FR2670404A1 - Device and method for automatic classification of independent vehicles on the move - Google Patents

Abstract

The device comprises a sensor assembly (CAP) placed in a predetermined relationship with respect to the roadway (CH), and comprising a first optical barrier (BO1) mounted at height transversally to the movement of the vehicle (VA), able to deliver first signals (S1) representing the shadow cast from the vertical edge of the vehicle (VA) which lies within the said barrier at every instant, and second means (BO2) which deliver second signals (S2) representing the speed of the vehicle (VA) on traversing the said first barrier (BO1). Pre-processing means (PUT) then combine the first (S1) and second (S2) signals so as to supply third signals (S3) representing the shadow cast by the vehicle (VA) in its entirety. A processing unit (UT) finally receives the third signals (S3) and, on the basis of predetermined information placed in memory, relating to predetermined criteria of the group comprising at least the length, the height, the number of axles, the height at the level of at least one axle, classifies the vehicle (VA) in question into one and only one from among several predetermined categories.

Description

Device and method for automatic classification on the fly of autonomous vehicles
The present invention relates to automatic classification on the fly of autonomous vehicles.

It finds an application in the automatic classification on the fly of autonomous vehicles at highway toll points.

Generally, the pricing of autonomous vehicles at highway toll points is based on the membership of the current vehicle in one and only one category from a pre-established set of possible categories for example motorcycle, car, car with trailer, pickup truck, and truck.

Most often, such storage is facilitated by the constraint imposed on the drivers of the vehicles themselves to drive their vehicle to the highway toll points associated with the category of their vehicle.

However, to avoid possible fraud, it is generally required as many toll points as there are categories and / or the intervention of at least one human operator who performs and / or controls such classification.

The Applicant has posed the problem of providing an automatic classification device on the fly of autonomous vehicles making it possible to remedy this drawback.

Thus, a first object of the invention is to provide an automatic classification device on the fly of autonomous vehicles making it possible to automatically and dynamically establish the category of an autonomous vehicle when it passes in front of the device.

Another object of the invention is to provide an automatic classification device on the fly of autonomous vehicles making it possible to be operational at all times, night and day.

Yet another object of the invention is to provide a device allowing dynamic automatic classification with a very short response time.

Finally, another object of the invention is to provide a device offering a reliable automatic classification.

The present invention relates to an automatic classification device on the fly of autonomous vehicles.

According to a general definition of 1 invention, the device comprises - a sensor assembly placed in a predetermined relationship with respect to the roadway, and comprising at least first means forming a first optical barrier with a network of photosensitive elements, raised in height transversely to the movement of the vehicle , and suitable for delivering first signals representing the shadow cast by the vertical section of the vehicle which fits into said barrier at any time, and second means suitable for delivering second signals representing the speed of the vehicle when crossing said barrier first barrier, - pre-processing means capable of combining the first and second signals to provide third signals representing the shadow cast of the vehicle as a whole, and - a processing unit receiving the third signals and clean, from predetermined information stored, relating to criteria predetermined res of the group comprising at least the length, the height, the number of axles, the height at the level of at least one axle, to place the vehicle in progress in one and only one among several predetermined categories.

According to a preferred embodiment of the invention, the second means form a second optical barrier with a network of photosensitive elements, mounted substantially horizontally transversely to the movement of the vehicle, and suitable for delivering the second signals representing the progress of the vehicle relative to the first barrier.

In practice, the second optical barrier is placed at a height relative to the roadway of the order of 20 cm.

Advantageously, the second optical barrier is subdivided into a second head optical barrier placed upstream of the first barrier, and having a predetermined length and into a second tail barrier downstream of the first barrier and having a predetermined length.

In the case of an application relating to the automatic classification on the fly of autonomous vehicles at motorway toll points, the length of the second head barrier satisfies a predetermined condition relating to a distance chosen between the first axle of a predetermined vehicle and the front end of said vehicle on the one hand, the length of the second tail barrier satisfies a predetermined condition relating to a chosen distance between the last axle of the predetermined vehicle and the rear end of said vehicle on the other part, while the sum of the respective lengths of the second head barrier and the second tail barrier satisfies a predetermined condition relating to a distance chosen between the first and last axles of said predetermined vehicle.

According to one aspect of the invention, the first means forming a first optical barrier comprise - a first light source capable of generating a first light beam - first focusing means for focusing said first light beam thus generated in the inputs of a first plurality of optical fibers, - a first pillar mounted in height, intended to support the outputs of the first plurality of optical fibers respectively distributed at a predetermined spacing along said first pillar, and respectively delivering first substantially horizontal light rays in a substantially plane vertical with respect to the roadway, and substantially transverse with respect to the movement of the vehicle, a second pillar mounted in height, transversely opposite the first pillar and intended to support the inputs of a second plurality of optical fibers respectively distributed at a pre-spacing determined along said second pillar to receive the shadow cast by the vertical edge of the vehicle which fits into said inputs of the second plurality of optical fibers at each instant, - first opto-electronic reception means with network of elements photosensitive connected to the outputs of the second plurality of optical fibers and capable of delivering the first signals representing the shadow cast by the vertical section of the vehicle thus received.

According to another aspect of the invention, the second means forming a second optical barrier comprise - a second light source suitable for generating a second light beam, - second focusing means for focusing said second light beam thus generated in the inputs of a third plurality of optical fibers, a first bar mounted horizontally and intended to support the outputs of the third plurality of optical fibers respectively distributed along said first bar at a predetermined spacing, and respectively delivering second light rays substantially horizontal in a plane substantially horizontal with respect to the roadway and substantially transverse with respect to the movement of the vehicle, - a second bar mounted horizontally, transversely opposite the first bar and intended to support the inputs of a fourth plurality of fibers optics respectively distributed along said second bar at a predetermined spacing to receive the shadow cast by the horizontal edge of the vehicle which is registered in said inputs of the fourth plurality of optical fibers at each instant, - second opto-electronic reception means to network of photosensitive elements connected to the outputs of the fourth plurality of optical fibers and capable of delivering the second signals representing the shadow cast of the horizontal section of the vehicle thus received, that is to say the advancement of the vehicle relative at the first barrier.

According to a variant of the present invention, the first means forming the first optical barrier comprise - a third light source capable of generating a third light beam, - third focusing means for focusing said third light beam thus generated in the inputs of a fifth plurality of optical fibers, - a third pillar mounted in height and intended to support the outputs of the fifth plurality of optical fibers respectively distributed along said third pillar at a predetermined spacing, and respectively delivering third substantially horizontal light rays in a plane substantially vertical with respect to the roadway, and substantially transverse with respect to the movement of the vehicle, a fourth pillar mounted in height, transversely opposite the third pillar and intended to support a first set of reflective elements distributed along the didi t fourth pillar for reflecting the shadow cast by the vertical edge of the vehicle which is reflected in said first set of reflective elements at all times, - a fifth pillar mounted vertically opposite the fourth pillar and intended to support the entrances of a sixth plurality of optical fibers respectively distributed along said fifth pillar at a predetermined spacing to receive the shadow cast by the vertical edge of the vehicle which fits into said inputs of the sixth plurality of optical fibers at each instant, and - fourth opto-electronic reception means with a network of photosensitive elements connected to the outputs of the sixth plurality of optical fibers and suitable for delivering the first signals representing the shadow cast by the vertical section of the vehicle thus received.

Advantageously, the fifth and sixth pluralities of optical fibers constitute one and only one seventh plurality of optical fibers, the outputs of which are supported by one and only one sixth pillar mounted in height and opposite transversely to the fourth pillar, and provision is made to have between the third focusing means and the inputs of the seventh plurality of optical fibers of the first separating optical means own on command to transmit the third light beam into the inputs of the seventh plurality of optical fibers and to transmit the drop shadow inscribed in the outputs of the seventh plurality of optical fibers to the fourth opto-electronic reception means.

According to a variant of the invention, the second means forming a second optical barrier comprise - a fourth light source suitable for generating a fourth light beam, - fourth focusing means for focusing said fourth light beam thus generated in the inputs of a eighth plurality of optical fibers, a third bar mounted horizontally and intended to support the outputs of the eighth plurality of optical fibers respectively distributed along said third bar at a predetermined spacing, and respectively delivering fourth substantially horizontal light rays in a plane that is substantially horizontal with respect to the roadway, and substantially transverse with respect to the movement of the vehicle, - a fourth bar mounted horizontally, transversely opposite the third bar and intended to support a second set of reflective elements distributed over thealong said fourth bar at a predetermined spacing to reflect the shadow cast by the horizontal edge of the vehicle which is reflected in said second set of reflective elements at all times, - a fifth bar mounted horizontally, transversely opposite to the fourth bar and intended to support the inputs of a ninth plurality of optical fibers respectively distributed along said fifth bar at a predetermined spacing to receive the shadow cast by the horizontal edge of the vehicle which fits into said inputs of the ninth plurality of optical fibers at each instant, - fifth opto-electronic reception means with network of photosensitive elements connected to the outputs of the ninth plurality of optical fibers and suitable for delivering the second signals representing the shadow cast by the horizontal edge of the vehicle thus received.

In practice, the eighth and ninth pluralities of optical fibers constitute one and only one tenth plurality of optical fibers, the outputs of which are supported by a single sixth bar mounted horizontally and opposite transversely to the fourth bar, and it provision is made to have between the fourth focusing means and the inputs of the tenth plurality of optical fibers second optical separating means specific to command to transmit the fourth light beam into the inputs of the tenth plurality of optical fibers and to transmit the shadow range registered in the outputs of said tenth plurality of optical fibers to the fifth opto-electronic reception means.

Advantageously, at least one of the first to fourth light sources consists of a light-emitting diode having a wavelength of the order of 600 to 800 nanometers.

According to another embodiment of the invention, the photosensitive elements are of the charge transfer type.

According to another important characteristic of the invention, the third signals representing the shadow cast of the vehicle as a whole are stored in a matrix memory of pixels distributed in N rows and P columns, and the processing unit comprises processing means images of said matrix memory suitable for extracting at least one vector component representative of at least one essential and distinctive characteristic of the shadow cast of the vehicle as a whole with a view to the automatic classification of said vehicle into one and only one of several predetermined categories.

In practice, the image processing means comprise - first means for cutting the matrix memory into a chosen number of horizontal bands each having a chosen number of lines, - first calculation means suitable for calculating the number of pixels at l excited state (value 1) for each horizontal band and - first means suitable for associating with the number of pixels in the excited state thus calculated a first vector component representative of the vertical distribution of the shadow cast of the vehicle as a whole.

According to another aspect of the invention, the image processing means comprise - second means for cutting the matrix memory into a chosen number of vertical bands each having a chosen number of columns, - second calculation means suitable for calculating the number of pixels in the excited state for each vertical strip, and - second means suitable for associating with the number of pixels in the excited state thus calculated a second vector component representative of the horizontal distribution of the shadow cast by the vehicle in his outfit.

According to another embodiment of the invention, the image processing means comprise - third means for cutting the matrix memory into a chosen number of blocks of rectangular configuration, - third calculation means suitable for calculating the number of pixels in the excited state for each block, and - third means suitable for associating with the number of pixels in the excited state thus calculated a first vector component representative of the surface distribution of the shadow cast by the vehicle as a whole.

According to another embodiment of the invention, the image processing means comprise fourth calculation means suitable for calculating the center of gravity of the matrix memory relative to the subset of its elements in the excited state (value 1).

According to another characteristic of the invention, the image processing means comprise: - first scanning means capable of scanning circularly at constant angle the outline of the shadow cast of the vehicle as a whole from a chosen point from the matrix memory, - first measurement means suitable for measuring the length of the segments at constant angle thus obtained having for first end the point chosen and for second end the intersection of said segments with the contour, and - fourth means suitable for associate with the distances of the segments thus measured a fourth vector component representative of the topology at constant angle of the shadow cast of the vehicle as a whole.

According to another characteristic of the invention, the image processing means comprise: - second scanning means suitable for scanning a selected point of the matrix memory from a plurality of equidistant points chosen from the contour of the drop shadow of the vehicle as a whole, - second measuring means suitable for measuring the length of the segments thus obtained having for the first end the chosen point of the matrix memory and for the second end the equidistant points chosen from the contour of the shadow cast by the vehicle in as a whole, and - fifth means suitable for associating with the distances from the segments thus measured a fifth vector component representative of the topology at constant distance from the shadow cast of the vehicle as a whole.

Advantageously, the image processing means also comprise means suitable for determining, for the shadow cast of the vehicle as a whole, the invariants by translation, by rotation and by homothety.

Finally, the image processing means comprise means suitable for determining the number of axles of the vehicle to be processed.

According to another essential characteristic of the device according to the invention, the processing unit is based on an automatic classification network of neuronal type, provided with information drawn from prior learning carried out, for each category of vehicles to be treated, on a selected subset of vehicles from this category, and tested on all of the vehicles to be treated.

In practice, the first to fifth vector components representative of the shadow cast by a vehicle to be treated, thus extracted from the matrix memory by the image processing means, are applied to the inputs of the automatic classification network of neural type according to an order. chosen which then ranks from these vector components, the current vehicle in one and only one category among several predetermined categories.

The invention also relates to a dynamic automatic classification method for autonomous vehicles.

According to the invention, the method comprises the following steps a) prepare a data set concerning all the vehicles to be treated, b) provide an automatic classification network of neural type, configured in learning mode, c) for each category of vehicles to be processed, using the data set thus prepared, carry out learning on a selected subset of vehicles of this category, d) using information from the learning thus carried out, test on all the vehicles to be treated and start again at step c) for poorly tested vehicles.

According to the invention, step a) comprises the following steps a1) prepare a set of drop shadows relating to all of the vehicles to be treated a2) extract at least one vector component representative of the essential and distinctive characteristic of each drop shadow and a3) constitute a data set concerning all of the vehicles to be treated from said vector component thus extracted.

Other characteristics and advantages of the invention will become apparent in the light of the detailed description below, and of the appended drawings in which - Figure 1 is a schematic view of a sensor assembly placed in predetermined relation to the roadway according to the invention - Figure 2 is a schematic view of the automatic classification device according to the invention in which the sensor assembly operates in transmission mode; and - Figure 3 is a schematic view of the automatic classification device according to the invention in which the sensor assembly operates in reflection mode.

In FIG. 1, the reference DC designates an automatic classification device on the fly of autonomous vehicles according to the invention.

The automatic classification device DC comprises a sensor assembly CAP placed in a predetermined relationship with respect to the roadway CH.

The CAP sensor assembly comprises a first optical barrier Bol with a network of photosensitive elements ELV, mounted vertically transverse to the movement of the vehicle VA.

The first optical barrier BO1 delivers first signals S1 representing the shadow cast by the vertical section of the vehicle VA which fits into said first barrier BO1 at each instant.

Second means BO2 deliver second signals S2 representing the speed of the vehicle when crossing the first barrier BO1. In practice, the second means BO2 form a second optical barrier with a network of photosensitive elements ELH, mounted substantially horizontally transversely to the movement of the vehicle VA.

It should be noted that the second signals S2 which represent the shadow cast by the horizontal section of the vehicle VA which is inscribed in the second barrier BO2 at each instant, also represent the progress of the vehicle
VA compared to the first Bol barrier.

PUT preprocessing means combine the first and second signals S1 and S2 to provide third signals S3 representing the shadow cast of the vehicle as a whole.

Finally, a processing unit UT receiving the third signals S3 ranks the vehicle VA in progress in one and only one among several predetermined categories based on predetermined information stored, relating to predetermined criteria of the group comprising at least the length, the height, the number of axles, the height at at least one axle.

In the case of an application relating to the automatic and dynamic classification of autonomous vehicles at motorway toll points, the second optical barrier BO2 is subdivided into a second head optical barrier BO2T placed upstream of the first barrier BO1, and in a second tail barrier BO2Q downstream of the first barrier BO1.

Advantageously, the second optical barrier BO2 is placed at a relatively low height relative to the roadway to allow detection of the front of any vehicle regardless of its height.

For example, the second optical barrier BO2 is placed at a height relative to the roadway of the order of 20 cm.

To avoid losing information on the front of the vehicle, the Applicant has defined conditions which must be verified by the respective lengths of the second head barrier BO2T and the second tail barrier BOQ2.

Thus, the length L1 of the second head barrier BO2T must verify a predetermined condition relating to a chosen distance between the first axle of a predetermined vehicle and the front end of said vehicle.

Next, the length L2 of the tail gate BO2Q must verify a predetermined condition relating to a distance chosen between the last axle of the predetermined vehicle and the rear end of said vehicle.

Finally, the sum of the lengths L1 and L2 must verify a predetermined condition relating to a distance chosen between the first and last axles of a predetermined vehicle.

For example, the length L1 of the second head barrier
BO2T has a value greater than 2.80 m which corresponds to the greatest length recorded between the first axle of a predetermined vehicle and the front end of this vehicle.

For example, the length L2 of the second tail barrier
BO2Q has a value greater than 4.70 m which corresponds to the greatest length recorded between the last axle of a predetermined vehicle and the rear end of said vehicle.

For example, the sum of the lengths L1 and L2 can be of the order of 9 m which corresponds to the greatest length recorded between two axles of a predetermined vehicle.

We now refer to FIG. 2 which schematically represents an automatic classification device according to the invention in which the sensor assembly CAP described with reference to FIG. 1 operates in transmission mode.

A light source SL1 generates a first light beam F1. Focusing means MF1 of the lens type for example focus said light beam F1 in the inputs of a plurality of optical fibers FO1 grouped together in the form of a plastic strand TO1. At the end of the strand TO1, the plurality of optical fibers FOl is subdivided into a first subfamily of optical fibers lFOl and into a second subfamily of optical fibers 2FOl.

The pillar P1 of a gantry PQ supports the outputs SlFOl of the optical fiber subfamily lFOl respectively distributed at a predetermined spacing along the pillar
P1. SlFOl outputs deliver light radiation
RL1V substantially horizontal in a plane substantially vertical with respect to the roadway CH, and substantially transverse with respect to the movement of the vehicle VA.

The pillar P2 of the portal PQ, transversely opposite to the pillar P1 supports the inputs EFO2 of a plurality of optical fibers F02 respectively distributed at a predetermined spacing along the pillar P2 to receive the shadow cast by the vertical section of the vehicle VA which s '' registered in the EFO2 inputs of the plurality of optical fibers
F02 at all times. The plurality of optical fibers F02 is for example wrapped in a strand TO2 of plastic material.

The outputs SF02 of the plurality of optical fibers F02 are respectively connected to optoelectronic reception means ROE1 with a network of photosensitive elements which deliver the first signals S1 representing the shadow cast by the vertical section of the vehicle VA.

The pillars P1 and P2 constitute the first optical barrier Bol which detects the presence of a VA vehicle and measures its instantaneous vertical cross-section. As soon as the presence of the vehicle VA is detected, the instantaneous vertical cross-section will be sampled in constant advance of the vehicle VA using information from the second optical barrier BO2.

The second optical barrier BO2 is substantially identical to the first barrier BO1, it differs from it only in its relative position relative to the roadway CH. More specifically, the second barrier BO2 is substantially disposed in a horizontal plane relative to the roadway CH while the first barrier BO1 is substantially arranged in a vertical plane with respect to the roadway CH.

Advantageously, the light source used by the optical barrier BO2 is the same as the light source SL1 used by the optical barrier BO1.

A horizontally mounted BH1 bar supports the outputs
S2FO1 of the subfamily of optical fibers 2FO2 respectively distributed according to a predetermined space along the bar BH1. The outputs S2FO1 respectively deliver light radiation RL1H substantially horizontal in a plane substantially horizontal with respect to the roadway CH and substantially transverse with respect to the movement of the vehicle VA.

A bar BH2 mounted horizontally, transversely opposite to the bar BH1 supports the EFO3 inputs of a plurality of optical fibers respectively distributed along said bar BH2 at a predetermined spacing to receive the shadow cast by the horizontal edge of the vehicle which fits into the inputs of the plurality of optical fibers
F03 at all times.

For example, the plurality of optical fibers F03 is wrapped in a strand T03 made of plastic.

Opto-electronic reception means with a network of photosensitive elements ROE2 connected to the outputs SF03 of the plurality of optical fibers F03 deliver the second signals S2 representing the shadow cast of the horizontal section of the vehicle thus received, that is to say tell the progress of the vehicle compared to the first optical barrier BO1.

As mentioned above, the sampling of the instantaneous vertical cross section of the vehicle is stopped when an absence of vehicle is detected by the optical barrier BO1.

We now refer to Figure 3.

We find in Figure 3 the essential and constitutive elements of the sensor assembly described with reference to Figure 2, they differ only in their operating mode which is this time in reflection mode while the operating mode of the sensor assembly described with reference to FIG. 2 is in transmission mode.

A light source SL2 generates a light beam F2 which is then focused by focusing means MF2 in the EFO4 inputs of a plurality of optical fibers
F04 grouped in the form of a plastic TO4 strand. At the end of the strand TO4, the plurality of optical fibers F04 is subdivided into a first subfamily of optical fibers lFO4 and into a second subfamily of optical fibers 2po4.

A pillar P3 mounted in height supports the outputs SlFO4 of the first subfamily of optical fibers 1F04 respectively distributed at a predetermined spacing along the pillar P3. The SlFO4 outputs respectively deliver the substantially horizontal RL2V light radiation in a plane substantially vertical with respect to the roadway
CH and substantially transverse to the displacement of the vehicle VA.

A pillar P4 mounted in height, transversely opposite the pillar P3 supports a set of reflective elements RF1 distributed at a predetermined spacing along the pillar P4 to reflect the shadow cast by the vertical edge of the vehicle which is thus reflected in the game of '' reflective elements
RF1 at all times.

The outputs SlFO4 of the plurality of optical fibers lFO4 receive the shadow cast by the vertical section of the vehicle thus reflected by the set of reflectors RF1 and which now fits into the outputs SlFO4 of the plurality of optical fibers 1F04 at each instant.

I1 is planned to have between the focusing means
MF2 and the EF04 inputs of the plurality of optical fibers
F04, own SPO optical separating means on command to transmit in one direction the light beam F2 in the inputs EFO4 of the plurality of optical fibers lFO4 and to transmit in the other direction the shadow cast in the outputs SlFO4 of the plurality from optical fibers 1F04 to opto-electronic reception means ROE3 with a network of photosensitive elements, the latter finally delivering the first signals S1 representing the shadow cast by the vertical section of the vehicle VA.

For its part, the second optical barrier BO2 comprises a light source for example identical to the light source
SL2 which generates a light beam F2 focused by the focusing means MF2 in the inputs EFO4 of the plurality of optical fibers F04.

BH3 bar mounted horizontally, supports outputs
S2F04 of the second subfamily of optical fibers 2F04 respectively distributed at a predetermined spacing along the bar BH3. The outputs S2F04 of the plurality of optical fibers 2FO4 respectively deliver light radiation RL2H substantially horizontal in a plane substantially horizontal with respect to the roadway CH, and substantially transverse with respect to the movement of the vehicle VA.

A BH4 bar mounted horizontally, transversely opposite to the BH3 bar supports a set of reflective elements
RF2 distributed along the bar BH4 at a predetermined spacing to reflect the shadow cast by the horizontal edge of the vehicle which is thus reflected in the set of reflective elements RF2 at all times.

S2FO4 outputs receive the drop shadow of the horizontal section of the vehicle thus reflected by the reflectors
RF2 and which is now part of the S2FO4 outputs of the second subfamily of 2FO4 optical fibers at all times.

The optical separating means SPO then transmit on command in one direction the light beam F2 in the inputs EFO4 of the plurality of optical fibers 2FO4 and transmit in the other direction the drop shadow inscribed in the outputs S2FO4 of the plurality of optical fibers 2FO4 towards the reception means ROE3, the latter finally delivering the second signals S2 representing the shadow cast of the horizontal section of the vehicle VA.

In practice, the light sources SL1 (FIG. 2) and SL2 (FIG. 3) consist of light-emitting diodes working in infrared with a wavelength of the order of 600 to 800 nanometers.

The PUT preprocessing means combine the first S1 and second S2 signals to provide third signals
S3 representing the shadow cast of the vehicle as a whole.

The third signals S3 are then stored in a matrix memory MM of pixels distributed in N lines and
P columns. The processing unit UT then processes said matrix memory MM to extract vector components representative of the essential and distinct characteristics of the shadow cast of the vehicle as a whole.

Finally, from the vector components thus extracted, and stored in memory, the processing unit UT ranks the current vehicle in one and only one among several predetermined categories.

As we will see in more detail below, the vector components relate to predetermined criteria of the group including in particular the length, the height, the number of axles, the height at the level of at least one axle.

For example in France we know five categories of autonomous vehicles.

The first category is made up of motorcycles.

This category is characterized by the fact that said vehicles have two axles and have a height above the front axle of less than 1 m.

The second category of vehicles is cars. This category is characterized by the fact that the vehicle has two axles and has a height above the front axle of less than 1.30 m.

The third category of vehicles consists of vans which are characterized by having two axles and having a height having a value greater than or equal to 1.30 m above the front axle.

The fourth category consists of vehicles that tow a trailer. This category is distinguished by the fact that the vehicle has a height having a value less than 1.30 m above the front axle and having more than two axles.

Finally, the fifth category is made up of trucks which are characterized by having a height having a value greater than 1.30 m and having more than two axles.

According to the invention, the automatic classification of a vehicle into one and only one category among the five categories mentioned above requires two steps a) the determination of vector components representative of the relevant characteristics of the shadow cast of the vehicle to be treated; and b) automatic classification from the vector components thus determined in one and only one category among several categories.

Initially, the vector components are extracted from the vehicle profile using various image processing that will be described in more detail below.

Secondly, the automatic classification from the vector components thus extracted can be carried out using neural techniques which are perfectly suited to the problem to be solved and which by their nature have the following main advantages - robustness to noise and to possible failures of the sensor assembly, and - learning the categories to be classified making it possible to generalize the system to a large number of applications and to overcome the problems linked to the evolution of the content of the categories and to the evolution categories.

We will now describe the various image processing techniques used to determine the relevant vector components of the drop shadow of a vehicle.

In general, the information used by the image processing means consists in processing a vector signature extracted from the binarized image of the vehicle processed.

 It is here to be noted that the pixels of the matrix memory MM can only take two values 0 and 1 for example, the value 0 represents the background of the image, and the value 1 represents the shape of the vehicle.

For example, the image processing means comprise - means for cutting the matrix memory MM into a chosen number of horizontal bands each having a chosen number of lines - calculation means suitable for calculating the number of pixels in the state excited for each horizontal strip; and means suitable for associating with the number of pixels in the excited state thus calculated a vector component representative of the vertical distribution of the shadow cast by the vehicle as a whole.

After the distribution of the material along the vertical axis.

 It is possible to carry out image processing to determine the distribution of the material along the horizontal axis.

In practice, the image processing means comprise - means for cutting the matrix memory MM into a chosen number of vertical bands each having a chosen number of columns - calculation means suitable for calculating the number of pixels in the state excited for each vertical strip; and means suitable for associating with the number of pixels in the excited state thus calculated a vector component representative of the horizontal distribution of the shadow cast by the vehicle as a whole.

Then, it is possible to determine the surface distribution of the shadow cast of the vehicle as a whole.

Thus for example, the image processing means comprise - means for cutting the matrix memory into a chosen number of blocks of rectangular configuration - calculation means suitable for calculating, the number of pixels in the excited state for each block ; and means suitable for associating with the number of pixels in the excited state thus calculated a vector component representative of the surface distribution of the shadow cast by the vehicle as a whole.

Another image processing consists for example in calculating the center of gravity of the matrix memory.

According to another aspect of the invention, the image processing means comprise - means suitable for scanning circularly at constant angle the outline of the shadow cast of the vehicle as a whole from a chosen point in the memory matrix, - means suitable for measuring the length of the segments thus obtained having for the first end the point chosen and for the second end the intersection of said segments with the contour; and means suitable for associating with the distances of the segments thus measured a vector component representative of the topology at constant angle of the shadow cast of the vehicle as a whole.

For example, the points chosen belong to the group formed by the center of gravity of the matrix memory, or any point vertically on an axle, or even a quarter of the height of the vehicle recorded on this same axle.

Advantageously, the processing means also comprise - means suitable for scanning a selected point of the matrix memory MM from a plurality of equidistant points chosen from the contour of the shadow cast by the vehicle as a whole, - means suitable for measure the length of the segments thus obtained having for the first end the chosen point of the matrix memory MM and for the second end the equidistant points chosen from the outline of the shadow cast of the vehicle as a whole, and - means suitable for associating with the distances of the segments thus measured a vector component representative of the topology at constant distance from the shadow cast of the vehicle as a whole.

In the same way as previously, the points chosen can be the center of gravity of the matrix memory, any point vertically on an axle, or at half the height of the vehicle recorded on this same axle.

 It is also possible that the image processing means comprise means suitable for determining the number of axles of the vehicle or also for determining for the shadow cast of the vehicle as a whole the invariants by translation, by rotation and by homothety.

According to the invention, the processing unit is based on an automatic classification network of neuronal type, provided with information drawn from prior learning carried out, for each category of vehicles to be treated, on a representative subset chosen from vehicles of this category, and tested on all the vehicles to be treated.

In practice, at least one of the vector components representative of the drop shadow thus extracted from the matrix memory by the image processing means is applied to the inputs of the automatic classification network of the neural type according to a chosen order. This network then ranks from the vector components thus extracted the vehicle to be treated in one and only one category among the five categories mentioned above.

For example, the neural network is the one sold by the Company
NESTOR (United States) under the reference NSD 1000.

In practice, the input vectors of the neural network are standardized vectors which are sensitive to the order in which they are formed.

For example, it is associated with each axis of the vector space one and only one input vector.

A reliable classification has been obtained by the Applicant by applying to the inputs of the neural network the vector components relating to the horizontal, vertical and surface distribution of the drop shadow.

Advantageously, the invention also relates to a dynamic automatic classification method for autonomous vehicles comprising the following steps a) preparing a data set concerning all of the vehicles to be treated, b) providing an automatic classification network of neural type, configured in learning mode, c) for each category of vehicles to be treated, using the data set thus prepared, carry out supervised learning on a selected subset of vehicles in this category, and d) using information drawn from the learning thus carried out, test on all of the vehicles to be treated and start again at step c) for poorly tested vehicles.

Supervised learning is understood to mean the fact that, with the previously prepared data, there are associated labels which correspond to the categories of vehicles to be classified.

Advantageously, step a) comprises the following steps a1) prepare a set of drop shadows relating to all the vehicles to be treated a2) extract vector components representative of the essential and distinct characteristics of each drop shadow; and a3) constitute a data set concerning all of the vehicles to be processed from said vector components thus extracted.

Claims (27)

Claims 1. Automatic classification device on the fly of autonomous vehicles, characterized in that it comprises: - a sensor assembly (CAP) placed in predetermined relation with respect to the roadway (CH), and comprising at least:  first means forming a first optical barrier (BOl) with a network of photosensitive elements (ELV), mounted vertically transverse to the movement of the vehicle (VA), capable of delivering first signals (S1) representing the shadow cast by the edge vertical of the vehicle (VA) which fits into said barrier at all times, and  second means (BO2) suitable for delivering second signals (S2) representing the speed of the vehicle (VA) when crossing said first barrier (BOl), and - pre-processing means (PUT) suitable for combining the first (S1) and second (S2) signals for supplying third signals (S3) representing the shadow of the vehicle (VA) as a whole, - a processing unit (UT) receiving the third signals (S3) and suitable for starting from predetermined information stored, relating to predetermined criteria of the group comprising at least the length, the height, the number of axles, the height at the level of at least one axle, to store the vehicle (VA) in courses in one and only one among several predetermined categories. 2. Device according to claim 1, characterized in that the second means form a second optical barrier (BO2) with a network of photosensitive elements (ELH), mounted substantially horizontally transversely to the movement of the vehicle (VA), clean in delivering the second signals (S2) representing the progress of the vehicle (VA) with respect to the first barrier (BOl). 3. Device according to claim 2, characterized in that the second optical barrier (BO2) is placed at a height relative to the floor of the order of 20 cm. 4. Device according to one of claims 2 and 3, characterized in that the second optical barrier (BO2) is subdivided into an optical head barrier (BO2T) placed upstream of the first barrier (BOl), having a predetermined length (L1) and in a tail barrier (BO2Q) downstream of the first barrier (BO1) and having a predetermined length (L2). 5. Device according to claim 4, characterized in that the length (L1) of the head barrier (BO2T) satisfies a condition relating to a distance chosen between the first axle of a predetermined vehicle and the front end of said vehicle, in that the length (L2) of the tail gate (BO2Q) satisfies a condition relating to a distance chosen between the last axle of said predetermined vehicle and the rear end of said vehicle and in that the sum (L1 + L2) of the respective lengths of the head gate (BO2T) and the tail gate (BO2Q) checks a condition relating to a distance chosen between the first and last axles of said predetermined vehicle. 6. Device according to any one of the preceding claims, characterized in that the photosensitive elements (ELV and ELH) are of the charge transfer type. 7. Device according to any one of the preceding claims, characterized in that the first means forming a first optical barrier (BOl) comprise: - a first light source (SL1) capable of generating a first light beam, - first means focusing (MF1) for focusing said first light beam (F1) thus generated in the inputs (EFO1) of a first plurality of optical fibers (lFOl), - a first pillar (P1) mounted in height and intended to support the outputs (S1FO1) of the first plurality of optical fibers (lFol) respectively distributed at a predetermined spacing along said first pillar (P1), and respectively delivering first light rays (RL1V) substantially horizontal in a plane substantially vertical with respect to the roadway (CH), and substantially transverse relative to the displacement of the vehicle (VA), - a second pillar (P2) mounted in height, transversely oppo sewn to the first pillar (P1) and intended to support the inputs (EFO2) of a second plurality of optical fibers (F02) respectively distributed at a predetermined spacing along said second pillar (P2) to receive the shade of the vertical section of the vehicle (VA) which is registered in said inputs (EFO2) of the second plurality of optical fibers (F02) at each instant, - of the first opto-electronic reception means (ROE1) with a network of photosensitive elements connected to the outputs (SF02) of the second plurality of optical fibers (F02) and suitable for delivering the first signals (S1) representing the shadow cast by the vertical section of the vehicle thus received. 8. Device according to any one of the preceding claims, characterized in that the second means forming a second optical barrier (BO2) comprise: - a second light source (SL1) capable of generating a second light beam (F1), - second focusing means (MF1) for focusing said second light beam (F1) thus generated in the inputs (EFO1) of a third plurality of optical fibers (2FOl), - a first bar (BH1) mounted horizontally and intended supporting the outputs (S2FO1) of the third plurality of optical fibers (2FO1) respectively distributed at a predetermined spacing along said first bar (BHl), and respectively delivering second light rays (RL1H) substantially horizontal in a substantially horizontal plane with respect to the roadway (CH) and substantially transverse with respect to the movement of the vehicle (VA), - a second bar (BH2) mounted horizontally, transversely opposite the first bar (BH1) and intended to support the inputs (EFO3) of a fourth plurality of optical fibers (F03) respectively distributed along said second bar (BH2) to receive the drop shadow of the horizontal slice of the vehicle which registers in said inputs (EFO3) of the fourth plurality of optical fibers at each instant, - second opto-electronic reception means (ROE2) with a network of photosensitive elements connected to the outputs (SF03) of the fourth plurality of optical fibers (F03) capable of delivering the second signals (F2) representing the shadow cast by the horizontal edge of the vehicle thus received, that is to say the advancement of the vehicle relative to the first barrier ( Bowl). 9. Device according to any one of claims 1 to 6, characterized in that the first means forming the first optical barrier (BOl) comprise - a third light source (SL2) capable of generating a third light beam (F2), - third focusing means (MF2) for focusing said third light beam (F2) thus generated in the inputs (EFO4) of a fifth plurality of optical fibers (lFO4), - a third pillar (P3) mounted in height and intended for supporting the outputs (SlFO4) of the fifth plurality of optical fibers (1F04) respectively distributed at a predetermined spacing along said third pillar (P3), and respectively delivering third light rays (RL2V) substantially horizontal in a plane substantially vertical relative to to the roadway (CH), and substantially transverse with respect to the movement of the vehicle (VA), - a fourth pillar (P4) mounted in height, tra nsversely opposite the third pillar (P3) and intended to support a first set of reflective elements (RFl) distributed along said fourth pillar (P4) at a predetermined spacing to reflect the shadow cast by the vertical section of the vehicle which is reflected in said first set of reflective elements (RF1) at each instant, - a fifth pillar (P3) mounted in height, transversely opposite the fourth pillar (P4) and intended to support the inputs (SlFO4) of a sixth plurality of fibers optics (lu04) respectively distributed along a predetermined axis along said fifth pillar (P3) to receive the shadow cast by the vertical edge of the vehicle which fits into said inputs (SlFO4) of the sixth plurality of optical fibers (IF04) at each instant, - fourth opto-electronic reception means with network of photosensitive elements (ROE3) connected to the outputs (EF04) of the sixth plurality of f optical ibres (lu04) and capable of delivering the first signals (S1) representing the shadow cast by the vertical section of the vehicle thus received. 10. Device according to claim 9, characterized in that the fifth and sixth pluralities of optical fibers (lu04) constitute one and only one seventh plurality of optical fibers (lu04) whose inputs and outputs (SlFO4) are supported by one and a single sixth pillar (P3) mounted in height and opposite transversely with respect to the fourth (pillar P4), in that it is planned to have between the third focusing means (MF2) and the seventh plurality of optical fibers (F04) first own optical separating means (SPO) on command to transmit the third light beam (F2) into the inputs (EFO4) of the seventh plurality of optical fibers and to transmit the drop shadow recorded in the outputs (SF104) of the seventh plurality of optical fibers to the fourth opto-electronic reception means (ROE3). 11. Device according to any one of claims 1 to 7, characterized in that the second means forming a second optical barrier (BO2) comprise - a fourth light source (SL2) capable of generating a fourth light beam (F2), - fourth focusing means (MF2) for focusing said fourth light beam (F2) thus generated in the inputs (EFO4) of an eighth plurality of optical fibers (2po4), - a third bar (BH3) mounted horizontally and intended to support the outputs (S2FO4) of the eighth plurality of optical fibers (2FO4) respectively distributed along said third bar (BH3) at a predetermined spacing, and respectively delivering fourth light rays (RL2H) substantially horizontal in a plane substantially horizontal with respect to the roadway (CH), and substantially transverse with respect to the movement of the vehicle (VA), - a fourth bar (BH4) mounted horizontally, transversely opposite the third bar (BH3) and intended to support a second set of reflective elements (RF2) distributed along said fourth bar (BH4) at a predetermined spacing to reflect the shadow cast by the horizontal section of the vehicle which is reflected in said second set of reflective elements (RF2) at each instant, - a fifth bar (BH3) mounted horizontally, transversely opposite the fourth bar (BH4) and intended to support the inputs (S2FO4) of a ninth plurality of optical fibers (2FO4) respectively distributed along said fifth bar (BH3) at a predetermined spacing to receive the shadow cast by the horizontal edge of the vehicle which is registered in said inputs (S2FO4 ) of the ninth plurality of optical fibers (2FO4) at each instant, - of the fourth opto-electronic reception means with a network of photosensitive elements (ROE3 ) connected to the outputs (EFO4) of the ninth plurality of optical fibers (2FO4) and capable of delivering the second signals (S2) representing the shadow cast of the horizontal section of the vehicle thus received. 12. Device according to claim 11, characterized in that the eighth and ninth pluralities of optical fibers (2po4) constitute one and only one tenth plurality of optical fibers (2FO4) whose inputs and outputs (S2FO4) are supported by a single and single sixth bar (BH3) mounted horizontally and transversely opposite with respect to the fourth bar (BH3), in that provision is made between the fourth focusing means (MF2) and the tenth plurality of optical fibers (2FO4) second optical separating means (SPO) capable of transmitting the fourth light beam (F2) into the inputs (EFO4) of the tenth plurality of optical fibers (EFO4) and transmitting the drop shadow inscribed in the outputs (S2FO4) from said tenth plurality of optical fibers (2FO4) to the fourth opto-electronic reception means (ROE3). 13. Device according to any one of the preceding claims, characterized in that one of the light sources (SL1 and SL2) consists of at least one light-emitting diode having a wavelength of the order of 600 to 800 nanometers. 14. Device according to any one of the preceding re vendications, characterized in that the third signals (S3) representing the shadow cast of the vehicle as a whole are stored in a matrix memory (MM) of pixels distributed in N lines and P columns, and in that the processing unit (UT) comprises means for processing images of said matrix memory (MM) capable of extracting vector components representative of the essential and distinctive characteristics of the shadow cast by the vehicle in as a whole for the automatic classification of said vehicle into one and only one among several predetermined categories. 15. Device according to claim 14, characterized in that the image processing means comprise - first means for cutting the matrix memory (MM) into a chosen number of horizontal bands each having a chosen number of lines, - first means of calculation suitable for calculating the number of pixels in the excited state (value I) for each horizontal strip, - first means suitable for associating with the number of pixels in the excited state thus calculated first a vector component representative of the distribution vertical of the shadow cast of the vehicle as a whole. 16. Device according to claim 14 or claim 15, characterized in that the processing means comprise: - second means for cutting the matrix memory into a chosen number of vertical bands each having a chosen number of columns, - second means of calculation suitable for calculating the number of pixels in the excited state (value 1) for each vertical strip, - second means suitable for associating with the number of pixels in the excited state thus calculated a second vector component representative of the horizontal distribution the shadow cast of the vehicle as a whole. 17. Device according to any one of claims 14 to 16, characterized in that the processing means comprise: - third means for cutting the matrix memory (MM) into a chosen number of blocks of rectangular configuration, - third means of calculation suitable for calculating the number of pixels in the excited state (value 1) for each block, - third means suitable for associating with the number of pixels in the excited state thus calculated a third vector component representative of the surface distribution of the shadow cast of the vehicle as a whole. 18. Device according to any one of the preceding claims 14 to 17, characterized in that the processing means further comprise means suitable for calculating the center of gravity of the matrix memory relative to the subset of its elements when excited state (value 1). 19. Device according to any one of the preceding claims 14 to 18, characterized in that the processing means comprise - first scanning means capable of sweeping circularly at constant angle the contour of the shadow cast by the vehicle as a whole at starting from a selected point of the matrix memory, - first measuring means suitable for measuring the length of the segments thus obtained having for first end the chosen point and for second end the intersection of said segments with the contour, - fourth means able to associate with the distances of the segments thus measured a fourth vector component representative of the topology of the shadow cast by the vehicle as a whole. 20. Device according to any one of the preceding claims 14 to 19, characterized in that the processing means comprise - second scanning means capable of scanning a chosen point of the matrix memory from a plurality of equidistant points chosen of the outline of the drop shadow of the vehicle as a whole, - second measurement means suitable for measuring the length of the segments thus obtained having for the first end the chosen point of the matrix memory and for the second end the equidistant points chosen from the outline of the shadow of the vehicle as a whole, - fifth means suitable for associating with the distances of the segments thus measured a fifth vector component representative of the topology of the shadow of the vehicle as a whole. 21. Device according to any one of the preceding claims 14 to 20, characterized in that the processing means comprise means suitable for determining for the shadow cast of the vehicle as a whole to treat the invariants by translation, by rotation, and by homothety. 22. Device according to any one of the preceding claims 14 to 21, characterized in that the processing means comprise means suitable for determining the number of axles of the vehicle. 23. Device according to one of the preceding claims, characterized in that the processing unit (UT) is based on an automatic classification network of neural type, provided with information drawn from a prior learning carried out, for each category of vehicles to be treated, on a selected subset of vehicles in this category, and tested on all of the vehicles to be treated. 24. Device according to claim 23, characterized in that at least one of the first to fifth vector components representative of the drop shadow thus extracted from the memory by the processing means is applied to the inputs of the automatic type classification network. neural which then ranks the current vehicle in one and only one category among several predetermined mine categories. 25. A dynamic automatic classification method for autonomous vehicles, characterized by the following steps: a) preparing a data set relating to all of the vehicles to be treated, b) providing an automatic classification network of neural type, configured in the learning, c) for each category of vehicles to be processed, using the data set thus prepared, carry out learning on a selected subset of the vehicles in this category, d) using the information taken from the learning thus carried out, test on all the vehicles to be treated and repeat step c) for poorly tested vehicles. 26. The method as claimed in claim 25, characterized in that step a) comprises the following steps al) preparing a set of shadows relating to all of the vehicles to be treated a2) extracting vector components representative of the essential and distinct characteristics of each drop shadow; and a3) constitute a data set concerning all of the vehicles to be processed from said vector components thus extracted. 27. Application of the device according to any one of the preceding claims to the automatic classification on the fly of autonomous vehicles at motorway toll kiosks.