DE102021105659A1 - Method, device and radar system for tracking objects - Google Patents

Abstract

Detection points can initially be observed over several radar images. The observed detection points that form the tracklets can be used to generate a segmentation of the current radar image by associating the tracklets with at least one object track that represents at least one object based on at least one feature parameter. The segmentation results from tracking detection points across multiple radar images (i.e. using tracking information) that is used for assigning detection points to objects (segmentation by tracking). A two-stage tracking approach can be implemented, where for a current radar image (new) detection points are assigned to tracklets, which can be understood as the first stage, and then the tracklets are assigned to object tracks, which can be understood as the second stage.

Description

FIELD OF REVELATION

The present disclosure relates generally to methods, systems, vehicles, and apparatus for tracking at least one object in measurement data from a radar system, and more particularly to a radar system configured to track at least one object in its measurement data.

BACKGROUND

In the field of autonomous or quasi-autonomous operation of vehicles such as airplanes, watercraft or land vehicles, especially automobiles, which can be manned or unmanned, sensing the vehicle's surroundings and tracking objects in the vehicle's surroundings can be crucial for sophisticated functionalities be considered. These functionalities can range from driver assistance systems with different levels of autonomy to fully autonomous driving of the vehicle.

In certain environments, several different types of sensors are used to sense a vehicle's surroundings, such as: B. monoscopic or stereoscopic cameras, light detection and ranging sensors (LiDAR) and radio detection and ranging sensors (radar). The different sensor types have different properties that can be used for different tasks.

Embodiments of the present disclosure relate to aspects of the processing of measurement data from radar systems, as a result of which the computationally intensive fusion of data from different sensor types can be avoided.

Radar systems can supply measurement data, in particular distance, Doppler and/or angle measurements (azimuth and/or elevation), with high precision in a radial direction. This allows both (radial) distances and (radial) speeds in a field of view of the radar system between different reflection points and the (respective) antenna of the radar system to be measured precisely.

Basically, radar systems transmit (emit) radar signals into the field of view of the radar system, where the radar signals are reflected from objects that are in the field of view of the radar system and are received by the radar system. The transmission signals are, for example, frequency-modulated continuous wave signals (FMCW). Radial distances can be measured using the time it takes the radar signal to travel, while radial velocities can be measured using the frequency shift caused by the Doppler effect.

By repeatedly sending and receiving the radar signals, radar systems can monitor the field of view of the radar system over time by supplying measurement data that includes a number of, in particular consecutive, radar images (radar frames).

A single radar image can be, for example, a range-azimuth image (range-azimuth frame) or a range-doppler-azimuth image (range-doppler azimuth frame). A range-doppler-azimuth-elevation image (range-doppler-azimuth-elevation frame) would also be conceivable if data is available in the elevation direction.

A large number of reflection points, which can form reflection point clouds, can be detected in each of the plurality of radar images. However, the reflection points or point clouds in the radar images do not per se contain a semantic meaning. Accordingly, a semantic segmentation of the radar images is necessary in order to be able to evaluate (“understand”) the scene surrounding the vehicle.

The segmentation of a radar image means that a meaning is assigned to the individual reflection points in the individual radar images. For example, the reflection points can be assigned to the background of the scene, the foreground of the scene, stationary objects such as buildings, walls, parked vehicles or parts of a street, and/or moving objects such as other vehicles, cyclists and/or pedestrians in the scene.

In the prior art, the semantic image segmentation is usually carried out in images that are recorded by a camera sensor (camera images), such as in "No Blind Spots": Full-Surround Multi-Object Tracking for Autonomous Vehicles using Cameras & LiDARs ' described by Akshay Rangesh and Mohan Trivedi.

In the case of camera images, it is advantageous for semantic segmentation that most of the light is diffusely reflected into the sensor, so that connected regions in the camera image can be observed. However, semantic segmentation in radar images is particularly difficult.

In general, radar systems observe specular reflections of the transmitted signals that emitted by the radar system, since the objects to be detected tend to have smoother reflection characteristics than the (modulated) wavelengths of the transmitted signals.

Consequently, the radar images obtained do not contain contiguous regions representing individual objects, but rather individual prominent reflection points (such as the edge of a bumper) distributed over regions of the radar image.

For tracking objects in the scene (in the field of view of the radar system), i.e. tracking an object over several images (frames), it becomes even more difficult, since the individual reflection points that can belong to an object depend on the radar image (radar frame ) to radar image (radar frame) may vary. This means that, for example, reflection points of a first radar image can disappear in a second, for example (immediately) following radar image, while other reflection points can appear in the second radar image.

In view of the above, the object of the embodiments of the present disclosure is to provide a method for tracking at least one object in radar images of a radar system and a corresponding radar system, wherein objects of interest are tracked in an efficient and reliable manner.

In particular, the objects of the embodiments of the present disclosure can be achieved by the method for tracking at least one object in measurement data of a radar system according to claim 1 .

SUMMARY OF THE INVENTION

• - Detektieren von Detektionspunkten in den Radarbildern;
• - Zuordnen der Detektionspunkte eines aktuellen Radarbildes zu einer Vielzahl von Tracklets, wobei jedes Tracklet eine Spur von mindestens einem über mehrere Radarbilder beobachteten Detektionspunkt ist; und
• - Zuordnen (insbesondere Gruppieren) der Tracklets basierend auf mindestens einem Merkmalsparameter zu mindestens einer Objektspur (die das mindestens eine zu verfolgende Objekt darstellt).

In particular, an object of the present disclosure is achieved by a method according to an embodiment of the present invention for tracking at least one object in measurement data of a radar system, which contain multiple, in particular consecutive, radar images (radar frames) captured by a radar system, comprising:

• - detecting detection points in the radar images;
• - Assigning the detection points of a current radar image to a plurality of tracklets, each tracklet being a trace of at least one detection point observed over a plurality of radar images; and
• - Assigning (in particular grouping) the tracklets based on at least one feature parameter to at least one object track (representing the at least one object to be tracked).

One aspect of the present disclosure is based on the idea that in the radar images initially detection points can be observed over several radar images, the observed detection points that form the tracklets can be used to generate a segmentation of the present radar image by the tracklets at least be assigned to an object track that represents at least one object based on at least one feature parameter.

The segmentation according to the method according to the invention results from following detection points over several radar images (i.e. using tracking information), which is used for assigning detection points to objects (segmentation by tracking).

Accordingly, the present method can be viewed as a two-stage approach, in which for a current radar image (new) detection points are assigned to tracklets, which can be viewed as a first stage (of the tracking method), and then the tracklets are assigned to object tracks, what can be seen as a second stage (of the prosecution process).

A method according to the present disclosure differs from the conventional approach, where each radar image is usually treated independently at one stage through segmentation without using tracking information for segmentation.

The detection of detection points can be understood as finding intensity peaks (intensity peaks) in the radar image, where the radar images can be understood as a three-dimensional intensity map, for example in an angle-range bin (angle-range bin) or an angle-doppler Range bin (angle-doppler range bin). The term "detection point" is not to be understood as a zero-dimensional point in the geometric sense, but preferably as an area of the above-mentioned intensity peak (e.g. related to an edge of a bumper or another structure, in particular an edge and / or corner of a vehicle, in particular an area with pronounced reflection). The detection point can include at least one, in particular multiple, resolution cells (for example pixels and/or voxels) in the radar image.

The measurement data recorded by a radar system can be a two-dimensional or multi-dimensional, complex-valued array that includes dimensions that include the (azimuth and/or elevation) angle, the (radial) velocity (also called Doppler) and the radial distance (also called distance called) may contain. For example, it is possible to use the magnitude in each angular-doppler-range bin to describe how much energy the radar sensor receives from each point in the scene (in the field of view) for a radial velocity.

Consecutive radar images can be understood as a plurality of radar images, each radar image (except for a first radar image) following another radar image in time. From a given (measured or determined) plurality of radar images, all radar images can be used or only a subset of the radar images.

In particular, obtaining and/or maintaining the tracklets and the object tracks is based on at least one dynamic system model, as a result of which a robust and precise tracking of the tracklets and the object tracks can be achieved.

The at least one dynamic system model can be used to estimate at which position in a radar image, in particular a subsequent radar image, a tracklet and/or an object track can be expected.

• - Prädizieren (Vorhersagen) eines oder mehrerer Parameter jedes Tracklets für das aktuelle Radarbild durch Propagieren des dynamischen Systemmodells, wobei die Parameter jedes Tracklets mindestens eine Position, insbesondere eine Position und eine Geschwindigkeit, vorzugsweise eine Position und eine Geschwindigkeit und eine Beschleunigung, und eine Kovarianz des Tracklets in einem Radarbild umfassen; und
• - Korrigieren der Parameter jedes Tracklets basierend auf den Detektionspunkten, die dem entsprechenden Tracklet zugeordnet sind.

Preferably, the method further includes the following steps:

• - Predicate (predict) one or more parameters of each tracklet for the current radar image by propagating the dynamic system model, the parameters of each tracklet at least a position, in particular a position and a speed, preferably a position and a speed and an acceleration, and a covariance of the tracklet in a radar image; and
• - Correcting the parameters of each tracklet based on the detection points associated with the corresponding tracklet.

The prediction is preferably carried out before the detection points are assigned to the tracklets and the correction is carried out after the detection points are assigned to the tracklets.

By predicting the parameters of the tracklets before assigning the detection points, it is possible to improve the assignment of the detection points to the tracklets. By correcting the parameters of the tracklets after assigning the detection points, it is possible to correct the inaccuracies of the model, which can improve the performance of the tracking solution.

In particular, in the step of assigning the detection points to the tracklets, a detection point is assigned to a tracklet if a position of the detection point lies within a gate of a tracklet; wherein new tracklets are always initialized from the detection points when the criterion for the assignment of a detection is not met for any of the existing tracklets, in particular when a position of a detection point is outside the gates of all existing tracklets.

The method described above for assigning detection points to tracklets is particularly simple and computationally easy. The gate of a tracklet can be an association area, such as a polytope (e.g. polygon, in particular tetragon, or polyhedron, in particular hexahedron), in particular a, preferably rectangular, parallelogram (e.g. parallelogram, in particular rectangle or square, or parallelepiped, in particular a cuboid or cube), a (hyper)ellipsoid or an ellipse or a (hyper)sphere or a circle (in particular depending on the dimensions of the corresponding image(s).

New tracklets are preferably initialized from unassigned detection points, which are detection points that are not assigned to any tracklet, i. H. the unassigned detection points are not in any gate of the tracklets (which is preferably the criterion for the assignment). This allows for a simple, but quick and effective method of initializing new tracklets.

Preferably, a gate for each tracklet is either fixed in size or adaptive in size. When it is adaptive, the size of the gate can correlate with the covariance of the tracklet, specifically such that the size of the gate increases as the covariance increases, and vice versa.

Accordingly, it is possible to vary the size of the gate of the tracklet depending on the covariance of the tracklet's predicted position in the current radar image. This further improves the assignment of the detection points to the tracklets, since the number of incorrect assignments between detection points and tracklets can be reduced, in particular when the position of a tracklet is more likely probability is predicted (which correlates with a higher covariance).

When assigning the detection points to the tracklets, it is preferred that a detection point is assigned to the tracklet whose position is closest to the detection point.

This allows for a simple but effective method of assigning the detection points to the tracklets.

Alternatively or additionally, a detection point can be probabilistically assigned to multiple tracklets when assigning the detection points to the tracklets. Preferably, probabilistic values indicating the probability that a detection point is associated with a tracklet are increased as the distance between the position of the detection point and the predicted position of the tracklet decreases, and vice versa.

For example, the Mahalanobis distance can be used as a measure for assigning the detection points to the tracklets.

The assignment of detection points to several tracklets is particularly advantageous in difficult assignment situations, e.g. B. when detection points are within two or more gates of different tracklets.

In particular, the feature parameter for assigning (grouping) the tracklets, based on which the tracklets are clustered into the object tracks, includes an overlap of the gates of the individual tracklets in at least the current radar image and/or a summed overlap of the gates of the individual tracklets in several previous radar images .

The overlap of the gates of the individual tracklets in at least the current radar image and/or a (cumulative) overlap of the gates of the individual tracklets in several previous radar images as feature parameters is a useful criterion for assigning (grouping) tracklets to object tracks.

In particular, using the summed overlap of the gates of the individual tracklets in several previous radar images as a feature parameter can increase the robustness of the assignment.

The association (grouping) of the tracklets is preferably performed by a clustering method, in particular by a Density-Based-Spatial-Clusteringot-Applications-with-Noise (DBSCAN) method, which is a simple but effective clustering method.

It is preferred that the method further includes the correction of parameters of the object track, in particular a position, a speed and/or an acceleration of the object track, by the parameters of the object track based on a predicted speed and/or a predicted acceleration of the tracklets of the corresponding object track are updated.

Accordingly, it is possible to improve the tracking solution of object tracks by using the predicted parameters of the tracklets, which are assigned to the object tracks, as measurements for the object tracks.

In particular, each tracklet includes metadata that includes at least a status of the tracklet, a track count value, and a lost track count value. The metadata of the tracklets preferably also contain an identification number which indicates which object track the tracklet is assigned to.

The status may differ by a state of the tracklet, e.g. one of a tracked and untracked (e.g. lost or untracked) state, or one of a preliminary state, a tracked state and a (at least temporarily) lost state, or one of a preliminary state, a tracked state, a (at least temporarily) lost state, and a completely lost state.

• - Aktualisieren der Metadaten der Tracklets; und
• - Initialisieren von Detektionspunkten als neue Tracklets, die nicht den bestehenden Tracklets zugeordnet sind,

wobei das Aktualisieren der Metadaten und das Initialisieren von Detektionspunkten als neue Tracklets nach dem Zuordnen der Detektionspunkte zu den Tracklets durchgeführt werden.In particular, the method further includes the following steps:

• - Updating the metadata of the tracklets; and
• - initialize detection points as new tracklets not mapped to existing tracklets,

wherein updating the metadata and initializing detection points as new tracklets are performed after associating the detection points with the tracklets.

Preferably, different filters are used to model the dynamics of the tracklets and to model the dynamics of the object tracks. The filter for modeling the dynamics of the tracklets is preferably less computationally intensive than the filter for modeling the dynamics of the object tracks.

An alpha-beta filter is preferably used for modeling the dynamics of the tracklets and a Kalman filter for modeling the dynamics of the object tracks, as a result of which a reasonable compromise between the computational effort and the performance of the tracking results can be achieved.

Alternatively, an alpha-beta filter is used to model the dynamics of the tracklets and the object tracks, which further reduces the computational effort.

Alternatively or additionally, an object model can be derived from a library of object models for each object group (object track) and a switching Kalman filter can be used to model the object tracks, with a switching state of the switching Kalman filter representing an object class.

The library of object models can e.g. B. object models for vehicles such as cars or trucks, cyclists or pedestrians.

This can make it possible to derive the distribution of the object over an area in the radar image, which can improve the accuracy of assigning the detection points to the tracklets as well as assigning the tracklets to the object tracks. Additionally or alternatively, the dynamic system models can thereby be switched in accordance with the object models, so that the modeling of the dynamics of the objects can be adapted to the object to be tracked.

The class-specific object models are preferably learned (derived) from data sets. It is also conceivable that an object classifier, such as e.g. B. a support vector machine (SVM), neural networks or a (deep) convolutional neural network (CNN), is used to classify an object trace into one of the object classes of the library.

In a further (independent or dependent) aspect of the present disclosure, an object of the disclosure is achieved in particular by a radar method, the (or, if an independent aspect: a) plurality of radar images contained in the measurement data, a first plurality of radar images acquired by a first radar unit, the measurement data further including a second plurality of radar images acquired by a second radar unit which is not associated (and/or not coherent) with the first radar unit, the first and the second plurality of radar images are (chronologically) synchronized and at least partially (spatially) overlap, the radar images containing distance, Doppler and angle measurements, with a multi-dimensional velocity vector for at least one, in particular for several, preferably for each detection point from the Doppler measurements , which appears in synchronized radar images d which is detectable from the first and the second plurality of radar images, the determination of the multi-dimensional velocity vector being based on the corresponding Doppler measurements of the first and the second radar unit. In this further aspect, the two-step tracking approach is preferred, but not mandatory (i.e. all following embodiments of the further aspect may refer to the two-step tracking or may refer to any other, e.g. single-step or conventional, tracking method ).

The multi-dimensional velocity vector can be used to speed up the initialization of new (object) tracks, which can be particularly beneficial when objects are rapidly appearing in the radar system's field of view.

In particular, the multi-dimensional velocity vectors are used in correcting parameters of a lane (a track), in particular in correcting the parameters of the tracklet (t1 to tm).

In particular, the multi-dimensional velocity vectors are used when updating the metadata of tracks (especially the tracklets).

Furthermore, they can be used as new tracks (tracklets) when initializing acquisitions (detection points), whereby updating the metadata of the tracks (tracklets) and initializing detection points as new tracks (tracklets) can be accelerated. In particular, a transition of the status of the respective track (tracklet) from a preliminary state to a tracked state can be accelerated.

Preferably, the status of a track (tracklet) is immediately changed from a preliminary state to a tracked state when the track (tracklet) is within a range of the position of a detection point for which a multidimensional vector is determined and when a comparative measure , in particular a sum of the inner product, the multi-dimensional velocity vector and the multi-dimensional velocity vectors of the neighboring multi-dimensional velocity vectors of the detection point is equal to or larger than a predetermined threshold value.

Accordingly, it can be determined whether the multidimensional speed vectors of adjacent detection points are at least approximately congruent. If the multi-dimensional velocity vectors of adjacent detection points are at least approximately congruent, it can be assumed that the adjacent detection points relate to a track (tracklet).

• - eine erste Radareinheit, die dazu ausgestaltet ist, eine Vielzahl von Radarbildern durch Senden und Empfangen von Radarsignalen zu detektieren, die an potentiellen zu verfolgenden Objekten in einem Sichtfeld der ersten Radareinheit reflektiert werden; und
• - eine Verfolgungsberechnungseinheit, die dazu ausgestaltet ist, die erfassten Radarbilder durch Ausführen der Schritte eines Verfahrens der oben beschriebenen Art zu verarbeiten.

An object of the present disclosure is also achieved by a radar system that is designed to track at least one or more objects in measurement data of the radar system that contain a large number of, in particular consecutive, radar images, using a method of the type described above :

• - a first radar unit configured to detect a plurality of radar images by transmitting and receiving radar signals reflected on potential objects to be tracked in a field of view of the first radar unit; and
• - a tracking calculation unit arranged to process the acquired radar images by performing the steps of a method of the type described above.

• - eine zweite Radareinheit, die dazu ausgestaltet ist, eine Vielzahl von Radarbildern durch Senden und Empfangen von Radarsignalen zu detektieren, die an potentiellen zu verfolgenden Objekten in einem Sichtfeld der ersten Radareinheit reflektiert werden.

Preferably, the radar system further includes:

• - a second radar unit configured to detect a plurality of radar images by transmitting and receiving radar signals reflected on potential objects to be tracked in a field of view of the first radar unit.

The field of view of the first radar unit and the field of view of the second radar unit preferably overlap at least partially.

The radar system may include one or more (e.g. at least two or at least four) antennas for transmitting and/or receiving radar signals. Each of the first and/or the second radar unit can comprise one or more (e.g. at least two or at least four) antennas for transmitting and/or receiving radar signals.

Furthermore, an object of the present disclosure is achieved by a vehicle in which a radar system of the type described above is installed, the vehicle being an aircraft and/or a watercraft and/or a land vehicle, preferably a motor vehicle, the vehicle being manned or can be unmanned.

Features and advantages associated therewith, which are described in connection with the method according to the invention for tracking at least one object in measurement data of a radar system, can be applied and transferred to the radar system or vehicle of the present disclosure. The method steps explained above can be implemented individually or in combination in the radar system or the vehicle as corresponding configurations (e.g. a control and/or computing unit).

Further advantageous embodiments can be found in the dependent claims.

character list

• 1 zeigt ein Ablaufdiagramm eines Beispiels des erfindungsgemäßen Verfahrens zur Verfolgung mindestens eines Objekts in einer Vielzahl von Radarbildern;
• 2 zeigt eine schematische Darstellung eines Beispiels für ein erfindungsgemäßes Radarsystem;
• 3A zeigt ein Beispiel für ein aktuelles Radarbild, in welchem der Schritt S1 des Verfahrens zur Verfolgung mindestens eines Objekts durchgeführt wird;
• 3B zeigt ein Beispiel für ein aktuelles Radarbild, in welchem Schritt S2 und Schritt S3 des Verfahrens zur Verfolgung mindestens eines Objekts durchgeführt wird;
• 3C zeigt ein Beispiel für ein aktuelles Radarbild, in welchem Schritt S4 des Verfahrens zur Verfolgung mindestens eines Objekts durchgeführt wird;
• 4 zeigt eine schematische Darstellung eines weiteren Radarsystems gemäß der Erfindung.

The present disclosure is explained in more detail below using non-limiting embodiments or examples with reference to the accompanying drawings. The figures show the following:

• 1 shows a flow chart of an example of the method according to the invention for tracking at least one object in a plurality of radar images;
• 2 shows a schematic representation of an example of a radar system according to the invention;
• 3A shows an example of a current radar image in which step S1 of the method for tracking at least one object is carried out;
• 3B shows an example of a current radar image, in which step S2 and step S3 of the method for tracking at least one object is carried out;
• 3C shows an example of a current radar image, in which step S4 of the method for tracking at least one object is carried out;
• 4 shows a schematic representation of another radar system according to the invention.

DETAILED DESCRIPTION

In 1 1 is a flowchart of an example of the method for tracking at least one object in a plurality of radar images, in accordance with the present disclosure. In this example, method steps S0 to S8 are carried out, with steps S1 to S8 forming a tracking loop. Y is the abbreviation for "yes", N is the abbreviation for "no", and E is the abbreviation for "end".

In method step S0, the tracking loop is initialized by creating empty lists for both the tracklets and the object tracks.

After the tracking loop has been initialized, the method steps of the tracking loop S1 to S8 are carried out until the last radar image to be processed is reached. When the last radar image to be processed is reached, the tracking loop is ended in step S8.

In the tracking loop, method step S1 is performed, which is the detection of detection points d in a first radar image f1. In this example, the radar images are (at least essentially) heatmaps in which the (radial) speed is plotted against the distance and angle measurements.

In practice, the radar images are pre-processed using methods for suppressing noise and artifacts and methods for extracting conspicuous regions in the radar image. These methods include, for example, estimating a background model and removing the modeled background from the radar image, or creating a mask that highlights desired areas and suppresses others through element-by-element multiplication.

For the pre-processing of the radar image z. For example, the Constant False Alarm Rate (CFAR) thresholding method can be used, in which a background model is estimated by local averaging. The basic idea here is that the noise statistics can be uneven across the array (the radar image).

Another possible variation of the above method would be Cell Averaging (CA) CFAR, in which a moving average is calculated while excluding an area in the center of the averaging window (guard cells) to avoid introducing a desired object into the background estimation is included. Also, Order-Statistic (OS) CFAR is a variant of CA-CFAR that uses a percentile operation instead of a mean operation.

In addition, methods that use time-adaptive background modeling can also be used to preprocess the radar frames. For example, a multi-tap filter, preferably a 5-tap filter, in particular a 3-tap filter, in particular a 2-tap filter with an infinite impulse response (IIR) filter can be used for temporal smoothing of the foreground detection in order to to avoid the need to recalculate the entire background model with each radar image.

After pre-processing the radar frames using a thresholding method as described above, the foreground areas in the radar frame are extracted. However, some glitches may still exceed the threshold. The glitches can be removed by assuming that all regions of interest can be viewed as three-dimensional blob shapes.

In this example, detecting the detection points involves finding local maxima in the pre-processed radar image after extracting the foreground regions in a radar image.

The detection points d1 to dn recorded as described above are then used as measurement values for the tracking in the further processing of the tracking loop.

In the first radar image there are no tracklets and no object trails because the tracking loop is initialized with empty lists of tracklets and object trails. Consequently, in the first radar image, the detection points d1 to dn are used to initialize the new tracklets t1 to tm, which carry a preliminary state.

In the following, the tracking loop is described according to an example of the present disclosure for a present non-first radar image fp where the list of tracklets t1 to tm is not empty.

In method step S2 of the tracking loop according to this example, one or a plurality of parameters of each tracklet t1 to tm for the current radar image fp is predicted (predicted) by propagating the dynamic system model.

$$v_{t|t-1}\leftarrow v_{t-1|t-1}$$

wobei µ_{t-1| t-1} die korrigierte Schätzung zum Zeitpunkt t - 1 angibt und µ_t|t-1 die vorhergesagte Schätzung zum Zeitpunkt t. Das Propagieren des Modells mit konstanter Beschleunigung kann wie folgt ausgedrückt werden: $${\mu }_{t|t-1}\leftarrow {\mu }_{t-1|t-1}+v_{t-1|t-1}+0.5\cdot {\alpha }_{t-1|t-1}$$

$$v_{t|t-1}\leftarrow c_{t-1|t-1}+{\alpha }_{t-|t-1}$$

$${\alpha }_{t|t-1}\leftarrow {\alpha }_{t-1|t-1}$$

wobei v_t-1lt-1 die korrigierte Geschwindigkeitsschätzung zum Zeitpunkt t - 1, v_t|t-1 die vorhergesagte Geschwindigkeitsschätzung zum Zeitpunkt t und a_t-1|t-1 die korrigierte Beschleunigungsschätzung zum Zeitpunkt t - 1 angibt.The parameters of each tracklet t1 to tm may include a position and a velocity, an acceleration and a covariance of the tracklet t1 to tm in a radar image. The dynamic system model can be a simple constant velocity model or a constant acceleration model. Propagating the model at constant speed can be expressed as: $${\mu }_{t|t-1}\leftarrow {\mu }_{t-1|t-1}+v_{t-1|t-1}$$

$$v_{t|t-1}\leftarrow v_{t-1|t-1}$$

where µ _{t-1| t-} 1 indicates the corrected estimate at time t - 1 and µ _t|t- 1 indicates the predicted estimate at time t. Propagating the constant acceleration model can be expressed as: $${\mu }_{t|t-1}\leftarrow {\mu }_{t-1|t-1}+v_{t-1|t-1}+0.5\cdot a_{t-1|t-1}$$

$$v_{t|t-1}\leftarrow c_{t-1|t-1}+a_{t-|t-1}$$

$$a_{t|t-1}\leftarrow a_{t-1|t-1}$$

where v _t-1lt-1 is the corrected velocity estimate at time t - 1, v _t|t-1 is the predicted velocity estimate at time t, and a _t-1|t-1 is the corrected acceleration estimate at time t - 1.

After the parameters of each tracklet t1 to tm have been predicted in method step S2, the detection points d1 to dn are assigned to the tracklets t1 to tm in method step S3.

The detection points d1 to dn can be assigned to the tracklets t1 to tm using a gating method. In the gating method, a gate that is an association area such as a a rectangle, square, ellipse or circle, around the (predicted) center position of each tracklet t1 to tm. In this example, the association region is a fixed-size ellipse.

After the assignment of detection points d1 to dn to the tracklets t1 to tm in method step S3, the tracklets t1 to tm are assigned to the object tracks g1 to gk in method step S4 on the basis of at least one feature parameter. In this example, tracklets t1 through tm are assigned to object tracks g1 through gk based on the overlap of the gates of tracklets t1 through tm, using a DBSCAN method to group tracklets t1 through tm into object tracks.

In method step S5, the metadata of the tracklets t1 to tm are updated. In this example, the status of tracklets t1 through tm is retained, where the status can include a tentative state, a tracked state, a lost state, and a completely lost state.

Furthermore, the metadata of the tracklets t1 to tm can include a track count value and a lost track count value as well as a unique identification number that identifies to which object track g1 to gk the tracklet t1 to tm belongs.

The status update rules allow a tracklet t1 to tm to be lost for multiple radar images when there are no detection points that can be associated with the tracklet. Finally, if the tracklet t1 to tm is lost for a predetermined period of time, the tracklet is considered to be in the completely lost state.

The track count value is incremented in each iteration of the tracking loop that tracklet t1 through tm can be tracked, i.e. H. detection points can be assigned to the tracklet t1 to tm. The Lost Track Count value is increased in each iteration of the tracking loop in which the tracklet cannot be tracked, i.e. H. no detection point can be assigned to the tracklet t1 to tm.

Newly initialized tracklets t1 to tm initially have a preliminary state until the track count value reaches a predetermined value.

A tracklet in a preliminary state with no associated detection point is immediately updated to the completely lost state and removed from the list of tracklets t1 to tm.

If a tracklet has a tracked state and no detection points d1 through dn can be assigned to the tracklet, the tracklet's status is updated to the lost state and the incrementing of the Lost Track Count value begins. As soon as a detection point can again be assigned to the tracklet t1 to tm, the status of the tracklet is set back to the tracked state and the track count value and the lost track count value are set to zero.

When a tracklet's Lost Track Count value reaches a predetermined value, it is updated to a completely lost state and removed from the list of tracklets.

In method step S6, each detection point that is not assigned to tracklets t1 to tm is used to initialize new tracklets that have a preliminary status as described above. A new tracklet is initialized for each (unassigned) detection point. Most such tracklets are bogus and are lost entirely after a few iterations of the tracking loop.

und der Alpha-Beta-Filter wird wie folgt angewendet: $$r_t={\overline{x}}_t-{\mu }_{t|t-1}$$

$${\mu }_{t|t}={\mu }_{t|t-1}+\alpha \cdot r_t$$

$$v_{t|t}=v_{t|t-1}+\beta \cdot r_t$$

$$a_{t|t}=a_{t|t-1}+2\gamma \cdot r_t$$

wobei α, β und 2γ ∈ [0,1] Adaptionsraten sind.In step S7, the parameters of all active tracklets t1 to tm (that is, tracklets with a tracked state) that were predicted (predicated) in step S2 are corrected. This correction is based on the associated detections, according to rules that are heuristic or that result as solutions to Bayesian filter equations corresponding to the assumed dynamic system models. In particular, synthetic observations are formed by a weighted averaging of the detection points: $${\overline{x}}_t=\frac{{\displaystyle {\sum }_{i=1}^N{w}_i{x}_t^i}}{{\displaystyle {\sum }_{i=1}^N{w}_i}}$$

and the alpha-beta filter is applied as follows: $${\mathrm{right}}_t={\overline{x}}_t-{\mu }_{t|t-1}$$

$${\mu }_{t|t}={\mu }_{t|t-1}+a\cdot {\mathrm{right}}_t$$

$$v_{t|t}=v_{t|t-1}+\beta \cdot {\mathrm{right}}_t$$

$$a_{t|t}=a_{t|t-1}+2g\cdot {\mathrm{right}}_t$$

where α, β and 2γ ∈ [0,1] are adaptation rates.

In method step S8 it is queried whether the tracking loop should be ended, e.g. B. if the current radar picture is the last radar picture to be tracked. If tracking is to be continued, the method steps S1 to S7 are carried out for the next radar image. If the current radar picture is the last radar picture to be tracked, the track loop is cancelled.

In connection with the order of the individual steps of the tracking loop, it should be noted that the order of the method steps in the present example is not limiting. Accordingly, the order of the individual process steps can be reversed if this makes technical sense.

In 2 1 is a schematic representation of an example of a radar system 100 in accordance with the present disclosure. The radar system 100 is designed to detect measurement data that includes a large number of (consecutive) radar images of a field of view FoV of the radar system 100 .

In the field of view FoV of the radar system 100, a first moving object O1 can be observed over several radar images, starting with a current radar image fp. In 2 are three different positions O1(fp), 01(fp+n, with n=1 in this example) and 01(fp+m, with m=3 in this example) of the moving object O1 along a trajectory T of the moving object O1 shown.

Furthermore, in 2 a stationary object O2 in the field of view FoV of the radar system 100 is shown.

A current radar image fp is in the 3A until 3C presented as a radar doppler visualization, with velocity presented as a heatmap over the range and angle dimension. In the 3A until 3C the background is extracted, for example by a method as described above.

Furthermore, the detection points d1 to d16 are detected after step S1 and in the until drawn as circles. The number of possible detection points according to the present disclosure is not limited to the number of detection points d1 to d18 in this example. The extracted background is in a broadly shaded area of the 3A until 3C shown.

In the in the 3A until 3C In the example shown, a finely shaded area, which is hatched from bottom right to top left, represents velocities that are around the zero point. In addition, a finely hatched area hatched from bottom left to top right in the in the 3A until 3C illustrated example represent velocities that are not equal to zero.

In 3B the model of the dynamic system is propagated according to method step S2, so that the predicted (mean) positions of the tracklets t1 to t6 are estimated. The predicted (mean) positions of tracklets t1 to t6 are in 3B shown as squares.

In 3B the gates a1 to a6 of the tracklets t1 to t6 are drawn with a dashed line. In the example of 3A until 3C the gates a1 to a6 of the tracklets t1 to t6 are shown as fixed-size ellipses. As explained above, the gates a1 to a6 of the tracklets t1 to t6 can also be adaptive in size.

The detection points d1 to d18 are assigned to the tracklets t1 to t6 according to method step S3 of the tracking loop, as explained above. The detection points d7 to d9 are unassigned detection points ud, since the detection points d7 to d9 do not lie within any gate a1 to a6 of the tracklets t1 to t6.

in the in 3B In the example shown, the detection point d13 lies within two gates a3 and a4. Accordingly, the detection point d13 is assigned to the tracklet t3 because the distance between the detection point d13 and the predicted (mean) position of the tracklet t3 is smaller than the distance between the detection point d13 and the predicted (mean) position of the tracklet t4.

In 3C the tracklets t1 to t6 are assigned to the object tracks g1 and g2 according to method step S4. In this particular example, a clustering technique is used to Cluster tracklets t1 to t6 into object tracks based on the overlap of gates a1 to a6 of tracklets t1 to t6 in the current radar image. The object tracks represent the objects to be tracked by the tracking loop. The centers of the object tracks g1 and g2 are in 3C shown as diamonds.

4 shows a schematic representation of another radar system 100 with a first radar unit 110 and a second radar unit 120. In this example, the radar system 100 further comprises a tracking calculation unit 130, which is configured to process the acquired radar images by performing the steps of the method as above explained performs.

It is noted that the tracking calculation unit 130 can also be part of at least one of the radar units 110 and 120 . In the example of 4 the radar units 110, 120 are designed to transmit the measurement data acquired by each radar unit 110, 120 to the tracking calculation unit 130. The radar units 110, 120 are non-associated radars, ie the radar units 110, 120 do not share any antennas in a larger antenna field, for example.

Radar units 110, 120 each have a field of view FoV-110, FoV-120, with fields of view FoV-110, FoV-120 of radar units 110, 120 overlapping at least partially (spatially) in a field of view FoV of radar system 100.

und wird von beiden Radareinheiten 110, 120 beobachtet. Starke, zuverlässige Reflexionspunkte des Objekts O1 in der Szene (Sichtfeld FoV) werden von beiden Radargeräten 110, 120 erfasst, um zwei radiale Geschwindigkeitskomponenten zu erhalten.A moving object O1 with an actual speed is located in the field of view FoV ${\overrightarrow{v}}_{01}$

and is observed by both radar units 110,120. Strong, reliable reflection points of the object O1 in the scene (field of view FoV) are detected by both radar devices 110, 120 in order to obtain two radial velocity components.

The two radial velocity components of each detection point d1 to d4 can be solved with a least squares solution to estimate a two-dimensional velocity vector. Accordingly, a multi-dimensional velocity vector can be estimated if the measurement data also include measurements in the elevation direction, for example.

The calculation of the two-dimensional or multi-dimensional velocity vector involves an interpolation operation between the range-angle coordinate grids of the radars 110, 120. For each radar image, a list of two-dimensional or multi-dimensional velocity vectors corresponding to the magnitudes of the radar images (e.g. the minimum of the sizes of the radar units 110, 120 ) rated accordingly.

bis ${\overrightarrow{v}}_{d4}$

. dargestellt.In 4 are four detection points d1 to d4 with the corresponding calculated two or more dimensional velocity vectors ${\overrightarrow{v}}_{\mathrm{i.e}1}$

until ${\overrightarrow{v}}_{\mathrm{i.e}4}$

. shown.

• Im Aktualisierungsschritt S5 ist es möglich, die Genauigkeit der Verfolgungsschleife zu verbessern. Insbesondere kann ein Übergang zwischen einem vorläufigen Zustand und einem verfolgten Zustand eines Tracklets beschleunigt werden, wenn sich das Tracklet innerhalb eines Bereichs um die Position eines Detektionspunkts befindet, für den ein zweidimensionaler Vektor bestimmt wird, und wenn ein Vergleichsmaß des mehrdimensionalen Geschwindigkeitsvektors und der mehrdimensionalen Geschwindigkeitsvektoren der benachbarten mehrdimensionalen Geschwindigkeitsvektoren des Detektionspunkts gleich oder größer als ein vorgegebener Schwellenwert ist.

The two or more dimensional velocity vectors can be included in the update step S5, in the initialization step S6 and in the correction step S7 of the tracking loop as follows:

• In the updating step S5, it is possible to improve the accuracy of the tracking loop. In particular, a transition between a preliminary state and a tracked state of a tracklet can be accelerated when the tracklet is within a range of the position of a detection point for which a two-dimensional vector is determined and when a comparative measure of the multi-dimensional velocity vector and the multi-dimensional velocity vectors of the neighboring multi-dimensional velocity vectors of the detection point is equal to or larger than a predetermined threshold.

For example, a two-dimensional or multi-dimensional velocity vector whose direction of movement corresponds to that of its neighbors generates a so-called hotspot around the corresponding detection point of the two-dimensional or multi-dimensional velocity vector.

The so-called hotspot can e.g. B. be an elliptical area, a rectangular area, a circular area or the like in Cartesian coordinates.

The two-dimensional or multi-dimensional velocity vectors can be sorted out by the position of the corresponding detection point around the two-dimensional or multi-dimensional velocity vectors to identify the neighbors of the two-dimensional or multi-dimensional velocity vectors.

The normalized inner product of the velocity vectors with the velocity vectors of the neighbor can be calculated and summed up as a standard of comparison hot _i . The summed, normalized inner product of the velocity vectors hot _i can then be compared to a predetermined threshold value hot _thresh to determine whether a hotspot is created or not: $$HO{t}_i=\left[\left({\displaystyle {\sum }_{j\in neiGH}\frac{v_i\cdot v_j}{{\Vert v_i\Vert }_2{\Vert v_j\Vert }_2}}\right)\ge HO{t}_{tH\mathrm{right}esH}\right]$$

The normalized interior product measures agreement about which direction the object is moving at that location. The sum is an accumulation metric ranging from -n (perfect mismatch) through 0 (expected value when directions are uniformly random) to n (perfect match) for a number n of velocity vectors.

Similarly, in the initialization step S6, tracklets can be initialized more quickly with a tracked state if two or more dimensional velocity vectors of observed detection points are available.

Tracklets or object tracks (groups of tracklets) can be initialized with a tracked state instead of a tentative state if the same measure of comparison applies.

So e.g. B. Both tracklets and object traces in a radar image can be initialized with a tracked state for an object as long as the observed detection points and enough two- or multi-dimensional velocity vectors in its neighborhood match.

In addition, the two-dimensional or multi-dimensional velocity vectors can be included in the correction step S7 of the tracking loop.

wobei ζ ∈ [0,1]. Dieser Schritt der Median-Aggregation und der konvexen Kombination kann auch verwendet werden, um die Geschwindigkeit eines Tracklets, das aufgrund von zwei- oder mehrdimensionalen Geschwindigkeitsvektoren (wie oben beschrieben) initialisiert wurde, warm zu starten.For example, update step S5 can be performed and then another (separate) update step based on the neighboring two or more dimensional velocity vectors. Here, the tracklets to be corrected contain velocity vectors as parameters, so the tracklet's center position and velocity vector can be used to gate the position and velocity. Then a median can be computed over the two-dimensional or multi-dimensional velocity vectors to discard further outliers and update the tracklet's velocity with an appropriately weighted convex combination: $$v_i\leftarrow \zeta \cdot v_i+\left(1-\zeta \right)\cdot me\mathrm{i.e}ian\left(vec{s}_i\right)$$

where ζ ∈ [0,1]. This step of median aggregation and convex combination can also be used to warm start the velocity of a tracklet that has been initialized due to two or more dimensional velocity vectors (as described above).

5 10 shows a system 1000 including an autonomous vehicle 1100 and a radar system 100 according to embodiments. The radar system 100 comprises a first radar unit 110 with at least one first radar antenna 111 (for transmitting and/or receiving corresponding radar signals), a second radar unit 120 with at least one second radar antenna 121 (for transmitting and/or receiving corresponding radar signals) and a tracking calculation unit 130.

The system 1000 may include a passenger interface 1200, a vehicle coordinator 1300, and/or a remote expert interface 1400. In certain embodiments, the remote expert interface 1400 allows a non-passenger entity to adjust and/or change the behavior settings of the autonomous vehicle 1100 . The non-passenger entity may be different from the vehicle coordinator 1300, which may be a server.

The system 1000 operates such that the autonomous vehicle 1100 is able to change and/or adjust driving behavior in response to parameters provided by the passenger (e.g., via the passenger interface 1200) and/or other interested parties (e.g., (e.g. via vehicle coordinator 1300 or remote expert interface 1400). The driving behavior of an autonomous vehicle may be modified according to explicit input or feedback (e.g., a passenger specifying a maximum speed or relative comfort level), implicit input or feedback (e.g., a passenger's heart rate), or any other suitable data or type the transmission of driving behavior preferences can be changed.

The autonomous vehicle 1100 is preferably a fully autonomous automobile, but may additionally or alternatively be any semi-autonomous or fully autonomous vehicle; e.g. a boat, an unmanned aerial vehicle, a driverless car, etc. Additionally or alternatively, the autonomous vehicle can be a vehicle that switches between a semi-autonomous state and a fully autonomous state, and thus the autonomous vehicles can have both attributes depending on the state of the vehicle a semi-autonomous vehicle and a fully autonomous vehicle.

The autonomous vehicle 1100 preferably includes an accelerator pedal interface that controls an engine throttle, engine speed (e.g., speed of an electric motor), or other mechanism mus controls to enable movement; a braking interface that controls the autonomous vehicle's brakes (or other motion-inhibiting mechanism); and a steering interface that controls the steering of the autonomous vehicle (e.g., by changing the angle of the wheels of the autonomous vehicle). The autonomous vehicle 1100 may additionally or alternatively include interfaces for controlling any other vehicle function; e.g. B. Windscreen wipers, headlights, turn signals, air conditioning, etc.

In addition, the autonomous vehicle 1100 preferably includes an on-board computer 1450.

The tracking calculation unit 130 may be at least partially located in and/or on the vehicle 1100 and may be (at least partially) integrated into the on-board computer 1450 and/or may be (at least partially) integrated into a computing unit in addition to the on-board computer 1450 . Alternatively or additionally, the locating unit can be integrated (at least partially) into the first and/or second radar unit 110, 120. When tracking unit 130 is (at least in part) additional to onboard computer 1450, it may be in communication with the onboard computer such that data may be transferred from tracking unit 130 to onboard computer 1450 and/or vice versa.

Additionally or alternatively, the tracking calculation unit 130 may be integrated (at least in part) into one or more or all of the passenger interface 1200, the vehicle coordinator 1300, and/or a remote expert interface 1400. In particular, in such a case, the radar system may include the passenger interface 1200, the vehicle coordinator 1300, and/or a remote expert interface 1400.

In addition to the one or two or more RADAR unit(s), the autonomous vehicle 1100 preferably includes a sensor unit 1500 (including, e.g., one or more or all computer vision ("CV") systems, LIDAR, wheel speed sensors , GPS, cameras, etc.).

The onboard computer 1450 can be implemented as an ADSC and functions to control the autonomous vehicle 1100 and processes collected data from the sensor unit 1500 and/or other sensors, in particular sensors provided by the radar units 110, 120 and/or data from the tracking calculation unit 130 to determine the state of the autonomous vehicle 1100 . Based on the vehicle state and programmed instructions, the onboard computer 1450 preferably modifies or controls the driving behavior of the autonomous vehicle 1100.

Driving behavior may include any information related to how an autonomous vehicle drives (eg, applying the brakes, accelerator, steering) when given a set of instructions (eg, a route or a plan). . Driving behavior may include a description of the controlled operation and movement of an autonomous vehicle and how the autonomous vehicle applies traffic rules during one or more driving sessions. The driving behavior can additionally or alternatively contain all information about how an autonomous vehicle calculates routes (e.g. prioritizing the fastest time over the shortest route), other operating behavior of the autonomous vehicle (e.g. operating lights, windshield wipers, settings of the traction control, etc.) and/or how an autonomous vehicle reacts to environmental influences (e.g. how an autonomous vehicle behaves when it rains or when an animal jumps in front of the vehicle). Some examples of elements that may contribute to driving behavior are acceleration restrictions, deceleration restrictions, speed restrictions, steering restrictions, suspension settings, route preferences (e.g. scenic routes, faster routes, not freeways), lighting preferences, "legal ambiguity" behavior (e.g. e.g., (e.g. whether a vehicle enters the intersection on green or waits at the intersection line), action profiles (e.g. how a vehicle turns, changes lanes or performs a driving manoeuvre), and action frequency constraints (e.g. how often a vehicle changes lanes).

The onboard computer 1450 controls the operation and the functionality of the autonomous vehicles 1100 and processes measurement data from the sensor unit 1500 and/or other sensors, in particular from the radar units 110, 120, and/or data from the tracking calculation unit 130 to states of the autonomous vehicles 1100 determine. Based on vehicle state and programmed instructions, onboard computer 1450 preferably modifies or controls the behavior of autonomous vehicles 1100. Tracking calculation unit and/or onboard computer 1450 is/are preferably a general purpose computer used for I/O communication with vehicle control systems and sensor systems is suitable, but may additionally or alternatively be any suitable computing unit. The onboard computer 1450 is preferably connected to the Internet via a wireless connection (e.g., via a cellular data connection). Additionally or alternatively, the onboard computer 1450 with any Number of wireless or wired communication systems to be connected.

The sensor unit 1500 preferably comprises location and driving sensors, e.g. B. Photodetectors, Cameras, SONAR, LIDAR, GPS, Inertial Measurement Units (IMUs), Accelerometers, Microphones, Strain Gauges, Pressure Switches, Barometers, Thermometers, Altimeters, etc.

In one embodiment, any number of electrical circuits can be made 5 , particularly as part of the tracking computing unit 130, the on-board computer 1450, the passenger interface 1200, the vehicle coordinator 1300 and/or the remote expert interface 1400 on a circuit board of an associated electronic device. The circuit board can be a general circuit board that can house various components of the electronic device's internal electronic system and can also provide connections for other peripheral devices. More specifically, the circuit board can provide the electrical connections that allow the other components of the system to communicate electrically. Any suitable processors (including digital signal processors, microprocessors, supporting chipsets, etc.), computer readable non-volatile memory elements, etc. may be suitably connected to the board based on particular configuration needs, processing requirements, computer design, etc. Other components such as Devices such as external storage, additional sensors, controllers for audio/video playback and peripherals can be connected to the board as plug-in cards, via cables or integrated into the board itself. In various embodiments, the functionality described herein may be implemented in emulation form as software or firmware running within one or more configurable (e.g., programmable) elements arranged in a structure that supports those functionality. The software or firmware that provides the emulation may be provided on a non-transitory, computer-readable storage medium that includes instructions that enable one or more processors to perform those functionalities.

Reference List

S0

step of initializing the tracking loop;

S1

method step of recognizing detection points in the current radar image;

S2

step of predicting one or more parameters of each tracklet;

S3

method step of assigning the detection points to the tracklets;

S4

Method step of assigning the tracklets to the object tracks;

S5

method step of updating the metadata of the tracklets;

S6

Method step of initializing new tracklets;

S7

step of correcting the parameters of the tracklets;

S8

step of querying whether the present radar image is the last radar image to be processed;

a1 to al

Gates (association areas) of the tracklets;

d1 to dn

detection points;

t1 to tm

tracklets;

g1 to gk

object traces;

you

unassigned detection points;

fp

current radar image;

fp+1

Radar image following the current radar image;

fp+2

another radar image following the current radar image;

O1, O2

first and second objects;

O1(fp)

first object in a current radar image;

T

trajectory;

FoV

field of view of the radar system;

FoV-110

field of view of the first radar unit;

FoV-120

field of view of the second radar unit;

actual speed of the object;

two or more dimensional velocity vectors of a detection point;

100

radar system;

110

first radar unit;

111

first radar antenna;

120

second radar unit;

121

second radar antenna;

130

tracking calculation unit

1000

system

1100

vehicle

1200

Passenger Interface 1200

1300

Vehicle Coordinator 1300

1400

Remote Expert Interface

1450

on-board computer

1500

sensor unit

The above description of the illustrated embodiments is not intended to be exhaustive or limited as to the precise forms disclosed. While specific implementations of, and examples for, various embodiments or concepts are described herein for purposes of illustration, various equivalent modifications are possible, as will be appreciated by those skilled in the relevant art. These modifications can be made in light of the above detailed description or figures.

Various embodiments may include any suitable combination of the embodiments described above, including alternative (or) embodiments of embodiments described above in the conjunctive form (and) (e.g., the "and" may be "and/or"). Additionally, some embodiments may include one or more articles of manufacture (e.g., non-transitory computer-readable media) having instructions stored thereon that, when executed, result in actions of any of the embodiments described above. Furthermore, some embodiments may include devices or systems having any suitable means for performing the various operations of the embodiments described above.

In certain contexts, the embodiments discussed herein may be applicable to automotive systems (particularly autonomous vehicles, preferably autonomous automobiles), industrial (safety-critical) applications, and industrial process control.

Additionally, certain embodiments discussed herein may be provided for tracking at least one object in measurement data from a radar system in digital signal processing technologies for medical imaging, automotive technologies for safety systems (e.g., stability control systems, driver assistance systems, braking systems, infotainment, and interior applications of any kind).

Portions of various systems for tracking at least one object in measurement data from a radar system as proposed herein may include electronic circuitry to perform the functions described herein. In some cases, one or more portions of the system may be provided by a processor that is specifically configured to perform the functions described herein. For example, the processor may include one or more application specific components, or it may include programmable logic gates configured to perform the functions described herein. The circuit can operate in the analog domain, in the digital domain, or in a mixed-signal domain. In some cases, the processor may be configured to perform the functions described herein by executing one or more instructions stored on a non-transitory, computer-readable storage medium.

In one embodiment, any number of the electrical circuits of the present figures may be implemented on a circuit board of an associated electronic device. The circuit board can be a general circuit board that can house various components of the electronic device's internal electronic system and can also provide connections for other peripheral devices. More specifically, the circuit board can provide the electrical connections that allow the other components of the system to communicate electrically. Any suitable processors (including digital signal processors, microprocessors, supporting chipsets, etc.), computer readable non-volatile memory elements, etc. may be suitably connected to the board based on particular configuration needs, processing requirements, computer design, etc. Other components such as Devices such as external storage, additional sensors, controllers for audio/video playback and peripherals can be connected to the board as plug-in cards, via cables or integrated into the board itself. In various embodiments, the functionality described herein may be implemented in emulation form as software or firmware running within one or more configurable (e.g., programmable) elements arranged in a structure that supports those functionality. The software or firmware that provides the emulation may be provided on a non-transitory, computer-readable storage medium that contains instructions that enable a processor to perform these functionalities.

Claims (21)

Method for tracking at least one object (o1, o2) in measurement data of a radar system (100), which contain several, in particular consecutive, radar images recorded by a radar system (100), comprising: - detecting detection points (d1 to dn) in the radar images (fp); - Assigning the detection points (d1 to dn) of a current radar image (fp) to a large number of tracklets (t1 to tm), each tracklet (t1 to tm) being a trace of at least one detection point (d1 to dn) observed over a number of radar images ; and - Assigning the tracklets (t1 to tm) based on at least one feature parameter to at least one object track (g1 to gk). procedure after claim 1 , wherein obtaining and/or maintaining the tracklets (t1 to tm) and the object tracks (g1 to gk) is based on at least one dynamic system model. A method according to any one of the preceding claims, further comprising: - Predicting one or more parameters of each tracklet (t1 to tm) for the current radar image (fp) by propagating the dynamic system model, the parameters of each tracklet (t1 to tm) having at least a position, in particular a position and a speed, preferably a position and a velocity and an acceleration, and a covariance of the tracklet in a radar image; and - Correcting the parameters of each tracklet (t1 to tm) based on the detection points (d1 to dn) associated with the corresponding tracklet (t1 to tm), where the prediction is performed before associating the detection points with the tracklets and the correcting is performed after the assignment of the detection points to the tracklets is carried out. Method according to one of the preceding claims, wherein when assigning the detection points (d1 to dn) to the tracklets (t1 to tm), a detection point is assigned to a tracklet if a position of the detection point is within a gate of a tracklet, with new tracklets from the detection points be initialized whenever the criterion for the assignment of a detection is not met for any of the tracklets present, in particular when a position of a detection point is outside the gates of all tracklets present. Method according to one of the preceding claims, in particular according to claim 4 , where a gate for each tracklet (t1 to tm) is either fixed in size or adaptive in size, where the size of the gate correlates with the covariance of the tracklet, in particular such that the size of the gate is increased when the covariance increases, and vice versa. Method according to one of the preceding claims, wherein when assigning the detection points (d1 to dn) to the tracklets (t1 to tm), a detection point is assigned to that tracklet which has a position which is closest to the detection point. Method according to one of the preceding claims, wherein when assigning the detection points (d1 to dn) to the tracklets (t1 to tm), a detection point is probabilistically assigned to a plurality of tracklets (t1 to tm), wherein probabilistic values indicating the probability that a detection point is associated with a tracklet are increased as the distance between the position of the detection point and the predicted position of the tracklet decreases, and vice versa. Method according to one of the preceding claims, wherein the feature parameter for grouping the tracklets (t1 to tm), on the basis of which the tracklets (t1 to tm) are clustered into the object tracks (g1 to gk), an overlap of the gates of the individual tracklets (t1 to tm) in at least the current radar image (fp) and/or a summed overlap of the gates of the individual tracklets (t1 to tm) in several previous radar images. Method according to one of the preceding claims, wherein the grouping of the tracklets (t1 to tm) is carried out using a clustering method, in particular using a DBSCAN method. Method according to one of the preceding claims, further comprising correcting parameters of the object track (g1 to gk), in particular a position, a speed and/or an acceleration of the object track (g1 to gk), by updating the parameters of the object track based on a predicted one Speed and/or a predicted acceleration of the tracklets of the corresponding object track. A method according to any one of the preceding claims, wherein each tracklet (t1 to tm) comprises metadata identifying at least one of a status of the tracklet (t1 to tm), a track count value and a lost track count value. Method according to one of the preceding claims, in particular according to claim 11 , further comprising: - updating the metadata of the tracklets (t1 to tm); and - initializing detection points as new tracklets that are not assigned to the existing tracklets (t1 to tm), wherein updating the metadata and initializing detection points as new tracklets are performed after assigning the detection points to the tracklets. Method according to one of the preceding claims, wherein an alpha-beta filter for modeling the dynamics of the tracklets and a Kalman filter for modeling the dynamics of the object tracks is used, or wherein an alpha-beta filter for modeling the dynamics of the tracklets and the Object tracks is used. Method according to one of the preceding claims, in particular according to one of Claims 1 until 12 , wherein for each object track (g1 to gk) an object model is derived from a library of object models and a switching Kalman filter is used to model the object tracks, with a switching state of the switching Kalman filter representing an object class. Method, in particular according to one of the preceding claims, for tracking at least one object (o1, o2) in measurement data of a radar system (100), which comprise several, in particular consecutive, radar images recorded by a radar system (100), comprising: - detecting detection points (d1 to dn) in the radar images (fp); wherein the plurality of radar images included in the measurement data is a first plurality of radar images acquired by a first radar unit (110), wherein the measurement data further includes a second plurality of radar images acquired by a second radar unit (120) are detected that is not associated with the first radar unit (110), the first and second plurality of radar images being synchronized and at least partially overlapping, the radar images containing distance, Doppler and angle measurements, the Doppler measurements being used to produce a multi-dimensional velocity vector is determined for at least one, in particular for several, preferably for each detection point that can be detected in synchronized radar images of the first and the second plurality of radar images, the determination of the multi-dimensional velocity vector on the corresponding Doppler measurements of the first and the second radar unit (110, 120) based. Method according to one of the preceding claims, in particular according to claim 15 , wherein the multi-dimensional velocity vectors are used when correcting parameters of a track, in particular when correcting the parameters of the tracklet (t1 to tm). Method according to one of the preceding claims, in particular according to claim 15 or 16 , wherein the multi-dimensional velocity vectors are used when updating metadata of a track and when initializing detection points as a new track, in particular when updating the metadata of the tracklets (t1 to tm) and when initializing detection points as new tracklets. Method according to one of the preceding claims, in particular according to claim 15 , 16 or 17 , wherein the status of a lane is immediately changed from a preliminary state to a tracked state when the lane is within a range around the position of a detection point for which a multidimensional vector is determined and when a comparison measure, in particular a sum of the inner product , the multi-dimensional velocity vector and the multi-dimensional velocity vectors of the neighboring multi-dimensional velocity vectors of the detection point is equal to or greater than a predetermined threshold, in particular wherein the status of a tracklet is immediately changed from a preliminary state to a tracked state when the tracklet moves within a range around the Position of a detection point is for which a multi-dimensional vector is determined, and when a comparative measure, in particular a sum of the inner product, the multi-dimensional velocity vector and the multi-dimensional velocity vectors of the adjacent multidimensional velocity vectors of the detection point is equal to or greater than a predetermined threshold. Radar system (100), which is designed to track at least one object (o1, o2) in measurement data of the radar system (100), which include a large number of, in particular consecutive, radar images, using the method according to any one of the preceding claims, having : - a first radar unit (110) which is designed to detect a plurality of radar images by transmitting and receiving radar signals which are present at potential objects to be tracked (o1, o2) reflected in a field of view of the first radar unit (110); and - a tracking calculation unit (130) arranged to process acquired radar images by performing the steps of the method according to any one of the preceding claims. Radar system (100) after claim 19 , further comprising: - a second radar unit (120), which is designed to detect a plurality of radar images by transmitting and receiving radar signals that are at potential objects to be tracked (o1, o2) in a field of view of the first radar unit (110) are reflected, wherein the field of view of the first radar unit (110) and the field of view of the second radar unit (120) at least partially overlap. Vehicle in which a radar system (100) after claim 19 or 20 is attached, the vehicle being an aircraft or watercraft or a land vehicle, preferably an automobile, the vehicle being either manned or unmanned.

Images