DEVICE AND PROCESS FOR THE ORIENTATION OF CONTAINERS

Abstract

A device and a process for the orientation of containers with a carousel rotating about a vertical axis with a circumferential plurality of rotating seats imparting a rotation to each container causing a rotation-revolution. A recording device with illuminator recording multiple sequential images of the container during rotation-revolution. A controller adapted learn the lateral surface of said container in a number of positions. Learning each of the N positions said container is positioned with a random initial orientation, and during rotation a number of images are recorded. The controller acquires one current image, processes a similarity function representing the similarity between each current image and the images recorded in the same position in said learning step, calculates the similarity functions and the angle corresponding to the maximum value of this sum is used for the orientation of said container.

Claims

1. A device for the orientation of containers comprising a rotating carousel rotating about a carousel vertical axis thereof having a circumferential plurality of seats each of which rotates about a seat axis to impart to each container housed therein a rotation about a container vertical axis and jointly with said rotating carousel a movement of rotation-revolution, a fixed illuminator of said container and a single fixed recording device recording multiple sequential images of said container during said movement of rotation-revolution, said recording device being in a position to acquire the light reflected from said container illuminated by said illuminator, an electronic controller comprising a learning device, a validation device and an operating device, the learning, validation, and operating devices operating together in a temporal sequence, said learning device being adapted to carry out a step of learning the entire lateral surface of one said container in a number N of positions within the field framed by said recording device, in which, when said carousel is stationary, for each learning in each of the N positions said container is positioned with a random initial orientation, and during a rotation of 360° about the container axis a number R of images of said container being acquired, said validation device being adapted to carry out a validation step in which said electronic controller acquires from said recording device only one current image of said container in each position N, said electronic controller processing a similarity function representing the similarity between each current image Ni and the R images recorded in the same position N in said learning step, phase synchronizing and summing said similarity functions; an operating means active during the rotation of said rotating carousel, in which for each of said containers said electronic controller acquires from said recording device only one current image in each position N, processes a similarity function representing the similarity between each current image Ni and the R images recorded in the same position N in said learning step, calculates and sums the similarity functions and the angle α.sub.m corresponding to the maximum value of this sum is used for the orientation of said container.

2. The device for the orientation of containers according to claim 1, wherein said angle α.sub.m is communicated by said electronic controller to a motor system moving the plates said rotating carousel, which, by imposing on said container a rotation of −α.sub.m, positions all of said containers according to the same angle.

3. The device for the orientation of containers according to claim 1, wherein said recording device uses cameras with sensors of the CMOS type adapted to allow a mechanism of recording sequential images with “successive crops” of the sensor, the sensor crop being set to pick up only the portion of pixels where said container is located at each time during its advancement.

4. The device for the orientation of containers according to claim 1, wherein said validation step is an automatic procedure, said electronic controller inferring all the information automatically on the basis of a statistical observation of a statistically necessary number of said containers in transit in the field framed by said recording device.

5. The device for the orientation of containers according to claim 1, wherein said validation step consists in the resolution of algorithms which allow obtaining a measure of image similarity between an image recorded by said recording device in a given position N during the passage of said container and all the R images forming part of the population learned during said learning step in that same given position N.

6. The device for the orientation of containers according to claim 1, wherein said validation algorithms are based on partial sums of the similarity functions of successive positions N.

7. The device for the orientation of containers according to claim 1, wherein said validation procedure produces, by said electronic controller, a drawing of a graph containing the similarity measure of an image recorded by said recording device in a given position N during the passage of said container during the operating phase with all the R images recorded by said recording device and saved by said electronic controller in said learning step with said carousel stationary in the same position N with said container in rotation about the container axis.

8. The device for the orientation of containers according to claim 7, wherein said graph contains an indication of the “peak”, said “peak” being the visual expression that allows an understanding of which image of the learned population is most similar to the current one, if there are other similar images in the population and in what proportion with respect to the main one, its “sharpening” determining the precision of the system, i.e. the similarity difference between the most similar image among the learned ones and the images in the adjacent degrees of rotation.

9. The device for the orientation of containers according to claim 7, wherein said graph is displayed by the electronic controller a screen for each container.

10. A process for the orientation of containers in a movement of rotation-revolution on a rotating carousel that provides for passing said containers in movement in front of an illuminator, recording, for each of said containers, from a single point of view and with a single recording device, a predetermined number R of images from N predetermined positions inside the field framed by said recording device, comprising processing by an electronic controller of a similarity function representing the similarity between each current image and the R images recorded in the same position N in an initial learning step with the carousel stationary and then validated and phase synchronized, said similarity functions being summed together and the angle α.sub.m corresponding to the maximum value of this sum being used to identify the orientation of the container.

11. The process for the orientation of containers according to the claim 10, wherein said angle α.sub.m is communicated by said electronic controller to a motor system moving said plates of said rotating carousel, which, by imposing on said container a rotation of −α.sub.m, positions all of said containers according to the same angle.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Additional features and advantages of the invention will become more apparent from the description of a preferred but non-exclusive embodiment of a device for the orientation of containers illustrated by way of non-limiting example in the appended drawings, in which:

[0069] FIG. 1 is a schematic view of the orientation device according to the invention;

[0070] FIG. 2 is a schematic side elevation view of the optical reconstruction station;

[0071] FIG. 3 schematically shows the number N of positions in which images of the container are detected;

[0072] FIG. 4 illustrates the learning step;

[0073] FIG. 5 shows the operating phase;

[0074] FIG. 6 shows a similarity graph;

[0075] FIG. 7 graphically shows the phase synchronized peaks acquired in the various positions of the container.

DETAILED DESCRIPTION

[0076] With particular reference to the above-described figures, the device for the orientation of containers is denoted in its entirety by the number 1.

[0077] The device 1 is preferably used for the orientation of transparent or opaque glass or plastic containers 4, typically at least partly cylindrical and/or conical in shape, intended to be labelled. The device 1 comprises a rotating carousel 2 rotating about a vertical axis L1 thereof and having a circumferential plurality of seats 3, each supporting a corresponding container 4 oriented with the vertical axis L2.

[0078] Each seat 3 rotates about its own axis L3, likewise vertical and coinciding with L2 so, as a result of the combined actuation of the carousel 2 and the seat 3, a movement of rotation-revolution is imparted to the container 4, during which it maintains its axis with a vertical orientation.

[0079] Conveniently, the carousel 2 has an angular position that is rigidly connected, during its movement, with the angular position of each of the seats 3.

[0080] In a pre-established area of the carousel 2 there is provided an optical reconstruction station, denoted in its entirety with the number 5, having an illuminator 6 of the individual container 4 and a recording device 20 for recording multiple sequential images of the individual container 4 during its motion of rotation-revolution.

[0081] In the solution illustrated, there is provided a single fixed illuminator 6 and a single fixed recording device 20, in particular a fixed frame camera.

[0082] The illuminator 6, positioned at the perimeter of the carousel 2, has a projection surface in front of which the containers 4 pass during the motion of rotation-revolution.

[0083] The containers 4 expose their entire lateral surface to the illuminator 6 during their passage in front of the latter.

[0084] As may be seen in FIG. 2, the optical principle used is that of the reflection of light on the outer surface of the container 4.

[0085] Essentially, the optical reconstruction station provides for the illuminator 6 and the recording device 20 to be positioned on the same side of the containers 4 which pass sequentially in front of them, and a mirror 22 which redirects the light reflected from the container 4 onto the recording device 20.

[0086] The illuminator 6 emits light from its emission surface facing towards the carousel 2, and the emitted light strikes an angular sector of the surface of the container 4, enabling the observation thereof.

[0087] The device further comprises a synchronization means 9 for synchronizing between the images recorded by the recording device 20 and the angular positions of the carousel 2 and those of the seats 3 of the containers 4, and an electronic control means for recording, processing and analyzing the images and the data.

[0088] As said, the present invention aims to orient the containers that pass through the Ni positions of the optical reconstruction station.

[0089] It is necessary to distinguish three distinct steps of use of the invention.

[0090] 1.—Learning: the invention “learns” a container 4.

[0091] In this step the carousel 2 is stationary (or in a non-operating position), and the container 4 is positioned successively in each of the N positions identified in the field of view of the recording device 20. In each of said Ni positions, the container is set into rotation by the seat 3 rotating about its axis L2 and Ri images are acquired over a complete revolution of the container 4 itself, one at every Δα=360°/R degrees.

[0092] Therefore, N×R images are learned.

[0093] 2.—Validation: the various learning operations carried out in the preceding step are “phase synchronized”; in this manner the geometric and motion parameters are learned.

[0094] The container 4 is made to pass in front of the orientation system with the carousel 2 in the operating mode: the container 4 rotates according to a precise law of motion while it passes across the field framed by the recording system 20 and N single images are acquired in the Ni pre-established positions.

[0095] By means of a specific algorithm, the N−1 offsets for re-synchronizing the similarity functions are calculated by the control system 10.

[0096] The process is automatic and requires no input from qualified personnel, only the starting of the carousel 2.

[0097] These first two steps, the learning step and the validation step, are carried out solely to create an “operating recipe” with a type of container 4 that was previously unknown.

[0098] Let f be the operator that returns a value of similarity between two images, let s=f(r) then be a similarity function with rϵ where 1≤r≤R, and sϵR, where −1≤s≤1, the discrete function that represents, in graph form, the similarity between the image of a container 4 in a defined position Ni and the population of R images learned in that defined position.

[0099] The similarity function s=f(r) thus takes on values between −1 and 1, where the two extreme values have the meaning of zero similarity and total coincidence, respectively.

[0100] The system computes N similarity functions for each container, one for every position Ni.

[0101] Let these be s=f(r.sub.i) with ϵ where 1≤i≤N.

[0102] Let S(r)=Σ.sub.t-1.sup.Nf(r.sub.i) be the sum function of the N similarity functions. Since the similarity functions each express the similarity of what is seen now with what was learned previously, and in relation to an angular sector of the container 4, the sum function expresses the whole container.

[0103] However, this can be true only assuming that the individual similarity functions are “phase synchronized” i.e. “superposable”, i.e. in the assumption that for every r the similarity functions express all of the similarities of the container rotated by the same amount, i.e. α=r.Math.Δα.

[0104] If this assumption is true, let M=max(S(r)) be the maximum value taken on by the sum function of the N similarity functions.

[0105] This value is formed at a certain r.sub.m.

[0106] Therefore, M=S(r.sub.m).

[0107] Thus α.sub.m=r.sub.m.Math.Δα is the angle by which the container 4 is rotated, and that is the angle that is communicated to the system for moving the plates 3, which, by imposing a rotation of −α.sub.m on the container, positions all of the containers according to the same angle.

[0108] As seen, at the basis of the orientation of the containers 4 there is the assumption that the similarity functions are “phase synchronized” in such a way that it makes sense to produce the sum thereof and that at the maximum value of the latter the resulting angle of the container 4 is read.

[0109] The process of “phase synchronizing” the similarity functions is called “validation”.

[0110] “Validation” is the automatic process that implements the control and management system 10 at the end of the learning step, and which by observing and recording the passage of a number C of randomly oriented containers 4 carries out the phase synchronization necessary for the subsequent operating phase.

[0111] At the passage of each of these C containers 4, which move across the field of view of the recording device 20 exactly as when in the operating mode, N single images are acquired and the N similarity functions are calculated.

[0112] Let these similarity functions be s.sub.i—f(r.sub.i) with

[0113] iϵ and 1≤i≤N where N is the number of the positions

[0114] rϵ and 1≤r≤R where R is the number of angular sample images over a revolution

[0115] sϵR and 1≤s≤1 where s is a similarity value

[0116] The object of this validation process is to calculate N−1 angular offsets which allow the N similarity functions to be made summable, hence superposable.

[0117] What is meant by “offset” of a similarity function will now be explained.

[0118] Let s.sub.i—f(r.sub.i) be the i-th similarity function relative to the i-th position.

[0119] The same function offset k.sub.i is:

[0120] s.sub.i—f(r.sub.i−k.sub.i) with

[0121] iϵ and 1≤i≤N where N is the number of the positions

[0122] rϵ and 1≤r≤R where R is the number of angular samples over a revolution

[0123] sϵR and −1≤s≤1 where s is the similarity value

[0124] kϵ and 1≤k≤R where R is the number of angular samples over a revolution which is equivalent to a “circular shift over the domain”.

[0125] Let these offsets be indicated with k: the purpose of the validation is to calculate the offset vector <k.sub.1, k.sub.2, k.sub.3, . . . k.sub.N-1> which enables the similarity functions to be phase synchronized, and thus gives meaning to their sum.

[0126] A generic similarity function, if it refers to an angular part of the container 4 in which unique information is present, generally possesses a “peak”.

[0127] If, for the sake of simplicity, we consider a container 4 completely covered by unique information, where all the similarity functions possess a peak, the phase synchronization of the functions means finding the series of offsets that leads all of the peaks to coincide in a same value of r for all of the functions.

[0128] The validation corresponds, in fact, to finding the N−1 offsets which, when applied to the similarity functions originating from a sufficiently large statistical sample of containers 4, maximize the sum of all the functions, of all the positions, for all the containers.

[0129] Formally, for a generic container cj there are N similarity functions

[0130] s.sub.ij=f(r−k.sub.i) with

[0131] iϵ and 1≤i≤N where N is the number of the positions

[0132] rϵ and 1≤r≤R where R is the number of angular samples over a revolution

[0133] sϵR and −1≤s≤1 where s is a similarity value

[0134] kϵ and 1≤k≤R where R is the number of angular samples over a revolution

[0135] jϵ and 1≤j≤C where C is the number of containers used

[0136] Let S(r).sub.<k1, k2, k3, . . . kN-1>=Σ.sub.j=1C (Σ.sub.i=1N f(r.sub.i−k.sub.i)) be the space of the sum functions of all the C×N similarity functions of all the C containers for all the N positions.

[0137] That is, given an offset vector <k.sub.1, k.sub.2, k.sub.3, . . . k.sub.N-1> one obtains a function S(r) that is a single similarity function, the sum function of all the C×N similarity functions of all the C containers for all the N positions.

[0138] Therefore, this function will have a maximum value at a given r.

[0139] Let that maximum value be M.sub.<k1, k2, k3, . . . kN-1>=max (.sub.r-1.sup.R(S(r).sub.<k1, k2, k3, . . . kN-1>)).

[0140] For every offset vector, therefore, there exists a maximum value M: the offset vector looked for is the one that produces the highest value of M.

[0141] An exhaustive execution of this calculation is simply impracticable, however. In the most common practical implementation with a number of containers C=30, a number of positions N=12, and a number of images R=resolution=360 images over a revolution, the general complexity of the calculation is the following:

max(.sub.r-1.sup.RS(r).sub.<k1,k2,k3, . . . kN-1>)=max(.sub.r-1.sup.R(Σ.sub.j=1C f(Σ.sub.i=1N f(r.sub.i−k.sub.i))))

[0142] It is a matter of performing 360.sup.11 calculations of the function S(r), then extracting the maximum value for all 360 values of r of the domain: the typical complexity of this calculation is therefore in the order of 360.sup.12 iterations.

[0143] Therefore, an integral part of the present invention is a strategy for calculating the N−1 offsets that allows the complexity of the calculation itself to be reduced.

[0144] This strategy is based on the partial sums of the positions.

[0145] A first reference position r1 is chosen, for which the similarity functions will thus not be offset, whereas a second reference position r2 is offset by all of the possible R values.

[0146] Let S(r).sub.<k1>=Σ.sub.j=1 C (f(r.sub.2−k.sub.1)+f(r.sub.1)) be a population of S(r), each for a different value of the offset k.sub.1.

[0147] For each of these functions the value Max M is found, and the k.sub.1 chosen corresponds to the maximum value M found, let it be k.sub.1_ok.

[0148] What has been done in this manner is to calculate the best offset k.sub.1 which maximizes the maximum of the sum of the similarity functions of the first two positions r1 and r2. Progressively, and in the same manner, once the offset between the first two positions r1 and r2 has been blocked, one proceeds to sum the first three for all of the possible offsets k2 of the third position r3:

[0149] S(r).sub.<k2>=Σ.sub.j=1C (f(r.sub.3−k.sub.2)+f(r.sub.2−k.sub.1_ok)+f(r.sub.1)), and in the same manner one chooses the offset k.sub.2_ok that makes the maximum value M of the function S(r) reach the highest value.

[0150] Similarly, the process is repeated for all the successive positions:

[0151] S(r).sub.<k3>=Σ.sub.j=1 C (f(r.sub.4−k.sub.3)+f(r.sub.3−k.sub.2_ok)+f(H)), . . . until the complete calculation of the offset vector:

[0152] <k.sub.1_ok, k.sub.2_ok, k.sub.3_ok . . . k.sub.(N-1)_ok>

[0153] In short, the “validation” process, i.e. the calculation of the N−1 angular offsets (expressed in sampling units r, which coincide with a degree in the case of sampling with 360 images, with half a degree in the case of 720 images, etc.), enables the system to know: [0154] geometry of the container; [0155] relative container—recording device—illuminator geometry; [0156] position of the different points of recording of an image of the container; [0157] initial angle of container learning in each recording point; [0158] movement of rotation-revolution that the container completes within the field framed by the recording device during the operating phase;

[0159] At the end of this process, the system is ready for the operation of orienting the containers 4.

[0160] 3.—Operation: with the carousel in the operating mode, every container 4 in transit, with a motion of rotation-revolution, and which arrives with a random angle in the field framed by the recording device 20, is photographed N times in the Ni established, known positions.

[0161] N graphs are drawn, representing the similarity between the current Ni photos and the Ni populations learned during the learning in each position. The functions are summed and the maximum peak is formed at a given angle α.sub.m.

[0162] This given angle α.sub.m is the angle that is communicated by the management and control system 10 to the motorized system that controls the motor of the plate 3 under the container 4, the result being that the container 4 itself is rotated and oriented in the desired manner in order to perform further operations thereon such as labelling, marking, and others.

[0163] In concrete terms: in the operating phase, at the passage of a container 4 with a motion of rotation-revolution within the field framed by the recording device 20, the same container is photographed in N positions, which are the same where the learning process was carried out. Each of the N recorded photos originating from the Ni positions is compared with the R images learned in that same position of the Ni positions in the learning step, and a similarity function is drawn.

[0164] This function, if the image it is associated with contains the information useful for orienting the container, will contain a “peak”.

[0165] All of the similarity functions associated with images containing unique information for the purposes of orientation contain a peak.

[0166] Such peaks have been previously phase synchronized with one another by the validation process. What occurs is the creation of a “sum” function of all the similarity functions.

[0167] This function, as some of the summed functions have a peak, and these being phase synchronized, will itself contain a peak at a specific value of r: let this be defined as r.sub.m.

[0168] The orientation of the container is then performed by determining the rotation angle am to which the value of r.sub.m makes reference, i.e. α.sub.m=r.sub.m.Math.Δα.

[0169] During operation, the graphic pattern of this sum function is displayed by the electronic control and management system 10 to the machine operators on the screen 100 for each container 4. The human eye is able to perceive the succession of pointed, sharpened shapes that follow one another in succession on the screen much better than if they were expressed by numbers.

[0170] The shape of this function is able to communicate a great deal of information: for example, if the apex becomes rounded or the slopes less steep, there is a loss of system precision due to differences in the containers or in the positions in which they are photographed, if the points are split into two (or bifurcate), this communicates a difference in the present movement of rotation-revolution compared to the one learned.

[0171] The application of the concept of “peak of similarity” is meant to be an integral part of the present invention; it enables: [0172] by means of the “validation” or “peak phase synchronization” process, all the geometry of the system and of the motion of the container 4 to be autonomously reconstructed without the entry of parameters; [0173] by displaying the sum function of the similarity functions, the good performance of the system to be communicated also to non-expert persons at a more “emotional” and analogue level rather than a cognitive and numerical one, given the visual ability of human beings to perceive shapes at a high speed, but not numbers.

[0174] Modifications and variants, in addition to the ones already mentioned, are naturally possible, such as, for example, lines for moving containers along linear or complex paths rather than on a circular carousel.

[0175] In practical terms, it has been observed that a device 1 for the orientation of containers 4 according to the invention is particularly advantageous because of its low cost from both a structural and a management point of view, given its compact size and its having only one viewing point, because of its operating simplicity and use by personnel who are not particularly qualified, because of the emotional indication of the proper functioning of the orientation system and real-time performance, also at high machine speeds and frequencies of the containers to be oriented.

[0176] A device for the orientation of containers thus conceived is susceptible of numerous modifications and variants, all falling within the scope of the inventive concept; moreover, all of the details may be replaced by technically equivalent elements.

[0177] In practice, all of the materials used, as well as the dimensions, may be of any kind, according to needs and the state of the art.