A method for vehicle environment mapping and corresponding system, vehicle and computer program product

Abstract

A method (1000) for vehicle (V) environment mapping, comprising the following operations: - receiving (1010) a set of input values ​​from a plurality of sensors, - applying (1030) a temporal fusion process to the set of input values, thereby obtaining a respective set of occupancy grid maps, - applying (1040, 1050, 1060) a data fusion process to the set of occupancy grid maps, thereby obtaining at least one fused occupancy grid map, characterized in that it includes: - detecting (1070) differences by comparing occupancy grid maps in the map set, thereby obtaining a set of detected differences, - processing (1090) at least one fused occupancy grid map, and outputting a fused occupancy grid map of a drivable space, the processing operation including performing arbitration of conflicts in the at least one fused occupancy grid map (1090a, 1090b). The composite fused occupancy grid map of the drivable space is supplied (IA) to a user circuit, such as a driver assistance interface.

Description

Technical Field

[0001] The present invention relates to map building technology and vehicle environment mapping.

[0002] One or more embodiments may involve processing at least one fused grid map to determine occupancy of objects, such as to facilitate detecting the location of obstacles and how to avoid contact with such obstacles. Background Art

[0003] Multi-sensor data fusion can facilitate combining information from different sources to form a unified picture, for example by providing a fused occupancy grid map representing the environment based on data from multiple sources.

[0004] Different mapping technologies can provide different types of maps, such as:

[0005] - are specialized for structured environments and therefore depend on the presence of such structure in the surroundings, such as feature-based models,

[0006] - Tailored for unstructured environments, such as occupancy grid maps.

[0007] These two approaches are by no means mutually exclusive; in fact they complement each other.

[0008] Occupancy grid maps are well suited for path planning and obstacle avoidance tasks. In contrast, feature-based maps may be well suited for localization purposes, where the relative poses of objects may be required to accurately estimate the vehicle's own position and orientation.

[0009] Occupancy grid maps (grid maps for short) represent the (vehicle surrounding) environment as a two-dimensional grid map with fixed size and spatial resolution. The accuracy of the grid maps can be defined according to the resolution of the discrete binary grid. In addition, they can be oriented in the direction of the longitudinal and transverse axes of the vehicle and can have edge portions placed at right angles to each other.

[0010] A grid map may be provided to represent the vehicle environment, the grid map being segmented into grid cells, each cell storing features describing the environment. For example, each cell may store a probability value / likelihood value, such as the probability of whether the cell is occupied. In practice, map construction may be formulated as an optimization problem that maximizes the probability that a particular grid cell is actually occupied (or free) given prior knowledge of the environment, vehicle position, and sensor measurements.

[0011] The grid map helps provide the driver assistance system with information about the presence of obstacles in the surrounding mapped environment and their spatial occupancy. Therefore, the driver assistance system can operate vehicle functions based on this information. For example, the driver assistance system can react to traffic conditions.

[0012] As known to those skilled in the art, a grid map may represent an environment in various coordinate systems, such as Cartesian or polar coordinates.

[0013] The environment may be scanned by a plurality of sensors / detectors (eg, Global Positioning System (GPS) antennas and / or Geographic Information System (GIS) antennas, Light Detection and Ranging (LIDAR), cameras, etc.).

[0014] Multiple grid maps can be generated from a single sensor data to provide relevant information about the environment.

[0015] It is an object of the present invention to combine (e.g. by data fusion) all sensor information with high precision and accuracy, in particular information relevant for example for driver assistance systems and methods for vehicles and automated vehicles, to fuse multiple occupancy grid maps into a unified coherent map.

[0016] Occupancy grid maps can help map the environment around the vehicle for possible obstacles, thereby determining the drivable space in such a way that an autonomous system that can estimate its own position can safely navigate through the mapped environment.

[0017] In contrast to single sensor mapping, grid maps constructed due to multi-sensor fusion rely heavily on sensor measurement weights. Conventional mapping approaches such as Bayesian Occupancy Filter (BOF) or Dempster-Shafer theory (DST) apply equal weights unless specified by the user. Empirically adjusted sensor weights are tailored for certain scenarios, especially when sensor readings are not contradictory. Summary of the invention

[0018] It is an object of one or more embodiments to overcome the inherent limitations of solutions available from the prior art.

[0019] According to one or more embodiments, this object is achieved by a method having the features specified in claim 1. One or more embodiments may relate to a corresponding system, a vehicle and a computer program product.

[0020] The claims form an integral part of the technical teaching provided herein regarding the various embodiments.

[0021] According to the solution described herein, the method comprises providing a fused grid map representation of the vehicle environment (eg in the form of an occupancy grid map) based on data from a plurality of sources.

[0022] In one or more embodiments, adaptive sensor weights may be applied to a multi-sensor fusion map, for example, iteratively. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Embodiments will now be described, by way of non-limiting example only, with reference to the accompanying drawings, in which:

[0024] - Figure 1 is an exemplary functional diagram of a vehicle according to an embodiment,

[0025] - Figure 2 is a view of an embodiment of an occupation grid map,

[0026] - Figure 3 yes Figure 1 Examples of possible arrangements of parts of the process,

[0027] - Figure 4 is an example of a possible view occupying a grid map according to an embodiment,

[0028] - Figure 5 and Figure 6 is an example of a possible arrangement of parts of the processing of FIG.

[0029] - Figure 7 yes Figure 1 Examples of possible views of embodiments of the process,

[0030] - Figure 8a , Figure 8b and Fig. 9 yes Figure 1 Examples of possible arrangements of parts of the process,

[0031] - Fig.10 is an example of an embodiment of a vehicle environment mapping method. DETAILED DESCRIPTION

[0032] In the following description, one or more specific details are shown to provide a deeper understanding of the examples of the embodiments of the present description. The embodiments can be obtained without one or more specific details or with other methods, components, materials, etc. In other cases, known structures, materials or operations are not shown or described in detail, so that some aspects of the embodiments will not be difficult to understand.

[0033] References to "an embodiment" or "one embodiment" in the framework of this specification are intended to indicate that a particular configuration, structure, or feature described in connection with the embodiment is included in at least one embodiment. Therefore, phrases such as "in one embodiment" or "in an embodiment" that may appear in one or more places in this specification do not necessarily refer to the same embodiment.

[0034] Furthermore, particular configurations, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0035] The references used herein are provided for convenience only and therefore do not define the degree or scope of protection of the embodiments.

[0036] Figure 1 An example of a functional diagram of a vehicle according to an embodiment of the solution described herein is shown. Such an embodiment comprises a vehicle equipped with a plurality of sensors S1, S2, S i The vehicle V and the driving assistance system 100 are configured to provide an arbitrated fused grid map set MM including at least one arbitrated fused occupancy grid map M based on the set of fused occupancy grid maps FF, wherein the fused occupancy grid map set FF includes the fused occupancy grid map M generated by the system 100 based on the sensors S1, S2, S i The measurement details at least one fused occupancy grid map F of the environment, the arbitrated fused grid map being, for example, a map that has been determined as part of all conflicts determined in the data fusion operation to obtain the at least one fused occupancy grid map F. Such a determined fused occupancy grid map M in the embodiments discussed herein corresponds to a map of the drivable space.

[0037] For example, the vehicle V may be an autonomous vehicle such as a driverless car. Other examples of vehicles may be car-like vehicles, autonomous robots, and the like.

[0038] Multiple sensor / detector devices S1, S2, S i Sensors known per se are included, such as lidar / radar sensors, cameras, GPS antennas, and sources using GPS information such as (GPS / GIS) background maps, etc., and the device is coupled to the system 100. Specifically, in the example shown, the sensors are lidar, cameras, and a GPS / GIS system that generates a (GPS / GIS) background map based on the detected position.

[0039] Sensors S1, S2, S i A collection can typically include sensors with the following characteristics:

[0040] - complementary functionality, such as covering different fields of view of the environment explored by the vehicle and / or in the physical working principles utilized,

[0041] - Coherent functionality, such as providing data about the same field of view and / or acquiring data on the same field of view through the same physical operating principle.

[0042] Typically, the system 100 provides the arbitrated map M to a user, for example as an aid for assisted driving, or to another system that uses the determined map M to control the vehicle V, such as a system for autonomous driving. Figure 1 It is shown that the system 100 may be coupled to an output interface IA. For example, the interface IA may be a user interface including a display unit that may be used to display such an arbitrated fused occupancy grid map M of the drivable space. Alternatively, the map data available at the output interface IA may be used and processed by another system.

[0043] The system 100 is configured to process data from sensors S1, S2, S i The measurement data ("readings") of the collection are Y1, Y2, Y i , providing the fused occupancy grid map F, E set FF to the output interface IA (outside the data fusion block 40).

[0044] As described above, the arbitrated fused occupancy grid map M of the drivable space available at interface IA may be used by the vehicle's control system to operate the vehicle itself, for example, to navigate the environment represented in the fused occupancy grid map M provided as output by the system 100 .

[0045] like Figure 1 As shown in the example, the system 100 generally includes a plurality of sensors S1, S2, S i , a time fusion phase 30 is combined, a data fusion phase 40 and an arbitration phase 90 .

[0046] In particular, multiple sensors S1, S2, S i The temporal fusion stage 30 receives the sensor readings Y1, Y2, Y i gather.

[0047] like Figure 1 As shown, the time fusion stage 30 set may include a first time fusion stage 301, a second time fusion stage 302, and an i-th time fusion stage 30i to process the time fusion data from a plurality of sensor readings Y1, Y2, Y i A single reading for each sensor in i ), thereby providing sensor occupancy grid maps G1, G2, G iThe collection is given as output as described below.

[0048] For example, the set of time fusion stages 30 may be distributed over the respective sensors, providing local decentralized processing of the sensor readings in a manner known per se (eg via software).

[0049] exist Figure 1 In the exemplary embodiment shown:

[0050] a first temporal fusion stage 301 receives a first data set Y1 of sensor readings from a first sensor S1 and provides a first map G1 as output,

[0051] a second time fusion stage 302 receives a second data set Y2 of sensor readings from a second sensor S2 and provides a second map G2 as output,

[0052] The i-th time fusion stage 30i receives the i-th data set Y of sensor readings from the i-th sensor i And provide the i-th map G i as output.

[0053] Therefore, according to the information from sensors S1, S2, S i The individual sensor readings Y1, Y2, Y i Each of the temporal fusion stages 301, 302, 30i in the set of temporal fusion stages 30 provides an occupancy grid map G1, G2, G3 represented in the vehicle frame. i gather.

[0054] Alternatively, the i-th sensor / source can also directly provide the grid map G i As a reading, this may be the case, for example, when it comes to acquiring GIS / GPS maps from a remote sensing source.

[0055] As described above, in a variant embodiment, the processed sensor readings Y1, Y2, Y i The operations can include:

[0056] -Coordinate system conversion operations, e.g. from sensor frame to vehicle frame,

[0057] -Time fusion operations, such as data standardization and data synchronization based on vehicle working hours,

[0058] -Data fusion operations, such as combining, associating, correlating, and estimating data.

[0059] Preferably, the time fusion stage 30 provides the collection of the maps G1, G2, G i At least one i-th map in the set G i According to the i-th sensor Si The types of measurements provided provide information about specific characteristics of the environment.

[0060] Figure 2 is a view of an exemplary occupancy grid map G0, such as an a priori grid map used for system initialization to process data from the i-th sensor.

[0061] The occupancy grid map usually consists of a number of N = n*m individual cells g ij k , where n is the number of rows and m is the number of columns, indexed by indices i and j respectively.

[0062] Specifically, as in the Cartesian coordinate system, any single cell g in the grid map G0 ij k Preferably it has fixed dimensions and a predetermined geometric shape, such as a square.

[0063] Occupancy grid map G0 = G 0,k can evolve over time and is therefore a function of time k.

[0064] Each cell g of the grid G0 ij can be a variable (e.g. a random variable) that can (at time k) assume a certain number of "state" values ​​with certain probabilities.

[0065] Each cell g of the grid in the illustrated embodiment ij k This includes random variable values ​​and associated (categorical) "state" values, for example:

[0066] -The value can be a probability value p(g ij k ), in order to estimate the probability that the part of the environment mapped by a single cell is correctly estimated as being “occupied” by the object,

[0067] -According to the probability value p(g ij k ) can assign a "state" label value from a list of possible state values ​​(e.g., idle / unknown / occupied) to represent the occupancy state of a portion of the environment mapped by a cell of the grid map.

[0068] In general, if the probability value stored in a cell is:

[0069] - below a certain threshold, such as p(g ij k )<0.5, classify the cell with a first state label (e.g., “idle”), which is assigned to the corresponding cell,

[0070] -Above a certain threshold, such as p(g ij k )>0.5, classify the cell with a second state label (e.g., “occupied”), which is assigned to the corresponding cell,

[0071] - is equal to a certain threshold, such as p(g ij k )≈0.5, classify the cell with a third state label (e.g., “unknown”), which is assigned to the corresponding cell.

[0072] Raster maps are usually described as two-dimensional grids, such as G k = {g ij k :i≤i≤n,1≤j≤m}.

[0073] For example, Figure 2 , Figure 4 and Figure 7 As shown, when a grid map is visualized on a display or in any graphical form, information regarding the classification label of each grid cell may be encoded as a pattern / color that fills the grid cell area.

[0074] With in each grid cell g ij k The probability distribution p(g ij k ) evolves over time as new sensor readings are acquired from the sensor, and at each time instant k, the sensor’s measurement Y is considered i And the vehicle's posture V is updated accordingly.

[0075] Figure 3 is an exemplary diagram showing an embodiment of the i-th time fusion processing stage 30i.

[0076] The temporal fusion process 30i involves aggregating the information received in the sensor readings, for example, by combining the i-th source S i The reading Y obtained at time k i,k With the i-th source S i The reading Y at the previous time k-1 i,k-1 To summarize.

[0077] Therefore, according to the information from sensors S1, S2, S i The individual sensor readings Y1, Y2, Y i Each time fusion stage 301, 302, 30i provides an occupancy grid map G1, G2, G i gather.

[0078] The temporal fusion processing stage 30i may include coordinate system conversion operations 301, 302, 30i from the sensor frame to the vehicle frame as follows:

[0079] - formatting the sensor readings into a vector with features associated with the preferred sensor model,

[0080] - Compute the inverse likelihood / sensor model to extract context-relevant information from the measurements,

[0081] - Account for measurement errors present in sensor readings.

[0082] like Figure 3 As shown, the i-th time period 30i is obtained from the i-th sensor S at a given time k. i Receives a dataset Y of sensor readings / measurements k , for example, with an n-dimensional vector Y k =(y 1 k ,y 2 k ,…,y n k ) in the form of.

[0083] Similarly, the vehicle state V k (e.g., its position) at a given time k may be known (e.g., stored in stage 312) and may be described as an m-dimensional vector, such as V k =(v 1 k ,v 2 k ,…,v m k ). For example, the vehicle V can be parameterized by the center of mass position and orientation angle of the vehicle V relative to the ground plane. k posture.

[0084] At a certain time k, the vehicle posture / state V k and sensor reading Y k Combining, for example, providing input in the form of likelihood values, such as p(Y k |V k ) to be input into the time fusion processing stage 30i.

[0085] As described in this paper, a Bayesian occupancy filter (BOF) is sufficient. For example, BOF is used to calculate the occupancy grid map G k The posterior probability p(G k |Y k ,V k ) to calculate the occupancy grid map G taking into account the new set of sensor readings iThe updated probability value of the grid cell.

[0086] In order to perform the temporal fusion process 30i, an initial occupancy grid map M0 is provided, for example stored / generated in stage 324. The initial grid M0 is initialized, for example:

[0087] - Classify the state of each cell as “unknown”, e.g., if no prior information about the vehicle’s surroundings is available, assign a probability value of 0.45 <p(g ij 0)<0.55, and / or

[0088] - If a priori information about the environment is known, load the grid cell values. For example, when a vehicle enters / exits a house garage, stage 303 may be operated by the autonomous control system to provide an initial occupancy grid map M0 based on the a priori known garage floor plan geometry.

[0089] In addition, the "previous" grid map G k-1 Generated / stored in stage 307 of the temporal fusion stage 30i, where the grid cell has a probability value calculated at time k-1 prior to time k at which the sensor reading was taken.

[0090] Note that at system startup, the initial grid M0 and the "previous" grid map M k-1 Can have equal values, such as M k-1 =M0.

[0091] The temporal fusion stage 30i fuses the prior map knowledge M0 and / or the “previous” grid map M k-1 , for example, in stage 322, with the measured likelihood p(y k |x k ) in the form of new data.

[0092] Bayes' theorem states that the grid map M k Cell g ij k The probability p(g ij k |y k ,x k ) is proportional to:

[0093] - "Previous" grid map M k-1 Cell g ij k-1 The probability p(g ij k-1 |y k-1 ,x k-1 ),as well as

[0094] -For example, the probability p(g ij k |y k ,x k ),

[0095] - Bayesian gain, f{p(not(g ij k )|y k ,x k )-p(g ij k |y k ,x k )}.

[0096] like Figure 5 As shown in , the Bayesian gain is calculated in the gain stage 520.

[0097] Note that the Bayesian gain is computed based only on the measurements, under the assumption of a priori unknown boundaries of the map.

[0098] exist Figure 3 In the exemplary embodiment shown, a forgetting factor 327 is also applied to the sensor occupancy grid, for example, to the "previous" map. The forgetting factor stage may introduce a parameter (e.g., an exponential decay factor) as a factor for increasing or decreasing the updated grid G. k In the previous grid G k-1 The design parameters of the weights on the grid are applied to the grid cell values.

[0099] Therefore, the temporal fusion process 30 can be formulated as a recursive solution to the following problem: Considering the environment G0, the vehicle position V k and sensor reading Y k The prior knowledge of a particular grid cell g k ij The probability of being actually occupied (or free) is maximized.

[0100] Note that at time k (denoted by G k ) at the grid map G (e.g. G = G i ) is the processing of the sensor reading at time k (e.g., Y i,k ) and the previous map G at time k-1 (denoted as G k-1 ) result. In addition, the previous map G k-1 It contains not only information related to time k-1 itself, but also information related to the entire evolution of the map from the time origin to time k-1 (e.g., from time 1 to time k-1).

[0101] In addition, according to the first-order Markov assumption, when the sensor occupancy grid map G is represented iWhen referring to the evolution of map G at a single "time frame" at time k, but not limited to the embodiment, the complete map history from time 1 to k-1 is represented / captured in a single map G at time k-1.

[0102] Figure 4 is from multiple sensors S1, S2, S i The multiple sensor readings Y1, Y2, Y i Possible sensor occupancy grid maps processed - frames G1, G2, G i In particular, Figure 4 The respective parts a1), a2), a3), a4) of part a) represent Figure 4 Enlarged views of the respective parts b1), b2), b3), b4) of part b).

[0103] The respective "state" labels categorized from the grid cell contents of the sensor occupation grid map may be changed depending on the sensor reading parameters / operating principle of each sensor.

[0104] In fact, the respective sensors occupy the grid maps G1, G2, G k Perhaps each provides information about a particular feature of the environment.

[0105] For instance, in the example shown:

[0106] The first occupancy grid map G1 is the result of processing the data from the vision sensors, thereby providing information about static and dynamic objects and the topology of the environment (eg a road with lanes).

[0107] - The second grid map G2 is the result of processing the data from the lidar sensor in order to provide information on the estimated distances of static and / or dynamic obstacles.

[0108] - Map G i It is the result of processing data from sensors represented by a Geographic Information System (GIS) to provide contextual information data about the environment (e.g. road areas / areas outside roads). As is known, GIS integrates remote sensing imagery, i.e. images obtained by measuring the earth using sensors on aircraft or satellites.

[0109] Specifically, after the sensor readings Y1, Y2, Y i The temporal fusion process 30i generates respective sensor occupancy grid maps G1, G2, G i Afterwards, for each time step k, the sensors occupy grid maps G1, G2, G iThe grid maps in the set are "fused" together in a fused map set FF comprising at least one "fused" occupied grid map F,E.

[0110] To this end, the raster fusion stage 40 receives the map sets G1, G2, G i , and by mapping the map sets G1, G2, G i (For example, the timing frame G 1,k , G 2,k , G i,k ) performs a fusion operation on the data and provides at least one "fused" occupancy grid map F, E as output.

[0111] Different approaches can be used for data fusion 40, which may include combining maps G1, G2, G i The data in the set (of time frames) are combined, associated, estimated, and correlated with each other.

[0112] These two methods are shown in Figure 8 and Fig. 9 Two embodiments are shown which are summarized to show how the data fusion block 40 is implemented.

[0113] For example, Figure 1 As illustrated, the Bayesian inference 50 shown in FIG8 or Fig. 9 Data fusion based on DS theory 60 is shown.

[0114] Therefore, the fused grid map can facilitate accurate prediction and / or detection of obstacle occupancy along the driving path of vehicle V, thereby facilitating online correction of the driving path of vehicle V through the driving assistance system / interface.

[0115] For example, when the vehicle is operated by an autonomous driving system (not visible in the figure), it can help avoid collisions between the vehicle and obstacles.

[0116] As described above, for example, the raster fusion process 40 may include at least one of the following:

[0117] - Bayesian inference / Bayesian occupation filter (BOF) fusion processing stage 50,

[0118] -DS theory (DST) fusion processing stage 60.

[0119] In practice, the Bayesian occupation filter (BOF) (see e.g. Figure 5 ) and DS Evidence Theory (DST) (see e.g. Figure 6 ) is applicable to the raster fusion processing stage 40. Although in FIG. 8 and Fig. 9, separate implementations are shown, but the raster fusion processing stage 40 and the arbitration stage 90 may include BOF and DST embodiments operating in parallel.

[0120] like Figure 5 As shown in the example, in BOF fusion 50, based on the weight W k Sets, such as W k =[w k，1 ,w k,2 ,w k,i ], the fusion processing formula converts the sensors occupying the grid maps G1, G2, G i The data in the grid cells of the grid map occupied by each sensor in the set are combined, indicating that the ... i The respective confidence levels in the accuracy of .

[0121] In general, the result of the BOF (grid fusion process) 50 can be expressed as a weighted (linear) combination of the sensor occupancy grid maps resulting from the temporal fusion, for example:

[0122]

[0123]

[0124] Unless specified in advance by the user and / or design, it is assumed that all sensors contribute equally to the BOF grid fusion process 50, that is, the maps are combined with equal weights, such as w k,i =w k .

[0125] In the classic Bayesian setting, we assume that in the grid cell The spatiotemporal relationship between them is evaluated by Bayes’ theorem as a combination of multiple sensor grids G k,i The authenticity (probability) of the map p(F k |G k ),Right now

[0126]

[0127] Here, the estimate of the map occupancy is compared with the likelihood p(G k |F k ) of each single sensor grid G k Make a comparison.

[0128] However, grid-based approaches do not have the ability to predict the behavior of objects, as this would require the implementation of feature recognition and data association techniques and would therefore rely solely on measurement weights. Similarly, as with temporal fusion, without model-based predictions, the expected change in the map from time k-1 to k is p(Fk |F k-1 )=0. Therefore, the Bayesian gain f{p(not(F k ))|G k )-p(F k |G k )} may not be able to determine the sensor weight w k . Empirically tuned sensor weights may be tailored only for certain cases and thus may not be suitable for general purpose mapping. Therefore, it becomes apparent that the sensor weights need to be recursively tuned during the raster fusion process.

[0129] So, for example, the true weight w of the i-th sensor k,i and the actual value of the weight The error between k,i It can be expressed as:

[0130]

[0131] Providing a feedback network of iteratively adjustable weights, for example, in a data set W provided by arbitration stage 90 as discussed below, can facilitate automatic sensor weights w during BOF fusion k Adjustment.

[0132] Alternatively, for example in Fig. 9 Box 60 (in Figure 6 In the DS Evidence Theory (described in detail in ), a set of combinatorial rules called DS Evidence Theory (DST) can be used. DST is developed to resolve the ambiguity that occurs between occupancy grids by utilizing combinatorial rules (i.e., evidence theory) applied on the evidence grid.

[0133] In fact, according to Bayes' theorem, the occupancy of cells is evaluated based on a single probability for each cell, while in DST, decisions about occupancy are made based on a trust function of four "qualities". Moreover, trust transitions in Bayes' theorem are only possible between two states and are subject to symmetry restrictions.

[0134] like Figure 6 As illustrated, the DST process 60 provides at least one occupancy grid map F k and / or complementary occupancy grid maps (e.g. not(F k ) and / or conflict grid map C k ) as output. The DST process 60 provides as output a fused raster map F k Set, not(F k ), C k Can be collectively referred to as the "evidence grid" E.

[0135] About Figure 2Similarly as discussed, the evidence grid map E has individual grid cells assigned probability values. Furthermore, the classification associated with the probability values ​​includes more than just a binary set of "states" of the cells: in fact, uncertainty is modeled as a belief function Bel(X).

[0136] The "confidence" about the occupancy of the environment (i.e., the evidence supporting the state claim, the degree to which the classification is believed to be true) is classified as occupied (O) or free (FR), resulting in the set of all possible states of the random variable X = {O, FR}. The set of possible subsets of X, the "frame of identification" (FOD for short), is denoted as The main goal of DST is to compute the confidence Bel(X), which is the probability of obtaining the support for the identification framework 2 X The belief function Bel(X) therefore represents the degree to which X is believed to be true and can be expressed as:

[0137]

[0138] in:

[0139] -X is a possible state 2 X Any state in, for example, X=O,

[0140] -m(A) is the value of the basic probability density function, also called the "quality" of A, for X For each subset A containing the state X, for example, if X = O, m(A) = m(O), m(O, FR), m(O, Ω),

[0141] -Bel(X) is 2 X The subset of X that explains all the evidence supporting claim X.

[0142] The content of the cells in the evidence grid map E (also called “likelihood”) is an indicator of how reliably the presence of an object / obstacle can be detected in the environment space surrounding the vehicle V.

[0143] The DST fusion process 60 generally begins with a basic mass assignment (BMA) phase 610, which includes a hard constraint-based process that redistributes the four masses to occupy the grid map cells p(g k,i ). The results of the BMA are then provided to the DS combination rules stage 690 as described below.

[0144] For example, two information sources, such as sensor occupancy grid maps G1 and G2, are connected by the XOR operator Combined into a fused trust function m with "joint quality", its value is usually calculated as:

[0145] For X≠Ω

[0146] For X = Ω

[0147] Given the commutative nature, the order in which the sensors are fused does not matter.

[0148] In the evidence approach, trust can shift between four states, where each transition has different meanings, dynamics, and importance. Although DST models higher confidence levels in cells rather than just a single probability value, the sensor grid weights w k It remains unresolved at this stage.

[0149] As shown in the embodiment of FIG. 8 referencing a data fusion block 50 performing BOF, in this embodiment the set of temporal fusion stages 30 are coupled to a grid fusion state 40 which includes a data fusion block 50 based on a Bayesian inference 50 and a difference assessment stage 70 .

[0150] The difference evaluation stage 70 receives the maps G1, G2, G i The detected differences D are collected and provided, the collection D containing information about the results of an operation of comparing the maps (eg, one-to-one). The comparison operation may be performed by any different technique suitable for identifying at least equal and different (ie, conflicting) cells between the maps.

[0151] For example, the set of detected differences D may include:

[0152] - the set of locations of cells that have conflicts between their respective sensor grid maps, as described below, and / or

[0153] - A set of values ​​indicating inconsistencies between pairs of sensors, for example, obtained by subtracting the probability value of a grid cell in a first sensor grid from the probability value of a corresponding grid cell in a second sensor grid.

[0154] For example, the above information may be combined in a matrix form.

[0155] Figure 7 is an example of a possible outcome of processing a map set in the difference detection stage 70 .

[0156] like Figure 7 As shown, part a) is an example time frame of the fused map F. Figure 7 A portion of it (bounded by the black box) is enlarged in parts b), c) and d).

[0157] Since different sensors provide different information that may conflict with each other, most of the grid cells in the fused maps F and E may have associated values ​​corresponding to highly conflicting or uncertain information about spatial occupancy, i.e.

[0158] Thus, the difference processing stage 70 may provide a set of detected differences D containing structured information about the presence of conflicts between groups occupying grid maps, in particular describing where the conflicts are and which groups (eg, pairs) of compared grid maps the conflicts relate to.

[0159] Specifically, the detected difference set D includes, for example:

[0160] -First difference Δ 12 , indicating an inconsistency between the first occupancy grid map G1 and the second occupancy grid map G2;

[0161] - Second difference Δ 2i , indicating the second occupied grid map G2 and the i-th occupied grid map G i inconsistencies between;

[0162] -i-th difference Δ i1 , indicating that the i-th occupied grid map G i Inconsistency with the first grid map G1.

[0163] For example, this information can be expressed as 12 , Δ 2i , Δ i1 The detected differences D are stored / encoded in the form of an array of , each storing the result of each comparison of a set of sensor grid cells, eg, containing a "conflict" value for each cell in the respective grid map cell.

[0164] The "conflict" value may be calculated by subtracting the probability value in a map cell from the probability value in a different map cell.

[0165] The effect of the difference processing stage 70 is to facilitate the processing of the difference between the sensors S1, S2, S i The readings provided are Y1, Y2, Y i The generated time fusion grid maps G1, G2, G i In fact, for the “conflicting” values ​​Δ in the difference D set, 12 , Δ 21 , Δ i1The knowledge of this can be used to "hint" the probability of "uncertain" cells of the "fused" maps F, E as "occupied" or "free" through the conflict arbitration stage process 90 to provide a "coordinated" or arbitrated drivable fused map M with a reduced number of conflicting cells.

[0166] like Figure 8a As illustrated, the data fusion stage 40 and the difference assessment stage 70 are coupled to an arbitration stage 90 which receives as input a map set FF which, in the case of the embodiment of FIG. 8 , comprises the BOF data fusion block 50 , a fused map F and a set of detected differences D.

[0167] exist Fig. 9 In the case of the embodiment of , the arbitration stage 90 ′ receives a set FF, which includes the fused map F and the evidence grid E.

[0168] Figure 8 Fig. 9 The illustrated arbitration stage 90, 90' basically comprises an occupation grid processing stage 85 (in Fig. 9 85′), receiving the fused map set FF as input and providing the arbitrated fused map set MM as output, such a stage 85 generally includes a conflict detection stage 83, a conflict resolution stage 87 and a conflict correction stage 89. Fig. 9 In the drawings, variant embodiments are indicated by 90', and corresponding blocks are indicated by a "primary" superscript, such as 85', 83', 89', 87', to illustrate possible different internal implementations of the corresponding functions.

[0169] Depending on the method used in the data fusion processing stage 40 (e.g., BOF fusion 50 and / or DST fusion 60), the conflict arbitration stage 90 may use different methods to resolve potentially conflicting (e.g., contradictory) information provided by sensors regarding the occupancy status of cells of the fusion grid F and / or the evidence grid E.

[0170] Specifically, in Figure 8a In the illustrated variant embodiment, the data fusion processing stage 40 and the BOF fusion processing stage 50 in the difference assessment stage 70 are coupled to the conflict arbitration stage 90 .

[0171] In the embodiment considered, the conflict arbitration stage 90 also comprises a situation evaluation stage 80 based on a probabilistic model or alternatively on coded rules and logical formulas to resolve conflicting cell values ​​in the fused map F by processing the detected set of differences D. Thus, the conflict arbitration stage 90 receives as input the fused map F from the BOF fusion processing stage 50 and the detected set of differences D from the difference detection stage 70.

[0172] 8 , the situation assessment stage 80 includes a conflict assessment stage 84 and a difference-based relationship modeling (DBRM) stage 82. The occupancy grid processing stage 85 includes a conflict detection stage 83, a conflict resolution stage 87 and a conflict correction stage 89.

[0173] The conflict detection stage 83 in the occupancy grid processing stage 85 receives the fused map F from the BOF processing stage 50 and provides as output the map F and / or the conflicting cell set CA and / or a trigger signal.

[0174] Specifically, the conflict detection stage 83 detects the grid cells in the input fusion map F. The operations of marking the conflicting cells CA set are performed within, which can include, for example:

[0175] - The grid cells of the fusion map F have probability values ​​for a certain number of BOF data fused for 50 cycles

[0176] -Grid cells Marked as uncertain within a certain time tk / Unknown / Conflict.

[0177] The conflict detection stage 83 may thus provide as output a set of marked cells CA to signal that a conflict has occurred. Note that multiple marked cells CA within the fused grid F may be marked at time k. Thus, the same marking may be enabled for each set of cells CA within the fused grid F. Alternatively, multiple marking values ​​may be associated with multiple grid cells at the same time k, for example, to distinguish higher "conflict levels" from lower "conflict levels".

[0178] The trigger signal may be generated periodically, ie at a given moment, or according to the probability value in the set of conflicting cells CA. The trigger signal may be used internally or provided to the user processing stage 87,84.

[0179] For example, the conflict detection stage 83 may receive multiple time frames of the fused map F from the BOF processing stage 50 in the data fusion process 40 before applying the marking. The conflict detection stage 83 may wait for a certain time k (e.g., a certain number of fusion processing cycles) before marking the set of conflicting cells CA, and / or wait for the cell probability values ​​to stabilize.

[0180] Alternatively, the existence of a non-empty set of conflicting cells CA itself may be used as a trigger signal.

[0181] In the embodiment considered, the conflict detection stage 83 provides said set of marked cells CA and / or fused maps F and / or trigger signals to a conflict assessment stage 84 in the situation assessment stage 80 and to a conflict resolution stage 87 in the occupancy grid map processing stage 85 .

[0182] In the exemplary embodiment of FIG. 8 , the conflict assessment stage 83 in the situation assessment stage receives as input:

[0183] - a set of detected differences D from the difference evaluation stage 70,

[0184] - a difference-based relational model DM provided by the DBRM stage 82 (e.g. stored in the memory 82),

[0185] - A collection of conflicting cells CA, and / or a fusion map F and / or a trigger signal.

[0186] The difference-based relational model DM provided by stage 82 represents a set of rules to be applied to conflicting grid cells in the fused grid map F and / or the set of conflicting cells CA to perform at least one of the following:

[0187] -Spatial evaluation of conflicting cells, such as cell positions,

[0188] -Evaluation of occupation probabilities, e.g. through difference resolution logic.

[0189] Therefore, the conflict assessment stage 84 is configured to use (eg, combine) the detected set of differences D and the difference-based relationship model DM, thereby providing a method to resolve conflicts that maintains the state of the grid cell "uncertain / unknown".

[0190] More specifically, in order to obtain information to "hint" the occupancy of the ambiguous cell as a first state (e.g., "occupied" ) or a second state (e.g. "idle" state), the conflict assessment phase 84 may employ difference resolution logic as described below.

[0191] With the above reference Figure 7 Similar to the discussion, a first exemplary parsing logic is developed by observing the existence of difference-based relationships between different sensor measurements.

[0192] exist Figure 7 In parts b), c) and d) of Figure 1, area A is a grid cell corresponding to the edge of the road (e.g., a lane). This information is confirmed as a lane by the visual sensor and as a (circular) boundary by the (GPS / GIS) background map. However, the lidar sensor may generate cells of the molten map Classified as "idle" because most of the beam energy is dispersed into the surrounding environment.

[0193] Therefore, if Figure 7 As shown, the difference D in the set of detected differences D between the visual sensor and the lidar (part c) and the background map (GIS / GPS) and the lidar (part d) 2i and D i1 have high discrepancy values, respectively, because the measurements of the two sensor pairs contradict each other. In contrast, the discrepancy Δ between the visual sensor and the GIS map (part b) 12 Very small, because the measurements from both sensors confirm each other.

[0194] Obstacle detection is a matter of concern, particularly with regard to accurately determining the presence of temporary guardrails that may appear on the road and / or lanes that define road boundaries. These objects may be a source of uncertainty as their presence / existence may create / be in conflict in at least one pair of sensor readings.

[0195] For each cell, the presence of the object can be evaluated based on the detected difference D by using the following logic / rules:

[0196] - If the difference Δ 2i and Δ 12 Represents the sensor pairs S1, S2 and S i , S2 are inconsistent, and Δ i1 Indicates the sensor pair S i , S1 are inconsistent, the cell is marked as "occupied".

[0197] - If Δ 2i Indicates the sensor pair S2, S i Inconsistent and Δ 12 Indicates that the sensor pairs S1 and S2 are consistent and Δ i1 Indicates the sensor pair S i , S1 are consistent, the cell is marked as "occupied",

[0198] - In any other case, the cell is marked as "free".

[0199] Thus, the "conflict" value is a parameter of the difference assessment model DM provided by the difference modeling stage 82. The "conflict" may be parameterized in the model DM provided by stage 82. For example, as discussed previously, "conflict" = 0.5 may mean that using the data from the first sensor, the grid cell g ij k is marked as “idle”, and according to the grid map from the second sensor reading, the grid cell g ijk Marked as "occupied".

[0200] The conflict assessment stage 83 provides information about the conflicting cells and the conflict type (e.g., in the data set CR) as an output of the situation assessment stage 80. Thus, as illustrated in FIG8 , the conflict resolution stage 87 in the occupancy grid processing stage 85 receives as input:

[0201] - the data set CR provided by the situation assessment stage 80, and

[0202] - A trigger signal provided by the conflict detection stage 83 and / or the fused map F and / or the set of positions and values ​​of the conflicting cells CA.

[0203] The conflict resolution stage 87 processes the inputs, such as evaluating the probability of conflict as spatiotemporal confidence in the sensor measurements. Thus, an adaptive weight value W may be calculated for further processing. For example, the weight value may include an adaptive weight value W that may be passed "back" to a previous stage, such as via a feedback network to the data fusion processing stage 40, particularly in stage 520 of the BOF fusion stage 50, to be applied to the sensor grid values ​​in the data fusion process.

[0204] The conflict resolution stage 87 processes the input, providing a set of values ​​CC.

[0205] In one or more embodiments, the set of values ​​CC represents a correction δF applied to the fused map F in the case of a feed-forward correction, or a weight correction δW in the case of a feedback correction.

[0206] This will thus "hint" (i.e., push or reallocate) the probability of the conflicting cell entering a defined state (e.g., occupied or vacant state). For example, the resolved grid cell value can be calculated based on a measure of the "conflict" value in the detected difference set D, for example by applying rules in the data set CR provided by the situation evaluation stage 80.

[0207] The output of the conflict resolution stage 87 is transmitted to the conflict correction stage 89.

[0208] Therefore, the conflict correction stage 89 receives:

[0209] - a set of values ​​CC from the conflict correction phase 87,

[0210] - rule dataset CR,

[0211] - A fused map F from the BOF fusion stage 50 in the data fusion processing stage 40 .

[0212] The conflict correction stage 89 provides as output a drivable fused map M, which can be provided to the driver assistance system / interface IA as described above. For example, the fused grid map M is obtained in the conflict correction stage 89 by merging and / or replacing the values ​​of the conflicting grid cells in the fused map F with the corresponding values ​​CC (e.g., resolved grid cell values).

[0213] In the feedforward or feedback variants, respectively, the processing in the conflict correction stage 89 can be expressed as: M = F - δF, the difference between the fused map F and the correction δF of stage 87; or That is, the set of previously calculated weight values The difference between the correction δW of stage 87.

[0214] Therefore, the fused grid map M helps to accurately predict and / or detect the occupancy of obstacles along the driving path of the vehicle V, thereby facilitating online correction of the driving path of the vehicle V through the driver assistance system / interface IA.

[0215] For example, when the vehicle is controlled by an autonomous driving system (not visible in the figure), it can help avoid collisions between the vehicle and obstacles.

[0216] exist Figure 8b In the example, to provide a simplified example, the sensor grid maps G1, G2, G i A grid cell having n=2 rows and m=1 columns, whose processing in the difference phase provides as result a set of detected differences D, for example:

[0217] - The first pair of sensors agree, e.g. Δ 12 =0,

[0218] - The second pair of sensors does not agree, e.g. Δ 2i =1,

[0219] - The third pair of sensors is inconsistent, for example Δ i1 =1.

[0220] Although different calculations may be performed in other embodiments to determine the difference, the calculations shown above, of course, represent simple examples.

[0221] The combination model DM processes the detected set of differences in a conflict assessment phase 84, providing a data set CR value, e.g.

[0222] In addition, the conflict resolution stage 87 may receive initial values ​​of the weights used in the fusion stage, e.g.

[0223] Therefore, the conflict resolution stage 87 may calculate updated weight values, for example to provide to the BOF fusion stage 50 in the raster fusion processing stage 40. For example, the weights may be calculated as follows:

[0224] CC=[w1,w2,w3]

[0225] in:

[0226] -w1=w2=u / u1+u2+u i ;

[0227] -w3=v / u1+u2+u3

[0228] as well as

[0229] -u=CA+CA*CC=0.55,

[0230] -v=CA-CC=0.11.

[0231] Therefore, regardless of the feedforward configuration or the feedback configuration, the occupancy probabilities of the sensor grid maps can be fused and weighted accordingly according to these weights in the fused grid map M.

[0232] In another variant embodiment of the arbitration stage 90, as in Figure 1 , 5 As can be seen in FIG. 8 , the value CC may include an adaptive weight W to be fed back to the BOF fusion process 50 in the data fusion process stage 40, for example, in particular in the stage 520 of the BOF fusion process. This facilitates iterative weighting of W, 520, 50 of the sensor occupancy grid map in the data fusion process 50, for example, to provide a weighted linear opinion pool fusion.

[0233] As mentioned above, in Fig. 9 In the illustrated variant embodiment, the DST fusion processing stage 60 in the data fusion processing stage 40 is coupled to the arbitration stage 90 .

[0234] Specifically, in Fig. 9 In the illustrated variant embodiment, the arbitration stage 90' ​​may include an evidence grid processing stage 85'.

[0235] like Fig. 9 As illustrated in , similar to that discussed previously, the evidence grid processing stage 85' includes a detection stage 83' and a resolution stage 87'.

[0236] In such Fig. 9In the illustrated variant embodiment, when receiving a set FF of fused maps that also includes evidence grid E from the DST process 60, the detection stage 83′ is configured to pass the evidence grid E to the resolution stage 87′, for example, each time the stage 83′ receives evidence grid E, a value is passed and the activation of the stage 87′ is triggered.

[0237] exist Fig. 9 In the illustrated variant embodiment, the conflict resolution stage 87 ′ includes, for example, a storage ratio conflict resolution (PCR) rule, which is suitable for use in conjunction with the DST grid fusion 60 .

[0238] In such Fig. 9 In the illustrated variant embodiment, the decision "logic" about the grid cell occupancy value (e.g., how to "prompt" the probability of a conflicting cell) can be based on a "trust function" Bel(X) with multiple values, such as four values ​​(quality)-(free, occupied, conflicting, unknown), different from the binary Bayesian value free / occupied. Therefore, since it is possible to distribute "trust" between the four "states", DST helps to improve the resolution of the mapping. In fact, this helps to represent the transitions that occur in the environment in a more detailed and realistic way and provide a more reliable map of the environment.

[0239] In fact, even if the GPS information indicates that there are no walls, unless the LiDAR sensor indicates otherwise, it is best to highlight this information by assigning a "conflict" value to distinguish it from the case where both GPS and LiDAR have no information about an area and therefore mark it as "unknown". In BOF fusion 50, in both cases of the previous example, the grid cells of the fused map F are assigned the same probability value (e.g., 0.5) and the same state (e.g., "unknown"). On the other hand, DST can distinguish more "evidence" states, such as the four states idle, occupied, conflict, and unknown.

[0240] However, it has been observed that DST fusion 60 may produce counter-intuitive results when combining information from (highly) conflicting sources, so PCR processing 87' can be used to cope with unpredictable "evidence" grid maps E, extending DST capabilities to automatically resolve potential conflicts.

[0241] For example, one or more embodiments may include an operation 87' of redistributing the conflicting masses, for example, in proportion to the non-empty sets according to all integrity constraints.

[0242] For example, the PCR stage 87′ is configured to calculate the combined combinations of all sources n for all subsets (free, occupied, conflicted, unknown) ∈ A, generating the basic DST formula as shown below:

[0243]

[0244] The total conflict quality can be expressed as follows:

[0245]

[0246] in:

[0247] - Addendum m in the above formula (X1∩X2∩…∩X n ) is the partial conflict mass.

[0248] Therefore, if Fig. 9 As illustrated, in one or more embodiments using the evidence grid E, each partial conflict mass is redistributed 89' to its subsets in proportion to the basic probability mass that has been allocated to these subsets in the conflict resolution phase 87' of the arbitration phase 90'.

[0249] Note again that Figures 8 and Fig. 9 The variant embodiments may be combined in another variant embodiment, where the arbitration stage 90 may be coupled to the BOF fusion process and the DST fusion 60 .

[0250] In fact, using the proportional conflict redistribution rule 87' (PCR for short) on the evidence grid map E and using the rule CR provided by the situation evaluation 80 of the difference D to resolve conflicts are not mutually exclusive approaches.

[0251] Both approaches can be combined in a variant embodiment, including, for example, in the arbitration phase 90:

[0252] - a plurality of specialized parallel phases, each performing a specific function, such as a pair of conflict resolution phases 87, 87';

[0253] - The same stage that performs both operations, for example the same processing stage 87 is equipped to operate on the DST evidence grid E with PCRs 87' and / or on the BOF fusion map F and the rule dataset CR.

[0254] The latter arrangement may prove useful, for example, if the processing stages 50 and 60 are included in the processing data fusion stage 40. In practice, the DST processing 60 and the BF processing 50 may be performed, for example, in parallel in the data fusion stage 40 to improve the consistency of the two methods and reduce each other's weaknesses, for example to promote improved robustness.

[0255] Thus, the conflict arbitration stage 90 may facilitate a consistent and reliable correction of conflicts in grid cells of at least one fused grid map F, E, thereby outputting a drivable fused map M.

[0256] It will be appreciated that the representation of the processing stages discussed above - in particular the data fusion stage 40, BOF fusion 50, DST fusion 60, arbitration stage 90 and / or 90' and difference detection stage 70 - as distinct entities is for the sake of clarity and simplicity only. In one or more embodiments, one or more of those stages may be integrated in the form of multi-function stages and / or circuits, such as in a single processor or DSP.

[0257] Fig.10 is an example diagram of a method 1000 of vehicle environment mapping in accordance with the described embodiments.

[0258] The method 1000 for mapping the environment of a vehicle V (e.g., a land vehicle) includes the following operations:

[0259] - At the beginning of the method operation, S1, S2, S i , and there are multiple sensors S1, S2, S i (e.g. including lidar, cameras and GPS / GIS);

[0260] -1010 indicates that multiple sensors S1, S2, S i Receive input values ​​Y1, Y2, Y i Operations on collections;

[0261] -1030 represents applying the (dispersed) time fusion process 30; 301, 302, 30i to the input values ​​Y1, Y2, Y i The operation of the set is used to obtain the occupied grid maps G1, G2, G i 's respective collections;

[0262] -1040 represents applying the data fusion process 40 to the occupancy grid maps G1, G2, G i The set operation is performed to obtain a set FF of fused occupancy grid maps F and E, for example, using a Bayesian occupancy filter process represented by 1050 or a DS evidence theory process represented by 1060,

[0263] -1070 indicates that by comparing the maps G1, G2, G i An operation of detecting differences (70) using occupancy grid maps in a set (e.g., 2-combinations of occupancy grid maps) to obtain a set D of detected differences,

[0264] -1090 represents the operation of processing 1090 the set of fused occupancy grid maps FF, outputting a set of fused occupancy grid maps MM of the drivable space, the processing operation including performing conflict arbitration 1090a, 1090b in the at least one fused occupancy grid map F, E. Operation 1090a corresponds to the operation performed by the embodiment of the arbitration block 90 in FIG. 8, while operation 1090b corresponds to Fig. 9 The operations performed by the embodiment of the arbitration block 90′ in FIG. 1 , which will also be described in detail below;

[0265] -IA represents an operation of providing the arbitrated fused occupancy grid map M of the drivable space to at least one driver assistance device, the driver assistance device IA being configured to operate according to the arbitrated fused occupancy grid map M of the drivable space.

[0266] As described above, operation 1090 processing particularly includes at least one of the following:

[0267] - feed forward processing 1090a, 1090b,

[0268] - Feedback processing 1090a, including adaptive weighting of sensor occupancy grid cells and / or sensor readings, for example by providing adaptive weights W to the data fusion processing stage 40 (for example to stage 520 of BOF fusion 50).

[0269] As described above, operation 1090a of the feed-forward processing specifically includes:

[0270] - For example, at block 83, in at least one fused map F or E, a set of grid cells classified as uncertain is marked within a certain period of time, thereby obtaining a marked set of grid cells,

[0271] - analyzing said set of detected differences D, for example in a block 84, providing a set of composite rules according to a stored model (for example the model DM) (for example stored in the block 82),

[0272] - applying said set of composite rules (in particular adaptive weighting rules) to the values ​​of said set of marked grid cells in at least one fused map F, for example in block 87, thereby providing a set of values ​​CC (for example composite / corrected grid cell values),

[0273] - providing said composite fused occupancy grid map (M) of the drivable space by incorporating said set of composite grid cell values ​​into marked grid cells in at least one fused map F, for example in box 89 .

[0274] As described above, operation 1090b processes, inter alia:

[0275] - For example, at block 83', transferring the set of grid cells in the evidence grid E that are classified as uncertain within a certain period of time,

[0276] - applying a proportional conflict redistribution of "quality" in the evidence grid E, for example in block 87', thereby providing a set of composite grid cell values ​​CC,

[0277] - providing said composite fused occupancy grid map (M) of the navigable space by merging said composite grid cell value set into a marked grid cell in the evidence grid E, for example in box 89'.

[0278] As described above, the method may include using blocks 50 and 90 for BOF or blocks 60 and 90′ for DST, for example, performing operations 1050, 1090a or 1060, 1090b. However, in another variant embodiment, the data fusion block 40 may include blocks 50 and 60, and the system 100 may include arbitration blocks 90, 90′, which means that the method may include and perform Fig.10 All operations shown, such as operations 1040, 1050, 1070, 1090a and 1090b.

[0279] In summary, one or more embodiments:

[0280] - provide a vehicle (e.g. land vehicle) environment map,

[0281] - includes temporal fusion and data fusion operations, e.g. via Bayesian occupancy filters and / or DS evidence theory,

[0282] - Temporal fusion processing can be performed locally in a single sensor, and single sensor grid maps can be compared in groups of two to analyze differences,

[0283] - further processing of the difference analysis and data fusion results through conflict arbitration processing, such as using difference-based relational models and / or proportional conflict redistribution,

[0284] - conflict arbitration may provide a composite fused grid map, e.g. with new values ​​for conflicting cells provided by the arbitration process,

[0285] The composite fused grid map may be provided to a driving assistance system, for example a vehicle equipped with the system, to perform the method according to the embodiment.

[0286] It will also be understood that the various individual implementation options illustrated throughout the drawings accompanying this specification are not necessarily intended to be employed in the same combination illustrated in the drawings. Therefore, one or more embodiments may employ these (i.e., non-mandatory) options individually and / or in different combinations relative to the combinations illustrated in the drawings.

[0287] The details and embodiments may vary (even significantly) with respect to what has been described purely by way of example, without prejudice to the underlying principle and without departing from the scope of protection, which is defined by the appended claims.

Claims

1. A method (1000) for mapping a vehicle (V) environment, comprising the following operations: - Receive (1010) data from multiple sensors (S1, S2, S i ) input values ​​(Y1, Y2, Y i )gather, - applying (1030) a time fusion process (30, 301, 302, 30i) to the input values ​​(Y1, Y2, Y i ) collection, thereby obtaining their respective occupancy grid maps (G1, G2, G i )gather, - applying (1040, 1050, 1060) a data fusion process (40, 50, 60) to the occupancy grid map (G1, G2, G i ) set, thereby obtaining a fused occupancy grid map set (FF) including at least one fused occupancy grid map (F), It is characterized in that A data fusion process (40) is applied to the occupancy grid maps (G1, G2, G i ) set includes a Bayesian inference / Bayesian occupancy filter BOF fusion processing stage, thereby obtaining a fused occupancy grid map (F); and includes comparing the occupancy grid maps (G1, G2, G i ) set to detect (1070) differences, thereby obtaining the detected differences (D, Δ 12 , Δ 2i , Δ i1 ) set, and the method comprises: processing (1090) the fused occupancy grid map set (FF), the processing (1090) comprising performing conflict arbitration (1090a, 1090b) in the fused occupancy grid map set (FF); and outputting an arbitrated fused occupancy grid map set (MM) comprising at least one arbitrated fused occupancy grid map (M), Wherein performing conflict arbitration (1090a, 1090b) in the fused occupancy grid map set (FF) comprises processing (85; 85') the fused occupancy grid map set (FF), comprising: - marking (83; 83') a set of grid cells (CA) in said at least one fused occupancy grid map (F); - provide (87; 87') a set of composite rules (CR); - applying said composite rule (CR) to at least one fused occupancy grid map (F) in said set of fused occupancy grid maps (FF); - applying the set of composite rules (CR) to grid cells in the at least one fused occupancy grid map (F) thereby providing a set of composite grid cell values ​​(CC); - providing (89) said arbitrated fused occupancy grid map (M), comprising incorporating said set of composite grid cell values ​​(CC) into at least one fused occupancy grid map (F); wherein a set of composite rules (CR) is provided comprising processing (80) at least one of said detected differences (D, Δ 12 , Δ 2i , Δ i1 ) set, including the differences (D, Δ 12 , Δ 2i , Δ i1 ) collection, the model is configured to be applied to the fused occupancy grid map (F) received from the BOF fusion processing stage.

2. The method according to claim 1, characterized in that Conflict arbitration is performed (1090a) including feedback processing.

3. The method according to claim 1, characterized in that A set of composite rules (CR) is provided including applying proportional conflict redistribution rules configured to be applied to the evidence fused raster map (E) received from the DS theory DST fusion processing stage.

4. The method according to claim 1, characterized in that: A data fusion process (40) is applied to the occupancy grid map (G1, G2, G i ) includes the Bayesian inference / Bayesian occupation filter BOF fusion processing stage and the DS theory DST fusion processing stage, as well as By comparing the occupancy grid maps (G1, G2, G i ) set to detect (1070) differences, thereby obtaining the detected differences (D, Δ 12 , Δ 2i , Δ i1 ) set; and processing (80) the detected differences (D, Δ 12 , Δ 2i , Δ i1 ) comprising analyzing (84) the detected differences (D, Δ 12 , Δ 2i , Δ i1 ) set, the model is configured to be applied to the fused occupancy grid map (F) received from the BOF fusion processing stage.

5. The method according to claim 1, characterized in that Applying the time fusion process (30) to the input values ​​(Y1, Y2, Y i ) collection includes multiple sensors (S1, S2, S i ) in each sensor (S1, S2, S i ) The sensor readings are processed locally and decentralized in their respective time fusion stages.

6. The method according to claim 1, characterized in that By comparing the occupancy grid maps (G1, G2, G i ) to detect (1070) differences in the occupancy grid maps in the set including evaluating the occupancy grid maps (G1, G2, G i ) between each pairwise combination of grid cell positions and values ​​in the set of pairwise combinations of groups.

7. The method according to claim 1, characterized in that The method includes providing a fused occupancy grid map of the drivable space to an interface (IA).

8. The method according to claim 1, characterized in that The vehicle (V) is a land vehicle.

9. The method according to claim 1, characterized in that: The arbitrated fused occupancy grid map set (MM) includes an arbitrated fused occupancy grid map (M) of a drivable space.

10. The method according to claim 1, characterized in that The set of grid cells (CA) comprises the set of grid cells (CA) classified as uncertain grid cells in the at least one fused occupancy grid map (F).

11. The method according to claim 1, characterized in that: The stored model (DM) comprises a difference-based relational model (82).

12. The method according to claim 2, characterized in that: The feedback processing includes providing adaptive weights W to the data fusion processing stage to include adaptive weighting of sensor occupation of grid cells and / or sensor readings (W; 520).

13. An environment mapping system (100), comprising: - At least one sensor (S1, S2, S i ), the at least one sensor being configured to collect at least one sensor reading, - a processing circuit coupled to at least one sensor (S1, S2, S i ) to receive the at least one sensor reading, - Driver assistance system interface, Features: The processing circuit is configured to provide a fused occupancy grid map of at least one drivable space to the driver assistance system interface using the method of any one of claims 1 to 12.

14. The system according to claim 13, characterized in that Multiple sensors (S1, S2, S i ) includes at least one of the following: - LiDAR, - GPS detector, providing contextual information, - Vision sensors.

15. A vehicle (V) equipped with a system (100) according to any one of claims 13 to 14 in combination with at least one driver assistance device, the driver assistance device being configured to operate in dependence on the arbitrated fused occupancy grid map (M).

16. A computer program product loadable into a memory of at least one processing circuit and comprising software code portions for performing the steps of the method according to any one of claims 1 to 12 when the product is run on the at least one processing circuit.

Images