Method and system for occupancy status detection

Abstract

The present disclosure relates to methods and systems for occupancy state detection. A computer-implemented method for occupancy state detection in an area at a predetermined point in time, the following steps are performed by computer hardware components: determining a probability distribution over a list of possible occupancy states of the area at a preceding point in time; determining measurement data related to the area at a predetermined point in time; and determining a probability distribution over a list of possible occupancy states of the area at the predetermined point in time based on the measurement data and the probability distribution over the list of possible occupancy states of the area at the preceding point in time.

Description

Technical Field

[0001] This disclosure relates to methods and systems for occupancy status detection. Background Technology

[0002] Various sensors, such as radar or LiDAR sensors, are used in automotive applications to monitor the vehicle's environment. Within a vehicle's environment, an area may be occupied by various objects, such as other vehicles, pedestrians, obstacles, etc., which can affect the safety and reliability of driver assistance systems. Therefore, it may be desirable to know whether an area is occupied, and if so, which occupancy state is valid for that area. However, reliable occupancy detection suffers from complexity and high computational workload.

[0003] Therefore, there is a need to provide effective methods and systems for occupancy status detection. Summary of the Invention

[0004] This disclosure provides methods, computer systems, vehicles, and non-transitory computer-readable media implemented by a computer. Embodiments are shown in the specification and accompanying drawings.

[0005] In one aspect, this disclosure relates to a computer-implemented method for detecting occupancy status in a region at a predetermined time point. The method includes the steps performed by computer hardware components: determining a probability distribution on a list of possible occupancy statuses of the region at a previous time point; determining measurement data associated with the region at the predetermined time point; and determining a probability distribution on the list of possible occupancy statuses of the region at the predetermined time point based on the measurement data and the probability distribution on the list of possible occupancy statuses of the region at the previous time point. The list of possible occupancy statuses includes static occupancy status, dynamic occupancy status, free space status, a first uncertain state between static and dynamic occupancy statuses, a second uncertain state between static and free space statuses, a third uncertain state between dynamic and free space statuses, and an unknown occupancy status.

[0006] In other words, the probability distribution of the region at previous time points is determined from a list of defined possible occupancy states (i.e., states describing the presence or absence of objects in the region). Previous time points can be immediately preceding a predetermined time point (in other words, there are no other time points between the previous time point and the predetermined time point). It should be understood that discrete time point sequences can be used, such as equidistant time points, time points every predetermined amount of seconds, such as per second, or every 1 / 10 of a second (i.e., 100 ms), etc. The predetermined time point can be the current time point or any time point. The predetermined time point can directly follow a previous time point (in other words, there are no other time points between the predetermined time point and the previous time point). Furthermore, measurement data of the region at the predetermined time point can be measured by a sensor. The sensor can be or may include a radar sensor and / or a LiDAR sensor. The probability distribution of the region at the predetermined time point can be determined from a list of defined possible occupancy states, wherein the probability distribution of the region at the predetermined time point can depend on the probability distribution of the region at previous time points and the measurement data of the region at the predetermined time point.

[0007] The area can be the environment of a robot or vehicle. The area can be variable in size, or it can be a specific area of ​​interest (or section), such as in front of the robot or vehicle.

[0008] Occupancy detection within a region can be understood as a combination of object detection and estimation of the possible states of the object. In other words, if an object is detected in a region, the region can be defined as occupied. Otherwise, if the region is not occupied, it can be defined as free space. If a region is occupied, the object in that region can have one possible state, namely an occupancy state. An occupancy state can be one of the defined possible occupancy states from the list of possible occupancy states described herein. A static occupancy state can describe a state in which the occupancy state can be in a stable or quasi-stable state, i.e., the occupancy state can remain constant relative to time (in other words: an object with a static occupancy state does not move relative to time). A dynamic occupancy state can describe a state in which the occupancy state can have behavior that changes relative to time (in other words: an object with a dynamic occupancy state can move relative to time). Therefore, the difference between static and dynamic occupancy states can depend on the behavior of each occupancy state relative to time. For each uncertain state described in the list of possible occupancy states, it may be unclear which of the two possible occupancy states the object has. For an unknown occupancy state, it may be unknown whether the region is occupied or free space.

[0009] A probability distribution can provide the probabilities of each state. The probabilities over all probabilities can sum to 1 (i.e., 100%), or the probabilities can sum to any different value such that the corresponding probabilities do not provide the probability itself, but the relative values ​​of the corresponding probabilities of states provide information about which state is more likely than another.

[0010] According to the implementation method, the probability distribution on the list of possible occupancy states of the area at a previous time point includes a predetermined initial state distribution on the list of possible occupancy states of the area.

[0011] According to the implementation method, the initial state distribution on the list of possible occupancy states of a region includes an equal distribution on the list of possible occupancy states of a region.

[0012] The equal distribution of the list of possible occupied states in this region means that each possible state has the same probability in the initial state.

[0013] According to the implementation, the probability distribution on the list of possible occupancy states of the area at the previous time point is determined based on the probability distribution on the list of possible occupancy states of the area at the previous time point and on the measurement data related to the area at the previous time point and at another time point before the previous time point.

[0014] In other words, the probability distribution of the region at a previous time point is determined on a defined list of possible occupancy states, where the probability distribution of the region at a previous time point depends on the probability distribution of the region at another previous time point and the measurement data of the region at that previous time point.

[0015] The other point in time can immediately precede the previous point in time (in other words, there is no other point in time between the other point in time and the previous point in time). The previous point in time can directly follow the other point in time (where there is no other point in time between the other point in time and the previous point in time).

[0016] There can be multiple other time points before another time point. Then, determining the probability distribution of the region at other time points can be performed in a manner similar to determining the probability distribution at a predetermined time point or determining the probability distribution at previous time points.

[0017] According to the implementation method, the probability distribution on the list of possible occupancy states of the area at a predetermined time point is also determined based on a transition matrix that includes multiple state transition probabilities between possible occupancy states.

[0018] The transition matrix can be the transition matrix of a Markov model, and methods according to various implementations can be performed based on a Markov model.

[0019] The transition matrix can indicate the probability of state transitions under the assumption that sensor data is unavailable.

[0020] Multiple state transition probabilities can be the corresponding state transition probabilities from each of the first occupied states of possible occupied states to the second occupied state of possible occupied states. The first occupied state can be the same as the second occupied state, or the first occupied state can be different from the second occupied state.

[0021] According to the implementation, the following state transition probabilities are different from zero: the state transition probability from the static occupancy state to the static occupancy state; the state transition probability from the static occupancy state to the first uncertain state between the static occupancy state and the dynamic occupancy state; the state transition probability from the static occupancy state to the second uncertain state between the static occupancy state and the free space state; the state transition probability from the dynamic occupancy state to the dynamic occupancy state; the state transition probability from the dynamic occupancy state to the first uncertain state between the static occupancy state and the dynamic occupancy state; the state transition probability from the dynamic occupancy state to the third uncertain state between the dynamic occupancy state and the free space state; the state transition probability from the free space state to the free space state; the state transition probability from the free space state to the second uncertain state between the static occupancy state and the free space state; and the state transition probability from the free space state to the third uncertain state between the dynamic occupancy state and the free space state. The state transition probability from the first uncertain state between the static occupancy state and the dynamic occupancy state to the first uncertain state between the static occupancy state and the dynamic occupancy state; the state transition probability from the first uncertain state between the static occupancy state and the dynamic occupancy state to the unknown occupancy state; the state transition probability from the second uncertain state between the static occupancy state and the free space state to the second uncertain state between the static occupancy state and the free space state; the state transition probability from the second uncertain state between the static occupancy state and the free space state to the unknown occupancy state; the state transition probability from the third uncertain state between the dynamic occupancy state and the free space state to the third uncertain state between the dynamic occupancy state and the free space state; the state transition probability from the unknown occupancy state to the unknown occupancy state (which can be equal to 1).

[0022] All other possible state transition probabilities can be zero, i.e., the state transition probability from a static occupied state to a dynamic occupied state, to a free space state, to a third uncertain state between a dynamic occupied state and a free space state, or to an unknown occupied state; the state transition probability from a dynamic occupied state to a static occupied state, to a free space state, to a second uncertain state between a static occupied state and a free space state, or to an unknown occupied state; the state transition probability from a free space state to a static occupied state, to a dynamic occupied state, to a first uncertain state between a static occupied state and a dynamic occupied state, or to an unknown occupied state; the state transition probability from a first uncertain state between a static occupied state and a dynamic occupied state to a static occupied state, to a dynamic occupied state, to a free space state, to a second uncertain state between a static occupied state and a free space state, or to a third uncertain state between a dynamic occupied state and a free space state; the state transition probability from a static occupied state to a free space state... The state transition probability from the second uncertain state to the static occupied state, or to the dynamic occupied state, or to the free space state, or to the first uncertain state between the static and dynamic occupied states, or to the third uncertain state between the dynamic and free space states; the state transition probability from the third uncertain state between the dynamic and free space states to the static occupied state, or to the dynamic occupied state, or to the free space state, or to the first uncertain state between the static and dynamic occupied states, or to the second uncertain state between the static and free space states; the transition from the unknown occupied state to the static occupied state, or to the dynamic occupied state, or to the free space state, or to the first uncertain state between the static and dynamic occupied states, or to the second uncertain state between the static and free space states, or to the third uncertain state between the dynamic and free space states.

[0023] In other words: There may be no state transition probability between a statically occupied state and a dynamically occupied state, or between a free-space state and an unknown occupied state. There may be no state transition probability between a statically occupied state, a dynamically occupied state, or a free-space state and an uncertain state, where the uncertain state is not between the corresponding statically occupied state, the corresponding dynamically occupied state, or the corresponding free-space state and a statically occupied state, the corresponding dynamically occupied state, or a free-space state. There may be no state transition probability from an uncertain state back to a statically occupied state, a dynamically occupied state, or a free-space state. Apart from the unknown occupied state itself, there may be no state transition probability from an unknown occupied state to any other possible state.

[0024] According to the implementation method, the transition matrix depends on the estimated velocity of objects in the region.

[0025] The estimated velocity of objects in a region can also be called the estimated cell velocity.

[0026] According to the implementation method, the probability distribution on the list of possible occupancy states of a region at a predetermined time point is also determined based on an observation matrix that includes the observed emission probability.

[0027] The observation matrix can be the observation matrix of a Markov model, and methods according to various implementations can be executed based on the Markov model.

[0028] The observation matrix can incorporate sensor data into this method.

[0029] The observed emission probability can be the corresponding observed emission probability of each of the probability distributions from the list of possible occupancy states of the region at a predetermined time point to the list of possible observed states. The list of possible observed states can depend on the sensor used for observation.

[0030] According to the implementation method, the area includes cells occupying a grid, which includes multiple additional cells.

[0031] In other words, the area can be represented by cells, where multiple cells can represent occupied grids. Each additional cell represents a corresponding additional area. The method described herein can be applied to each of these additional cells (in other words, for each additional area). The cells and each additional cell can be in one of the possible states, and the states can differ for different cells.

[0032] According to the implementation method, the measurement data is determined based on LIDAR sensors, and the measurement data includes uncertainty information or free space information between static occupancy and dynamic occupancy.

[0033] LiDAR sensors can measure the range or distance between the sensor and a vehicle or object. Measurement data from a LiDAR sensor can also include the azimuth and elevation angles of the vehicle or object relative to the sensor. The measurement data recorded from a LiDAR sensor can be very detailed and can include fine and accurate information about objects at a distance. Ambient lighting may not affect the quality of the information captured by a LiDAR, so daytime and nighttime results may not suffer any performance loss due to interference from factors such as shadows, sunlight, or headlight glare.

[0034] According to the implementation method, the measurement data is determined based on radar sensors, and the measurement data includes static occupancy information, dynamic occupancy information, uncertainty information between static occupancy and dynamic occupancy, or free space information.

[0035] Radar sensors are unaffected by adverse or severe weather conditions, operating reliably in dark, wet, or even foggy weather. Radar sensors can identify the distance, direction, and relative speed of a vehicle or object. Direction can be determined by azimuth and elevation angles, which can be measured by the radar sensor.

[0036] According to the implementation, the observation matrix depends on at least one of the detection range rate from the radar sensor and the distance from the radar sensor.

[0037] Approach velocity can be a rate, i.e., relative velocity. Approach velocity can describe the rate at which a vehicle or object moves toward or away from a radar sensor.

[0038] In another aspect, this disclosure is directed to a computer system comprising multiple computer hardware components configured to perform multiple or all of the steps of the computer-implemented methods described herein.

[0039] A computer system may include multiple computer hardware components (e.g., a processor, such as a processing unit or processing network, at least one memory, such as a memory cell or memory network, and at least one non-transitory data storage device). It should be understood that additional computer hardware components may be provided and used to perform the steps of the computer-implemented methods within the computer system. The non-transitory data storage and / or memory cell may include computer programs that instruct the computer, for example, to use the processing unit and at least one memory cell to perform some or all of the steps or aspects of the computer-implemented methods described herein.

[0040] In another aspect, the present invention relates to a vehicle comprising the computer system and sensors described herein, wherein the measurement data is determined based on the output of the sensors. The sensors may be radar systems and / or LIDAR systems.

[0041] The vehicle can be a car, truck, or motorcycle, and the sensors can be mounted on the vehicle. The sensors can be pointed at an area in front of, behind, or to the side of the vehicle. As the vehicle moves, the sensors can capture measurement data.

[0042] On the other hand, this disclosure relates to a non-transitory computer-readable medium comprising instructions for performing several or all of the steps or aspects of the computer-implemented methods described herein. The computer-readable medium may be configured as: an optical medium, such as an optical disc (CD) or digital versatile disc (DVD); a magnetic medium, such as a hard disk drive (HDD); a solid-state drive (SSD); a read-only memory (ROM), such as flash memory; etc. Furthermore, the computer-readable medium may be configured as data storage accessible via a data connection such as an Internet connection. The computer-readable medium may, for example, be an online database or cloud storage.

[0043] This disclosure also relates to a computer program for instructing a computer to perform some or all of the steps or aspects of the computer-implemented methods described herein.

[0044] Using the methods and systems described in this paper, Hidden Markov Models can be used as a means of distinguishing between dynamic and static occupancy within the occupancy grid approach framework. For example, by defining the possible states and transitions of each cell and modeling the sensor data observation matrix, individual cells of the occupancy grid can be processed separately. Attached Figure Description

[0045] This document describes exemplary embodiments and functions of the present disclosure in conjunction with the following schematically illustrated figures:

[0046] Figure 1 It is a graphical representation of a Markov chain;

[0047] Figure 2 It is a visualization of the calculation of a single probability distribution;

[0048] Figure 3 The state transition diagram is shown;

[0049] Figure 4 This is a flowchart illustrating a method for occupancy status detection according to various embodiments; and

[0050] Figure 5 It is a computer system having multiple computer hardware components, configured to perform steps of computer-implemented methods for object detection according to various embodiments. Detailed Implementation

[0051] Object occupancy detection can be applied to various technical fields. For example, object detection in path planning and collision avoidance in robotics applications or in driver assistance systems in the automotive industry can utilize occupancy detection-based methods and systems.

[0052] The environment of a robot or vehicle may include multiple different objects in different states, such as a static state for buildings and trees, a dynamic state for moving vehicles and pedestrians, or even an unknown state if the behavior of the object cannot be identified or measured. To detect the different states of objects, the environment of the robot, vehicle, etc., can be divided into uniformly spaced regions of multiple cells using a grid. Each cell of the grid can be represented by a binary random variable indicating the presence of an obstacle, object, part of an obstacle, or part of an object at that location in the environment. For each of these cells or regions, an occupancy state can be estimated, specifying whether the cell is occupied. This method can be called occupancy grid mapping. The occupancy grid method can compute approximate posterior estimates of these random variables.

[0053] In other words, occupancy mesh mapping refers to a set of methods in probabilistic robotics, for example, for mobile robots or vehicles, that solve the problem of generating a graph from noisy and uncertain sensor measurement data under the assumption that the robot's pose or the vehicle's position is known.

[0054] To reliably detect occupancy states in dynamic environments, robust models can be used to obtain accurate occupancy state estimates. For example, methods based on Markov chain models can be used. Markov chain models can predict that the probability of switching from the current state will not change based on previous states. In probability theory, a Markov model can be a stochastic model used to simulate pseudo-randomly varying systems. It can be assumed that future states depend only on the current state and not on events that occurred before it (i.e., the assumption has Markov properties). This assumption allows for reasoning and computation using models that would otherwise be difficult to handle. Therefore, in the fields of predictive modeling and probabilistic prediction, it may be desirable for a given model to exhibit Markov properties.

[0055] An enhancement to the Markov chain method can be a Hidden Markov Model (HMM). An HMM can be a statistical Markov model, where it can be assumed that the modeled system is a Markov process X with unobservable (“hidden”) occupied states. An HMM can assume the possible existence of another process, observation Y, whose behavior (probabilistically) depends on process X. The main idea of ​​an HMM method can be to extract information about process X by observing process Y.

[0056] Hidden Markov Models (HMMs) can be constructed using occupancy grid diagrams created for observation at different discrete time steps (t = 0, 1, 2, ...). At each discrete time step t, the system can be in a certain internal (“hidden”) occupancy state h. t And it can be based solely on h t To emit measurements (which can also be called observations) mt The system can transition from time step t to time step t+1 to a new occupied state h. t+1 Furthermore, this process can be repeated. This can be called a Markov chain, and... Figure 1 The diagram is shown in the image.

[0057] Figure 1 A graphical representation of Markov chain 100 is shown. Figure 1 Divided into two parts by a thick line, the area above the thick line represents the actual state (also known as the hidden layer), and the area below the thick line represents the measurement (also known as the observation). The actual state includes multiple occupied states h1 102, h2 104, h3 106, and h4 108 at a discrete time step t. These occupied states are connected in such a way that the occupied state at a given time point depends on the occupied state at a previous time point; that is, the time point of the previous time point is exactly before this time point. The connections between occupied states are indicated by arrows q. 12 126, q 23 128, q 34 130 and q 45 132 indicates that the arrow q 12 126, q 23 128, q 34 130 and q 45 132 represents the occupancy state h from the previous time point t-1. t-1 Occupancy status h up to the current time point t The transfer. For example, arrow q 34 Connection 130 is in occupancy status h3 106 and occupancy status h4 108. Furthermore, Figure 1 The various occupancy states h1 102, h2 104, h3 106, and h4 108 at the top are connected by arrows. Figure 1 The corresponding measurements at the bottom are m1 110, m2 112, m3 114, and m4 116. Arrows b1 118, b2 120, b3 122, and b4 124 indicate if an occupancy state h exists. t Then the measurement m occurs t .

[0058] To apply the Hidden Markov Model (HMM) framework, some variables and operators can be predefined. It can be assumed that the list of possible occupancy states Ω is finite. The list of occupancy states Ω can include {S, D, F, SD, SF, DF, SDF}, where an occupancy state can be defined as a static occupancy state S, a dynamic occupancy state D, a free-space state F, a first uncertain state SD between static and dynamic occupancy states S and D, a second uncertain state SF between static and free-space states S and F, a third uncertain state DF between dynamic and free-space states D and F, and an unknown occupancy state SDF.

[0059] The initial state distribution p(Ω1=ω) on the list of possible occupancy states of the region can be predetermined.

[0060] State transition probability p ij (Ω t+1 =ω'|Ω t =ω), ω'∈Ω can be defined as indicating the probability that the occupied state ω' will occur if a previous occupied state ω exists. These are labeled Figure 2 The rightward arc in the middle.

[0061] We can assume that a set of observations O is finite. The probability of observation, b(O) t =o|Ω t =ω), o∈O can indicate the probability of observing o if an occupied state ω exists. These annotations... Figure 2 The downward arc in the middle.

[0062] If Ω and O are finite, then the initial state distribution p can be represented as a vector p of size |Ω|.

[0063] exist Figure 2 In the middle, the state transition probability p ij It can indicate if the state ω is occupied. i If it appeared at a previous time point t-1, then the occupancy state ω at the current time point t is... j The possibility of it occurring.

[0064] The transition probabilities form a transition matrix P of size |Ω|×|Ω|, where each row i is a polynomial of the next state given the current state i. Similarly, the observation emission probabilities form an observation matrix Q of size |Ω|×|O|, where each row i is a polynomial distribution over the observation given the occupancy state.

[0065] b0, P, and Q can be used together to form an HMM model.

[0066] The probability of the j-th occupied state is α t The forward computation of (j) can be performed in each step using the following formulas:

[0067]

[0068] Figure 2 A visualization of the calculation of a single probability distribution 200 is shown. This calculation can be performed by a computer-implemented method using the occupancy status detection of a region at a predetermined time point t. The computer hardware components perform the following steps: determining the probability distribution α on a list of possible occupancy statuses of the region at a previous point t-1. t-1 (1) 204, α t-1 (2) 206, α t-1 (3) 208 to α t-1 (N)210; Determine the measurement data (in other words: observation) related to the region at a predetermined time point t. t 212; and based on measurement data o t The probability distribution α on the list of possible occupancy states of the region at time 212 and previous time points. t-1 (1) 204, α t-1 (2) 206, α t-1 (3) 208 to α t-1 (N)210, determine the probability distribution α on the list of possible occupancy states of the region at a predetermined time point t. t (j)202;

[0069] In other words: the j-th probability distribution α at the predetermined time point t t (j)202 can be obtained by considering all previous probability distributions α at the previous time point t-1. t-1 (1) 204, α t-1 (2) 206, α t-1 (3) 208 to α t-1 (N)210 uses the state transition probability p 1j 214, p 2j 216, p 3j 218 and p Nj 220 weighted and further multiplied by the observed emission probability b j (o t Add 222 together to calculate, where o t 212 is the observation at the predetermined time point t.

[0070] The area can be represented by cells, where multiple cells can represent an occupied grid. Each cell from the multiple cells of the occupied grid can be processed as a separate system. The steps of the above method can be performed in each cell from the multiple cells of the occupied grid. Each cell from the multiple cells of the occupied grid can be observed using radar sensors or LIDAR sensors.

[0071] If the list of possible occupancy states Ω is finite and is defined, for example, as Ω = {S, D, F, SD, SF, DF, SDF}, then the possible transitions between occupancy states can be graphically represented in the form of a chart. Figure 3 The diagram illustrates the state transition graph 300. Possible occupancy states Ω can be defined as follows: static occupancy state S, dynamic occupancy state D, free space state F, a first uncertain state SD between static occupancy state S and dynamic occupancy state D, a second uncertain state SF between static occupancy state S and free space state F, a third uncertain state DF between dynamic occupancy state D and free space state F, and an unknown occupancy state SDF. Possible occupancy states are connected by multiple arrows representing multiple state transition probabilities between possible occupancy states. Those state transition probabilities between possible occupancy states can be defined as presented in the following transition matrix P:

[0072]

[0073] The probabilities provided in the transition matrix P can be viewed as equivalents of decay or forgetting factors (which can specify that knowledge of the state will decay or be forgotten in the absence of measurement data).

[0074] As shown in the above formula, the following state transition probability is different from zero and is defined as: the state transition probability p from static occupied state S to static occupied state S2. S→S 302. The state transition probability p of the first uncertain state SD between static occupancy state S and dynamic occupancy state D. S→SD 304. The state transition probability p of the second uncertain state SF between the static occupied state S and the free space state F. S→SF 306. The state transition probability p from dynamically occupied state D to dynamically occupied state D D→D 308. The state transition probability p of the first uncertain state SD between the dynamic occupancy state D and the static occupancy state S and the dynamic occupancy state D. D→SD 310. The state transition probability p of the third uncertain state DF between the dynamically occupied state D and the free space state F. D→DF 312. The state transition probability p from free space state F to free space state FF→F 314. The state transition probability p of the second uncertain state SF between the free space state F and the static occupied state S and the free space state F. F→SF 316. The state transition probability p of the third uncertain state DF between the free space state F and the dynamically occupied state D and the free space state F. F→DF 318. The state transition probability p from the first uncertain state SD between static occupancy state S and dynamic occupancy state D to the first uncertain state SD between static occupancy state S and dynamic occupancy state D. SD→SD 320. The state transition probability p from the first uncertain state SD between the static occupancy state S and the dynamic occupancy state D to the unknown occupancy state SDF. SD→SDF 322. The state transition probability p from the second uncertain state SF between the static occupied state S and the free space state F to the second uncertain state SF between the static occupied state S and the free space state F. SF→SF 324. The state transition probability p from the second uncertain state SF between the static occupied state S and the free space state F to the unknown occupied state SDF. SF→SDF 326. The state transition probability p from the third uncertain state DF between the dynamically occupied state D and the free space state F to the third uncertain state DF between the dynamically occupied state D and the free space state F. DF→DF 328. The state transition probability p from the third uncertain state DF between the dynamically occupied state D and the free space state F to the unknown occupied state SDF. DF→SDF 330, and the state transition probability p from the unknown occupied state SDF to the unknown occupied state SDF. SDF→SDF 332, where the state transition probability p SDF→SDF 332 equals 1. All other entries are zero, meaning the state transition probability is zero, therefore there is no state transition probability between corresponding states.

[0075] For all state transition probabilities that may be non-zero, the following condition must be met:

[0076]

[0077]

[0078] If these conditions are met, it can be shown that, without measurement updates, an unknown occupancy state SDF will be reached after an infinite number of steps, which is the desired behavior. In other words, after an infinite number of iterations (iteration number n→∞) (where only the transition matrix P is applied without measurement updates), all state transition probabilities are concentrated in the unknown occupancy state SDF. In other words, in the "infinite powers" of the transition matrix P, the state transition probability from any occupancy state to the unknown occupancy state SDF is 1, and all other state transition probabilities are set to 0:

[0079]

[0080] If calculated, the transition matrix P can depend on the estimated velocity v of objects in the region, i.e., P = P(v). This parameterization can provide better predictions of region occupancy status.

[0081] Radar sensors and LiDAR sensors can possess different characteristics that can be used to detect potential information, which can lead to different sets of observations for radar sensors and LiDAR sensors. The set of observations for a radar sensor can be defined as O. radar = {S, D, SD, F}, and the observation set for a LiDAR sensor can be defined as O LIDAR ={SD,F}, where the abbreviations have the same meaning as described herein, namely, static occupancy state S, dynamic occupancy state D, free space state F, and the first uncertain state SD between static occupancy state S and dynamic occupancy state D.

[0082] This may mean that, based on LiDAR sensors, only the uncertainty or free space information between static and dynamic occupancy (in raycasting mode) can be derived, without distinguishing between static occupancy state S and dynamic occupancy state D. Raycasting mode can be an inverse sensor model calibration, which can model not only the occupancy area but also the free space between occupancy areas. In this mode, free space is modeled in the cell under the ray between the sensor origin and the detection. Given sensor measurements, the inverse sensor model enables the estimation of occupancy and empty areas. Based on radar sensor detection, if the detection can be classified as dynamic or static, information about static and dynamic occupancy can be extracted. Otherwise, only the uncertainty or free space information between static and dynamic occupancy (in raycasting mode) can be obtained from the detection. For the method described in this paper, pre-filtering of sensor information is not required; in other words, the method described here can even work without pre-filtering sensor information.

[0083] For each sensor, i.e., for radar sensors and for LiDAR sensors, the observation matrix Q can be defined.radar and observation matrix Q LIDAR Add to the observation matrix Q radar The observed emission probability q for each term can depend on the distance r from the detection of the radar sensor, the same as the "classical" inverse sensor model (ISM), and the detection of the nearby velocity. Based on the above values, a distinction can be made between static occupancy state S and dynamic occupancy state D:

[0084]

[0085] Observation matrix Q LIDAR It can depend on the distance r from the LIDAR sensor:

[0086]

[0087] Figure 4 A flowchart 400 illustrating a method for occupancy detection according to various embodiments is shown. At 402, a probability distribution on a list of possible occupancy states of a region at a previous time point can be determined. At 404, measurement data associated with the region at a predetermined time point can be determined. At 406, a probability distribution on a list of possible occupancy states of the region at a predetermined time point can be determined based on the measurement data and the probability distribution on the list of possible occupancy states of the region at a previous time point, wherein the list of possible occupancy states includes: static occupancy state S, dynamic occupancy state D, free space state F, a first uncertain state SD between static occupancy state S and dynamic occupancy state D, a second uncertain state SF between static occupancy state S and free space state F, a third uncertain state DF between dynamic occupancy state D and free space state F, and an unknown occupancy state SDF.

[0088] According to various implementations, the probability distribution on the list of possible occupancy states of the region at a previous time point includes a predetermined initial state distribution on the list of possible occupancy states of the region.

[0089] According to various implementations, the initial state distribution on the list of possible occupancy states of the region includes an equal distribution on the list of possible occupancy states of the region.

[0090] According to various implementations, the probability distribution on the list of possible occupancy states of the area at the previous time point is determined based on the probability distribution on the list of possible occupancy states of the area at the previous time point and on the measurement data related to the area at the previous time point and at another time point before the previous time point.

[0091] According to various embodiments, the probability distribution on the list of possible occupancy states of the region at the predetermined time point is further determined based on a transition matrix that includes multiple state transition probabilities between possible occupancy states.

[0092] According to various implementations, the following state transition probabilities are not zero: state transition probability 302 from static occupancy state S to static occupancy state S; state transition probability 304 from static occupancy state S to the first uncertain state SD between static occupancy state S and dynamic occupancy state D; state transition probability 306 from static occupancy state S to the second uncertain state SF between static occupancy state S and free space state F; state transition probability 308 from dynamic occupancy state D to dynamic occupancy state D; state transition probability 310 from dynamic occupancy state D to the first uncertain state SD between static occupancy state S and dynamic occupancy state D; state transition probability 312 from dynamic occupancy state D to the third uncertain state DF between dynamic occupancy state D and free space state F; state transition probability 314 from free space state F to free space state F; state transition probability 316 from free space state F to the second uncertain state SF between static occupancy state S and free space state F; state transition probability 317 from free space state F to the third uncertain state DF between dynamic occupancy state D and free space state F. 8; The state transition probability from the first uncertain state SD between static occupancy state S and dynamic occupancy state D to the first uncertain state SD between static occupancy state S and dynamic occupancy state D is 320; The state transition probability from the first uncertain state SD between static occupancy state S and dynamic occupancy state D to the unknown occupancy state SDF is 322; The state transition probability from the second uncertain state SF between static occupancy state S and free space state F to the second uncertain state SF between static occupancy state S and free space state F is 324; The state transition probability from the second uncertain state SF between static occupancy state S and free space state F to the unknown occupancy state SDF is 326; The state transition probability from the third uncertain state DF between dynamic occupancy state D and free space state F to the third uncertain state DF between dynamic occupancy state D and free space state F is 328; The state transition probability from the third uncertain state DF between dynamic occupancy state D and free space state F to the unknown occupancy state SDF is 330; The state transition probability from the unknown occupancy state SDF to the unknown occupancy state SDF is 332, which equals 1.

[0093] According to various implementations, the transition matrix depends on the estimated velocity of objects in the region.

[0094] According to various implementations, the probability distribution on the list of possible occupancy states of a region at a predetermined time point is also determined based on an observation matrix that includes the observed emission probability.

[0095] According to various implementations, the area includes cells occupying a grid, which in turn includes multiple additional cells.

[0096] According to various implementations, measurement data is determined based on LIDAR sensors, and the measurement data includes uncertainty information or free space information between static occupancy and dynamic occupancy.

[0097] According to various implementation methods, measurement data is determined based on radar sensors, and the measurement data includes static occupancy information, dynamic occupancy information, uncertainty information between static and dynamic occupancy, or free space information.

[0098] According to various implementations, the observation matrix depends on at least one of the detection proximity velocity from the radar sensor and the distance from the radar sensor.

[0099] Each of steps 402, 404, and 406, as well as the further steps described above, can be performed by computer hardware components.

[0100] The methods and systems described herein can be used to provide occupancy status detection.

[0101] Figure 5 A computer system 500 with multiple computer hardware components is illustrated, said multiple computer hardware components being configured to perform steps of a computer-implemented method for occupancy state detection according to various embodiments. The computer system 500 may include a processor 502, a memory 504, and a non-temporary data storage device 506. A camera 508 and / or a distance sensor 510 (e.g., a radar sensor or a LiDAR sensor) may be provided as part of the computer system 500 (e.g., ...). Figure 5 (as shown), or it can be provided outside the computer system 500.

[0102] Processor 502 can execute instructions provided in memory 504. Non-temporary data storage device 506 can store computer programs, including instructions that can be transferred to memory 504 and then executed by processor 502. Camera 508 and / or distance sensor 510 can be used to determine measurement data, such as measurement data related to an area at a predetermined time point, as described herein.

[0103] Processor 502, memory 504, and non-temporary data storage device 506 may be connected to each other, for example, via electrical connection 512 (e.g., cable or computer bus) or via any other suitable electrical connection, to exchange electrical signals. Camera 508 and / or distance sensor 510 may be connected to computer system 500, for example, via an external interface, or may be provided as part of the computer system (in other words: internal to the computer system, for example, via electrical connection 512).

[0104] The terms “connection” or “link” are intended to include direct “connection” (e.g., via a physical link) or direct “link” as well as indirect “connection” or indirect “link” (e.g., via a logical link).

[0105] It should be understood that the above description of one of the methods can be similarly applied to computer system 500.

[0106] List of reference numerals

[0107] 100 Markov chains

[0108] 102 Occupied status h1

[0109] 104 Occupied status h2

[0110] 106 Occupied status h3

[0111] 108 Occupied status h4

[0112] 110 Measure m1

[0113] 112 Measuring m2

[0114] 114. Measuring m3

[0115] 116. Measuring m4

[0116] 118 arrows

[0117] 120 arrows

[0118] 122 arrows

[0119] 124 arrows

[0120] 126 arrows

[0121] 128 arrows

[0122] 130 arrow

[0123] 132 arrows

[0124] Calculation of 200 single probability distributions

[0125] 202 Probability distribution at a predetermined time point

[0126] 204 Probability distribution at previous time points

[0127] 206 Probability distribution at previous time points

[0128] 208 Probability distribution at previous time points

[0129] 210 Probability distribution at previous time points

[0130] 212 Observations at predetermined time points

[0131] 214 State transition probability

[0132] 216 State transition probability

[0133] 218 State transition probability

[0134] 220 State transition probability

[0135] 222 Observe the launch probability

[0136] 300 State Transition Diagram

[0137] S Static Occupancy Status

[0138] D Dynamic Occupancy Status

[0139] F Free space state

[0140] SD First Uncertainty

[0141] SF Second Uncertainty

[0142] DF Third Uncertainty

[0143] SDF Unknown Occupancy Status

[0144] 302 State transition probability

[0145] 304 State transition probability

[0146] 306 State transition probability

[0147] 308 State transition probability

[0148] 310 State transition probability

[0149] 312 State transition probability

[0150] 314 State transition probability

[0151] 316 State transition probability

[0152] 318 State transition probability

[0153] 320 State transition probability

[0154] 322 State transition probability

[0155] 324 State transition probability

[0156] 326 State transition probability

[0157] 328 State transition probability

[0158] 330 State transition probability

[0159] 332 State transition probability

[0160] 400 shows a flowchart of a method for occupancy state detection according to various embodiments.

[0161] 402 Steps to determine the probability distribution at previous time points

[0162] 404 Steps for determining measurement data

[0163] 406. Steps to determine the probability distribution at a predetermined time point

[0164] 500 Computer systems according to various implementation methods

[0165] 502 processor

[0166] 504 memory

[0167] 506 Non-Temporary Data Memory

[0168] 508 camera

[0169] 510 Distance Sensor

[0170] 512 connections

Claims

1. A computer-implemented method for determining a probability distribution in a region at a predetermined time point using a Hidden Markov Model, the method comprising the following steps performed by computer hardware components: - Determine the probability distributions (204, 206, 208, 210) on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the region at previous time points, where, The area includes cells occupying a grid, and the occupying grid includes multiple additional cells; - Use sensors to determine measurement data related to the region at the predetermined time point; as well as - Based on the measurement data and the probability distributions (204, 206, 208, 210) on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the area at the previous time point, determine the probability distribution (202) on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the area at the predetermined time point. The list of possible occupancy states (S, D, F, SD, SF, DF, SDF) includes: static occupancy state (S), dynamic occupancy state (D), free space state (F), a first uncertain state (SD) between the static occupancy state (S) and the dynamic occupancy state (D), a second uncertain state (SF) between the static occupancy state (S) and the free space state (F), a third uncertain state (DF) between the dynamic occupancy state (D) and the free space state (F), and an unknown occupancy state (SDF). Based on the probability distribution of the region at the predetermined time point, perform at least one of the following: (a) plan a path for the robot, including avoiding robot collisions; (b) detect objects for vehicle driving.

2. The method according to claim 1, in, The probability distributions (204, 206, 208, 210) on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the region at the previous time point include a predetermined initial state distribution on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the region.

3. The method according to claim 2, in, The initial state distribution on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the region includes an equal distribution on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the region.

4. The method according to claim 1, in, The probability distributions (204, 206, 208, 210) on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the area at the previous time point are determined based on the probability distributions on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the area at the previous time point and at another time point prior to the previous time point.

5. The method according to claim 1, in, The probability distribution (202) on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the region at the predetermined time point is further determined based on a transition matrix including multiple state transition probabilities (302, 304, 306, 308, 310, 312, 314, 316, 318, 320, 322, 324, 326, 328, 330, 332) between possible occupancy states (S, D, F, SD, SF, DF, SDF).

6. The method according to claim 5, wherein, The following state transition probabilities (302, 304, 306, 308, 310, 312, 314, 316, 318, 320, 322, 324, 326, 328, 330, 332) are different from zero: - The state transition probability from the static occupancy state (S) to the static occupancy state (S) (302); - The state transition probability (304) of the first uncertain state (SD) from the static occupancy state (S) to the static occupancy state (S) and the dynamic occupancy state (D). - The state transition probability (306) of the second uncertain state (SF) between the static occupied state (S) and the free space state (F). - The state transition probability (308) from the dynamic occupancy state (D) to the dynamic occupancy state (D). - The state transition probability (310) of the first uncertain state (SD) between the dynamic occupancy state (D) and the static occupancy state (S) and the dynamic occupancy state (D). - The state transition probability (312) of the third uncertain state (DF) between the dynamic occupancy state (D) and the free space state (F). - The state transition probability from the free space state (F) to the free space state (F) (314); - The state transition probability (316) of the second uncertain state (SF) between the free space state (F) and the static occupied state (S) and the free space state (F). - The state transition probability (318) of the third uncertain state (DF) from the free space state (F) to the dynamic occupied state (D) and the free space state (F). - The state transition probability (320) from the first uncertain state (SD) between the static occupancy state (S) and the dynamic occupancy state (D) to the first uncertain state (SD) between the static occupancy state (S) and the dynamic occupancy state (D). - The state transition probability (322) from the first uncertain state (SD) between the static occupancy state (S) and the dynamic occupancy state (D) to the unknown occupancy state (SDF). - The state transition probability (324) from the second uncertain state (SF) between the static occupied state (S) and the free space state (F) to the second uncertain state (SF) between the static occupied state (S) and the free space state (F). - The state transition probability (326) from the second uncertain state (SF) between the static occupied state (S) and the free space state (F) to the unknown occupied state (SDF). - The state transition probability (328) from the third uncertain state (DF) between the dynamic occupied state (D) and the free space state (F) to the third uncertain state (DF) between the dynamic occupied state (D) and the free space state (F). - The state transition probability (330) from the third uncertain state (DF) between the dynamic occupancy state (D) and the free space state (F) to the unknown occupancy state (SDF). - The state transition probability (332) from the unknown occupancy state (SDF) to the unknown occupancy state (SDF) is equal to 1.

7. The method according to claim 5, in, The transition matrix depends on the estimated velocity of objects in the region.

8. The method according to claim 1, in, The probability distribution (202) on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the region at the predetermined time point is further determined based on an observation matrix that includes the observed emission probability (222).

9. The method according to claim 1, in, The measurement data is determined based on LIDAR sensors, and the measurement data includes uncertainty information or free space information between static occupancy and dynamic occupancy.

10. The method according to claim 1, in, The measurement data is determined based on radar sensors and includes static occupancy information, dynamic occupancy information, uncertainty information between static and dynamic occupancy, or free space information.

11. The method according to claim 10, in, The probability distribution (202) on the list of possible occupancy states (S, D, F, SD, SF, DF, SDF) of the region at the predetermined time point is further determined based on an observation matrix that includes the observed emission probability (222), and The observation matrix depends on at least one of the detection proximity velocity to the radar sensor and the distance to the radar sensor.

12. A computer system (500) comprising a plurality of computer hardware components configured to perform the steps of a computer-implemented method according to any one of claims 1 to 11.

13. A vehicle comprising the computer system (500) according to claim 12 and the sensor.

14. A non-transitory computer-readable medium comprising instructions for performing a computer-implemented method according to any one of claims 1 to 11.

Images