US-12623658-B2 - Method for assisting in keeping a vehicle on the road in the event of lane narrowing or widening

US12623658B2US 12623658 B2US12623658 B2US 12623658B2US-12623658-B2

Abstract

A method is for detecting the route taken by a motor vehicle in a particular zone within which the number of lanes increases or decreases. The method includes recording position data characterizing the position of the motor vehicle in relation to the lane separation lines marked on the road and calculating, via a recurrent neural network, at least one parameter for the detection of the particular zone, the recorded position data being supplied as input to the recurrent neural network.

Inventors

RENAUD DEBORNE
RAPHAEL QUILLIARD

Assignees

AMPERE S.A.S.

Dates

Publication Date: 20260512
Application Date: 20221124
Priority Date: 20211210

Claims (14)

1 . A method, implemented by a system installed on a motor vehicle with an autonomous driving capability, for detecting, on a road being travelled by the motor vehicle, a singular zone within which a number of traffic lanes changes, the method comprising: capturing, by an image-capturing device, images of a road being travelled by the motor vehicle, reading, based on the captured images, position data characterizing a position of the motor vehicle with respect to lane separation lines marked on the road; and computing, by a recurrent neural network, at least one parameter of detection of the singular zone, the read position data being delivered as input to said recurrent neural network, wherein the position data includes (i) a distance between a main path of the road and a left separation line beside the motor vehicle, (ii) a distance between the main path of the road and a right separation line beside the motor vehicle, (iii) a distance between the main path of the road and a left separation line ahead of the motor vehicle, and (iv) a distance between the main path of the road and a right separation line ahead of the motor vehicle.
2 . The detecting method as claimed in claim 1 , wherein at least two parameters are computed, including a parameter of detection of an increase in the number of traffic lanes, and a parameter of detection of a reduction in the number of traffic lanes.
3 . The detecting method as claimed in claim 1 , wherein said parameter is a level of probability that a zone of the road is the singular zone.
4 . The detecting method as claimed in claim 1 , wherein the recurrent neural network comprises an input layer, an output layer, and at least a first hidden layer comprising an attention mechanism.
5 . The detecting method as claimed in claim 4 , wherein the attention mechanism is of LSTM type.
6 . The detecting method as claimed in claim 4 , wherein said first hidden layer comprises several tens of neurons.
7 . The detecting method as claimed in claim 4 , wherein the output layer is an activation layer and at least one densifying other hidden layer, which is located after said first hidden layer and in which a number of neurons is equal to a number of neurons in the output layer, is provided.
8 . A method for assisting with keeping a motor vehicle in a traffic lane of a road, comprising: the detecting method as claimed in claim 1 ; and determining a steering-angle setpoint for steered wheels of the motor vehicle, depending on the computed parameter of detection of the singular zone.
9 . The assisting method as claimed in claim 8 , wherein, when said parameter indicates that the number of traffic lanes is increasing, the assisting method further comprises selecting, from two right and left lane separation lines located on either side of the motor vehicle, a first line that diverges the least from the motor vehicle and/or from a path of the motor vehicle, and wherein in said determining, the steering-angle setpoint is determined depending on the position of said first line, and independently of the position of the other of the two right and left lane separation lines.
10 . The assisting method as claimed in claim 9 , wherein, in said determining, the steering-angle setpoint is determined by assuming that a width of the traffic lane measured before the singular zone does not vary within the singular zone.
11 . The assisting method as claimed in claim 8 , wherein, when said parameter indicates that the number of traffic lanes is decreasing, the steering-angle setpoint is determined depending on the positions of the right and left lane separation lines located after the singular zone.
12 . The detecting method as claimed in claim 1 , wherein said at least one parameter includes: a parameter of probability of occurrence of road widening, a parameter of probability of occurrence of road narrowing, and a parameter of probability of absence of road widening or narrowing.
13 . The detecting method as claimed in claim 1 , wherein the reading and computing steps are performed at each of a plurality of regular time increments.
14 . A method, implemented by a system installed on a test vehicle with an autonomous driving capability, for parameterizing a computer to assist with keeping a motor vehicle in a traffic lane of a road, comprising: carrying out, in a real environment, a training trip during which the test vehicle is driven over road segments comprising singular zones; capturing, by an image-capturing device, images of a road being travelled by the test vehicle, reading and storing, based on the captured images, position data relating to a position of the test vehicle with respect to lane separation lines marked on a segment of road travelled and predetermined information indicating whether said road segment is a singular zone within which a number of traffic lanes increases or decreases; building a database containing, for various successive times, said position data and said stored information; training a recurrent neural network making it possible to determine at least one parameter of detection of the singular zone of the road within which the number of traffic lanes increases or decreases, said recurrent neural network receiving said database by way of training datum; and saving the recurrent neural network to the computer of the motor vehicle, wherein the position data includes (i) a distance between a main path of the road and a left line beside the test vehicle, (ii) a distance between the main path of the road and a right line beside the test vehicle, (iii) a distance between the main path of the road and a left line ahead of the test vehicle, and (iv) a distance between the main path of the road and a right line ahead of the test vehicle.

Description

TECHNICAL FIELD OF THE INVENTION The present invention generally relates to the field of advanced driver-assistance systems. It more particularly relates to a method for detecting, by means of a computer located on-board a vehicle being driven on a road, each singular zone of the road within which the number of traffic lanes increases or decreases. It also relates to a method for assisting with keeping the motor vehicle in its traffic lane and to a method for parameterizing the on-board computer. PRIOR ART Some motor vehicles are currently equipped with an image-capturing device designed to detect the position of road line markings delineating traffic lanes. The position of the lines, around and in front of the vehicle, makes it possible to implement guidance systems to assist with driving. These guidance systems may for example be LCA systems (LCA standing for Lane Centering Assist) that keep the vehicle equidistant between a right line and a left line. Road widening, for example due to a splitting of a traffic lane into two, an intersection or a freeway exit, may nevertheless cause these guidance systems to make mistakes. Thus, in the case where a single traffic lane splits into two parallel traffic lanes, there is a risk of the LCA system guiding the vehicle to straddle the two new lanes, before abruptly returning the vehicle to one of the two lanes when the marking between these two lanes appears. The same type of problem is liable to occur when a road narrows. Poor detection of line position may also cause guidance systems to make mistakes. For example, if the vehicle confuses a shadow or reflection with a line, the guidance system may cause the vehicle to follow this shadow or reflection and therefore guide the vehicle out of its lane. Generally, lane widening and narrowing may fool guidance systems and/or cause them to follow paths that are sub-optimal in terms of safety, distance travelled or energy consumed. To limit these drawbacks, document FR3109920 discloses a method comprising steps of: detecting, on the basis of the variation in the distance separating the right and left lines, widening of said lane, then, when widening is detected,acquiring the path of the motor vehicle,selecting, from the right and left lines, a line diverging least from the path, andguiding the motor vehicle depending on said position of the selected line. This method, although it often proves effective, sometimes malfunctions in ways that it would be desirable to prevent. PRESENTATION OF THE INVENTION More precisely, the present invention provides a solution that allows widening and narrowing of a road to be better detected. Thus, according to the invention, a method is provided for detecting a singular zone within which the number of traffic lanes increases or decreases, comprising steps of: reading position data of the motor vehicle being driven on the road with respect to lane separation lines marked on the road, andcomputing, by means of a recurrent neural network, at least one parameter of detection of a singular zone, the read position data being delivered as input to said recurrent neural network. Thus, the invention makes provision to employ a very particular type of neural network, namely a recurrent neural network, to detect singular zones. Specifically, it turns out that this type of network, once trained, allows highly satisfactory detection as the vehicle moves along a road and the width of the road varies. The use of a non-recurrent neural network would not allow results as reliable as those obtained here to be achieved. This solution, using neural networks, is perfectly applicable in many situations, for example in countries where traffic drives on the right or left of the road. It gives good results when it is coupled with the LCA function, not only on roads where the number of traffic lanes actually varies, but also on roads where the number of traffic lanes remains constant but where certain line markings are not sufficiently visible, so that the computer located on board the vehicle mistakenly perceives a variation in the number of traffic lanes. Specifically, by virtue of the invention, the number of errors in detection of singular zones is reduced. Furthermore, in case of error, the guidance solution employed in the LCA function makes it possible to pass through the singular zone without any problem and without requiring this function to be deactivated. In this way, in practice, it is rare for the LCA function to be disabled randomly and in a manner which is uncomfortable to the driver. The following are other advantageous and non-limiting features of the detecting method according to the invention, which features may be implemented individually or in any technically possible combination: at least two parameters are computed, including a parameter of detection of an increase in the number of traffic lanes, and a parameter of detection of a reduction in the number of traffic lanes;the recurrent neural