The invention relates to checking based on image processing and more particularly checking the taking of obj ects.

The invention applies advantageously but in a nonlimiting way to checking the taking of medicines.

In the state of the art, there are a number o f so lutions for checking the taking of medicines.

The patent application FR2650426 describes a system with drawers containing medicines. This system has liquid crystal screens for displaying prescriptions and an alarm for notifying the patient of the time to take medicines.

The patent application FR 271 768 1 describes a device with a casing provided with alarms, these alarms being set by virtue o f the control keys.

The patent application EP 1453466 describes a storage device that makes it possible to take medicines according to posology. This device comprises a weekly doser, an inclined plane, and semi-inclined daily dosers .

The patent application FR 2841360 described a checking device in which a prescription is transmitted by mobile telephone, the transmission being secured by the use of the telephone number and o f the code of the vital card. There is also provided a validation o f the compatibilities of the medicines by virtue of an exchange of data between the mobile telephone and a central file which contains the data relating to the incompatibilities and the images of the medicines.

The patent application FR2920297 describes a system comprising cells sealed by a sheet comprising an electrical continuity rupture device for detecting that the cell has been opened.

The international application WO 2009/023858 relates to the management of a medication using mobile telephones. It comprises a management program capable o f identifying and authenticating input/output images of the medication so as to confirm a correct medication performed by the patient. The international application WO 2008/085607 describes a device for dispensing medicines comprising a remote control system. This device comprises :

compartments for storing the medicines,

an image capture appliance positioned to capture an image o f the interior of each compartment,

a communication mo dule for transmitting the captured image to a central control station.

In these documents, there is no provision for a system for checking the taking of medicines by a patient which is incorporated in the appliance for dispensing these medicines. Also, in these documents, the check is performed remotely and it does not make it possible to perform a check in real time for a better reliability of the checking system.

According to one implementation and embodiment, there is proposed a standalone system for checking dispensing that does not entail exchanges with a central station.

According to another implementation and embodiment, there is proposed a real time checking system that is simple and that can be implemented inexpensively.

In particular, there is proposed a checking system which is suitable for elderly people, or handicapped people who have heavy medication.

According to one aspect, there is proposed a device for detecting obj ects, in particular forming a portable element that is capable, for example, of being carried in the pocket of a user, comprising a vessel intended to contain the objects to be detected, means for capturing at least one image of the vessel, and means for processing said at least one captured image.

The processing means comprise detection means for detecting obj ects of said at least one captured image, extraction means for extracting characteristics of each detected obj ect, generation means for generating a list of the characteristics of each detected obj ect and a memory for storing said generated list, the memory also being configured to store a first reference list of obj ect characteristics and the processing means also being capable of generating a second list o f characteristics from a captured image, the detection device also comprising means for comparing obj ects configured to compare the characteristics o f each obj ect of the second list with, respectively, the characteristics o f each obj ect of the reference list.

Thus, a device is provided which can be portable to be transported about the person easily. Furthermore, such a device is standalone energy-wise because it can be fitted with a built-in battery. It is also standalone in its operation because it does not necessarily have to be fitted with connection means. It also does not have to be remotely controlled from the outside.

The processing means are also capable of generating said reference list from a captured reference image.

According to one embodiment, the extraction means comprise segmentation means configured to segment the unrecognized obj ects out of the detected obj ects and perform a second obj ect detection for each unrecognized object.

According to another embodiment, the memory is configured to store a list of characteristics o f known obj ects, and the segmentation means also comprise shape recognition means configured to compare the characteristics o f each detected obj ect with, respectively, the characteristics o f each known obj ect in order to identify the recognized obj ects out of the detected obj ects .

According to yet another embodiment, the device comprises signalling means configured to signal the result o f the comparison obtained from the comparison means.

The processing means can also comprise background detection means capable of detecting if at least one obj ect is present within the vessel in a captured image.

The computation resources are not used unnecessarily to make a recognition if no obj ect is detected. The device may also comprise control means capable o f interrupting the processing means in the case where no obj ect is detected within the vessel.

The processing means may also comprise means for preprocessing the captured image.

According to another embodiment, the vessel has a dark and matt bottom.

The dark colour makes it possible for the medicines to be distinguished more easily from the background. The matt tint provides a reduction of the glare which could be confused with the medicines.

Thus, the detection o f the background and o f the obj ects are more accurate.

The vessel may also have a bottom comprising a number o f identical patterns .

The identical reproduction of a pattern with a spatial frequency makes it possible to improve the detection of the bottom o f the vessel.

Moreover, the vessel may have a relief bottom to prevent medicines from being stacked one on top of the other.

The device may also comprise means for displacing the objects contained in the vessel, such as vibrators that can be actuated, for example, by the user.

In another embodiment, the vessel comprises a transparent bottom, the device comprising lighting means arranged under the vessel and configured to light the vessel through its transparent bottom.

The device may also comprise an optical system, for enabling said means for capturing at least one image of the vessel to capture an image o f said obj ects to be captured.

The device may be configured in size and in weight to make it portable.

The device may comprise means for wedging the obj ects in said vessel. According to another aspect, there is proposed a use of the device defined hereinabove, as pill dispenser containing medicines forming said obj ects to be detected.

According to yet another aspect, there is proposed a method for detecting obj ects contained in a vessel, in which at least one image o f the vessel is captured and said at least one captured image is processed.

According to a general characteristic of this aspect, the processing comprises a detection o f the objects of said at least one image, an extraction o f the characteristics o f each detected obj ect, a generation of a list of the characteristics of each detected obj ect and a storage of said list of the characteristics, this method also comprising the storage of a first reference list o f obj ect characteristics, a capture of an image, a processing of said captured image so as to store a second list of characteristics, and a comparison o f obj ects comprising a comparison o f the characteristics o f each obj ect of the second list with, respectively, the characteristics o f each obj ect of the reference list.

The method is pyramidal, in the sense that the same simple processing is performed on each of the detected regions of the image and for each o f the regions, a more complex processing is provided.

The more complex processing also comprises the implementation o f the same simple processing on segmented regions of the image.

Furthermore, making a comparison on a list of extracted characteristics and not directly on the image allows for a simp le, accurate and rapid comparison.

According to one implementation, the method comprises a capture of a reference image, a processing of said reference image so as to store said reference list of characteristics.

According to another implementation, the processing comprises a segmentation of the unrecognized obj ects, out of the detected obj ects, and a second obj ect detection for each unrecognized obj ect.

According to another implementation, the method comprises a storage of a list of characteristics of known obj ects, and the processing also comprises a shape recognition comprising a comparison o f the characteristics o f each detected obj ect with, respectively, the characteristics o f each known obj ect, in order to identify the recognized obj ects out of the detected obj ects .

According to another imp lementation, the method comprises a signalling of the result of the comparison obtained from the comparison step .

According to yet another implementation, the method also comprises a first signalling if at least one obj ect of said image captured so as to store a second list of characteristics is unrecognized, and a second signalling if, fo llowing a comparison o f the characteristics o f each obj ect of the second list with, respectively, the characteristics o f each obj ect of the reference list, at least one obj ect of the second list is different from the obj ects of the reference list.

According to yet another implementation, the metho d comprises a displacement of the obj ects contained in the vessel fo llowing said first signalling, a capture of a new image, a processing of said new captured image so as to store a new list of characteristics, and a comparison o f obj ects comprising a comparison of the characteristics o f each obj ect of the new list with, respectively, the characteristics o f each obj ect of the reference list.

According to another implementation, the method comprises a capture of another image of the vessel, a processing of said other captured image, the processing comprising a detection o f the background of the vessel and a third signalling if at least one obj ect is detected within the vessel in said other processed image.

The processing may comprise a detection of the background o f the vessel and is interrupted in the case where no obj ect is detected within the vessel.

The processing may also comprise a step for preprocessing said at least one captured image.

Thus, the subsequent steps are simpler because the image is processed beforehand to reduce the number of calculations made during subsequent processing operations. This preprocessing step also makes it possible to improve the relevance of the results obtained during subsequent processing operations .

Other advantages and features of the invention will become apparent from studying the detailed description of implementations and embodiments which are in no way limiting, and the appended drawings in which:

- Figure 1 illustrates one embo diment of an obj ect detection device according to the invention;

- Figure 2 schematically illustrates a method for checking the dispensing o f medicine;

- Figure 3 illustrates an image processing method according to the invention;

- Figure 4 illustrates one embodiment of the processing system; and

- Figure 5 illustrates another embodiment o f the obj ect detection device.

Figure 1 represents an obj ect detection device 1 comprising at least one vessel 320, or pill dispenser, for receiving obj ects, in particular medicines. This obj ect detection device 1 comprises image capture means 301 , input means 302, signalling means 303 , and an electronic control unit ECU, such as, for example, a microprocessor.

This ECU comprises control means 3 1 1 , processing means 3 12 and a memory 3 13. The role of each of these elements is described in the fo llowing Figures 2 to 4.

Figure 2 illustrates the main steps o f one implementation o f a method for checking the taking of medicines, according to the invention. These steps are also described with reference to Figure 3 which represents one embodiment.

After a first initialization step (step 101 ) and the first placement by the user of the medicines ( 102), in a vessel, a first image is captured ( 104) . The user may, for example, be a pharmacist, a doctor, a nurse, or the patient himself. This first image is called reference image because, as will be indicated below, the validation o f the posology will be performed according to the processing of this first image.

In the case where the vessel comprises a number o f compartments 320 used to receive the obj ects to be detected, as is illustrated in Figure 1 , the image capture step is performed for each o f the compartments. Each compartment (morning, midday, evening) o f each o f the drawers (Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, Sunday) is situated under the image capture means 301 . For example, the image capture means may be of monochrome or colour type .

The photographing o f the compartments and the identification of the compartments are carried out by control means 3 1 1 according to information from the input means 302.

For example, in the case o f a system with drawers, the input means 302 comprise a system o f degree-of-opening sensors placed on each o f the drawers to deduce therefrom the position o f a compartment of a drawer relative to the image capture means 301 . A complementary image processing system makes it possible to determine the height of the vessel to deduce therefrom the corresponding day.

According to another example, the day and the timeband may also be entered by the user by means o f touching the input means .

The capture of the reference image for each compartment can be triggered via the control means 3 1 1 by pressing on a dedicated button CO (step 103) o f the input means 302. The capture of the reference image may also be triggered via the control means 3 1 1 by the system o f degree-of-opening sensors of the input means.

Fo llowing the capture of the reference image or images (step 104) by the capture means 301 , a processing step is performed on each of the images (step 105) .

The steps of the processing method are illustrated in Figure 3 and the implementation means are described in Figure 4.

In order to improve the computation steps during captured image processing operations, it is possible to perform steps 201 , 102 for preprocessing the captured image beforehand. For example, it is possible to make a lighting correction on the captured image (step 201 ) . Then, for example, the sharpness is corrected (step 202) . These two steps are performed by the preprocessing means 401 . The preprocessing means 401 may also, for example, be configured to make corrections to the colours and the contrast of the captured image.

Then, the detection means 402 perform a background detection (step 203) . According to this step, if, in the image, only a background is detected, then the second control means interrupt the processing (step 204) . In the contrary case, shape detection means 403 perform steps for obj ect detection (step 205) and shape analysis (step 206) . The obj ect detection step (step 205) comprises a region detection which consists in searching for detected pixels and an agglomeration o f the similar and adj acent pixels. The processing of an image also comprises a shape analysis step (step 206) . This shape analysis 206 and the subsequent steps are performed independently on each of the detected regions. This is illustrated in Figure 3 by the numerous vertical arrows .

Then, the recognition means 404 perform a step (step 208) for recognizing these shapes by comparison with a list o f characteristics of known obj ects. As an exemp lary embodiment, said list o f characteristics o f known obj ects is stored in the memory 3 13. This list of known obj ects comprises, for example, a square, a circle, an equilateral triangle, a half-disc, and a number of other oblong and semi-oblong shapes. In practice, in the case o f conventional medicines which must be placed appropriately by the user, that is to say without stacking, the obj ects that do not observe these criteria are not medicines or are medicines balanced on an edge or stacked with other medicines. Furthermore, this list of characteristics of known obj ects comprises, for each known obj ect, different characteristics. These characteristics may be shape characteristics, such as, for example, an obj ect width, a height, an obj ect diameter; or colorimetric characteristics, such as, for example, a tint, a contrast, a luminance, a colour; or even object type characteristics, such as, for example, a square, a circle, an ellipse, etc.

It may be noted that this list of characteristics of known objects is also a list of objects because it comprises characteristics of a number of known objects.

During this shape recognition step 208, if a detected region whose analysed shape is recognized by a comparison of objects, then the addition means 405 add, to a list of medicines LP, the detected object, considered to be a recognized object, whose shape is recognized (step 209), as well as all the extracted characteristics, such as, for example, the geometrical shape, the size, the colour, etc.

The comparison of objects consists in comparing the characteristics of the detected objects with, respectively, the characteristics of each object of the list of characteristics of known objects.

This comparison of objects is differentiated from a simple comparison of images by the images being compared pixel by pixel.

This list of medicines LP is a list of characteristics of the detected objects, also denoted list of detected objects. This list of medicines LP is generated following a step for extraction of the characteristics of the detected objects 211.

The list LP is, for example, stored in the memory 313.

In the case where the shape recognition could not culminate on the region detected and analysed, then, when possible, the segmentation means 406 perform a segmentation (step 213) on said region to identify the recognized objects and the unrecognized objects. For this, the region is, for example, split into two and the processing operations are performed independently on the two regions.

Following the segmentation step 213, the means for extracting characteristics 407 extract as many characteristics as possible (step 211) from the recognized region. Furthermore, the means for extracting characteristics are also configured to extract characteristics from the unrecognized and non-segmentable region, and in this case these characteristics can be used to identify obj ects that have complex characteristics.

Then, a step 212 for generation o f the list of medicines LP is performed, said list comprising the list of each detected obj ect, recognized or unrecognized, and their characteristics extracted in the preceding extraction step 21 1 .

Fo llowing this processing operation, a test 106 is then carried out by the control means 3 1 1 . If there is at least one element in the list LP then a signal (step 1 17) is indicated to the user. The signalling is performed by signalling means 303 and is triggered by the control means 3 1 1 . As an exemplary embodiment, these signalling means may be a loudspeaker or an LCD screen. They may also comprise a touch sensitive screen and, in this case, the signalling means are also the input means 302.

With this signal, the user is prompted to ensure that all the medicines are correctly placed (not balanced, not stacked) for example by gently knocking the vessel. It is also possible to consider actuating a vibrator incorporated in the device to modify the position o f the medicines. It is also possible to modify the lighting, by adjusting lighting means 304, and to capture another image to improve the identification o f the medicines. It is also possible to use the audible warning so that the user can manually modify the position of the medicines. If these modifications still do not make it possible for the unrecognized obj ect or obj ects to be recognized, these unrecognized obj ects are then considered as complex obj ects.

The method then continues with the step 104.

In the case where the conditions C I of the test are satisfied, that is to say, if there is at least one element in the list LP, then this list LP is stored in the memory 3 1 3 by the control means 3 1 1 which trigger the continuation o f the method.

In order to check that the user has correctly filled the vessel, that is to say, when he replaces the medicines in the vessel, a second image o f the vessel is captured, said second image is processed to establish a second list of characteristics, or of obj ects, and the obj ects of this second list are compared with the objects of the first list established from the reference image. This comparison o f obj ects is performed, by a comparison o f the characteristics o f the obj ects of the second list with, respectively, the characteristics of each obj ect of the first list.

In particular, after the medicines have been replaced in each o f the compartments of the vessel, a capture of a second image (step 109) is performed fo llowing the triggering (step 108) by the control means via the input means 302. For example, the input means comprise a third dedicated button RM. Then, the second captured image is processed by the processing means 3 12. The processing operation is identical to the one performed previously. Fo llowing this processing operation, a comparison o f obj ects 1 1 1 is triggered by the control means 3 1 1 . This comparison o f obj ects 1 1 1 makes it possible to validate whether the posology is correct, that is to say, whether the user has filled the compartments with the medicines corresponding to those of the reference image.

During this comparison of obj ects 1 1 1 , the obj ects of the image that has just been captured are compared with the obj ects of the first list. If the second list is empty or is not identical to the first list o f obj ects, then the replacement of the medicines is not correct. The control means 3 1 1 trigger the signalling means. With this second signal, the user is prompted to ensure that the replacement of medicines is correct, for example, by correctly placing all the necessary medicines or by removing the surplus or even by ensuring that all the medicines are correctly placed (not in balance, not stacked), for example by gently knocking the vessel. The method then continues with the step 1 12.

In the case where the conditions C2 of the comparison o f obj ects 1 1 1 are satisfied, that is to say, if the second list is not empty and if it corresponds to the list o f obj ects of the first list, that is to say that the obj ects of the second list are identical to the obj ects of the first list, then the method continues with the medicine-taking step (step 1 12) . The step 1 12 corresponds to the step of taking of the medicine by the user from a compartment of the vessel. For this, the drawer o f the corresponding day must be opened and the timeband is placed facing the capture means. After the medicine has been taken, the input means trigger the capture of a third image. To this end, the input means may comprise a dedicated button PM. The input means may also comprise a system of degree-o f-opening sensors placed on each of the drawers as indicated previously.

Then, the third captured image is processed by the processing means 3 12. The processing operation is identical to the one performed previously. Fo llowing this processing operation, a comparison o f obj ects 1 16 is triggered by the control means 3 1 1 . If there is at least one obj ect in the third list, then a third signal (step 1 19) is indicated to the user. To this end, the control means 3 1 1 also trigger the signalling means 303. With this third signal, the user is prompted to ensure that the compartment is indeed empty by, for example, finishing all the medicines in the compartment.

In the case where the conditions C2 of the comparison o f obj ects are satisfied, that is to say, if the third list is not empty, then the method continues according to two alternatives. Either there are still compartments from which the user must take his medicines and the medicine-taking step 1 12 is repeated, or the takings of medicines have been performed and the dispenser must be refilled. To this end, the medicine-replacement step 1 07 is performed once again.

The methods o f Figures 2 and 3 are respectively implemented in the control means 3 1 1 and in the processing means 3 12.

Figures 1 and 4 illustrate an architecture according to which the image capture means 301 , the input means 302, the signalling means 303 , and the ECU (electronic control unit) cooperate to perform the check on the dispensing of the medicines. The ECU comprises the control means 3 1 1 , the memory 3 13 and the image processing means 3 12. The latter comprise the preprocessing means 401 , the background detection means 402, the second control means 408 , the shape recognition means 404, the detection means 403 , the addition means 401 , the segmentation means 402 and the characteristics extraction means 408.

The means mentioned hereinabove are all incorporated in the device for detecting objects 1 , which renders the device independent of any external additional checking system.

The means mentioned hereinabove can be implemented in the form of so ftware modules, for example, in one or more computers .

Figure 5 shows another embodiment of a device for detecting obj ects 600. This device 600 comprises a vessel 601 that has a number of compartments 602 intended to receive medicines 603 . The compartments 602 are provided with a bottom which is transparent and which has surface irregularities 604, pyramidal for example. These surface irregularities make it easier to separate the medicines and make it possible to make the image processing simpler and more reliable.

Moreover, the compartments 602 include a patterned foam 605 for wedging the medicines and making it easier to extract the bottom of the vessel during image processing.

Furthermore, the device 600 comprises a rotary lid 606 which includes a total opening, that is to say, one which opens all the compartments 602, for the filling o f the vessel 601 and an opening for each compartment for the daily taking of medicines .

The device 600 is also fitted with a single fixed camera 607, lighting systems 608 , 609 for lighting under the vessel 601 and for lighting the compartments 602 through their transparent bottom, and an optical system 610, for example a spherical mirror, which makes it possible to limit the height of the device 600, while allowing said camera 607 to take an image of the obj ects of all the compartments o f the vessel.

Also shown are the light paths 700, 701 between the compartments 602 and the optical system 61 0 on the one hand, and between the optical system 610 and the camera 607 on the other hand. Thus, the device 600 has a height which can be low and which makes it possible to manufacture a device 600 with little bulk, for example which may be carried in a pocket of a user.

The device 600 also comprises a ECU comprising control means, processing means and a memory described in the preceding Figures 2 to 4.

In particular, this device 600 operates in any position since the medicines are wedged with the help of the fo ams and the image capture can be done in all positions.

Thus, a device for detecting obj ects that is portable is provided, that is to say, one that has a small size and low weight, designed to be easily carried about the person, in the pocket for example.

According to possible enhancements of the invention, during the image processing operations, co lour may also be used, allowing for simpler region detection and segmentation steps.