Systems and methods for tracking an extended state of a moving object using a composite measurement model

Abstract

A tracking system for tracking an expansion state of an object is provided. The tracking system includes at least one processor and a memory having stored therein instructions that, when executed by the at least one processor, cause the tracking system to perform a probabilistic filter that iteratively tracks a confidence of the expansion state of the object, wherein the confidence is predicted using a motion model of the object and further updated using a composite measurement model of the object. The composite measurement model includes a plurality of probability distributions constrained to lie on a contour of the object and has a predetermined relative geometric mapping to a center of the object. Further, the tracking system tracks the expansion state of the object based on the updated confidence of the expansion state.

Description

Technical Field

[0001] This disclosure generally relates to vehicle object tracking, and more specifically, to a system and method for tracking the expanded state of an object using measurements of the object. Background Technology

[0002] The control systems employed by vehicles (such as autonomous and semi-autonomous vehicles) predict safe movements or paths to avoid collisions with obstacles such as other vehicles or pedestrians. In some scenarios, vehicles are also configured to sense their surroundings using one or more sensors, such as road edges, pedestrians, and other vehicles. Some of these sensors include ultrasonic sensors, cameras, and LiDAR sensors used in existing advanced driver assistance systems (ADAS).

[0003] Vehicle control systems rely on automotive radar measurements to track the object states (including kinematic states) of other vehicles in order to control the vehicle. Extended object tracking (EOT), which involves multiple measurements per scan, has shown improved object tracking compared to traditional point object tracking, which includes only one measurement per scan. This is achieved by enhancing the object state from kinematic state only to both kinematic and extended states. The extended state provides the size and orientation of the tracked object. To achieve this, the spatial distribution (i.e., how automotive radar measurements are spatially distributed around the object) needs to be captured along with sensor noise. Current methods involve a framework of fixed points on a rigid body, which requires an inflexible data association between the fixed point set and the automotive radar detection, even for single object tracking. Spatial models, such as contour models and surface models, bypass this cumbersome data association step.

[0004] For automotive radar measurements, contour models reflect the distribution of measurements along the contour of an object (e.g., a rigid body), while surface models assume radar measurements are generated from the inner surface of a two-dimensional shape. Examples of contour models include simple rectangular shapes as well as more general star-convex shapes modeled by random hypersurface models or Gaussian process models. Some surface models, such as Gaussian-based elliptical models and hierarchical Gaussian-based elliptical models, are computationally much simpler than contour models, which require more degrees of freedom to describe more complex shapes. However, object measurements are affected by noise and only receive reflections from the object's surface. Therefore, the models described above do not capture real-world automotive radar measurements.

[0005] Therefore, there is a need for a system and method to track the kinematic and extension states of an object by capturing real-world automotive radar measurements. Summary of the Invention

[0006] Some implementations aim to provide a system and method for tracking the expansion state of an object. The expansion state of an object includes a kinematic state indicating one or a combination of the object's center position and velocity, and an extension state indicating one or a combination of the object's size and orientation. The object's center is one or a combination of an arbitrarily chosen point, the object's geometric center, the object's center of gravity, the center of a vehicle's rear axle, etc. Sensors (e.g., automotive radar) are used to track the object (e.g., a vehicle). In implementations, automotive radar can provide direct measurements of radial velocity, long operating range, small size in millimeter or Asia-Pacific Hertz bands, and high spatial resolution.

[0007] In point object tracking, a single measurement is received per scan from the vehicle. Point object tracking only provides the vehicle's kinematic state (position). Furthermore, a probabilistic filter using a measurement model with a kinematic state distribution is used to track the vehicle. In extended object tracking (EOT), multiple measurements are received per scan. These multiple measurements are constructed around the vehicle in space. Extended object tracking provides both the vehicle's kinematic state and extended state. A probabilistic filter using a measurement model with an extended state distribution is used to track the vehicle.

[0008] However, real-world automotive radar measurement distributions reveal a complex interplay of multiple reflections from a vehicle. This complexity complicates the design of a proper measurement model. Consequently, conventional measurement models are only applicable to kinematic states, not extended states.

[0009] Therefore, in some implementations, spatial models such as contour models and surface models are used to capture real-world automotive radar measurements. However, the aforementioned spatial models are inaccurate. Some implementations are based on the understanding that real-world automotive radar measurements are distributed around the edges or surfaces of an object (vehicle) with a specific volume, resulting in surface volume models. Therefore, some implementations aim to develop surface volume models that resemble and capture real-world automotive radar measurements. The surface volume model strikes a balance between contour models and surface models, possessing more realistic features while maintaining EOT accuracy.

[0010] In particular, in the implementation, a composite measurement model (a surface volume model) is determined based on the principles of contour models and surface models. The composite measurement model includes multiple probability distributions constrained to lie on the object contour and has a predetermined relative geometric mapping to the object center. The multiple probability distributions are used to cover the measurement extension along the object contour.

[0011] Composite measurement models are composite in several ways. For example, a composite measurement model has a composite structure, i.e., multiple probability distributions. Additionally, a composite measurement model has composite components, i.e., functions of multiple probability distributions, functions of contours, and their relationships. Furthermore, a composite measurement model has composite properties, i.e., multiple probability distributions are based on measurements, thus representing a data-driven approach to model generation, while the contours are based on modeling the shape of an object (e.g., the shape of a vehicle) using principles of physics-based modeling.

[0012] Furthermore, the composite measurement model utilizes different principles for modeling expansion states; that is, it combines the principles of contour models and surface models. As a result, the composite measurement model better represents the physical properties of the tracked object while simplifying measurement assignment. Additionally, the multiple probability distributions of the composite measurement model are more flexible than the single distribution of the surface model, better describing the object's contour and more flexibly interpreting measurements from different angles or views of the object.

[0013] Some implementations are based on the understanding that, theoretically, the shape of the contour is unrestricted, and multiple probability distributions can lie on the contour. However, in practice, these assumptions are incorrect for tracking expansion states. Instead, the object's contour is predetermined, and multiple probability distributions are fitted to the contour, rather than the contour being fitted to multiple probability distributions.

[0014] Offline (i.e., in advance) learning of the composite measurement model. The composite measurement model can be learned in a unit coordinate system or a global coordinate system. Some implementations are based on the understanding that learning the composite measurement model in a unit coordinate system is beneficial because it simplifies calculations and makes the composite measurement model unaware of the object's dimensions. Each of multiple probability distributions (represented as ellipses) can be assigned a measurement probabilistically. Measurements associated with an ellipse can be called elliptic assigned measurements.

[0015] According to some implementations, the composite measurement model learned offline is used for online tracking of the expansion state of an object (i.e., real-time tracking of the expansion state of the object). Some implementations are based on the understanding that the probabilistic properties of the composite measurement model can be advantageously aligned with probabilistic multiple hypothesis tracking (PMHT) methods. For example, this alignment allows for the implementation of probabilistic filters using at least a variant of the Kalman filter. For example, one implementation uses the unscented Kalman filter-probabilistic multiple hypothesis tracking (UKF-PMHT) method. The unscented Kalman filter (UKF) is used to transform the composite measurement model from a unit coordinate system to a global coordinate system. The probabilistic multiple hypothesis tracking (PMHT) method is then applied to assign the measurement at the current time step to different probability distributions in a probabilistic manner and update the expansion state of the object.

[0016] According to the implementation, given the expansion state of an object, the covariance matrix corresponding to the previous time step, and the object's motion model, the expansion state of the object at the current time step and the covariance matrix corresponding to the expansion state can be predicted. In one implementation, the motion model can be a coordinated turning (CT) motion model with extreme velocities. In some other implementations, a coordinated turning (CT) motion model with extreme velocities is used for the kinematic state, and a constant model with process noise and small covariance is used for the extension state (i.e., length and width), since the length and width are unlikely to change over time. The predicted expansion state of the object can be referred to as the prediction confidence of the expansion state because the prediction is probabilistic. Some implementations are based on the understanding that the predicted expansion state of the object may be inaccurate. To correct the predicted expansion state of the object, a composite measurement model in a unit coordinate system is used. However, the predicted expansion state is in a global coordinate system. Therefore, to align the composite measurement model in the unit coordinate system with the predicted expansion state, it is necessary to transform the composite measurement model from the unit coordinate system to the global coordinate system. In particular, it is necessary to transform the elliptical assignment measurement in the unit coordinate system to the global coordinate system.

[0017] Some implementations are based on the understanding that this transformation can be achieved using an unscented transformation function (or UKF). To this end, in some implementations, sigma points are generated for the ellipse (i.e., the probability distribution of the composite measurement model). "Ellipse" and "probability distribution" are used interchangeably and mean the same thing. Furthermore, the sigma points are propagated into the unscented transformation function, thus determining the predicted measurement in the global coordinate system corresponding to the ellipse-assigned measurement in the unit coordinate system. Additionally, the covariance matrix corresponding to the predicted measurement is determined based on the predicted measurement. Similarly, the measurements in the global coordinate system corresponding to the ellipse-assigned measurements associated with the remaining ellipses are determined. To this end, a predicted extended state model is obtained, where the composite measurement model is aligned according to the predicted extended state.

[0018] In addition, measurements at the current time step are received. Some implementations are based on the understanding that multiple probability distributions (ellipses) can be processed independently of each other. This independent processing allows for consideration of different perspectives for probing the expansion state of an object. To accommodate this independent processing, some implementations treat different probability distributions among the multiple probability distributions as belonging to different objects. Furthermore, some implementations are based on the understanding that soft probability assignment (i.e., probability assignment of measurements to different probability distributions) is more advantageous than hard deterministic assignment. Therefore, measurements are assigned to each probability distribution with corresponding associated probabilities. Measurements with corresponding associated probabilities associated with each probability distribution are called "synthetic measurements."

[0019] Furthermore, in some implementations, a composite centroid and a composite covariance matrix are determined for each probability distribution based on the corresponding composite measurements. Additionally, the prediction confidence of the expanded state is updated using the composite measurements associated with each probability distribution. For example, a probabilistic filter, such as a Kalman filter, can be used to update the prediction confidence of the expanded state with the composite measurements associated with each probability distribution to generate an updated expanded state for the object.

[0020] Therefore, one embodiment discloses a tracking system for tracking the expansion state of an object, the expansion state comprising a kinematic state indicating a combination of the position and velocity of the object's center and an extension state indicating a combination of the object's size and orientation. The tracking system includes: at least one processor; and a memory storing instructions that, when executed by the at least one processor, cause the tracking system to receive measurements associated with at least one sensor, wherein the at least one sensor is configured to detect a scene including the object using one or more signal transmissions to generate one or more measurements of the object per transmission; a probabilistic filter that iteratively tracks the confidence of the object's expansion state, wherein the confidence is predicted using a motion model of the object and updated using a composite measurement model of the object, wherein the composite measurement model includes multiple probability distributions constrained to lie on the object's contour and having a predetermined relative geometric mapping to the object's center; wherein, in each iteration of the iterative tracking, the confidence of the expansion state is updated based on the difference between a predicted confidence and an updated confidence, wherein the updated confidence is estimated based on the probability of measurements fitted to each of the multiple probability distributions and mapped to the object's expansion state based on the corresponding geometric mapping; and the expansion state of the object is tracked based on the updated confidence of the expansion state.

[0021] Therefore, another embodiment discloses a tracking method for tracking the expansion state of an object, the expansion state including a kinematic state indicating one or a combination of the position and velocity of the object's center and an extension state indicating one or a combination of the object's size and orientation. The tracking method includes: receiving measurements associated with at least one sensor, wherein the at least one sensor is configured to detect a scene including the object using one or more signal transmissions to generate one or more measurements of the object per transmission; performing a probabilistic filter to iteratively track the confidence of the object's expansion state, wherein the confidence is predicted using a motion model of the object and updated using a composite measurement model of the object, wherein the composite measurement model includes multiple probability distributions constrained to lie on the object's contour and has a predetermined relative geometric mapping to the object's center; wherein, in each iteration of the iterative tracking, the confidence of the expansion state is updated based on the difference between a predicted confidence and an updated confidence, wherein the updated confidence is estimated based on the probability of measurements fitted to each of the multiple probability distributions and mapped to the object's expansion state based on the corresponding geometric mapping; and tracking the object's expansion state based on the updated confidence of the expansion state.

[0022] A non-transitory computer-readable storage medium embodied thereon has a program executable by a processor to perform a method for tracking an extended state of an object, wherein the extended state includes a kinematic state indicating one or a combination of the position and velocity of the object's center and an extension state indicating one or a combination of the object's size and orientation. The method includes the steps of: receiving measurements associated with at least one sensor, wherein the at least one sensor is configured to detect a scene including the object using one or more signal transmissions to generate one or more measurements of the object per transmission; executing a probabilistic filter that iteratively tracks the confidence of the object's extended state, wherein the confidence is predicted using a kinematic model of the object and updated using a composite measurement model of the object, wherein the composite measurement model includes multiple probability distributions constrained to lie on the object's contour and having a predetermined relative geometric mapping to the object's center; wherein, in each iteration of the iterative tracking, the confidence of the extended state is updated based on the difference between a predicted confidence and an updated confidence, wherein the updated confidence is estimated based on the probability of measurements fitted to each of the multiple probability distributions and mapped to the object's extended state based on the corresponding geometric mapping; and tracking the extended state of the object based on the updated confidence of the extended state.

[0023] The currently disclosed embodiments will be further described with reference to the accompanying drawings. The drawings shown are not necessarily to scale, but rather focus on illustrating the principles of the currently disclosed embodiments. Attached Figure Description

[0024] [ Figure 1A ] Figure 1A This provides a schematic overview of the principles for tracking the expansion state of an object according to some implementation methods.

[0025] [ Figure 1B ] Figure 1B This provides a schematic overview of the principles for tracking the expansion state of an object according to some implementation methods.

[0026] [ Figure 1C ] Figure 1C This provides a schematic overview of the principles for tracking the expansion state of an object according to some implementation methods.

[0027] [ Figure 2 ] Figure 2 A block diagram of a tracking system for tracking the expanded state of an object, according to some embodiments, is shown.

[0028] [ Figure 3A ] Figure 3A Various schematic diagrams are shown illustrating the confidence level of using a composite measurement model to track the expansion state of an object according to some implementations.

[0029] [ Figure 3B ] Figure 3B Various schematic diagrams are shown illustrating the confidence level of using a composite measurement model to track the expansion state of an object according to some implementations.

[0030] [ Figure 3C ] Figure 3C Various schematic diagrams are shown illustrating the confidence level of using a composite measurement model to track the expansion state of an object according to some implementations.

[0031] [ Figure 4 ] Figure 4 A flowchart is shown illustrating a method for learning parameters of a composite measurement model according to some embodiments.

[0032] [ Figure 5A ] Figure 5A This diagram illustrates the transformation of training data collected from different motions of different objects into a common unit coordinate system according to some implementations.

[0033] [ Figure 5B ] Figure 5B This diagram illustrates the transformation of training data collected from different motions of different objects into a common unit coordinate system according to some implementations.

[0034] [ Figure 6 ] Figure 6 A block diagram is shown of an expectation maximization (EM) method for learning parameters of a composite measurement model according to some implementations.

[0035] [ Figure 7A ] Figure 7A A flowchart of an unscented Kalman filter-probabilistic multiple hypothesis tracking (UKF-PMHT) algorithm according to some implementations is shown.

[0036] [ Figure 7B ] Figure 7B A block diagram illustrating the steps performed according to some implementations to calculate the predictive measurement and covariance matrix is ​​shown.

[0037] [ Figure 7C ] Figure 7C A block diagram illustrating the steps performed according to some embodiments to calculate the synthetic measurement and the synthetic covariance matrix is ​​shown.

[0038] [ Figure 7D ] Figure 7D A block diagram illustrating the steps performed according to some implementations to update the extended state and covariance matrix.

[0039] [ Figure 8A ] Figure 8A A schematic diagram of a vehicle is shown, including a controller that communicates with a system employing principles of some implementation methods.

[0040] [ Figure 8B ] Figure 8B The following are shown according to some embodiments. Figure 8A A schematic diagram illustrating the interaction between the system's controller and the vehicle's controller.

[0041] [ Figure 8C ] Figure 8C A schematic diagram is shown of an autonomous or semi-autonomous controlled vehicle that generates control inputs using some implementation methods. Detailed Implementation

[0042] In the following description, numerous specific details are set forth for illustrative purposes in order to provide a thorough understanding of this disclosure. However, it will be apparent to those skilled in the art that this disclosure may be practiced without these specific details. In other instances, apparatuses and methods are shown only as block diagrams to avoid obscuring this disclosure.

[0043] As used in this specification and claims, the terms "for example" and "such as," as well as the verbs "comprising," "having," "including," and other verb forms thereof, when used in conjunction with a list of one or more components or other items, shall each be interpreted as open-ended, meaning that the list should not be considered as excluding other additional components or items. The term "based on" means at least partially based on. Furthermore, it will be understood that the wording and terminology used herein are for descriptive purposes and should not be considered limiting. Any headings used within this description are for convenience only and have no legal or limiting effect.

[0044] Figure 1A, Figure 1B and Figure 1C This diagram illustrates a schematic overview of some principles used in some implementations for tracking the extended state of an object. The extended state of an object includes a kinematic state indicating one or a combination of the object's center position and velocity, and an extension state indicating one or a combination of the object's size and orientation. The object's center can be one or a combination of an arbitrarily chosen point, the object's geometric center, the object's center of gravity, the center of a vehicle's rear axle, etc. A sensor 104 (e.g., automotive radar) is used to track the object (e.g., vehicle 106). In point object tracking 100, a single measurement 108 is received from vehicle 106 per scan. Point object tracking 100 only provides the kinematic state (position) of vehicle 106. Furthermore, a probabilistic filter with a measurement model having a kinematic state distribution is used to track vehicle 106. In extended object tracking (EOT) 102, multiple measurements 110 are received per scan. The multiple measurements 110 are spatially constructed around vehicle 106. EOT 102 provides both the kinematic state and the extension state of vehicle 106. A probabilistic filter with a measurement model having an extension state distribution is used to track vehicle 106.

[0045] However, as Figure 1B As shown, the distribution of real-world automotive radar measurements 112 indicates that the multiple reflections from vehicle 106 are complex. Due to this complexity, the design of the measurement model becomes complicated. Therefore, conventional measurement models are only applicable to kinematic states and not to extended states.

[0046] Therefore, in some implementations, such as contour model 114 (e.g. Figure 1C The spatial model (shown) and surface model 116 are used to capture real-world automotive radar measurements 112. However, the aforementioned spatial model is inaccurate. Some implementations are based on the understanding that real-world automotive radar measurements 112 are distributed around the edges of an object (vehicle 106) with a specific volume, which results in a surface volume model. To this end, some implementations are based on the goal of developing a surface volume model 118 that resembles and captures real-world automotive radar measurements 112. The surface volume model 118 balances the contour model 114 and surface model 116, having more realistic features while maintaining EOT accuracy.

[0047] In particular, in this implementation, a composite measurement model 120 (a surface volume model) is determined based on the principles of contour model 114 and surface model 116. The composite measurement model 120 includes multiple probability distributions 122 geometrically constrained to the object contour 124. Figure 1CIn this model, the geometric constraint is that the centers of multiple probability distributions lie on the contour. The composite measurement model has a predetermined relative geometric mapping to the center of the object. Multiple probability distributions 122 are used to cover the measurement extension along the object contour 124.

[0048] The composite measurement model 120 is composite in multiple ways. For example, the composite measurement model 120 has a composite structure, namely, multiple probability distributions 122. Additionally, the composite measurement model 120 has composite components, namely, functions of the multiple probability distributions 122, functions of the contour 124, and their relationships. Furthermore, the composite measurement model 120 has composite properties, namely, the multiple probability distributions 122 are based on measurement, thus representing a data-driven approach to model generation, while the contour 124 is based on modeling the shape of an object (e.g., the shape of a vehicle) using principles of physics-based modeling.

[0049] Furthermore, the composite measurement model 120 utilizes different principles for modeling the expansion state; that is, the composite measurement model 120 combines the principles of the contour model 114 and the surface model 116. As a result, the composite measurement model 120 better represents the physical properties of object tracking while simplifying measurement assignment. Additionally, the multiple probability distributions 122 of the composite measurement model 120 are more flexible than the single distribution of the surface model 116 and can be configured to better describe the contour 124, and also more flexibly interpret measurements from different angles or views of the object.

[0050] Some implementations are based on the understanding that, theoretically, the shape of contour 124 is unrestricted, and multiple probability distributions 122 can lie on contour 124. However, in practice, these assumptions are incorrect and useless for tracking expansion states. Instead, the contour 124 of the object is predetermined, and multiple probability distributions 122 are fitted to contour 124, while non-contour 124 is fitted to multiple probability distributions 122. This allows the physical structure of the object to be reflected during the update phase of the probability filter.

[0051] The composite measurement model 120 is learned offline (i.e., in advance). The composite measurement model 120 can be learned in a unit coordinate system or a global coordinate system. Some implementations are based on the understanding that learning the composite measurement model 120 in a unit coordinate system is advantageous because it simplifies calculations and makes the composite measurement model 120 unaware of the object's dimensions. Each of the plurality of probability distributions 122 (represented as ellipses) can be assigned a measurement probabilistically. Measurements associated with an ellipse can be referred to as elliptic assigned measurements.

[0052] Some implementations are based on the understanding that the expansion state of an object can be tracked online (i.e., in real time) using a composite measurement model 120. Specifically, various implementations use a probabilistic filter to track the confidence level of the object's expansion state, wherein the confidence level of the object's expansion state is predicted using the object's motion model and updated using the object's composite measurement model 120.

[0053] Figure 2 A block diagram is shown of a tracking system 200 for tracking the extended state of an object using a composite measurement model 120 (shown in the previous figures) according to some embodiments. The object may be a vehicle, such as (but not limited to) a car, bicycle, bus, or truck. Additionally, the vehicle may be autonomous or semi-autonomous. The extended state includes the object's kinematic state and its extension state. According to some embodiments, the kinematic state corresponds to the object's motion parameters, such as velocity, acceleration, heading, and turning rate. In some other embodiments, the kinematic state corresponds to the position of the object having its motion parameters. The tracking system 200 may include a sensor 202 or be operatively connected to a set of sensors to detect a scene using one or more signal transmissions. The one or more signal transmissions are then configured to generate one or more measurements of the object per transmission. According to some embodiments, the sensor 202 may be automotive radar. In some embodiments, the scene includes a moving object. In some other embodiments, the scene may include one or more objects, including both moving and stationary objects.

[0054] Tracking system 200 may have multiple interfaces for connecting tracking system 200 to other systems and devices. For example, network interface controller (NIC) 214 is adapted to connect tracking system 200 to network 216 via bus 212, network 216 connecting tracking system 200 to a set of sensors. Through network 216 (wirelessly or via wired), tracking system 200 receives reflection data of one or more signal transmissions for one or more measurements of each transmitted object. Additionally or alternatively, tracking system 200 includes output interface 220 configured to submit control input to controller 222.

[0055] The tracing system 200 also includes a processor 204 configured to execute stored instructions and a memory 206 storing instructions executable by the processor 204. The processor 204 may be a single-core processor, a multi-core processor, a computing cluster, or any number of other configurations. The memory 206 may include random access memory (RAM), read-only memory (ROM), flash memory, or any other suitable memory system. The processor 204 is connected to one or more input and output devices via a bus 212. Furthermore, the tracing system 200 includes a storage device 208 adapted to store different modules including instructions executable by the processor 204. The storage device 208 may be implemented using a hard disk drive, an optical disk drive, a thumb drive, a drive array, or any combination thereof.

[0056] Storage device 208 is configured to store a motion model 210a of the object and a composite measurement model 210b of the object (e.g., composite measurement model 120). Processor 204 is configured to iteratively execute a probabilistic filter to iteratively track the confidence level of the expanded state of the object predicted using the motion model 210a and updated using the composite measurement model 210b. (Refer to below) Figure 3A , Figure 3B and Figure 3C Describe in detail the confidence tracking of the object's expanded state.

[0057] Figure 3A This diagram illustrates the calculation of predicted measurements and covariance matrices according to some implementations. Given an object's expanded state 300, the covariance matrix corresponding to a previous time step, and the object's motion model, processor 204 predicts the object's expanded state 302 at the current time step and the covariance matrix corresponding to expanded state 302. The object's expanded state 300 corresponding to a previous time step is represented as x. k-1|k-1 The predicted extended state 302 of the object is represented as x. k|k-1 The extended state 300 includes various kinematic states, for example, x = [x m ,y m ,v,ψ,ω] T , where (x m ,y m ) T ψ is the center of the object, v is the vehicle's maximum speed, ψ is the orientation angle, and ω is the turning rate. In an alternative embodiment, in addition to the kinematic state, the expansion state 300 also includes the extension state, for example, x = [x m ,y m ,v,ψ,ω,l,w] TWhere l and w are the length and width of the object, respectively. Similarly, predicting the expansion state 302 includes predicting the kinematic state and / or the extension state. In one implementation, the motion model may be a coordinated turning (CT) motion model with extreme velocities. In some other implementations, a CT motion model with extreme velocities is used for the kinematic state, and a constant model with small covariance process noise is used for the extension state (i.e., length and width), since the length and width are unlikely to change over time.

[0058] The predicted extended state 302 of the object can be referred to as the prediction confidence of the extended state because the prediction is probabilistic. Some implementations are based on the understanding that the predicted extended state 302 of the object may not be accurate enough to generate a predictive measurement of the extended state, since an accurate spatial model of the vehicle radar measurement is required. Therefore, in some implementations, a composite measurement model 304 in a unit coordinate system learned offline is used. To align the composite measurement model 304 in the unit coordinate system with the predicted extended state 302, the composite measurement model 304 needs to be transformed from the unit coordinate system to the global coordinate system for the predicted extended state 302. In particular, the elliptical assignment measurement in the unit coordinate system needs to be transformed to the global coordinate system.

[0059] Some implementations are based on the understanding that this transformation can be implemented using an unscented transformation function 308. To this end, in one implementation, the processor 204 generates sigma points for the ellipse 306 (i.e., the probability distribution of the composite measurement model 304). "Ellipse" and "probability distribution" are used interchangeably and have the same meaning. Furthermore, the sigma points are propagated to the unscented transformation function 308, which is a function of the predicted expansion state 302, thus determining the predicted measurement in the global coordinate system corresponding to the elliptical assignment measurement of the ellipse 306 in the unit coordinate system. Additionally, the covariance corresponding to the predicted measurement is determined based on the predicted measurement. Similarly, the measurements in the global coordinate system corresponding to the elliptical assignment measurements associated with the remaining ellipses are determined. For this purpose, a predicted expansion state model 310 is obtained, in which the composite measurement model 304 is aligned according to the predicted expansion state 302. Furthermore, composite measurements are determined for the various probability distributions of the predicted expansion state model 310, as referred to below. Figure 3B As described.

[0060] Figure 3BA schematic diagram is shown illustrating the determination of synthetic measurements for various probability distributions of a model 310 predicting an expanded state, according to some embodiments. Processor 204 receives measurement 312 (indicated by a crosshair) at the current time step. Some embodiments are based on the understanding that multiple probability distributions 314a-314h (ellipses) can be processed independently of each other (e.g., in parallel). This independent processing allows for consideration of different perspectives on the expanded state of the probed object. To accommodate this independent processing, some embodiments treat different probability distributions among the multiple probability distributions 314a-314h as belonging to different objects. Furthermore, some embodiments are based on the understanding that soft probability assignment (i.e., probability assignment of measurement 312 to different probability distributions) is more advantageous than hard deterministic assignment. Soft probability assignment avoids the abrupt assignment of hard assignment while maintaining the correlation dimension linear with respect to the ellipse and the number of measurements.

[0061] To this end, processor 204 assigns measurement 312 to probability distribution 314 with associated probabilities. Similarly, processor 204 assigns measurement 312 to each of probability distributions 314a-314h with corresponding associated probabilities. A measurement having a corresponding associated probability associated with each of the plurality of probability distributions 314a-314h is called a “synthetic measurement”.

[0062] Furthermore, for probability distribution 314a, processor 204 determines the composite centroid 316a and the composite covariance matrix defining the extension 316b based on the composite measurement associated with probability distribution 314a. Similarly, for probability distribution 314e, processor 204 determines the composite centroid 318a and the composite covariance matrix defining the extension 318b based on the composite measurement associated with probability distribution 314e. Likewise, for probability distribution 314h, processor 204 determines the composite centroid 320a and the composite covariance matrix defining the extension 320b based on the composite measurement associated with probability distribution 314h. In this way, the composite centroid and composite covariance matrix are determined for each probability distribution. Furthermore, the confidence level of the predicted extension state is updated using the composite measurement associated with each probability distribution, as referred to below. Figure 3C As described.

[0063] Figure 3C This diagram illustrates the confidence level of the prediction for updating the expanded state 302 according to some embodiments. The processor 204 uses a probabilistic filter, such as a Kalman filter, to update the confidence level of the prediction for the expanded state 302 with synthetic measurements associated with the respective probability distributions, to generate an updated expanded state x for the object. k|k 322. The expansion state x of object updates k|k322 can be referred to as the updated confidence level of the expansion state. Furthermore, the updated confidence level of the expansion state is used to update the tracking confidence level. In the implementation, the tracking confidence level is updated based on the difference between the predicted confidence level and the updated confidence level. Additionally, the processor 204 tracks the expansion state of the object based on the updated tracking confidence level of the expansion state.

[0064] As described above, offline learning is used to develop a composite measurement model 304 for tracking the expansion state of an object. The offline learning and characteristics of the composite measurement model 304 are described below.

[0065] For example, the composite measurement model 304 includes L Gaussian components (i.e., ellipses) whose component mean lies on the profile. In an implementation, the profile can be a B-spline curve. B-spline curves are advantageous because they provide greater control flexibility for closed profiles. Furthermore, because B-spline curves satisfy the strong convex hull property, they offer finer shape control. For a given μ... l The range of the center Σ l The various ellipses can be determined using correlation probabilities. Assign N k Given a measurement assigned to an ellipse, the likelihood function is given as:

[0066]

[0067] in

[0068]

[0069]

[0070] (1) and (2) correspond to the sample mean and spread of the l-th ellipse. N represents a Gaussian distribution, and W is a Wishart distribution.

[0071] Some implementations are based on the understanding that the probability distribution of the composite measurement model 304 can be represented using a Gaussian distribution for better alignment with the probability filter. For example, in some implementations, the probability distribution is defined as a random matrix model (RMM) in the probability space (Ω, P, F), where the sample space Ω is a set of matrices. Random matrices are advantageous for representing multidimensional probability distributions, and the parameters of the probability distribution represented as an RMM can be shown using an elliptical shape. According to an implementation, using all L ellipses and given the assignment of measurements to the ellipses, L random matrix models are defined as...

[0072]

[0073] Where it is assumed that the mixing weight π l equal to π l= 1 / L.

[0074] Furthermore, assume that the center of the ellipse is located at the point of d. On the defined B-spline curve

[0075]

[0076] in Let B be the j-th control point, m+1 be the number of control points, and B be the number of control points. j,d (r) is a basis function with parameter r. By forcing μ l =c(r l ), where r l μ represents the center of the l-th ellipse l The corresponding parameters, the B-spline chain elliptical model (i.e., composite measurement model 304) are defined as follows:

[0077]

[0078] The parameters of the B-spline elliptical model (i.e., the composite measurement model 304) are the measurement quantity N of each component and the control points of the B-spline curve. and the covariance matrix of each component

[0079] Figure 4 A flowchart illustrating a method for learning parameters of a composite measurement model 304 according to some embodiments is shown. In step 400, the method includes receiving training data 400 comprising different measurements of different motions of different objects. In step 402, the method includes transforming the training data 402 to a common coordinate system.

[0080] Some implementations are based on the understanding that various statistical methods (e.g., the Expectation-Maximization (EM) method) can be used to learn the parameters of the composite measurement model 304 offline based on training data and contour knowledge of the object to be tracked. To this end, in step 404, the method includes learning the parameters of the composite measurement model 404 from the training data using statistical methods such as the EM method.

[0081] Figure 5A and Figure 5BThis diagram illustrates the transformation of training data collected from different motions of different objects to a common unit coordinate system, according to some implementations. Different measurements collected from tracking different trajectories 500 and 502 are transformed to corresponding object-centered (OC) coordinate systems 504 and 506. The transformed measurements are then aggregated 508. In some implementations, measurements are collected for the motion of similar types of objects (e.g., the motion of vehicles of similar categories). For example, for each trajectory, the implementation transforms measurements from global coordinates (GC) to object-centered (OC) coordinates from each time step and aggregates OC measurements from all trajectories of vehicles of similar size (e.g., cars).

[0082] Next, as Figure 5B As shown, the implementation transforms the aggregated OC 508 measurements to a unit coordinate (UC) system 510. In some implementations, the transformation to the UC system is performed using various normalization techniques that allow machine learning to be performed using the transformed training data. Furthermore, the measurements in the unit coordinate system 510 are used as training data for learning the parameters of the composite measurement model 304.

[0083] Figure 6 A block diagram is shown of an EM method for learning parameters of a composite measurement model 304 according to some embodiments. In some embodiments, measurements in a unit coordinate system 510 are used as training data 600, which is represented by the EM method as follows: In some other embodiments, the received aggregated OC measurement 508 is transformed to a unit coordinate system centered on the object by applying the following coordinate transformation. The origin is set and oriented so that the x-axis points face forward of the object:

[0084]

[0085] in S is the rotation matrix that functions as the orientation angle ψ, and S = diag(l, w) is the scaling matrix.

[0086] Training data 600 in the unit coordinate system and such as control points and range The initial parameter 602 is the input data for the EM method. The EM method consists of two main steps: the expectation step 604 and the maximization step 606.

[0087] The expected step 604 is to update the hidden random variable. First, the posterior association probability of each measurement is calculated as follows:

[0088]

[0089] Where μl and 4Σ l This is the mean and covariance matrix of each component. A scale factor of 4 is used to approximate a uniform distribution, and λ is the probability of outliers in the uniform distribution. Then, the remaining hidden variables... You can use (1) and (2) to update respectively.

[0090] Maximizing step 606 updates the model parameters θ={p} based on the Q function of (5). j ,Σ l}

[0091]

[0092] A B-spline curve in matrix-vector form can be rewritten as μ l =B l p, where nl = [B 0，d (rl), ...B m，d (r l )] T and in and Let these represent the control inputs in the x and y coordinates, respectively. By setting the derivative of Q(θ) (relative to θ) to 0, the control input can be given as p = H. + M, where H + yes Moore-Penrose inverse, and

[0093] and

[0094]

[0095] Furthermore, in p and Σ l The estimates are iterated until a convergence criterion 608 is reached. The convergence criterion 608 can be the predetermined likelihood in (8), the relative change of the estimated parameters on successive iterations being less than a predefined value, or a predetermined maximum number of iterations.

[0096] According to some implementations, an offline-learned composite measurement model is used to track the expansion state of an object online, i.e., to track the expansion state of the object in real time. Some implementations are based on the understanding that the probabilistic properties of the composite measurement model can be advantageously aligned with probabilistic multiple hypothesis tracking (PMHT) methods. For example, this alignment allows for the implementation of probabilistic filters using at least a variant of the Kalman filter. For instance, one implementation uses the unscented Kalman filter-probabilistic multiple hypothesis tracking (UKF-PMHT) method. The unscented Kalman filter (UKF) is used to transform the composite measurement model from a unit coordinate system to a global coordinate system. The probabilistic multiple hypothesis tracking (PMHT) method is then applied to assign the measurement at the current time step to different elliptic components probabilistically and update the expansion state of the object.

[0097] In the implementation, given an offline-learned composite measurement model and assuming the measurement x in the unit coordinate system... μ Relative to the l-th ellipse N(μ) l ,Σ l Distribution, corresponding measurement h in the global coordinate system l,k (x k|k-1 ) is defined as

[0098]

[0099] In addition to the fact that all augmentations are given by prediction states (k|k-1) with corresponding prediction distributions (e.g., Gaussian distributions), and s = diag(l k|k-1 ,w k|k-1 ) is defined in the same way as (6).

[0100] Some implementations are based on the understanding that since the transformation in (10) is nonlinear, especially with respect to the predicted orientation angle, the unscented transformation (UT) can be used to determine h. l,k (x k|k-1 mean The covariance matrix X l Therefore, with x μ Expand the predicted expansion state to Where n a =9. Then, determine 2n. a +1 weighted samples (i.e., sigma points) such that they can describe x aug true mean The covariance matrix P aug :

[0101]

[0102]

[0103]

[0104] Where κ is a proportionality parameter, such that κ+n a ≠0, and Let the i-th row of the matrix representing the square root of A be denoted by . Then, each sigma point propagates through the nonlinear function (10), i.e., B i =h l,k (A i ), and h l,k (x k|k-1 The first two moments are calculated as follows:

[0105]

[0106]

[0107] In the global coordinate system, the measurement z assigned to the l-th ellipse i It can be defined as

[0108] z i =h l，k (X k|k-1 )+n l (15)

[0109] in It corresponds to the reflection center, and n l ~N(0,R) is the measurement noise.

[0110] Measurement of a given time k And the L offline learning values ​​obtained at time step k PMHT is used to assign measurements to individual elliptic components. Unlike the general PMHT algorithm, which processes the association of measurements with objects and updates the kinematic states of multiple objects at consecutive time steps, the PMHT algorithm here is applied to process the association of measurements with ellipses and updates the kinematic and extension states of a single object at the current time step k. Specifically, PMHT uses the EM algorithm to assign soft measurements to ellipses, which in turn creates composite measurements for each component. Mathematically, the weights associated with the measurements and ellipses are... Synthetic measurement and the corresponding composite covariance matrix C zz The following derivation

[0111]

[0112]

[0113]

[0114] In addition, the expansion state and C are calculated and measured during the UT process (10). xz The covariance between them, and the filter gain is calculated as Update the expanded state x based on the l-th measurement formula in (15) k,l The covariance matrix C l,n PMHT iterates between the desired step and the maximization step until a predefined maximum number of iterations N is reached. iter In each iteration n, the expansion state x is updated incrementally through each component (i.e., on l) in the order of (10) and (16)-(18). k,l The covariance matrix C l,n The overall UKF-PMHT tracking algorithm is described below.

[0115] Figure 7A A flowchart of a UKF-PMHT tracking algorithm according to some embodiments is shown. The UKF-PMHT tracking algorithm is executed by processor 204. The UKF-PMHT tracking algorithm includes two phases: a prediction phase 700 and an update phase 704. In the prediction phase 700, a motion model is used to predict the confidence level and corresponding covariance matrix of the object's expansion state. Furthermore, in block 702, by setting the iteration index n = 1, and x... 1,1 =x k|k-1 and C 1,1 =C k|k-1 Initiate the 704 iteration of the update phase.

[0116] In the update phase 704, box 706, the predicted measurement and covariance matrix are calculated based on the offline-learned measurement model 712. Figure 7B A block diagram is shown illustrating the steps performed according to some implementations to calculate the predictive measurement and covariance matrix. In block 720, sigma points are generated for the first ellipse (i.e., l = 1) of the composite measurement model 712 and the most recently updated expansion state using equations (11) and (12). In box 722, the sigma point is propagated through the affine transformation given by (10). In box 724, the predictive measurement and covariance matrix are calculated for the first ellipse according to equations (13) and (14), respectively.

[0117] Return to reference Figure 7A In box 708, given the measurement at time k 714, calculate the composite measurement and composite covariance matrix. Figure 7CA block diagram is shown illustrating the steps performed according to some embodiments to calculate the synthetic measurement and the synthetic covariance matrix. In block 726, for the first ellipse, the measurement-ellipse association weights are calculated according to equation (16). In block 728, for the first ellipse, the synthetic measurement and the synthetic covariance matrix are calculated using equations (17) and (18), respectively.

[0118] Return to reference Figure 7A In box 710, update the expansion state x. l,n The covariance matrix C l,n Here, the subscript l is the elliptic index. Figure 7D This illustrates, according to some embodiments, the process of updating the expansion state x for the first ellipse (i.e., l = 1). l,n The covariance matrix C l,n The flowchart illustrates the steps. In box 730, the cross-covariance matrix is ​​calculated as follows.

[0119]

[0120] In box 732, calculate the Kalman filter gain. The Kalman filter gain is given by the following formula.

[0121]

[0122] In box 734, the expansion state is x l,n The covariance matrix C l,n Updated to

[0123]

[0124]

[0125] Furthermore, the same function given in blocks 706, 708, and 710 is performed for the second ellipse (i.e., l = 2) of the composite measurement model 712. For the second ellipse, the updated expansion state x... l,n The covariance matrix C l,n Used to calculate the predictive measurement and covariance matrix. In other words, the predictive measurement and covariance matrix are calculated using the latest updated expansion state and covariance matrix. Similarly, the same functions given in boxes 706, 708, and 710 are executed for the remaining ellipses of the composite measurement model 712. To this end, multiple internal iterations l = 1…L are performed to complete the iteration of update phase 704, where L is the number of ellipses in the composite measurement model 712. The iteration of update phase 704 (n = 1) completes the execution of the expansion state of the object at time k. Furthermore, in the next iteration (i.e., n = 2), update phase 704 is executed to produce the updated expansion state. Update phase 704 is executed iteratively until convergence criterion 716 is reached. Convergence criterion 710 can be a predetermined maximum number of iterations N.iter Once the convergence criterion 716 is reached, output the updated expansion state 718. and corresponding covariance matrix

[0126] Figure 8A A schematic diagram of a vehicle 800 is shown, which includes a controller 802 communicating with a tracking system 200 employing principles of some embodiments. The vehicle 800 can be any type of wheeled vehicle, such as a bus, coach, or rovers. Additionally, the vehicle 800 can be an autonomous or semi-autonomous vehicle. For example, some embodiments control the movement of the vehicle 800. Examples of movement include lateral movement of the vehicle controlled by a steering system 804 of the vehicle 800. In one embodiment, the steering system 804 is controlled by the controller 802. Alternatively or additionally, the steering system 804 may be controlled by the driver of the vehicle 800.

[0127] In some embodiments, the vehicle may include an engine 810, which may be controlled by a controller 802 or by other components of the vehicle 800. In some embodiments, the vehicle may include an electric motor instead of an engine 810, which may be controlled by a controller 802 or by other components of the vehicle 800. The vehicle may also include one or more sensors 806 to sense the surrounding environment. Examples of sensors 806 include rangefinders such as radar. In some embodiments, the vehicle 800 includes one or more sensors 808 to sense its current motion parameters and internal state. Examples of one or more sensors 808 include a Global Positioning System (GPS), an accelerometer, an inertial measurement unit, a gyroscope, an axis rotation sensor, a torque sensor, a deflection sensor, a pressure sensor, and a flow sensor. The sensors provide information to the controller 802. The vehicle may be equipped with a transceiver 812, which enables the controller 802 to communicate with the tracking system 200 of some embodiments via wired or wireless communication channels. For example, the controller 802 receives control input from the tracking system 200 via the transceiver 812.

[0128] Figure 8BA schematic diagram illustrating the interaction between controller 814 and controller 802 of a vehicle 800 according to some embodiments is shown. For example, in some embodiments, controller 814 of the vehicle 800 is a steering controller 816 and a brake / throttle controller 818 that control the rotation and acceleration of the vehicle 800. In this case, controller 802 outputs control commands to controllers 816 and 818 based on control inputs to control the kinematic state of the vehicle. In some embodiments, controller 814 also includes a higher-level controller, such as a lane-keeping assist controller 820 that further processes the control commands from controller 802. In both cases, controller 814 utilizes the outputs (i.e., control commands) of controller 802 to control at least one actuator of the vehicle (e.g., the vehicle's steering wheel and / or brakes) to control the movement of the vehicle.

[0129] Figure 8C A schematic diagram is shown of an autonomous or semi-autonomous controlled vehicle 822 generating control inputs using some implementations. The controlled vehicle 822 may be equipped with a tracking system 200. In some implementations, the controlled vehicle 822 tracks the expansion state of various obstacles 824, and subsequently generates control inputs based on the tracked expansion state of the obstacles. In some implementations, the control inputs include commands specifying one or a combination of values ​​for the steering angle and rotational speed of the vehicle's wheels, and measurements include one or a combination of values ​​for the vehicle's rotational speed and acceleration.

[0130] The generated control inputs are designed to keep the controlled vehicle 822 within the specific boundaries of the road 826 and to avoid other uncontrolled vehicles (i.e., obstacles 824 for the controlled vehicle 822). For example, based on the control inputs, the autonomous or semi-autonomous controlled vehicle 822 may, for example, overtake another vehicle on the left or right, or instead remain behind another vehicle in the current lane of the road 826.

[0131] The following description provides exemplary embodiments only and is not intended to limit the scope, applicability, or configuration of this disclosure. Rather, the following description of exemplary embodiments will provide those skilled in the art with a feasible description for implementing one or more exemplary embodiments. Various changes to the function and arrangement of the elements will be conceived without departing from the spirit and scope of the subject matter set forth in the appended claims.

[0132] Specific details are set forth in the following description to provide a thorough understanding of the embodiments. However, it will be understood by those skilled in the art that embodiments may be practiced without these specific details. For example, systems, processes, and other elements in the disclosed subject matter may be shown as components in block diagram form to avoid obscuring the embodiments with unnecessary detail. In other instances, well-known processes, structures, and techniques may be shown without unnecessary detail to avoid obscuring the embodiments. Furthermore, similar reference numerals and designations in the various figures indicate similar elements.

[0133] Furthermore, the various embodiments can be described as processes, depicted as flowcharts, data flow diagrams, structural diagrams, or block diagrams. Although flowcharts can describe operations as a sequential process, many operations can be performed in parallel or simultaneously. Additionally, the order of operations can be rearranged. A process may terminate upon completion of its operations, but may have additional steps not discussed or included in the figures. Moreover, not all operations in any specifically described process may appear in all embodiments. A process may correspond to a method, function, program, subroutine, subroutines, etc. When a process corresponds to a function, the termination of the function may correspond to the function returning to the calling function or the main function.

[0134] Furthermore, implementations of the disclosed subject matter can be carried out, at least partially, manually or automatically. They can be performed, or at least assisted by, using machines, hardware, software, firmware, middleware, microcode, hardware description languages, or any combination thereof, for manual or automatic implementation. When implemented in software, firmware, middleware, or microcode, program code or code segments that perform the required tasks can be stored in a machine-readable medium. The processor can then execute the required tasks.

[0135] Furthermore, the various methods or processes outlined herein can be encoded as software that can be executed on one or more processors employing any of a variety of operating systems or platforms. Additionally, such software can be written using any of a variety of suitable programming languages ​​and / or programming or scripting tools, and can also be compiled into executable machine language code or intermediate code that executes on a framework or virtual machine. Typically, in various implementations, the functionality of program modules can be combined or distributed as needed.

[0136] Implementations of this disclosure can be specifically embodied as a method, examples of which have been provided. Actions performed as part of this method can be ordered in any suitable manner. Therefore, implementations can be constructed that perform actions in a different order than those shown, which may include performing some actions simultaneously, although they are shown as sequential actions in the illustrative embodiments.

[0137] Although this disclosure has been described with reference to specific preferred embodiments, it will be understood that various other adjustments and modifications may be made within the spirit and scope of this disclosure. Therefore, aspects of the appended claims cover all such variations and modifications that fall within the true spirit and scope of this disclosure.

Claims

1. A tracking system (200) for tracking the expansion state of an object, said expansion state comprising a kinematic state indicating one or a combination of the position and velocity of the center of the object and an extension state indicating one or a combination of the size and orientation of the object, said tracking system comprising: At least one processor (204); as well as A memory (206) stores instructions that, when executed by the at least one processor (204), cause the tracking system (200) to: Receive measurements associated with at least one sensor (202), wherein the at least one sensor (202) is configured to detect a scene including the object via one or more signal transmissions, the one or more signal transmissions being configured to generate one or more measurements of the object per transmission; A probabilistic filter is executed to iteratively track the confidence of the expanded state of the object, wherein the confidence is predicted using a motion model (210a) of the object and updated using a composite measurement model (210b) of the object, wherein the composite measurement model (210b) includes multiple probability distributions constrained to lie on the contour of the object and has a predetermined relative geometric mapping to the center of the object, wherein in each iteration of the iterative tracking, the confidence of the expanded state is updated based on the difference between the predicted confidence and the updated confidence, wherein the updated confidence is estimated based on the probability of the measurement fitted to each of the multiple probability distributions and mapped to the expanded state of the object based on the corresponding geometric mapping; and The expansion state of the object is tracked based on the updated confidence level of the expansion state.

2. The tracking system according to claim 1, wherein, The processor is also configured to transform the composite measurement model from the unit coordinate system to the global coordinate system based on the confidence level of the prediction for the expanded state or the iteratively updated confidence level using an unscented transformation function.

3. The tracking system according to claim 1, wherein, The processor is also configured to assign the plurality of measurements to the different probability distributions by treating the different probability distributions among the plurality of probability distributions as belonging to different objects.

4. The tracking system according to claim 3, wherein, The processor is also configured to use probabilistic multiple hypothesis tracking (PMHT) to perform the assignment of the plurality of measurements to the different probability distributions that are considered as the different objects, thereby assigning the plurality of measurements to the different probability distributions.

5. The tracking system according to claim 1, wherein, The composite measurement model is learned based on the expectation-maximization (EM) method.

6. The tracking system according to claim 1, wherein, One or more parameters of each probability distribution are represented by a stochastic matrix model (RMM) in a two-dimensional (2D) probability space.

7. The tracking system according to claim 1, wherein, The outline of the object corresponds to a B-spline curve.

8. The tracking system according to claim 1, wherein, The confidence level of the predicted vehicle state is used to align the composite measurement model with the measurement using an unscented transformation.

9. The tracking system according to claim 1, in, The processor is configured to determine the control input of the vehicle controller based on the extended state of the object's tracking, and to control the vehicle according to the control input; and The vehicle is operationally connected to the tracking system according to claim 1.

10. A tracking method for tracking the expanded state of an object, wherein, The expansion state includes a kinematic state indicating one or a combination of the position and velocity of the center of the object and an extension state indicating one or a combination of the size and orientation of the object, and the method includes the following steps: Receive measurements associated with at least one sensor, wherein the at least one sensor is configured to detect a scene including the object using one or more signal transmissions, to generate one or more measurements of the object per transmission; A probabilistic filter is executed to iteratively track the confidence of the expanded state of the object, wherein the confidence is predicted using a motion model of the object and updated using a composite measurement model of the object, wherein the composite measurement model includes multiple probability distributions constrained to lie on the contour of the object and has a predetermined relative geometric mapping to the center of the object, wherein in each iteration of the iterative tracking, the confidence of the expanded state is updated based on the difference between the predicted confidence and the updated confidence, wherein the updated confidence is estimated based on the probability of the measurement fitted to each of the multiple probability distributions and mapped to the expanded state of the object based on the corresponding geometric mapping; and The expansion state of the object is tracked based on the updated confidence level of the expansion state.

11. The tracking method according to claim 10, wherein, The tracking method further includes the following steps: transforming the composite measurement model from the unit coordinate system to the global coordinate system based on the confidence level of the prediction for the expanded state or the iteratively updated confidence level using an unscented transformation function.

12. The tracking method according to claim 10, wherein, The tracking method further includes the following steps: assigning the multiple measurements independently to the different probability distributions by treating the different probability distributions among the multiple probability distributions as belonging to different objects.

13. The tracking method according to claim 12, wherein, The tracking method further includes the following steps: using probabilistic multiple hypothesis tracking (PMHT) to assign the plurality of measurements to the different probability distributions that are considered as different objects, thereby assigning the plurality of measurements to the different probability distributions.

14. The tracking method according to claim 10, wherein, The composite measurement model is learned based on the expectation-maximization (EM) method.

15. A non-transitory computer-readable storage medium having a program implemented thereon, the program being executable by a processor to perform a method for tracking the extended state of an object, wherein, The expansion state includes a kinematic state indicating one or a combination of the position and velocity of the center of the object and an extension state indicating one or a combination of the size and orientation of the object, and the method includes the following steps: Receive measurements associated with at least one sensor, wherein the at least one sensor is configured to detect a scene including the object using one or more signal transmissions, to generate one or more measurements of the object per transmission; A probabilistic filter is executed to iteratively track the confidence of the expanded state of the object, wherein the confidence is predicted using a motion model of the object and updated using a composite measurement model of the object, wherein the composite measurement model includes multiple probability distributions constrained to lie on the contour of the object and has a predetermined relative geometric mapping to the center of the object, wherein in each iteration of the iterative tracking, the confidence of the expanded state is updated based on the difference between the predicted confidence and the updated confidence, wherein the updated confidence is estimated based on the probability of the measurement fitted to each of the multiple probability distributions and mapped to the expanded state of the object based on the corresponding geometric mapping; and The expansion state of the object is tracked based on the updated confidence level of the expansion state.

Images