Ascertaining a number of traffic lanes and lane markings on road segments

Abstract

A method is described for creating observation data, in particular by at least one vehicle, traveled road segments being ascertained by the vehicle, lanes of the road segments traveled by the vehicle being ascertained by the vehicle, and the ascertained road segments together with the ascertained traveled lanes being transmitted as observation data from the vehicle to an external server unit. A method for ascertaining a number of traffic lanes, to a system, to an external server unit, and to a control unit are also described.

Claims

1. A method for automatically maintaining a digital map of roads in an up-to-date form using a plurality of vehicles that each (1) carries out a host lane localization that identifies (a) respective lanes on which the respective vehicle is traveling over time and (b) respective pairs of lane markings of the respective lanes and (2) transmits pieces of information identifying the lanes and pairs of lane markings, the method comprising: receiving, by an external server unit, the pieces of information transmitted by the vehicles; maintaining, by the external server unit, for each of a plurality of road segments of the digital map a respective matrix that includes a respective count for each of the pairs of lane markings separately for each of the respective lanes identified in the pieces of information transmitted by the plurality of vehicles; for each of the respective lanes identified in the pieces of information transmitted by the plurality of vehicles, incrementing, by the external server unit, the respective counts of the pairs of lane markings each time the respective count is identified for the respective lane in the transmitted pieces of information; using a trained neural network of the external server unit to perform, for each of the road segments, a respective lane classification that is updated over time based on values of the counters of the respective matrix of the respective road segment; and updating, by the external server unit, the digital map based on the lane classifications.

2. The method as recited in claim 1, wherein the host lane localization is carried out using camera-based systems of the vehicles.

3. The method as recited in claim 1, wherein the identifications of the respective pairs of lane markings includes an identification of a respective number and/or kind of lane markings on a right side and a left side of the respective vehicle.

4. The method as recited in claim 1, wherein the pieces of information are transmitted to the external server unit in batches after respective completed trips.

5. The method as recited in claim 1, further comprising removing, by the external server unit using the neural network, measuring errors from at least one of the matrices.

6. The method as recited in claim 1, wherein a road is divided into the road segments, the road segments being of equal length or of different lengths.

7. The method as recited in claim 1, wherein the lane classification includes classifying a number of lanes present in the lane segment based on the values of the counters associated with respective lane numbers identified in the pieces of information received by the external server unit from the vehicles.

8. A system for automatically maintaining a digital map of roads in an up-to-date form, the system comprising: an external server unit; and a plurality of vehicles communicatively coupled to the external server unit wherein: each of the vehicles includes a communication unit, at least one sensor, and a control unit that is configured to: carry out a host lane localization using the at least one sensor by which the control unit identifies (a) respective lanes on which the respective vehicle is traveling over time and (b) respective pairs of lane markings of the respective lanes; and transmit, to the external server unit and using the communication unit, pieces of information identifying the lanes and pairs of lane markings; the external server unit is configured to maintain for each of a plurality of road segments of the digital map a respective matrix that includes a respective count for each of the pairs of lane markings separately for each of the respective lanes identified in the pieces of information transmitted to the external server unit by the plurality of vehicles; for each of the respective lanes identified in the pieces of information transmitted to the external server unit by the plurality of vehicles, the respective counts of the pairs of lane markings is incremented each time the respective count is identified for the respective lane in the transmitted pieces of information; and the external server unit is configured to perform for each of the road segments a respective lane classification that is updated over time based on values of the counters of the respective matrix of the respective road segment, and update the digital map based on the lane classifications.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 shows a schematic representation of a system according to one specific embodiment of the present invention.

(2) FIG. 2 shows a schematic diagram to illustrate a method according to one specific embodiment according to the present invention.

(3) FIG. 3 shows an exemplary observation matrix for a road segment.

(4) FIG. 4 shows a schematic road segment for which the observation matrix was created based on observations by vehicles from FIG. 3.

(5) In the figures, the same design elements in each case have the same reference numerals.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

(6) FIG. 1 shows a schematic representation of a system 1 according to one specific embodiment of the present invention.

(7) System 1 includes a plurality of vehicles 2, which transmit observation data to an external server unit 4.

(8) Vehicles 2 include a camera-based measuring device 6. Measuring device 6 is designed in the form of a front camera having corresponding activation. Measuring device 6 is connected to a control unit 8 in a data-conducting manner.

(9) Control unit 8 includes an integrated GPS sensor which is used, together with measuring device 6, for a lane-accurate self-localization.

(10) As a result of measuring device 6, vehicles 2 are able to identify which roadway markings are situated on the left side and right side of vehicle 2 on the roadway. Based on the GPS sensor and/or based on the ascertained roadway marking, vehicle 2 is able to estimate on which traffic lane it is presently situated. The roadway marking may preferably take place by an evaluation of video data or image data of measuring device 6. Measuring device 6 may be a camera-based measuring device 6 for this purpose. Furthermore, vehicle 2 may ascertain the presently traveled or already covered road segment.

(11) The ascertained measuring data of measuring device 6 and of control unit 8 may be transmitted as observation data via a wireless communication link 10 to the external server unit. A communication unit 12 of control unit 8 establishes communication link 10 with a communication unit 14 of server unit 4.

(12) The transmitted observation data of a plurality of vehicles 2 are transformed in processing unit 16 into road segment-specific observation matrices and subsequently analyzed by a neural network 18. The results of neural network 18 may be assigned by external server unit 4 to a digital road map 20.

(13) FIG. 2 shows a schematic diagram to illustrate a method 22 according to one specific embodiment according to the present invention.

(14) Each vehicle 2 records 23 the segment traveled by this vehicle 2. For this purpose, it is recorded for the driven sequence of roads in each case on which lane vehicle 2 was situated. The roads may be known to vehicle 2 from an onboard map, for example. The information as to the lane on which vehicle 2 passed the road may take place by a host lane estimation. Furthermore, lane changes may be registered by control unit 8. With the aid of measuring device 6, vehicle 2 may ascertain what lane markings are present on both sides of vehicle 2. The host lane estimation may, for example, take place with the aid of GPS data and/or by an evaluation of video data or image data, which were recorded, for example, by camera-based measuring device 6.

(15) The host lane estimation and the lane marking identification may be subject to errors. The corresponding road segments subject to errors which are recorded by vehicles 2 all have ascertained observation data R. In particular, observation data R may include a great amount P of all traveled road segments, a great amount S of all possible lanes of the road segment, possible geographical positions P, and possible lane markings M. As a result, the following relationship applies:
R∈(P×S×G×M×M)

(16) Lane marking M is taken into consideration twice here, so that a distinction is made between a possible left-side and right-side lane marking. Possible lane markings M may, for example, be solid, dotted, solid in color and the like. Geographical position G may be implemented in the form of WGS85 coordinates.

(17) For example, it is possible to infer, based on the following observation, data of a possible lane-accurate route r subject to errors

(18) r=((p.sub.0, s.sub.0, g.sub.0, m.sub.1,1, m.sub.1,2)

(19) (p.sub.0, s.sub.1, g.sub.1, m.sub.2,1, m.sub.2,2)

(20) (p.sub.1, s.sub.2, g.sub.2, m.sub.1,3, m.sub.1,3)

(21) (p.sub.1, s.sub.3, g.sub.3, m.sub.4,1, m.sub.4,2)

(22) that a vehicle 2 at g.sub.0 on road p.sub.0 used lane s.sub.0, and at g.sub.1 on p.sub.0 changed from lane s.sub.0 to s.sub.1. Thereafter, at g.sub.2 a change was carried out from road p.sub.0 lane s.sub.1 to road p.sub.1 lane s.sub.2, and at g.sub.3 on p.sub.1 from lane s.sub.2 to lane s.sub.3. In this, on (p.sub.0, s.sub.0), m.sub.1,1 was identified as the left marking and m.sub.1,2 as the right marking, m.sub.2,1 and m.sub.2,2 were identified for (p.sub.0, s.sub.1). In this, p.sub.0, p.sub.1 correspond to ∈ P, s.sub.0, s.sub.1, s.sub.2, s.sub.3 to ∈ S, g.sub.0, g.sub.1, g.sub.2, g.sub.3 to ∈ G, and m.sub.1,1, m.sub.1,2 to ∈ M.

(23) These observation data R are subsequently transmitted 24 from vehicles 2 to external server unit 4. Observation data R may be collected in server unit 4. Server unit 4 includes an electronic road map 20. Road map 20 neither has to be up-to-date nor has to include pieces of information about the roads for this purpose. As a result of method 22, all relevant pieces of information about the lanes of road map 20 may be generated in an automated manner.

(24) Each road in digital map 20 which is to be annotated with pieces of lane information is divided 25 into road segments in its longitudinal direction. For example, the road segments may have a length of 15 m. The road segments may also have a length which is adapted as a function of the situation.

(25) Thereafter, the observation data are extracted road segment-wise for each of these road segments and transformed 26 into observation matrices B. For example, observation data R may include a plurality of elements r.sub.1, . . . , r.sub.m E R, where m=10000, for example, these elements r describing observations on the corresponding road segment. The pieces of information from each road segment form a dedicated observation matrix.

(26) Each observation matrix includes entries B.sub.i,j. The entries of observation matrix B represent frequencies with which lane-lane marking combinations (i, j) occurred in the observations. FIG. 3 shows such an observation matrix B. In particular, the observation matrix shows lanes S identified by vehicles 2 in the form of numbers, and roadway markings M identified on both sides.

(27) Observation matrix B shown in FIG. 3 was ascertained from a road segment 21 shown schematically in FIG. 4. Road segment 21 is designed as a six-lane road.

(28) Observation matrix B is subsequently analyzed 27 by a neural network 8 which assigns these observations to one of multiple classes. Each of these classes describes an option as to how a road may be unambiguously divided into lanes S using different lane markings M. If the class of the road segment 21 is known, the corresponding properties regarding the number of the lanes and kind of lane markings may be added 28 to the digital map 20.

(29) Upon entry of an observation matrix B, neural network 18 may output the corresponding road class on which these observations have been made. A finite number of road classes exist, each road class describing an unambiguous combination of lane markings.

(30) Used neural network 18 may be designed as a so-called “feedforward neural network” having multiple hidden levels. Moreover, the observation matrices may be normalized in a value range [0, 1]. For example, neural network 18 may include x.Math.y input nodes and o output nodes, o representing the number of all (useful) road classes, x the maximum number of lanes, and y the number of all combinations of road markings of one lane. According to observation matrix B shown in FIG. 3, x=15 and y=9. The road classes are completely and unambiguously numbered even before the training phase of neural network 18 and are each assigned to exactly one of the output nodes of neural network 18.