Occupancy grid with compact grid levels

Abstract

Method for representing the environment of a vehicle in an occupancy grid, wherein the data are acquired with at least one environment detection system, characterized in that information is stored in a further level only if the information of the grid cell in the first level has a predefinable value, wherein further information is stored in a further level only if a grid cell in a first level is marked as occupied.

Description

[0001] The invention relates to the technical field of representing a vehicle environment, in particular for controlling a driver assistance system.

[0002] A common approach to representing the vehicle's surroundings is an occupancy grid that divides the environment into cells and stores a classification for each cell as "driveable" or "occupied". In addition to driveability, classifications based on other characteristics can also be stored, such as reflected radar energy.

[0003] Besides its good compressibility, an advantage of the occupancy grid is its high level of abstraction, which also enables the fusion of data from various sensors, such as stereo cameras, radar, lidar, or ultrasound. A disadvantage of this abstraction, for example through object classification, is the loss of information required by some applications (driver assistance systems). This includes, for example, height information for estimating the curb's path or for detecting edges on barriers that are incorrectly identified as lane markings.

[0004] Other methods for representing a vehicle's environment do not store a classification of surrounding objects, but rather raw data. This can be done, for example, in a grid where reflected energy is stored for each cell. Disadvantages of representing the environment with raw data include the inability to fuse data from different sensors if they do not provide the same information. For instance, a radar sensor system measures reflected intensity, while a camera sensor system can measure the height of an object quite accurately. Furthermore, the representation is less compressible because the values ​​of neighboring cells typically differ.

[0005] Besides a grid, other simplified representations of a vehicle's environment are known, such as the so-called Stixel world. This is based on polar coordinates; for each polar "ray of view," the radial distance and height of the adjacent object are stored. While this reduces the amount of data, the covered area is limited, and the chosen coordinate system makes sensor fusion more difficult (in addition to differing data).

[0006] DE 102006056835 A1 discloses a method for grid-based processing of sensor signals.

[0007] DE 10049229 A1 describes a method, in particular for improving the reaction capability of vehicles or vehicle drivers when hazards occur in traffic.

[0008] DE 102007012458 A1 shows a method for object formation for environment modeling.

[0009] The object of the present invention is to provide an improved method for representing a vehicle environment in a grid-based environment model.

[0010] The task is examined through the characteristics of independent claims.

[0011] A method for representing a vehicle's environment in an occupancy grid is described, wherein the data is acquired using at least one environment detection system. A key inventive feature is that environment information is stored in more than one layer within at least one grid cell, and that information is only stored in a further layer if the information in the grid cell of the first layer has a predefinable value.

[0012] According to the invention, further information, preferably height information, is only stored in a further layer if a grid cell in a first layer is marked as occupied.

[0013] For a drivable area, the height is irrelevant, as the roadway is on an assumed ground level and therefore does not need to be stored. This can be achieved, for example, by setting it to a predefined value such as 0 or irrelevant.

[0014] Height is only stored for "occupied" cells that represent obstacles or objects. Because this representation means that many cells have the same value, namely irrelevant or 0, etc., the height layer can be significantly compressed.

[0015] Methods for compressing an environmental model are described, for example, in our application WO 2013 / 060 323 A1. The elevation information can be stored with minimal additional resource consumption using the method claimed here, enabling efficient storage and transmission of the environmental model.

[0016] Another advantage of the method presented here is that sensors that do not provide height information, such as a radar sensor system without elevation resolution, can also be integrated into the environment model. In this case, the evaluation of the radar measurements only contributes to the probabilities of "passable" and "occupied," which corresponds to the first level (Level 1) and thus increases the reliability of the classification of obstacles or objects. An improvement in height estimation would not occur through fusion in this case; however, for many applications, an accurate height estimation is less important than a reliable classification as "passable" or "occupied."

[0017] In a preferred embodiment of the invention, at least two environmental sensing systems are provided to acquire information for the occupancy grid. If a first environmental sensing system detects a grid cell as occupied and a second environmental sensing system detects the same grid cell as free, the information from the next level, in particular the height information, is used to decide whether the grid cell is in the free or occupied state.

[0018] Particularly when the Dempster-Shafer theory is used for the occupancy grid, but also in other cases, a conflict can be detected when multiple sensors are fused, if one sensor system identifies the grid cell as occupied and another as passable. Elevation data provides additional information for resolving these conflicts. For example, a curb might be detected as an obstacle by the camera but not by the radar. The camera's elevation data provides information indicating that the object is below the height that the radar system can reliably detect.

[0019] In a further embodiment of the invention, height information is entered into the occupancy grid depending on road height information.

[0020] For sensors that provide height information, such as stereo cameras, monocular cameras with optical flow analysis, lidar, or radar systems with multiple vertical detection ranges, an estimation of the ground plane is necessary to determine the areas where height information does not need to be stored. For example, non-planar ground planes can be estimated based on data from a stereo camera. After noise removal, the current disparity values ​​in the vehicle's orientation are compared with those of a planar environment. Depending on the distance, a higher tolerance is allowed to also capture downward or upward gradients suitable for driving. The area considered suitable for driving is approximated. Based on this approximation, the heights of adjacent objects relative to the road surface can be calculated and stored.

[0021] The elevation information allows for further levels of interpretation. For example, the following situation analysis can be made: While not passable under normal circumstances, it provides an alternative route for emergencies, such as protecting human lives. In an emergency, it might be more advantageous to drive into a ditch or over a high curb than to collide with oncoming traffic or pedestrians. Future automated driving systems may accept damage to the vehicle as a necessary evil to protect lives. Therefore, additional information is an important step in this direction.

[0022] Instead of or in addition to altitude information, further data can be stored in another layer; this could include, for example, intensities from a radar or lidar sensor, but especially also the states of moving objects, in particular a classification of whether they are moving, speed, lateral speed, acceleration or direction of movement.

[0023] Fig. Figure 1 shows an embodiment of the invention with an occupancy grid comprising two additional levels. The occupied cells in level 1 represent a lateral boundary and a leading vehicle. In the vertical level (level 2), data relating to the lateral boundaries and the leading vehicle are stored. In level 3, which contains kinematic data, only the area of ​​the leading vehicle is populated with information. The latter two levels can therefore be compressed significantly.

[0024] For the transmission of kinematic data, calculating the differences between successive kinematic planes, possibly followed by entropy coding, is particularly suitable. Since the kinematic data typically change only slightly (low accelerations / low jerk), a non-equal distribution of the differences is to be expected, with a high probability of low values, which can be readily compressed using entropy coding.

Claims

[1] Method for representing a vehicle's environment in an occupancy grid, wherein the data are acquired using at least one environment detection system, characterized by , that information is only stored in a further level if the information of the grid cell in the first level takes on a predefinable value, whereby further information is only stored in a further level if a grid cell in a first level is marked as occupied. [2] Method according to any one of the preceding claims, characterized by , that height information is stored in the next layer. [3] Method according to any of the preceding claims, wherein at least two environment detection systems are provided to acquire information for the occupancy grid, characterized by, that if a first environmental detection system recognizes a grid cell as occupied and a second environmental detection system recognizes the same grid cell as free, the information from the next level is used to decide whether the grid cell is in the state free or occupied. [4] Method according to any one of the preceding claims, characterized by , that height information is entered into the occupancy grid depending on road height information. [5] Method according to any one of the preceding claims, characterized by that at least one of the following pieces of information is stored in a further layer i) Intensities from a radar or lidar sensor ii) State of a moving object, in particular a classification as moving, speed, lateral speed, acceleration or direction of motion [6] Driver assistance system comprising at least a sensor system for detecting a vehicle environment and an evaluation or control unit for controlling a driver assistance function, which has an electronic memory on which a method according to one of the preceding claims is stored, and a processor for executing a method according to one of the preceding claims.

Images