GB2593215A

GB2593215A - Method for detecting a drivable space

Info

Publication number: GB2593215A
Application number: GB2004069.7A
Authority: GB
Inventors: Gatina Claudiu
Original assignee: Continental Automotive GmbH; Continental Automotive Romania SRL
Current assignee: Continental Automotive GmbH; Continental Automotive Romania SRL
Priority date: 2020-03-20
Filing date: 2020-03-20
Publication date: 2021-09-22
Also published as: GB202004069D0

Abstract

The invention refers to method for detecting a drivable space in an environment of a vehicle, comprising the following steps: providing an occupancy grid around an origin point corresponding to the vehicle position, comprising a plurality of grid cells, each grid cell being classified as occupied or free based on a probability of occupancy corresponding to a associated position in the environment of the vehicle, starting from an initial grid corner, detecting a first occupied cell as a reference cell with a reference vertex, looping around the reference cell to detect further reference cells and vertices, meaning adjacent occupied cells and their respective vertices and contain them into connected components that contain maximal horizontal and vertical chains of occupied cells adjacent to the respective reference cells, computing edge points for each connected component, connecting the computed edge points starting from an initial corner of the occupancy grid, and obtaining a free space polygon, reducing the obtained free space polygon to a drivable space polygon visible from the origin point.

Description

Method for detecting a drivable space The present invention generally relates to a method for detecting a drivable space in an environment of a vehicle, in connection with an advanced driver assistance system.

For autonomous navigation or driving assistance in environments with surrounding mobile objects, vehicles have to estimate continuously the drivable space, namely the free space in front of the vehicle where it is possible to define a safe trajectory. Radar and vision systems have been widely used for this purpose and dedicated algorithms have been developed. For example, patent applications published under E23029602A, E23514648A, W02013087067A, W02016165704A, US2019228237A address the detection of free space information by the use of occupancy grid-based maps. Such occupancy grids allow an easy fusion of data from different sensors, but usually they have large memory requirements. However, the published solutions are subject to further improvement, since the environment representation has to be accurate and characterized by a low computational cost.

The underlying problem is to find -in a fast and reliable manner 25 -a visibility polygon that encloses the area visible (and therefore drivable) from a point in an occupancy grid.

Therefore, the object of the invention is to solve the deficiencies of the mentioned prior art and to provide a method 30 able to compute the drivable free space in front of a vehicle based on an occupancy grid.

This object is achieved according to the invention by means of the technical characteristics mentioned in the independent claim, namely a method for detecting a drivable free space.

Further advantageous embodiments are the subject matter of the dependent claims.

The subject-matter of the present invention is a method for detecting a drivable space, said method comprising the following 10 sequence of steps: - providing an occupancy grid around an origin point corresponding to the vehicle position, comprising a plurality of grid cells, each grid cell being classified as occupied or free based on a probability of occupancy; -startingfromaninitialgridcorner, detecting a first occupied cell as a reference cell; - looping around the reference cell to detect adjacent occupied cells and contain them into connected components, meaning closed polygons that contain maximal horizontal and vertical chains of 20 occupied cells adjacent to the reference cell; - computing edge points for each connected component; - considering the grid corners as edge points, connecting all the grid corners with the computed edge points starting from the initial corner of the occupancy grid, and obtaining a free space 25 polygon; - reducing the obtained free space polygon to a polygon visible from the origin point and providing the visible polygon as drivable space.

The main advantage of the invention is the fact that the drivable space is detected in a fast and efficient manner, using very little extra memory. The method only loops around the connected components (i.e. polygons) from the occupancy grid, without wasting much time inside them. The output of this method (a polygon visible from a point) is optimized for extraction from the occupancy grid; moreover, it is a more precise alternative to other models of free space detection (for example, to the fan model of free space detection).

Further special features and advantages of the present invention can be taken from the following description of an advantageous embodiment by way of the accompanying drawings: -Fig. 1 shows a flowchart of a method for detecting a drivable space underlying an embodiment of the present invention; - Fig. 2 shows an image provided with an occupancy grid mapping an environment of a vehicle, - Fig. 3 shows the occupancy grid wherein a reference grid cell is taken as pivot for further looping, according to an embodiment of the invention; -Fig. 4 shows a drawing of a section of the occupancy grid, wherein a reference vertex is taken as pivot for further looping, according to an embodiment of the invention; - Fig. 5 is a schematic image of a drivable space detected according to the invention.

Referring now to Fig. 1, there is shown a flowchart for performing a method for detection for drivable space in an environment of a vehicle, according to invention. In the following, the individual steps for detecting the drivable space will be explained in more detail.

In a first step Si an occupancy grid is provided around an origin point corresponding to the vehicle position. Such an occupancy grid is provided by a driving assistance system and represent the environment by metric grids of cells and estimate the probability of any cell for being,occupied" depending on sensor readings. Such sensors may comprise, for example, optical sensors, radar sensors, ultrasonic sensors or other sensors capable to scan the environment and provide occupancy information.

Fig. 2 shows an exemplary image of such an occupancy grid OG provided around vehicle V. Grid cells not only contain information about the presence of objects Ob, but also about the absence (free-space) in terms of probability values. When such a probability value assigned to a grid cell is higher than a predetermined threshold, then the grid cell is considered as "occupied"; when the probability value is below the threshold, then the grid cell is considered as "free". Usually, such an occupancy grid is a two-dimensional map projected from a three-dimensional representation of environment (for example, an image taken by a camera); along the probability values for each grid cell (not illustrated), there is possible to show the distance in meters (as for example, 10, 20, ...70m) in a coordinate system X-Y having as origin the vehicle position, where X is a longitudinal direction (the same with the vehicle V's direction) and Y is a lateral direction. Grid resolution, shape and position relative to the origin point are user-defined but may not change over time when processing image series.

In a second step 82, there are searched occupied cells, starting from an initial corner of the grid. Actually, this step consists into associating a graph to the grid, according to the following assumptions: to each occupied cell a vertex is associated (vertex being the consecrated term in geometry for corner, designating an angular point of a polygon. For consistency reasons, the term vertex is used in this description from now on); two vertices are connected through an edge if and only if the corresponding cells are adjacent (meaning having an edge in common); a chain of cells is a list of adjacent cells; a subset of cells is connected if and only if a corresponding sub-graph is connected; and a subset of cells is a connected component if and only if it is a maximal connected subset of cells -otherwise said, a connected component is a polygon that consists of non-intersecting edges joined pair-wise to form a closed path.

There is a triple consisting of a cell in the grid, a specific vertex of the cell and a specification answering the question: is the pivot the cell or the vertex? Fig. 3 shows how the reference cell is used as pivot. When a first occupied cell is detected, it becomes the reference cell; the search for occupied cells continues by looping around the reference cell in a counterclockwise direction, starting from its reference vertex (for example, the bottom right corner of the reference cell) and finding the next vertex, referred to in Fig. 3 as,moving vertex". Considering an edge determined by the reference vertex and the moving vertex as a,moving edge", it is searched if another grid cell, adjacent to the moving edge, is occupied or unoccupied (free or outside the grid). The answer to this question opens two options: - when the adjacent cell is free or outside the grid, then the reference vertex is updated to the moving vertex, whereas the reference cell remains the same; - when the adjacent cell is occupied, then the reference vertex is used as pivot; in this case, looping around this vertex is made in a clockwise direction, startinc with the reference cell, until is reached a cell which is free or outside the grid.

Fig. 4 shows how the reference vertex is used as pivot. Ina third step 33 the search continues by looping around the reference cell to detect further reference cells and vertices, meaning adjacent occupied cells and their respective vertices and contain them into connected components. For each occupied cell around the pivot vertex, the maximal vertical and horizontal chains of cells that are occupied are marked as visited and contained in a dedicated connected component. When, while looping around, it is reached a cell which is free or outside the grid, the pivot moves back to the cell, and the reference cell becomes the last occupied cell when looping around the reference vertex. During this step, the looping ends when the initial reference vertex is reached again, successfully computing the outline (closed polygon) of a connected component from the grid. It can be proved that any reference vertex has at least one unoccupied adjacent cell, so this step never loops indefinitely.

After all connected component from the grid are computed, a fourth step follows, namely 34, of computing edge points for each connected component, meaning that extreme boundary points are computed. In fact, the boundary points are the projections from the origin to the grid margins of a leftmost and a rightmost points of each polygon, which are added as vertices of the respective polygon. Ina fifth step 35, considering an initial corner of the grid as an edge point too, the computed edge points are linked and a free space polygon is obtained. Finally, a sixth step 36 is performed, reducing the obtained free space polygon to a polygon visible from the origin point and providing the visible polygon as a drivable space, actually the output of the method itself.

Fig. 5 shows the output of the method for detecting a drivable space, where the edge points are referenced by Ep, and the drivable space is the dotted polygon DS. Outside this polygon are tipically multiple other polygons (as for example, Fl, P2 illustrated in Fig. 2 or P in Fig. 5). In order to obtain such a visible polygon, an angular sweep may be used, for example.

A computer implemented program (an algorithm) written in C++ and "translating" the method steps is fast (runs in 0(n logn), where n is the number of edge points computed from the occupancy grid). A run time analysis shows that the algorithm based on the above-described method (polygon model) runs in 0.551ms and uses very little extra memory (0(n)).

However, while certain embodiments of the present invention have been described in detail, those familiar with the art to which this invention relates will recognize various alternative designs and embodiments for practicing the invention as defined by the following claims.

List of reference numbers X Longitudinal axis Lateral axis OS Occupancy grid Ob Objects detected as occupied grid cells 21, P2, P Polygons modeling objects or non-drivable space Ep Edge points DS Drivable space V Vehicle

Claims

Patent claims 1. A method for detecting a drivable space in an environment of a vehicle, comprising the following steps: (31) providing an occupancy grid around an origin point corresponding to the vehicle position, comprising a plurality of grid cells, each grid cell being classified as occupied or free based on a probability of occupancy; (52) starting from an initial grid corner, detecting a first 10 occupied cell as a reference cell with a reference vertex; (53) looping around the reference cell to detect further reference cells and vertices, meaning adjacent occupied cells and their respective vertices and contain them into connected components that contain maximal horizontal and vertical chains of occupied cells adjacent to the respective reference cells; (54) computing edge points for each connected component; (55) connecting the computed edge points starting from an initial corner of the occupancy grid, and obtaining a free space polygon; (56) reducing the obtained free space polygon to a drivable space 20 polygon visible from the origin point.
2. Method for detecting a drivable space, according to claim 1, wherein the occupancy grid comprises a two-dimensional representation of grid cells specifying a probability of occupancy corresponding to an associated position in the environment of the vehicle.
3. Method for detecting a drivable space, according to claims 1 and 2, wherein the looping around the reference cell is made in 30 a counterclockwise direction starting from an initial reference vertex.
4. Method for detecting a drivable space, according to claims 1 and 3, wherein while looping around the reference cell, it is reached a free cell or a cell outside the grid, a next vertex in the counterclockwise direction is updated as reference vertex.
5. Method for detecting a drivable space, according to claims 1 5 and 3, wherein, while looping around the reference cell, an occupied cell is reached, the reference vertex remain the initial reference vertex.
6. Method for detecting a drivable space, according to claim 5, 10 wherein the looping continues around the initial reference vertex in a clockwise direction until a free cell or a cell outside the grid is reached.
7. A non-transitory computer readable recording medium, 15 comprising computer executable instructions which, when executed by a processor, cause the processor to perform a method according to claim 1.
8. An advanced driving assistance system configured to perform 20 driving assistance for a vehicle based on the drivable space provided by a method according to claim 1.