Object Tracking Based on Motion Model

Abstract

In a method for object tracking, a first state of the object (5), including a first direction of motion (6a) of a point (8a) to be tracked, is estimated based on a motion model. Ambient sensor data representative of the object (5) to be tracked is generated, and a geometric orientation of the object (5) to be tracked is determined based on the ambient sensor data. The point (8a) to be tracked is shifted in relation to the deviation of the geometric orientation from the first direction of motion, and a second state of the object (5) to be tracked is determined in relation to the shifted point (8b).

Description

Technical Field

[0001] The present invention relates to a method for object tracking, which includes, using at least one computing unit, estimating a first state of an object to be tracked, which includes a first motion direction of a point to be tracked whose position is predetermined with respect to the object to be tracked, based on a predetermined motion model for the object to be tracked, and generating ambient sensor data representing the object to be tracked using an ambient sensor system. The present invention further relates to a method for at least partially automatically guiding a host vehicle (Ego-Fahrzeug), an electric vehicle guidance system, and a computer program product.

Background Art

[0002] In the automatic or partially automatic guidance of a host vehicle, that is, for example, in the context of a driving assistance system or an autonomous or partially autonomous driving function, for example, tracking an object such as another road user or another vehicle around the host vehicle from the perspective of the host vehicle is a major task for enabling safe and reliable autonomous or partially autonomous driving or driving assistance. In order to estimate the future state of an object to be tracked, for example, an iterative method is used that assumes a known motion model that approximately describes the dynamic behavior of the object to be tracked. A widely used and conceivable motion model is the so-called single-track model.

[0003] In the publication R. Schubert et.al.: “Comparison and evaluation of advanced motion models for vehicle tracking“, 2008, 11th International Conference on Information Fusion, 2008, pp. 1-6, various further motion models suitable for object tracking in the automotive context are presented and compared.

[0004] For state prediction and verification or refinement, for example, the Kalman filter method (Kalman-Filterverfahren) or a method derived from the Kalman filter method, for example, the extended Kalman filter method (erweitertes Kalman-Filterverfahren), the unscented Kalman filter method (Unscented Kalman-Filterverfahren), etc. are used.

Summary of the Invention

[0005] The object of the present invention is to improve the accuracy of object tracking based on a motion model, in particular, to improve the accuracy of tracking an external vehicle from the perspective of the host vehicle based on a motion model.

[0006] The above object is solved by each subject matter of the independent claims. Advantageous developments and preferred embodiments are the subject matter of the dependent claims.

[0007] The present invention is based on the idea that by shifting the points of the object to be tracked that are tracked based on a motion model or whose state is estimated based on a motion model, the resulting motion direction of the shifted points better matches the geometric orientation of the object to be tracked as obtained based on ambient sensor data.

[0008] In one aspect of the present invention, a method for object tracking is presented. In this method, at least one computing unit, particularly at least one computing unit of a host vehicle, is used to estimate a first state of an object to be tracked, which is particularly located around the host vehicle, based on a predefined motion model for the object to be tracked. Herein, the first state includes a first motion direction of a point to be tracked. A surround sensor system, particularly a surround sensor system of the host vehicle, is used to generate surround sensor data representing the object to be tracked. At least one computing unit is used to identify a geometric orientation of the object to be tracked based on the surround sensor data. The point to be tracked is shifted using at least one computing unit in relation to a first deviation of the geometric orientation from the first motion direction. At least one computing unit is used to identify a second state of the object to be tracked in relation to the first state and the shifted point.

[0009] The point to be tracked has a predefined position with respect to the object to be tracked. Herein, the point to be tracked may be located on or within the object, but may also be located outside the object. That is, the point to be tracked is, in particular, a virtual point that is tracked based on the motion model.

[0010] The state of the object to be tracked, particularly the first state and the second state, includes the corresponding motion direction of the object to be tracked and, in some cases, also includes other model parameters of the point to be tracked, such as speed, acceleration, and / or position, etc. According to the motion model, the motion direction of the state, that is, particularly the motion direction of the first state, corresponds to the motion direction of the point to be tracked. However, within the scope of the method according to the present invention, since this point shifts to the shifted point, the second state includes, for example, the motion direction of the shifted point instead of the motion direction of the point to be tracked.

[0011] The direction of movement of the object to be tracked is a particularly well-suited model parameter. On the one hand, the direction of movement can be estimated within the scope of the movement model, and on the other hand, it can be measured directly or indirectly based on the vehicle's surrounding sensor system, for example, based on a camera, a lidar system or a radar system. In contrast, in order to obtain corresponding measured values regarding the direction of movement based on the surrounding sensor data, in particular, computer vision algorithms or other algorithms can be used for automatic perception.

[0012] Generally, the movement of the point to be tracked consists of a translational movement and a rotational movement. The first direction of movement can be understood in particular as the direction of the translational movement of the point to be tracked.

[0013] The first direction of movement and the geometric orientation may be given, for example, by the corresponding angles in a known coordinate system. Thus, the first deviation corresponds in particular to the angle difference or the absolute value of the corresponding angle difference.

[0014] The geometric orientation of the object to be tracked may be given, for example, by a certain direction fixedly predefined with respect to the object to be tracked. The geometric orientation of the vehicle may be given, for example, by the longitudinal axis of the vehicle or the direction of the longitudinal axis. The geometric orientation can be specified, for example, at least approximately by the orientation or direction of a bounding figure, also referred to as a "Bounding Box" in English, which can be specified by at least one computing unit based on the surrounding sensor data. The bounding figure may correspond, for example, to a rectangle or a cuboid surrounding the object to be tracked in a depiction based on the surrounding sensor data, that is, for example, in a corresponding camera image, or in a corresponding lidar point cloud or radar point cloud.

[0015] When observing various points on an object to be tracked, or various points with predefined positions relative to the object to be tracked, the directions of movement of these points generally differ in the case of the general movement of the object to be tracked. Correspondingly, since the direction of movement of the point to be tracked generally deviates from the geometric direction, the first deviation generally differs from zero.

[0016] Therefore, by considering the first deviation to identify the second state of the object to be tracked, a more accurate and reliable estimation of another state of the object to be tracked becomes possible. In contrast, the shift of the point to be tracked to obtain the shifted point may be larger, for example, as the first deviation is larger.

[0017] The estimation of the first state can be performed, for example, based on the Kalman filter algorithm or other mathematical estimation algorithms, particularly iterative estimation algorithms. Such methods usually include estimating the state of the object to be tracked, particularly based on a motion model, and refining or improving the estimated state by considering measurement values, particularly ambient sensor data. Thereby, while being able to be based on the motion model, corresponding actual measurement values can also be detected to consider deviations from the ideal behavior or the expected behavior. Here, in the context of the present invention, the first state may correspond to, for example, the state estimated based on the motion model, and the second state may correspond to the state refined or improved based on the ambient sensor data. In such a case, the first state and the second state are particularly related to the same time interval or the same iteration step.

[0018] Here, according to the present invention, the refinement step, i.e., the step for specifying the second state of the object to be tracked, is not executed based on the direction of motion and the point to be tracked estimated as part of the first direction of motion and the first state, but is executed based on the shifted point. By taking into account the deviation of the first direction of motion from the geometric orientation, in particular, the second deviation of the direction of motion of the shifted point from the geometric orientation can be smaller than the first deviation, or can be equal to zero in an ideal case. In this way, a better match between the refined estimated state and the motion model is achieved. Thus, overall, especially when the above-described method steps are repeatedly executed, errors during object tracking can be reduced, and accordingly, the accuracy and reliability of object tracking can be increased. Finally, this enhances the safety of the driving assistance function or the safety of the function for automatic or partially automatic driving of the host vehicle based on the output of the object tracking method.

[0019] Here, the improvement in accuracy according to the present invention increases as the first deviation of the first direction of motion from the geometric orientation of the object to be tracked increases. Thus, the improvement in accuracy can be more powerful especially when the object is larger.

[0020] Depending on the embodiment, the surround sensor system can include one or more subsystems, for example one or more cameras, one or more lidar sensor systems, and / or one or more radar sensor systems. Correspondingly, the surround sensor data can include one or more camera images, one or more lidar point clouds, and / or one or more radar point clouds.

[0021] According to at least one embodiment of the method for object tracking according to the present invention, in particular using at least one computing unit, the current radius of motion of the point to be tracked is specified based on the first state. In this case, the shift of the point to be tracked is performed in relation to the current radius of motion.

[0022] In this specification, both the point to be tracked and the shifted point are located on an arc corresponding to the current radius of motion. In particular, at least one computing unit can identify the center point of the circle of the arc and the current radius of motion based on the first state.

[0023] The shift of the point to be tracked is performed in relation to the first deviation and the current radius of motion. In particular, the greater the first deviation and the greater the current radius of motion, the greater the shift is performed. This can at least partially compensate for the geometric directional deviation of the shifted point from the second direction of motion with respect to the geometric directional deviation of the point to be tracked from the first direction of motion.

[0024] According to at least one embodiment, the first state includes the translational speed of the object to be tracked, in particular the point to be tracked, and the angular speed of the point to be tracked. The current radius of motion is in particular specified as the ratio of the translational speed to the angular speed, i.e., as the quotient of the translational speed and the angular speed, using at least one computing unit.

[0025] The translational speed is in particular parallel to the first direction of motion in this specification, and the angular speed corresponds to the angular speed about the center point of the circle corresponding to the current radius of motion. In this way, the current radius of motion can be reliably identified or estimated.

[0026] According to at least one embodiment, the shift of the point to be tracked is performed along an arc having a radius equal to the current radius of motion. In other words, the point to be tracked and the shifted point are located on one arc.

[0027] According to at least one embodiment, the first deviation is specified as the first angular difference between the geometric direction and the first direction of motion. The shift is performed by an arc section of the arc, and this arc section has a length L given by L = D * R, where D represents the first angular difference and R represents the current radius of motion.

[0028] Therefore, when the first angular difference is small, a very accurate shift can be performed so that the first deviation is compensated as much as possible, that is, the second deviation of the second movement direction of the shifted point with respect to the geometric orientation is approximately equal to zero. Therefore, a particularly good agreement with the motion model is achieved.

[0029] According to at least one embodiment, the second state includes the second movement direction of the shifted point. The second deviation of the geometric orientation from the second movement direction is smaller than the first deviation and is particularly equal to zero or approximately equal to zero.

[0030] According to at least one embodiment, the second movement direction is equal to or approximately equal to the geometric orientation.

[0031] In other words, in this case the second deviation is at least approximately equal to zero. Therefore, a good agreement with the motion model is achieved.

[0032] According to at least one embodiment, a point cloud is generated based on the ambient sensor data or the ambient sensor data includes a point cloud. A part of the point cloud representing the object to be tracked is identified. A bounding figure surrounding a part of the point cloud is determined, in particular, using at least one computing unit, where the bounding figure has a predefined geometric shape. The geometric orientation of the object to be tracked corresponds to the spatial orientation of the bounding figure.

[0033] By predetermining the geometric shape of the bounding figure, for example, by the corresponding symmetry or, in the case of a polygon, by the number of sides, the geometric orientation can be correspondingly uniquely defined in accordance with the spatial orientation of the bounding figure. In particular, the bounding figure is a rectangle or a cuboid. Here, depending on the embodiment, the aspect ratio of the rectangle or cuboid may be predefined or variable.

[0034] The point cloud corresponds in particular to a lidar point cloud or a radar point cloud. The point cloud, in this specification, contains a plurality of points. A part of the point cloud representing an object to be tracked corresponds to a subset of the point cloud. Identifying or discriminating a part of the point cloud representing an object to be tracked can be achieved, for example, by using a clustering method.

[0035] Such an embodiment has the advantage that the corresponding bounding figure can be identified based on a known method, thereby accurately identifying and compensating for the first deviation. In particular, by using a bounding figure in the shape of a rectangle or a cuboid, reproducible and reliable results for the identification of the geometric orientation can be obtained.

[0036] According to at least one embodiment, a camera image is generated based on the ambient sensor data, or the ambient sensor data includes a camera image. In the camera image, a bounding figure surrounding the depiction of the object to be tracked is identified, and the bounding figure has a predefined shape. The geometric orientation of the object to be tracked corresponds to the spatial orientation of the bounding figure.

[0037] Regarding the bounding figure, in this application, the above also applies equally to the bounding figure of a part of the point cloud. In such an embodiment, the bounding figure can be identified using at least one computing unit, for example, based on an object identification algorithm. This can be, for example, an algorithm based on machine learning, for example, an algorithm based on a trained artificial neural network. In contrast, a number of structures corresponding to, for example, the so-called YOLO algorithm are known.

[0038] According to at least one embodiment, a method based on a Kalman filter is used to identify the second state.

[0039] The method based on the Kalman filter may correspond to, for example, the Kalman filter method, the extended Kalman filter method, the unscented Kalman filter method, or other methods derived from the Kalman filter method.

[0040] In the present application, in the iterative step of the Kalman filter method, first, a first state is estimated. This is also referred to as prediction. Next, for example, a Kalman gain coefficient or the like is determined, and based on this, this prediction is improved in order to determine a second state, particularly in relation to the corresponding measurement value, here the ambient sensor data. This can also be referred to as refinement or refinement. However, according to the present invention, this refinement is not performed based on the original point to be tracked, but is performed based on the shifted point. In this way, more accurate object tracking can be achieved by an established method or the like based on the Kalman filter.

[0041] According to another aspect of the present invention, a method for at least partially automatically guiding a host vehicle, particularly an automobile, is presented. Using the host vehicle, particularly using an electronic vehicle guidance system of the host vehicle including an ambient sensor system and at least one computing unit, a method for object tracking according to the present invention is implemented. Using a control unit of the host vehicle, particularly a control unit of the electronic vehicle guidance system, for example, a control unit of at least one computing unit, at least one control signal for at least partially automatically guiding the host vehicle is generated in relation to a second state of the object to be tracked.

[0042] For example, at least one control signal is supplied to at least one corresponding actuator of the host vehicle, and then this actuator can at least partially automatically guide the host vehicle based on the at least one control signal or implement at least partial automatic guidance. Alternatively or additionally, at least one control signal can also be used for driving assistance for the driver of the host vehicle.

[0043] According to another aspect of the present invention, an electronic vehicle guidance system for a host vehicle is provided. The electronic vehicle guidance system has at least one computing unit configured to estimate a first state of an object to be tracked, particularly around the host vehicle, based on a predefined motion model for the object to be tracked, the first state including a first motion direction of a point to be tracked. The electronic vehicle guidance system has a surrounding sensor system for the host vehicle, the surrounding sensor system being configured to generate surrounding sensor data representing the object to be tracked. The at least one computing unit is configured to identify a geometric orientation of the object to be tracked based on the surrounding sensor data, shift the point to be tracked in relation to a first deviation of the geometric orientation from the first motion direction, and identify a second state of the object to be tracked in relation to the first state and the shifted point.

[0044] Another embodiment of the electronic vehicle guidance system directly results from various embodiments of the method for object tracking according to the present invention or from various embodiments of the method for at least partially automatically guiding the host vehicle according to the present invention, and vice versa. In particular, the electronic vehicle guidance system according to the present invention is configured to implement the method according to the present invention or implements such a method.

[0045] According to another aspect of the present invention, a computer program comprising instructions is provided. When these instructions are executed by the electronic vehicle guidance system according to the present invention, particularly by at least one computing unit of the electronic vehicle guidance system, these instructions cause the electronic vehicle guidance system to implement the method for object tracking according to the present invention or the method for at least partially automatically guiding the host vehicle according to the present invention.

[0046] According to another aspect of the present invention, a computer-readable storage medium storing the computer program according to the present invention is provided.

[0047] The computer program according to the present invention and the computer-readable storage medium according to the present invention can be interpreted as each computer program product including instructions.

[0048] An electronic vehicle guidance system can be understood as an electronic system configured to guide or control an automobile in a fully automatic or fully autonomous manner, particularly without requiring intervention by a driver for control. In this case, the automobile or the electronic vehicle guidance system spontaneously and fully automatically performs all necessary functions, such as steering operations, braking operations, and / or acceleration operations required in some cases, observation and detection of road traffic, and related necessary reactions. In particular, the electronic vehicle guidance system can be used to implement a fully automatic driving mode or a fully autonomous driving mode of an automobile according to level 5 of the classification compliant with SAE J3016. The electronic vehicle guidance system can also be understood as an advanced driver assistance system (ADAS) that assists a driver in a partially automated driving or a partially autonomous driving of an automobile. In particular, the electronic vehicle guidance system can be used to implement a partially automated driving mode or a partially autonomous driving mode of an automobile according to any one of levels 1 to 4 of the classification compliant with SAE J3016. Here, and hereinafter, "SAE J3016" refers to the corresponding standard of the June 2018 version.

[0049] Therefore, at least partially automated vehicle guidance can include guiding an automobile according to a fully automatic driving mode or a fully autonomous driving mode of level 5 compliant with SAE J3016. At least partially automated vehicle guidance can also include guiding an automobile according to a partially automated driving mode or a partially autonomous driving mode according to any one of levels 1 to 4 compliant with SAE J3016.

[0050] A computing unit can be understood as a data processing device in particular, that is, the computing unit can process data for executing computing operations in particular. These operations may include operations for indirectly accessing a data structure, such as a conversion table LUT (short for "look-up table" in English), in some cases.

[0051] The computing unit can particularly include one or more computers, one or more microcontrollers, and / or one or more integrated circuits, such as one or more application-specific integrated circuits, ASICs (short for "application-specific integrated circuit" in English), one or more field-programmable gate arrays, FPGAs, and / or one or more system-on-a-chip, SoCs (short for "system on a chip" in English). The computing unit may also include one or more processors, such as one or more microprocessors, one or more central processing unit units, CPUs (short for "central processing unit" in English), one or more graphics processing unit units, GPUs (short for "graphics processing unit" in English), and / or one or more signal processors, particularly one or more digital signal processors, DSPs. The computing unit may also include a physical or virtual combination of a computer or other units described above.

[0052] In various embodiments, the computing unit includes one or more hardware interfaces and / or software interfaces and / or one or more storage units.

[0053] The memory unit may be configured as a volatile data memory, for example, a dynamic memory with random access, as DRAM (short for "dynamic random access memory" in English), or a static memory with random access, as SRAM (short for "static random access memory" in English), or as a non-volatile data memory, for example, a read-only memory, as ROM (short for "read-only memory" in English), a programmable read-only memory, as PROM (short for "programmable read-only memory" in English), an erasable read-only memory, as EPROM (short for "erasable read-only memory" in English), an electrically erasable read-only memory, as EEPROM (short for "electrically erasable read-only memory" in English), a flash memory or flash EEPROM, a ferroelectric random access memory with random access, as FRAM (short for "ferroelectric random access memory" in English), a magnetoresistive random access memory with random access, as MRAM (short for "magnetoresistive random access memory" in English), or a phase change random access memory with random access, as PCRAM (short for "phase-change random access memory" in English).

[0054] Within the scope of the present disclosure, when a component of an electronic vehicle guidance system according to the present invention, in particular at least one computing unit or control unit of the electronic vehicle guidance system, is described as being configured, formed, designed, etc. to perform a specific function, to achieve a specific effect, or to be used for a specific purpose, it can be understood that this component, beyond the basic or theoretical usability or suitability of the component for this function, effect, or purpose, can specifically and actually perform or realize this function, achieve this effect, or be used for this purpose by means of corresponding adaptation, programming, physical design, etc.

[0055] An object recognition algorithm can be understood as a computer algorithm that is capable of identifying one or more objects within a provided input image by defining corresponding bounding figures or bounding boxes (in English, "bounding boxes") and assigning a corresponding object class to each of these bounding boxes, where the object class can be selected from a predefined set of object classes. In the present application, the assignment of an object class to a bounding box can be understood as being provided with a corresponding confidence value or probability regarding the fact that the object identified within the bounding box belongs to the corresponding object class. For example, the algorithm can provide such a confidence value or probability for a given bounding box for each object class. The assignment of an object class can include, for example, the selection or provision of the object class having the maximum confidence value or maximum probability. Alternatively, the algorithm can also define only the bounding boxes without assigning a corresponding object class.

[0056] Further features of the present invention will become apparent from the claims, the drawings, and the description of the drawings. Features and combinations of features mentioned in the above description, as well as features and combinations of features mentioned and / or shown in the description of the drawings hereinafter and / or in the drawings, may be included not only in each of the combinations described but also in other combinations of the present invention. In particular, the present invention also includes configurations and combinations of features that do not have all the features initially formulated in the claims. Furthermore, the present invention includes configurations and combinations of features that exceed or deviate from the combinations of features represented in the citation of the claims.

Brief Description of the Drawings

[0057]

Figure 1

Figure 2

Embodiments for Carrying Out the Invention

[0058] FIG. 1 schematically shows a host vehicle 1 having an exemplary embodiment of an electronic vehicle guidance system 2 according to the present invention. Further, an object 5 to be tracked around the host vehicle 1 is shown. The object 5 to be tracked is, in particular, another vehicle, merely schematically shown as a rectangle.

[0059] The electronic vehicle guidance system 2 includes a computing unit 3, which may be configured as, for example, a control device, an ECU of the host vehicle 1, or may be part of the control device. The electronic vehicle guidance system 2 further has a surrounding sensor system 4a, 4b, for example, a camera 4b and / or a lidar system 4a and / or a radar system (not shown).

[0060] The vehicle guidance system 2 can implement a method for object tracking according to the present invention. A corresponding flowchart of such a method is schematically shown in FIG. 2.

[0061] In step S1 of this method, for example, the initial state of the object 5 to be tracked is determined based on a predetermined motion model, for example, a single-track model. For this purpose, the calculation unit 3 can use, for example, the ambient sensor data of the ambient sensor systems 4a, 4b. In step S2, the first state of the object 5 to be tracked is estimated or predicted using the calculation unit 3 based on a predetermined motion model. Here, the first state includes the first motion direction 6a of the point 8a to be tracked of the object 5 to be tracked. Further, in step S2, ambient sensor data representing the object 5 to be tracked is generated using the ambient sensor systems 4a, 4b.

[0062] In step S3, the calculation unit 3 determines the geometric orientation of the object 5 to be tracked based on the ambient sensor data. For this purpose, the calculation unit 3 can apply, for example, an object detection algorithm to the camera image of the camera 4b and / or apply a clustering algorithm to the point cloud of the lidar system 4a. In this way, the calculation unit 3 can obtain a bounding figure surrounding the object 5 to be tracked. In FIG. 1, a rectangle is shown as the bounding figure 7. In this case, the geometric orientation corresponds to a predetermined spatial orientation of the bounding figure 7. For example, the geometric orientation is parallel to one side of the bounding figure 7, particularly to the long side of the rectangle.

[0063] Based on the ambient sensor data, the calculation unit 3 further determines the current radius of motion of the object to be tracked, and the current radius of motion corresponds to the radius of circle 9. Then, the point 8a to be tracked is shifted along circle 9 by an arc section 10 of length L = D * R by the calculation unit 3, thereby generating a shifted point 8b. Here, R corresponds to the radius of circle 9, and D corresponds to the angular difference between the first direction of motion 6a of the point 8a to be tracked and the geometric orientation of the object 5 to be tracked. Here, the second direction of motion 6b of the shifted point 8b is similarly approximately equal to the geometric orientation.

[0064] In step S4, the calculation unit can calculate a correction coefficient based on the shifted point 8b, particularly based on the second direction of motion 6b. For example, if an approach based on a Kalman filter is pursued, the correction coefficient may correspond to the corresponding Kalman amplification coefficient. Next, in step S4, the state of the object 5 to be tracked is updated in relation to the correction coefficient identified in relation to the shifted point 8b as well.

[0065] In this way, it is possible to achieve that the second direction of motion 6b used to update the state of the object to be tracked at least approximately coincides with the geometric orientation of the object 5 to be tracked, thereby enabling more accurate object tracking.

[0066] In various embodiments of the present invention, the point to be tracked is shifted to a point corresponding to a point that particularly does not have a lateral velocity, that is, a point that moves in the direction of the geometric orientation of the object to be tracked.

[0067] The object to be tracked is, in particular, a vehicle, for example a motor vehicle having two axes. If there is no wheel slip in the lateral direction, the shifted point may be located, for example, on the uncontrollable axis of the vehicle, i.e. the axis equipped with an uncontrollable wheel. If there is wheel slip in the lateral direction but the vehicle is not rotating and the vehicle has an axis that is controlled and an axis that is not controlled, the shifted point may be located, for example, between these two axes. In addition to the lateral wheel slip, if there is also rotation of the vehicle and / or in the general case where the two axes are controllable axes, the shifted point may also be located outside the vehicle.

[0068] Generally, the shifted point can be identified by minimizing the geometric orientation deviation of the object to be tracked from the direction of motion, based on the motion model.

[0069] In many embodiments, an object filter based on the corresponding state vector, for example an extended Kalman filter, can be used. The state vector can include, for example, the two-dimensional position coordinates of the point to be tracked, the translational velocity, the yaw rate, the yaw angle, etc. The calculation unit can identify the current radius of motion based on the motion model or based on the ratio of the translational velocity to the angular velocity of the object. Similarly, the center point of the corresponding circle may be identified, where the connecting line between the center point of the circle and the point to be tracked is perpendicular to the direction of motion of the point to be tracked. Then, in order to identify the point that is optimally shifted for tracking, the point to be tracked can be shifted on the circle by an angle corresponding to the angular difference between the geometric orientation and the direction of motion.

Claims

1. A method for object tracking, comprising: estimating, using at least one computing unit (3), a first state of an object (5) to be tracked, including a first motion direction (6a) of a point (8a) to be tracked, based on a predetermined motion model for the object (5) to be tracked; generating, using a surrounding sensor system (4a, 4b), surrounding sensor data representing the object (5) to be tracked; In the method, using the at least one computing unit (3) to identify a geometric orientation of the object (5) to be tracked based on the surrounding sensor data; shifting the point (8a) to be tracked, using the at least one computing unit (3), in relation to a first deviation of the geometric orientation from the first motion direction (6a); characterized in that, using the at least one computing unit (3), a second state of the object (5) to be tracked is identified in relation to the first state and the shifted point (8b).

2. Identifying a current radius of motion of the point (8a) to be tracked based on the first state; Performing the shift of the point (8a) to be tracked in relation to the current radius of motion. The method according to claim 1, characterized by the above.

3. The first state includes a translational velocity of the point (8a) to be tracked and an angular velocity of the point (8a) to be tracked; Identifying the current radius of motion as a ratio of the translational velocity to the angular velocity. The method according to claim 2, characterized by the above.

4. The method according to claim 2, characterized in that the shift of the point (8a) to be tracked is performed along an arc (9) having a radius equal to the current radius of motion.

5. Identifying the first deviation as a first angular difference between the geometric orientation and the first motion direction (6a); Performing the shift by an arc section (10) having a length L = D * R, where D represents the first angular difference and R represents the current radius of motion. The method according to claim 4, characterized by the above.

6. The second state includes a second motion direction (6b) of the shifted point (8b), and a second deviation of the geometric orientation from the second motion direction (6b) is smaller than the first deviation. The method according to claim 1, characterized by the above.

7. The method according to claim 6, characterized in that the second motion direction (6b) is equal to the geometric orientation.

8. generating a point cloud based on the ambient sensor data, or the ambient sensor data includes the point cloud, identifying a part of the point cloud representing the object (5) to be tracked, identifying a bounding figure (7) having a predefined shape that encloses the part of the point cloud, wherein the geometric orientation of the object (5) to be tracked corresponds to the spatial orientation of the bounding figure (7) The method according to claim 1, characterized in that.

9. generating a camera image based on the ambient sensor data, or the ambient sensor data includes the camera image, identifying, in the camera image, a bounding figure (7) having a predefined shape that encloses the depiction of the object (5) to be tracked, wherein the geometric orientation of the object (5) to be tracked corresponds to the spatial orientation of the bounding figure (7) The method according to claim 1, characterized in that.

10. The predefined shape of the bounding figure (7) corresponds to a rectangle, and the spatial orientation of the bounding figure (7) is parallel to one side of the rectangle, or The predefined shape of the bounding figure (7) corresponds to a cuboid, and the spatial orientation of the bounding figure (7) is parallel to one side of the cuboid The method according to claim 8, characterized in that.

11. The method according to claim 1, characterized in that a method based on a Kalman filter is used to identify the second state.

12. In a method for at least partially automatically guiding a host vehicle (1), using the host vehicle (1) to implement the method for object tracking according to any one of claims 1 to 11, using a control unit of the host vehicle (1) to generate at least one control signal for at least partially automatically guiding the host vehicle (1) in relation to the second state of the object (5) to be tracked The method, characterized in that.

13. An electronic vehicle guidance system (2) for a host vehicle (1), wherein the electronic vehicle guidance system (2) is At least one computing unit (3) configured to estimate a first state of an object (5) to be tracked, including a first movement direction (6a) of a point (8a) to be tracked, based on a predefined movement model for the object (5) to be tracked. A surround sensor system (4a, 4b) configured to generate surround sensor data representing the object (5) to be tracked. In an electronic vehicle guidance system (2) having the above. The at least one computing unit (3) Identifies a geometric orientation of the object (5) to be tracked based on the surround sensor data. Shifts the point (8a) to be tracked in relation to a first deviation of the geometric orientation from the first movement direction (6a). Identifies a second state of the object (5) to be tracked in relation to the first state and the shifted point (8b). An electronic vehicle guidance system (2), characterized in that it is configured as described above.

14. The surround sensor system (4a, 4b) includes a camera and / or a lidar system and / or a radar system. The electronic vehicle guidance system (2) according to claim 13.

15. A computer program product comprising instructions, When the instructions are executed by the electronic vehicle guidance system (2) according to claim 13 or 14, the instructions cause the electronic vehicle guidance system (2) to perform the method according to claim 1. A computer program product.

Images