FR3047589A1 - METHOD AND DEVICE FOR ESTIMATING THE LATERAL POSITION OF A VEHICLE IN A CIRCULATION PATH - Google Patents

Abstract

The invention relates to a method for estimating the lateral position of a vehicle in a traffic lane, comprising the following steps: - the capture of images (10) of the ground continuously, in front and / or at the rear of the vehicle (100), by means of at least one adapted camera, said images (10) corresponding to a ground surface having a width at least equal to a taxiway width and a depth, - the reconstructing an aerial view (11, 12) of the ground by integrating the captured images (10) over time, - detecting at least one line (21, 22) forming a boundary marking a traffic lane in said reconstructed aerial view (11, 12), estimating the relative position of the vehicle (100) with respect to said at least one line (21, 22) in the reconstructed aerial view (11, 12) ).

Description

METHOD AND DEVICE FOR ESTIMATING THE LATERAL POSITION OF A VEHICLE IN A CIRCULATION PATH

[001] The invention belongs to the field of devices for assisting the driving of vehicles, as well as that of autonomous vehicles.

[002] The invention thus relates, in general, devices and methods of driving assistance, as well as such devices and processes implemented in autonomous vehicles.

[003] The invention relates more particularly to a device and a method for estimating the lateral position of a vehicle in a traffic lane, in particular in order to be able to warn the driver in the event of a potentially unwanted crossing of a vehicle. line of demarcation of a taxiway, or for the purpose of processing by an autonomous management system of the pipeline.

[004] Today, motor vehicles include many driver assistance systems. These systems are commonly referred to by the acronym ADAS, well known to those skilled in the art, for Advanced Driver Assistance Systems, meaning advanced driver assistance systems.

[005] Such driving assistance systems equip more and more motor vehicles. Most of the functions provided by these ADAS systems require the vehicle to be able to detect the lane in which it is located.

[006] It should be noted that, for an application in an autonomous vehicle, the reliability, accuracy and availability of position information in the track are immediately crucial for the autonomous driving function is usable.

[007] Today, some vehicles are equipped with a front camera that can provide information on the position of the vehicle in a traffic lane. In addition, methods based on the use of cameras located in various parts of the bodywork of certain vehicles - cameras to assist maneuvers - are currently being developed in order to overcome some of the faults of the front camera, whose quality of the provided information is altered, as is known, by glare phenomena, limited in scope by the state of traffic and weather, but also because of the hypothesis used, and generally incorrect, of "ground plane> >, or because of the hypothesis used, and generally incorrect, horizontal attitude of the vehicle, ... etc.

[008] The document FR2947769_A1 describes a system comprising a camera located at the rear of the vehicle, for tracking the path of the vehicle to send an alert to the driver in case of deviation from an ideal trajectory predetermined. This document is representative of the state of the art.

[009] Thus, in practice, according to the state of the art, the lateral position of a vehicle in a traffic lane is information exploited by several driving assistance functions. These can be informative functions, such as an unintentional line crossing warning: in this case, the system observes the evolution of this position over time to estimate a crossing of the line and to alert the driver. It can also be automation functions that enslave the vehicle to its lateral position in the taxiway, in order to maintain it.

Today, the information relating to the position of the vehicle in a traffic lane is generally determined from the images captured by a front camera. This makes it possible to have long-range images and thus to anticipate the curves. The principle consists in detecting in the image the presence of the lines of separation between traffic lanes and estimating their equations in the reference system of the vehicle, thanks to calibration information of the camera, such as its field, its resolution, its position, its orientation with respect to the vehicle ... etc.

Similar systems and methods are currently developed, following the same principle, but using images from sensors other than the front camera, in particular from a plurality of cameras.

In any case, the existing systems are always directly dependent on the visibility of the lines of separation between traffic lanes in the camera field. Indeed, the detection of the lines and the correct estimation of their equations in the reference system of the vehicle involve the visibility of a sufficient portion of said lines, with minimal contrast. In certain unfavorable conditions, for example at night, in fog or rain, in case of sun or in case of heavy traffic, the visibility of the lines of separation between traffic lanes is not always assured and the performances systems are therefore degraded, the determination of the position of the vehicle in its lane may even be impossible.

Such a disadvantage is particularly detrimental, especially in the context of an implementation planned for a vehicle driving automation system, for which the availability of this information is crucial.

In addition, for all these known or currently developed systems, the camera calibration parameters for calculating the relative position of the lines relative to the vehicle must be finely adjusted. This makes them sensitive to any "decalibration" caused by a shock on a camera, a change of attitude of the vehicle ... etc. Likewise, these systems consider that the ground is a horizontal plane in the reference system of the vehicle, which constitutes an overall erroneous assumption.

In summary, currently developed methods provide, at best, a local correction based on what cameras exploited capture in their fields of vision.

Some major drawbacks of known systems for establishing the position of a vehicle in a traffic lane remain. Thus, the most recent methods do not make it possible to solve the problems due to the temporary absence of marking on the ground, for example linked to the presence of a discontinuous line, whose marking on the ground is not visible at any moment by the cameras implemented, local inconsistencies due to the presence of works, multiple markings, road joints, etc., errors in re-projections due to drifting calibrations, changes in attitude of the vehicle, the unevenness of the soil ... etc.

To mitigate these disadvantages, it may be envisaged to implement temporal filtering and probabilistic approaches but such approaches are by definition based on assumptions and involve finely calibrated systems. Above all, such filtering introduces an additional risk of obtaining an erroneous estimate of the position of the vehicle in a traffic lane.

Thus, to effectively solve the problem raised above, without causing new disadvantages, the present invention proposes to integrate over a period of time images of the ground captured by at least one camera, so as to constitute a " history "of the route taken by the vehicle, potentially over several hundred meters, to eliminate errors due to inconsistencies or local unavailability.

For this purpose, the method for estimating the lateral position of a vehicle in a traffic lane, according to the invention, comprises reconstituting an aerial view obtained by storing, as and when moving of the vehicle, a portion of the ground nearest the vehicle. The use of captured images of a near ground portion of the vehicle provides better image quality for the detection of ground markings, and the storage artificially extends the range of the camera. The detection of line (s) separating the traffic lanes is carried out on the reconstituted aerial view. The aerial view being formed in the vehicle mark, the detected lines can directly estimate the relative lateral position of said vehicle relative to said lines.

More specifically, the subject of the invention is a method for estimating the lateral position of a vehicle in a traffic lane, comprising the following steps: - the capture of images of the ground continuously, at the front and / or rear of the vehicle, by means of at least one adapted camera, said images corresponding to a ground surface having a width at least equal to a taxiway width and a depth, the reconstitution of an aerial view of the ground by integrating the captured images into time, the detection of at least one line forming a marking of the ground of delimitation of a traffic lane in said reconstituted aerial view, - the estimation of the relative position of the vehicle relative to the at least one line in the reconstructed aerial view.

Thus, thanks to the method according to the invention, it is possible to have a reconstituted aerial view corresponding to a long section of the road to detect the separation lines, which minimizes the sensitivity of the process vis-à-vis - local anomalies and provides better detection of the line (s) separating the traffic lanes. The method according to the invention also allows a better evaluation of the lateral position of a vehicle in a traffic lane.

Advantageously, the estimate of the relative position of the vehicle with respect to said at least one line in the reconstructed aerial view comprises the assembly of successive captured images corresponding to a ground surface having a depth, said depth being a function the speed of the vehicle.

Preferably, the detection of at least one line forming a ground marking delimiting a traffic lane in the reconstituted aerial view comprises the implementation of a Hough transform.

According to one embodiment, the step of detecting at least one line forming a ground marking delimiting a traffic lane in said reconstituted aerial view comprises a substep of time filtering.

Preferably, the step of estimating the relative position of the vehicle with respect to said at least one line in the reconstituted aerial view comprises a substep of time filtering.

According to one embodiment, the method according to the invention further comprises a driver warning step when the vehicle crosses said at least one line.

Advantageously, said track width may be equal to five meters.

Advantageously, said depth of the ground surface of the captured images may be less than two meters.

The invention also relates to a device for estimating the lateral position of a vehicle in a traffic lane, comprising at least one camera for continuously taking images of the ground, in front and / or or at the rear of the vehicle, said images corresponding to a ground surface having a width at least equal to a taxiway width and a depth, and a computing unit configured to perform the reconstruction of an aerial view of the ground by integrating captured images into time, detecting at least one line forming a ground marking of a taxiway in said reconstituted aerial view, and estimating the relative position of the vehicle with respect to said at least one line in the reconstructed aerial view, according to the method as briefly described above.

According to one embodiment, said at least one camera is a camera disposed at the rear of the vehicle, able to restore the driver of the vehicle images of the ground at the rear of the vehicle, the captured images corresponding to a surface ground with a width between 4 meters and 6 meters and a depth of between 1.5 meters and 2.5 meters.

Other features and advantages of the invention will appear on reading the detailed description of the embodiments of the invention, given by way of example only, and with reference to the drawings which show: FIG. 1 , the block diagram of the steps of the method according to the invention, • Figure 2, the diagram graphically representing the process of estimating the position of the vehicle in its lane, in accordance with the invention.

In what follows, the described embodiments focus more particularly on an implementation of the device according to the invention within a motor vehicle. However, any implementation in a different context, especially in any type of land vehicle, is also covered by the present invention.

Furthermore, a main embodiment of the method according to the invention comprises the implementation of a reversing camera and / or parking aid, capturing images of the ground at the rear and near the vehicle. However, other configurations of systems implementing the method according to the invention are envisaged, including implementing one or more cameras disposed at other locations of the vehicle, for example at the front.

Figure 1 shows functional blocks embodying the steps of the method according to the invention.

Thus, the method according to the invention provides a first step 1 of capturing images of the ground at the front and / or rear of the vehicle, continuously. According to one embodiment, the captured images come from cameras used elsewhere for parking assistance, namely a camera at the rear and possibly a camera at the front capturing images of the ground corresponding to a floor surface exhibiting a width corresponding at least to a taxiway width, about 5 meters, and a depth of between 1.5 meters and 2 meters, with a typical field of view of the order of 130 °.

From these captured images, the method according to the invention comprises a step 2 of reconstituting an aerial view of the road.

It should be noted that, said aerial view is not intended to be displayed, a fine alignment is not required. According to one embodiment, it is nevertheless provided for a registration of the images as a function of the displacement of the estimated vehicle by means of odometric data obtained using suitable sensors measuring wheel tops or the steering wheel angle for example.

According to one embodiment, each captured image, corresponding to a portion of the ground, is applied to the reconstituted aerial view according to the speed of the vehicle. Thus, considering the use of a single camera disposed at the rear of the vehicle, at a speed of v km / h, the vehicle travels v / 3.6 m / s; in fact, the portion of the image captured at the rear of the vehicle and applied to the reconstructed aerial view must cover the area of the k * v / 3.6 / n meters behind the vehicle, where n is the number of frames per second , and k a "margin" multiplier (for example, k = 1.5).

From there, at 130 km / h and captured 30 images / s, it is necessary to have captured images representing a ground surface having a depth of 1.8 meters. In general, it can thus be observed that the depth of the ground surface of the images captured by the camera, at the front and / or at the rear of the vehicle can always be less than 2 meters.

It should be noted that, according to one embodiment for limiting the computing power required, the number of images applied per second on the reconstituted aerial view is adapted according to the speed of the vehicle so as to obtain images always corresponding to the same floor area.

Furthermore, according to one embodiment, there is provided a recovery between images captured successively, in order to facilitate the detection of separation lines of the traffic lanes and improve the connections between successive images.

From the reconstructed aerial view, the method according to the invention provides a step 3 of detecting one or more ground marking lines embodying the delimitation between the traffic lanes.

There are known techniques for detecting lines in an image. For example, the detection of lines in the reconstructed aerial view can be achieved by implementing a Hough transform. The Hough transform makes it possible to identify all the lines passing through a point (Y, X) of the ground. Conversely, for a pair of coefficients (a, b) defining a line, the Hough transform makes it possible to identify the set of points (X, Y) situated on this line. The coefficients 'a' and 'b' respectively correspond to the angle formed by the line with the Y axis, and the distance from the line to the origin.

It should be noted however that the lines delimiting a traffic lane are not perfect lines. Generally, it is about clothoids, that is to say lines with continuous curvature. In order to optimally exploit the process according to the invention, it is planned to transpose the Hough principle to the analysis of such clothoids.

Indeed, a vehicle in which the method according to the invention is implemented evolves a priori within a traffic lane, substantially parallel to the lines which delimit it. According to one embodiment, the detection of these lines can thus be approximated, in the Hough space, with a search for lines whose coefficients 'a' are close to 0, and the coefficients 'b' are smaller than the distance to which it is desired to detect lines of separation of the traffic lanes, on either side of the vehicle.

This filtering makes it possible to limit the cases of false positives. In the case of a generalized Hough transformation, applied to the clothoids, the principle and the objective of the filtering remain indeed identical.

Moreover, the markings on the ground materializing the lines of separation of the traffic lanes are always clearer than the roadway. Therefore, another filtering consists in considering only the gradients corresponding to clear lines on a darker background, and of minimum size. This simple filtering also makes it possible to limit the cases of false positives.

According to one embodiment, there is further provided a temporal filtering of the line detection algorithm. With reference to FIG. 2, the aerial view 11, reconstituted from captured images 10 integrated in time, becomes the aerial view 12 after temporal filtering, allowing a better identification of the lines 21, 22 for separation between traffic lanes. It should be noted that any other algorithm for detecting lines on the reconstructed aerial view can also be implemented in the context of the method according to the invention.

From the reconstituted aerial view, in which is detected at least one line of separation between traffic lanes, and preferably two lines, on either side of the vehicle, the method according to the invention provides a step 4 estimation of the relative position of the vehicle relative to said line or lines, to deduce the lateral position of said vehicle in its driving lane.

Indeed, when several lines marked by a ground marking are detected, the detection of the lane in which the vehicle evolves is summarized in the identification of the relevant lines, corresponding to the lines closest to the left and to right of said vehicle.

Therefore, by choosing as the origin center of the vehicle, the distance of the vehicle 100 to each relevant line 21, 22 can be calculated in the reference frame R of the vehicle, with reference to Figure 2.

Alternatively, by calculating the width of the track, and considering the center of the track as the origin, the lateral position of the vehicle in the track is for example reduced to a value between -1, reflecting the presence of the vehicle at horse on the left line, and 1, reflecting the presence of the vehicle on horseback on the line on the right.

According to another embodiment, shown in Figure 2, choosing as the origin 0 the center of the vehicle, the lateral position of the vehicle 100 in the lane is calculated according to the width of the vehicle, so -1 corresponds to the presence of the left wheels on the left line, and 1 to the presence of the right wheels on the right line. Thus, the values between -1 and 1, for example 0.55 as in FIG. 2, correspond to a vehicle inside its taxiway, while the values beyond -1 and 1 translate a vehicle straddling two lanes of traffic.

It should be noted that any other algorithm for estimating the relative lateral position of the vehicle with respect to the lines of separation of the traffic lanes, detected in the reconstructed aerial view in accordance with the steps described above, can also be implemented. in the context of the process according to the invention.

In addition, according to one embodiment, provision is made for the implementation of a temporal filtering step, as for the detection of lines in the reconstituted aerial view.

In summary, the present invention provides a method for a reliable estimate of the lateral position of a vehicle vis-à-vis a traffic lane.

The method according to the invention provides in particular the detection of lines of separation between traffic lanes on an aerial view reconstituted over time, in particular by means of one or more camera capturing images corresponding to a ground surface exhibiting a width at least equal to a taxiway width, and a depth potentially less than 2 meters. The captured images are integrated over time to form the reconstructed aerial view.

Thus, thanks to the method according to the invention, it is possible to have a reconstituted aerial view corresponding to a long section of road to detect the lines, which minimizes the sensitivity of the process vis-à-vis the local anomalies and allows better detection of the line (s) separating the traffic lanes.

The risk of glare of the camera implementation is further greatly reduced due to the fact that it is not necessary to capture images more than 2 meters from the vehicle, as explained above.

Furthermore, to implement this method, the invention also provides a system comprising a camera and a suitable computing unit. As described above, said camera may for example be a camera of the type used in existing systems for parking assistance, since the field of view required to implement the method according to the invention is limited to a surface area. soil of the order of 5 meters wide and less than 2 meters deep. Such cameras have a high resolution, and the integration over time of such captured images allows, in the method according to the invention, to obtain a high resolution reconstructed aerial view also, typically of the order of 2 cm / pixel.

The method according to the invention also makes it possible to detect situations in which the channels widen, shrink or disappear.

It is specified, moreover, that the present invention is not limited to the examples described above and is capable of numerous variants accessible to those skilled in the art.

Claims (10)

Claims: A method for estimating the lateral position of a vehicle in a traffic lane, comprising the following steps: - capturing images (10) of the ground continuously, in front and / or at the rear of the vehicle (100), by means of at least one adapted camera, said images (10) corresponding to a ground surface having a width at least equal to a taxiway width and a depth, - the reconstitution of a aerial (11, 12) view of the ground by time integration of captured images (10) over time, detecting at least one line (21, 22) forming a taxiway bounding mark in said reconstructed aerial view (11, 12), estimating the relative position of the vehicle (100) with respect to said at least one line (21, 22) in the reconstructed aerial view (11, 12). The method of claim 1, wherein estimating the relative position of the vehicle (100) with respect to said at least one line (11, 12) in the reconstructed aerial view includes the assembly of captured images (10). ) corresponding to a ground surface having a depth, said depth being a function of the speed of the vehicle (100). The method according to claim 1, wherein detecting at least one line (21, 22) forming a taxiway boundary marking in the reconstituted aerial view (11, 12) comprises work of a Hough transform. 4. Method according to one of the preceding claims, wherein the step of detecting at least one line (21, 22) forming a ground marking delimiting a traffic lane in said reconstituted aerial view (11, 12) includes a substep of time filtering. 5. Method according to one of the preceding claims, wherein the step of estimating the relative position of the vehicle (100) relative to said at least one line (21, 22) in the reconstituted aerial view (11, 12 ) includes a substep of time filtering. 6. Method according to one of the preceding claims, further comprising a driver warning step when the vehicle (100) crosses said at least one line. 7. Method according to one of the preceding claims, said track width being equal to five meters. 8. Method according to one of the preceding claims, said depth of the ground surface of the captured images being less than two meters. 9. Device for estimating the lateral position of a vehicle in a traffic lane, comprising at least one camera for continuously taking images (10) of the ground, in front and / or at rear of the vehicle, said images (10) corresponding to a ground surface having a width at least equal to a taxiway width and a depth, and a calculation unit configured to perform the reconstruction of an aerial view of the ground ( 11, 12) by integrating the captured images (10) in time, the detection of at least one line (21, 22) forming a ground marking of a traffic lane in said reconstituted aerial view (11, 12), and estimating the relative position of the vehicle (100) with respect to said at least one line (21,22) in the reconstituted aerial view (11, 12), according to the method according to one of the preceding claims . 10. Device according to the preceding claim said at least one camera is a camera disposed at the rear of the vehicle, adapted to restore to the driver of the vehicle (100) images (10) of the ground at the rear of the vehicle, the captured images. (10) corresponding to a floor area having a width of between 3 meters and 5 meters and a depth of between 1.5 meters and 2.5 meters.