Grid-based environment model for a vehicle

Abstract

Method for a sensor system for environmental detection for a motor vehicle with evaluation unit, wherein a grid-based environmental model is calculated and at least one discrete value is assigned to each grid cell and Discretization is achieved by comparing environmental data and at least one threshold value. characterized by the fact that at least one discrete value of the grid is transmitted via a data transmission system in a vehicle to an evaluation or control unit.

Description

[0001] The invention lies in the technical field of environmental detection using sensors in a vehicle. The sensor data is suitably processed to provide an environmental model for further applications, particularly driver assistance functions.

[0002] State-of-the-art technology includes a grid-based environment model. This involves dividing a vehicle's environment into cells, and storing one or more features for each cell to describe the environment. Furthermore, object-based environment modeling methods that provide the position data of detected objects are also state-of-the-art.

[0003] One advantage of grid-based methods is that the entire environment of the vehicle is described. Information from sensor data is provided in the environment model for clear, occupied, and unknown areas, rather than only for areas occupied by objects (object-based methods). The explicit modeling of clear areas is becoming increasingly important, as many newer assistance functions, such as evasive maneuvering assistants, require information about a clear area that can be used as maneuvering space for the vehicle. Grid-based methods for describing the environment generally require a larger amount of data than object-based methods, and thus, an application in the vehicle requires greater memory resources and transmission bandwidth.

[0004] WO 2007 / 028 932 A1 shows a method for steering assistance comprising a sensor for object detection.

[0005] DE 10 2007 012 458 A1 relates to a method for object formation for environment modelling, in which handling of information from a sensor is carried out taking into account a segmentation threshold that is at least partially non-linear.

[0006] DE 10 2010 006 828 A1 describes a method for creating a model of the environment of a vehicle.

[0007] DE 10 2007 013 023 A1 discloses a recursive method for creating a raster map with grid cells, each of which is assigned a total value of a probability for the occupancy of the respective grid cell by at least one object.

[0008] It is the object of the present invention to compress the data of a grid-based environment model in such a way that use in the vehicle and, in particular, transmission of the environment model data via conventional vehicle bus systems between control units is possible.

[0009] The problem is solved by the features of the independent claims.

[0010] A method is claimed for a sensor system for environment detection in a motor vehicle, wherein a grid-based environment model is calculated. A grid-based environment model is based on dividing the environment of a vehicle into cells and storing a characteristic describing the environment for each cell. Storing raw sensor data or storing a classification for each cell as a probability value, e.g., storing the probability that a cell is occupied or unoccupied, requires a high storage capacity, which also requires a high-bandwidth bus system for transmission to or from a control unit.

[0011] According to the invention, each grid cell is assigned at least one discrete value (class). In particular, the discrete value or class is a measure of whether an object is located at the position represented by the grid cell and whether this object can be overrun, or the probability of the object being overrun.

[0012] The discretization or assignment of a class is achieved by evaluating environmental data, which contains information about detected environmental objects, and at least one threshold. The number of thresholds can be defined arbitrarily and influences the number of possible classes. With one threshold, a maximum of two discrete values ​​or classes can be defined, and with two thresholds, a maximum of three discrete values ​​or classes can be defined.

[0013] Preferably, a lossless compression method is applied to the discrete values ​​of a grid, in particular prior to transmission via a data transmission system in the vehicle. This also includes grid values ​​that were compressed prior to transmission.

[0014] In a preferred embodiment of the invention, compression is achieved by decorrelation of the temporal dependence, i.e., by calculating the difference between temporally successive discrete values ​​of the grid cells. In particular, only the difference values ​​are transmitted via a data transmission system in the vehicle.

[0015] In a further positive embodiment of the invention, a lossy compression method is applied to the discrete values ​​of a grid, in particular before transmission via a data transmission system in the vehicle.

[0016] In particular, the discrete values ​​of the grid, which represent areas further away from the vehicle, are compressed more strongly.

[0017] Preferably, alternatively or additionally, the discrete values ​​of the grid which represent features facing away from the vehicle are compressed more strongly. For example, the rear side of a construction site wall which is arranged facing away from the vehicle is compressed more strongly than the front side of the construction site wall which is arranged facing the vehicle. A particularly computationally efficient embodiment of the invention provides that the assignment of the discrete value or class is carried out on the basis of the environmental data with the aid of an assignment table, wherein the assignment table is stored in the memory of the evaluation unit. In a preferred embodiment of the invention, in order to change the content of a plurality of grid cells, the assignment table is changed. As a result of the changed assignment rule which corresponds to a discrete value or class,By assigning a changed value range of the environment data to a class, the content of all cells can be changed at once without recalculating and overwriting the discrete values ​​in each grid cell.

[0018] In a preferred embodiment of the invention, the environment model based on discrete values ​​in the grid is transmitted to an evaluation or control unit via a data transmission system in a vehicle. The data transmission system is preferably a bus system in the vehicle that connects at least two evaluation or control units. Preferably, one evaluation unit creates the grid-based environment model, and another evaluation or control unit uses the environment model to control a driver assistance function.

[0019] The invention claimed here comprises a sensor system for object detection for a vehicle with a first computing and evaluation unit on which a method as described above is stored.

[0020] In particular, a second evaluation or control unit and a data transmission system are provided, wherein the first evaluation or control unit is connected to the second evaluation or control unit in a vehicle via the data transmission system.

[0021] In a preferred embodiment of the invention, the first evaluation or control unit is provided for creating an environment model and the second evaluation or control unit is provided for controlling a driver assistance system.

[0022] The invention is explained in more detail below using exemplary embodiments and illustrations.

[0023] A grid-based environment model is based on dividing the environment of a vehicle into cells and storing a characteristic describing the environment for each cell. Storing raw sensor data or storing a classification for each cell as a probability, e.g., the probability that a cell is occupied or unoccupied, requires a high storage capacity, which also requires a high-bandwidth bus system for transmission to or from a control unit. A direct application of a compression method to a calculated grid often does not result in a high compression factor, since the probabilities of neighboring cells often differ only minimally. An example of such a grid is in Fig. 1 shown on the left.

[0024] The use of environmental data, particularly in a vehicle for a driver assistance system, generally requires a binary decision in which a decision is made based on a threshold value for the probability as to whether the cell is occupied and thus not passable by a vehicle, or free and thus passable by a vehicle. Discrete decision classes are therefore important which, for example, indicate the occupied / free states for a grid cell with a numerical value or similar. In a positive embodiment of the invention, the creation of the environmental model is carried out by a first evaluation or control unit by calculating the binary values ​​of the grid cells, or discrete values ​​of the grid cells in the case of more than two decision classes, and then transmitted to a second evaluation or control unit.

[0025] In this exemplary embodiment, the second evaluation and control unit is used to control driver assistance functions, such as the output of a braking, steering, light control, or warning signal, and the first evaluation and control unit is the evaluation unit of a sensor system for environmental detection. After discretization or classification of the values ​​stored in the grid cells, large areas with high spatial correlation are available. This is exemplified in Fig. 1. In Fig. 1 on the left is a grid before the discretization of the grid values ​​and in Fig. 1 shows a grid with discretized values. In Fig.1 on the right, there are now two states: empty or filled grid cells, which form contiguous regions. Here, high data compression can be achieved by applying well-known compression methods for spatial decorrelation, such as run-length coding or quad-trees, to the discrete values.

[0026] The values ​​of the grid cells are updated at predetermined time intervals. There is generally a high temporal correlation between successive values ​​of the grid cells in binary (discretized) representation, since even with the integration of new measurement data, although the probability in a cell changes, the assignment to a discrete class based on the threshold value often does not. In a preferred embodiment of the invention, a further strong compression of the data is achieved by decorrelation of the temporal dependence, ie the subtraction of successive grids. This is exemplified in Fig. 2. In Fig. 2 on the left shows a grid with discretized values ​​of the measurement cycle n. In Fig.2 On the right, a grid with discretized values ​​of the following measurement cycle n+1 is shown. If one now calculates the difference between the corresponding grid cells "Difference Grid (n+1) - Grid (n)", one finds that only the discretized value of 3 grid cells has changed. These cells are in the Fig. 2 marked on the right with a bold border.

[0027] In addition to a lossless compression method, or even on its own, a lossy method can be used in a further exemplary embodiment. This can be used to further reduce the data rate, but also to maintain a constant data rate after compression, as is typically required for automotive applications. Areas further away from the vehicle are compressed more heavily, since the required accuracy of the environment modeling decreases with the distance to the vehicle (e.g. parking assistance compared to lateral guidance on motorways). Furthermore, features facing the vehicle are particularly relevant, so that alternatively or additionally features facing away from the vehicle can be compressed more heavily (e.g. front / rear of a construction site wall). A four-element tree is particularly suitable for this, in which areas further away or closer to the vehicle are compressed more heavily.Such lossy compression can be achieved in areas facing away from the vehicle by limiting the tree depth.

[0028] By transmitting only differential values, further computing time can be saved on the application side, i.e., during evaluation or function control. In particular, the differential data can be transmitted for recalculations compared to the previously valid values.

[0029] An important point of the invention is therefore the application of thresholding before the application of compression methods, since only after thresholding is there a high spatial and temporal correlation, which enables high compression factors.

[0030] It is advantageous that by shifting the threshold value generation from the function to the calculation of the environment model, compression methods can be effectively applied to the grid-based environment model, thus enabling a transmission of the grid-based environment model via automotive bus systems.

[0031] Further advantages can be a reduction in memory requirements in the function control unit as well as a reduction in the required CPU resources, since after applying a compression method, contiguous areas that are equally classified can be processed with one operation instead of cell by cell. For example, in Fig. right to calculate two new values ​​instead of 25 in Fig. left.

[0032] In a preferred embodiment of the invention, the assignment of the discrete value is carried out based on the environmental data using an assignment table, which is stored in the memory of the evaluation unit. The assignment table thus stores the threshold values ​​that enable the assignment to a discrete value. An assignment table is provided here as an example. Table 1 Cell value 0 % - 25 % 25 % - 50 % 50 % - 75 % 75 % - 100 % Class 1 2 3 4

[0033] In particular, to change the values ​​of a plurality of grid cells, preferably all grid cells, only the assignment table with the threshold values ​​is changed. In addition to a linear mapping for the discretization as shown in Table 1, a problem-adapted discretization (e.g., logarithmic) can be used, which uses large discretization steps in less relevant regions of the continuous input values ​​and can thus resolve the relevant regions more finely with the same number of discrete classes.

[0034] To change the contents of all cells, the value assignment in the table is changed. By assigning the new values ​​in Table 2 to the cells, the probability is reduced in all cells without having to adjust the cells in the grid. Table 2 Cell value 0 % -25% 25 % -50% 50%-75% 75 %-100 % Class 1,2 3 4 5

[0035] Since a discrete domain of definition as the input of the application functions for the update results in a finite number of elements in the value domain, it is possible to calculate the values ​​in advance and store them in tables, thus saving further computing time. Another advantage of representing functions in tables is the easy verification of the input and output values. Furthermore, special cases can be handled by appropriate entries in the table.

Claims

[1] Method for a sensor system for environment detection for a motor vehicle with an evaluation unit, wherein a grid-based environment model is calculated and at least one discrete value is assigned to each grid cell and a discretization is carried out by comparing environmental data and at least one threshold value characterized by , that at least one discrete value of the grid is transmitted via a data transmission system in a vehicle to an evaluation or control unit. [2] Method according to claim 1, characterized by that a lossless compression method is applied to the discrete values ​​of a grid. [3] Method according to claim 2, characterized by that a compression of the data is achieved by decorrelation of the temporal dependence. [4] Method according to one of the preceding claims, characterized in that a lossy compression method is used. [5] Method according to claim 4, characterized by that the discrete values ​​of the grid, which represent areas further away from the vehicle, are compressed more [6] Method according to claim 4, characterized by that the discrete values ​​of the grid representing features facing away from the vehicle are more compressed. [7] Method according to one of the preceding claims, characterized by that the environment model and preferably a data compression is calculated before data transmission in a vehicle. [8] Method according to one of the preceding claims, characterized by that the assignment of the discrete value is carried out on the basis of the environmental data with the aid of an assignment table, whereby the assignment table is stored in the memory of the evaluation unit. [9] Method according to claim 1, characterized by that to change the values ​​of a plurality of grid cells, in particular all grid cells, only the assignment table is changed. [10] Device comprising a sensor system for object detection for a vehicle, a first computing and evaluation unit on which a method according to one of claims 1-9 is stored [11] Device according to claim 10, further comprising a second evaluation or control unit and a data transmission system, wherein the first evaluation or control unit is connected to the second evaluation or control unit in a vehicle via the data transmission system. [12] Device according to claim 10 or 11, wherein the first evaluation or control unit is provided for creating an environment model and the second evaluation or control unit is provided for controlling a driver assistance system

Images