artificial movement that does not correspond to the actual movement of the object. This offset makes it difficult to associate all reflection points with the same object. Furthermore, as the object moves, new reflection points appear on the object and other reflection points disappear. To mitigate this problem, the following novel technique is used.

[0075] As a first step, short-term hypotheses are used to model the dynamic / short-term movement of each detection from frame to frame. Tracked / filtered detections are previously referenced tracklets. As a next step, several tracklets (at least one, but a number greater than one is desirable) are clustered. This cluster is then tracked by a long-term hypothesis to model the movement underlying the object. The individual tracklets and groups are preferably modeled using alpha / beta-kalman filter rules and using birth and death rules. The Kalman filter uses a series of measurements observed over a long period to generate an estimated value of an unknown variable, which tends to be more accurate than an estimated value based on only a single measurement. The accuracy is due to the joint probability distribution over the variables for each time frame. An example of a birth rule is that there needs to be reflections in several, for example three, frames in order to be tracked as a tracklet. Similarly, an embodiment of a death rule is that for a tracklet to die, the corresponding detection of the tracklet must fail in three consecutive frames. The technical advantage of this method is that new and / or disappearing reflection points are accurately modeled and smaller fluctuations are compensated for. Similar techniques are used for an object (grouped or clustered tracklets) to be present in / enter / exit the scene.

[0076] It is advantageous to use a filter without constraints on the movement of the short-term hypothesis filtering, for example using a hovercraft object motion filter for the short-term hypothesis filter. In the case of the long-term hypothesis filter, it is preferable to use a filter with a more limited motion model, for example a vehicle motion model.

[0077] Figure 10 shows an exemplary median background estimate on a sparse range-angle grid according to some embodiments. As described above, the background estimate can be performed for range-angle using the velocity (Doppler) median. The median background estimate on a sparse range-angle grid includes a moving median on a 2D grid.

[0078] In some embodiments, the median background estimate for sparse range-angle includes sorting the median over eight points surrounding the origin of the square in question. In other embodiments, any number of two-dimensional points can be used. In still other embodiments, an excess mean metric can be utilized, which is also within the scope of the present disclosure.

[0079] Peak detection, thresholding, and verification are disclosed as follows. Thresholding does most, but not all, of the work of extracting the foreground region in the radar cube. Some spurious peaks exceed the threshold. These can be removed by assuming that all regions of interest are properly described as 3D blob shapes. This consists of finding all local maxima in the following way.

[0080]

Number

[0081] Figure 11 is a schematic diagram of an exemplary radar system that scores these peaks as the sum of magnitudes within a 5×5×5 neighborhood centered on each bin and then masks peak regions where the score falls below a minimum value. This can reduce the number of local maxima from 100 - 1,000 to 30 - 100.

[0082] The above advantages are the robustness that false peaks are removed by the verification step. Further, usability is improved by the peak regions (not only peaks) that can be used by the machine learning engine in the latter half of the pipeline.

[0083] One or more of the foregoing embodiments require performing a median operation on the velocity degrees of freedom, but an operator can be executed on any parameter to simplify (compress) the data for further analysis. For example, a median or other average value (moving average, or others) can be used with respect to range, angle, and / or tangential velocity to attenuate the data matrix.

[0084] FIG. 11 is a schematic diagram of an exemplary radar system according to some embodiments. The radar system includes a transmitter, a duplexer, a low-noise amplifier, a mixer, a local oscillator, a matched filter, an IF filter, a second detector, a video amplifier, and a display.

[0085] Thus, although some aspects and embodiments of the technology of the present application have been described, it will be understood by those skilled in the art that various changes, modifications, and improvements will readily occur to them. Such changes, modifications, and improvements are intended to be within the spirit and scope of the technology described in the present application. For example, those skilled in the art will readily envision various other means and / or structures for performing the functions described herein and / or obtaining one or more of the results and / or advantages. Each such variation and / or modification is considered to be within the scope of the embodiments described herein.

[0086] The above embodiments can be implemented in any of numerous ways. One or more aspects and embodiments of the present application related to the execution of a process or method can utilize program instructions executable by a device (e.g., a computer, a processor, or other device) to execute or control the execution of the process or method.

[0087] In this regard, various inventive concepts can be embodied as a computer-readable storage medium (or multiple computer-readable storage media) encoded with one or more programs (e.g., computer memory, one or more compact discs, optical discs, magnetic tapes, flash memories, circuit configurations in a field programmable gate array or other semiconductor device, or other tangible computer storage media), and when executed on one or more computers or other processors, execute a method of implementing one or more of the various embodiments described above.

[0088] One or more computer-readable media may be portable so that one or more programs stored thereon can be loaded onto one or more different computers or other processors to implement various ones of the above-described aspects. In some embodiments, the computer-readable media may be non-transitory media.

Claims

1. collecting radar signals reflected from objects within the field of view; forming a range-angle-Doppler bin representing a three-dimensional object within the field of view; removing background noise within the range-angle-Doppler bin using a local median operation that is a median operation across a selected dimension of the range-angle-Doppler bin; masking a low-energy peak region of the background by removing radial velocity values in the selected dimension to form a sparser range-angle two-dimensional grid than the grid of the three-dimensional object; A method comprising the steps of:

2. The method according to claim 1, further comprising an operation of utilizing peak region verification of the range-angle-Doppler bin to score peaks in a region centered on each bin to remove false peaks.

3. The method according to claim 1, further comprising the step of upsampling the values of the sparse range-angle two-dimensional grid using bilinear interpolation.

4. The method according to claim 1, wherein the selected dimension is the Doppler dimension.

5. The method according to claim 1, further comprising threshold masking for discarding data below a threshold.

6. The method according to claim 1, further comprising background estimation for estimating a background model.

7. The method according to claim 1, wherein the radar signal reflected from an object within the field of view is processed to extract the presence or absence of a reflection point detection.

8. The method according to claim 7, further comprising the step of applying short-term tracking rules to the reflection point detection to form tracklets, each tracklet corresponding to a reflection point detection that is tracked over several frames.

9. The method according to claim 8, further comprising the step of applying long-term tracking rules by clusters of tracklets to track an object having several reflection point detections.

10. The method according to claim 9, further comprising the step of applying a birth rule to the tracklet to establish the birth of the tracklet, the birth rule including that a reflection point exists in several frames.

11. The method according to claim 9, further comprising the step of applying an extinction rule to the tracklet to establish extinction of the tracklet, wherein the extinction rule includes failing to detect a reflection point in several consecutive frames.

12. The method according to claim 9, further comprising the step of applying a birth rule to a group which is a cluster of tracklets to establish the birth of the group, wherein the birth rule includes the presence of reflection points in several frames.

13. The method according to claim 9, further comprising the step of applying an extinction rule to a group which is a cluster of tracklets to establish the extinction of the group, wherein the extinction rule includes failing to detect a reflection in several consecutive frames.

14. The method according to claim 9, further comprising the step of applying a Kalman filter to the tracklet.

15. The method according to claim 9, further comprising the step of applying a Kalman filter to a group which is a cluster of tracklets.

Images