System and method for constructing three-dimensional model of dentition of at least one user

Abstract

The invention relates to a system for constructing a three-dimensional model of a dentition of at least one user, comprising: at least one intraoral device for placement in an oral cavity of the at least one user; at least one camera assembly disposed in the at least one intraoral device for capturing at least one stereoscopic image of the oral cavity of the at least one user; and at least one processing device receiving at least one stereoscopic image; wherein the at least one processing device comprises at least one trained neural network that analyzes the at least one stereoscopic image to estimate at least one depth map; and wherein the at least one processing device further comprises at least one positioning and mapping module that sequentially integrates the at least one stereoscopic image and the at least one depth map.

Description

TECHNICAL FIELD

[0001] The present invention relates to a system and method for constructing a three-dimensional model of the dentition of at least one user in the field of dentistry.

[0002] The system and method of the present invention are implemented by using passive stereo technology combined with neural networks, which makes the intraoral imaging device only need to contain the camera necessary to complete the task, while all data processing is performed outside the intraoral device, thus significantly reducing its complexity and cost compared to prior art solutions.

[0003] In this sense, the system of the present invention mainly comprises: at least one intraoral device to be placed inside the mouth of at least one user; at least one set of cameras arranged in the at least one intraoral device for capturing at least one stereo image inside the mouth of the at least one user; and at least one processing medium that receives the at least one stereo image; wherein the at least one processing medium comprises at least one trained neural network that analyzes the at least one stereo image to estimate at least one depth map; and wherein the at least one processing medium further comprises at least one localization and mapping module that integrates the at least one stereo image and the at least one depth map, in turn, into a generated three-dimensional model.

[0004] On the other hand, the method of the present invention mainly comprises the following steps: capturing at least one stereo image by means of at least one set of cameras arranged in at least one intraoral device; receiving the at least one stereo image by means of at least one processing medium; analyzing the at least one stereo image to estimate at least one depth map by means of at least one trained neural network comprised by the at least one processing medium; and integrating the at least one stereo image and the at least one depth map, in turn, into a generated three-dimensional model by means of at least one localization and mapping module comprised by the at least one processing medium.

[0005] Based on the system and method of the present invention, not only can the accessibility of such technology to dental patients be improved by simplifying the physical devices used for these procedures, but also a more accurate solution is provided than the solutions that use traditional modeling techniques. BACKGROUND

[0006] In dentistry and its different specialties, the use of techniques capable of modeling the dentition of patients is a widely disseminated practice due to the rapidity and simplicity of its modeling process.

[0007] Most of these solutions consider the use of an intraoral device or scanner that contains at least one camera that uses known three-dimensional modeling principles, such as confocal microscopy, active stereo or structured light, combined with algorithms that are usually executed inside the device itself.

[0008] In this sense, confocal microscopes are the most complex technology in terms of hardware usage and therefore more expensive to manufacture and implement. This technology estimates the depth value of an image by changing the focus of the light source and using a lens to filter out the out-of-focus light rays. For each focus setting, the camera receives light rays from the surface area that is in focus. Since the focus of the light rays is controllable and the geometry of the lens is known, there is a relationship that allows the depth value of the in-focus area to be estimated.

[0009] On the other hand, scanners that use active stereo technology capture pairs of images through stereo cameras while projecting a pattern onto the surface. Then, an algorithm calculates the correspondence between the two images using the projected pattern. This technology requires the setting of a pattern projector inside the scanner or intraoral device, which also makes its structure more complex.

[0010] Finally, scanners that use structured light technology project a known light pattern onto the surface and capture an image with a camera. Subsequently, by observing how the light pattern deforms on the surface, an algorithm infers the topology and depth of this surface. Again, scanners that use this technology must have additional hardware, which makes them more expensive to manufacture.

[0011] In addition, since most of these solutions are processed inside the intraoral scanner, great care must be taken to avoid damage to the device, since possible repairs can generate high costs.

[0012] Therefore, there is an increasing need for a system and method that not only simplifies the construction of scanners or intraoral devices by using fewer components and separating the data acquisition and processing processes of the intraoral device, but also a system and method that can significantly improve the quality of three-dimensional modeling, since the state of the art still has a lot of room for improvement in this regard.

[0013] In the patent field, there are solutions that point to three-dimensional modeling devices or systems for the dentition in the dental field. For example, US patent application US20140146142A1 describes a three-dimensional measuring device aimed at measuring without structures or active light projection, comprising an image acquisition device and a data processor for the images. The image acquisition device is capable of capturing at least two images simultaneously or almost simultaneously, one of which is completely or partially contained in the other. The contained image describes a narrower field of view than the other image and has a higher precision than the other image.

[0014] In this sense, although the document US20140146142A1 describes a device with at least one camera that processes the information outside the device, specifically on a dedicated computer, it does not mention the use of a neural network previously trained with a dental arch model in the information processing process, mentioning only the use of several algorithms at different stages of the process. Since the processing in the document US20140146142A1 is carried out on the computer of the professional who carries out the operation, the algorithms used cannot be too complex, since this would require a computer with high processing capacity, which would make the device usable only by a few people. Therefore, this document fails to provide an algorithm that can obtain results superior to the state of the art, since its processing capacity would be limited to the processing capacity of the user's computer, which does not occur in the present invention, since in the present invention, since the processing medium is located outside the intraoral device (it can be hosted in the cloud), the processing capacity can be significantly increased by using, for example, a neural network trained with a large number of parameters, significantly improving the results of the three-dimensional modeling.

[0015] Another example is the disclosure in international patent application WO2021250091A1, which describes an automatic segmentation method of a dental arch, comprising obtaining a three-dimensional surface of the dental arch, to obtain a three-dimensional representation containing a set of vertices, generating virtual views from the three-dimensional representation, projecting the three-dimensional representation onto each two-dimensional virtual view to obtain an image representing each vertex in the virtual view, processing each image using a deep learning network, back-projecting each image in order to assign to each vertex in the three-dimensional representation the pixels that appear in each image and correspond to it, and assigning to each vertex one or more probability vectors to determine the dental tissue class to which each vertex belongs.

[0016] When comparing the description of document WO2021250091A1 with the present application, it can be observed that, although the document describes an intraoral device that uses trained neural network technology for image processing, the algorithm is arranged within the processing module in the intraoral device, which greatly limits the processing capacity of the device, limiting the complexity of the neural network used. This does not occur in the present invention, in which all processing is carried out outside the intraoral device, thus not limiting the processing capacity and improving the quality of the three-dimensional modeling, allowing the reconstruction of the three-dimensional model to be obtained practically in real time. In addition, the structure of the intraoral device is simplified, since it only has the cameras and related circuitry necessary to obtain the images.

[0017] As can be seen from the above documents, the purpose of the vast majority of such devices and systems is not to simplify the structure of the intraoral device, thus saving economic costs, but to solve other types of problems, such as avoiding the use of structured light. Although there are solutions that perform data processing outside the intraoral device or that use neural networks for image analysis, these solutions are very far from the one described in the present invention, which, in addition to providing a simplified structure and use of the device, allows performing the processing on a server located in the cloud, which, together with a device that does not require the user to have on-site processing capabilities, allows significantly improving the quality of the three-dimensional modeling of the dentition, since more complex processing algorithms can be used in combination with highly parameterized, trained neural networks to analyze the received images and thus determine the corresponding depth maps.

[0018] Therefore, it is necessary to provide a system and method that not only provides a simpler, less costly intraoral device, but also improves the quality and speed of the three-dimensional modeling of the patient's dentition so that the dentist responsible for the operation can obtain the results in real time. Furthermore, a solution is needed that allows avoiding the use of complex and costly equipment for data processing, the present invention achieves the aforementioned objectives by receiving the information obtained by the intraoral device directly from the latter (or via a computer or similar electronic device as an intermediary) through a server and viewing the results. This and other advantages associated with other aspects of the technology will be described in more detail below. SUMMARY

[0019] The present invention relates to a system and method for constructing a three-dimensional model of the dentition of at least one user in real time, which simplifies the structure of the intraoral device used while improving the quality and speed of the three-dimensional modeling.

[0020] According to a first preferred embodiment of the present invention, a system for constructing a three-dimensional model of the dentition of at least one user comprises: - at least one intraoral device placed in the mouth of at least one user; - at least one set of cameras arranged in said at least one intraoral device for capturing at least one stereoscopic image of the mouth of said at least one user; and - at least one processing medium that receives at least one stereoscopic image; wherein said at least one processing medium comprises at least one trained neural network that analyzes said at least one stereoscopic image to estimate at least one depth map; and wherein said at least one processing medium further comprises at least one positioning and mapping module that sequentially integrates said at least one stereoscopic image and said at least one depth map into a generated three-dimensional model.

[0021] The system of the present application operates according to passive stereo technology. This means that it uses pairs of synchronized images and algorithms that exploit trained neural networks that can estimate the depth values of the scanned surface without projecting anything onto it, unlike the common prior art solutions.

[0022] In scenes with little texture and / or reflections, such as a patient's mouth, traditional passive stereo algorithms usually perform poorly. In these cases, the poor performance is caused by traditional algorithms that look for key points in each image and then try to match the key points in one image with the key points in the other image (this is called stereo matching). In the case of little texture and / or reflections, the key points are very blurry and the algorithm makes many mistakes in the process.

[0023] The present application uses an algorithm that is enhanced by using a highly complex neural network, instead of a traditional stereo matching algorithm. This neural network estimates the depth values of the images without using explicit key points and achieves a much higher accuracy than traditional stereo matching algorithms. To achieve this accuracy, the neural network must be trained with highly realistic synthetic images and depth values.

[0024] In this sense, by using this neural network, the algorithm of the present application requires high processing power. To avoid making the intraoral device more complex and therefore more expensive, it was decided to perform said processing outside the device, either in an external electronic equipment suitable for it or in a server set up in the cloud. Based on the above, thanks to the work of the neural network, it is possible to solve the two problems presented in this application, such as simplifying the hardware used, while improving the quality of the three-dimensional modeling of the patient's dentition that the user can obtain in real time.

[0025] According to another embodiment of the present application, the at least one processing medium further comprises at least one post-processing module that removes at least one noisy depth value from the three-dimensional model and recalculates the pose of the at least one set of cameras. Said post-processing is performed using all the information captured during the scan performed by the intraoral device. The recalculation of the pose of the at least one set of cameras indicates that, as the images are captured and the depth values are estimated, the system of the present application estimates the position in which the at least one set of cameras is located. This is done to know how to combine the different depth values. When estimating the position of the at least one set of cameras, a cumulative error is also generated that must be corrected in the post-processing phase.

[0026] According to another embodiment of the present application, the system further comprises at least one receiving device that receives at least one stereo image from the at least one set of cameras and sends it to the at least one processing medium.

[0027] According to another embodiment of the application, the at least one receiving device is at least one of a computer, a laptop, a tablet and a smartphone.

[0028] According to another embodiment of the application, at least one processing medium is arranged in the at least one receiving device. This allows the user to perform data processing on their own equipment, provided they have equipment with the necessary processing capacity.

[0029] According to another embodiment of the application, the at least one receiving device comprises at least one display interface. The display interface allows the user to see in real time the images acquired by the intraoral device and / or the final three-dimensional model of the dentition.

[0030] According to another embodiment of the application, the at least one processing medium is arranged in the cloud. This enables users who do not have sufficient processing capacity to host the processing medium and the neural network to choose to send the information collected by the intraoral device directly to a cloud server, or to use a receiving device (which, as mentioned previously, can be a computer, a tablet, etc.) as an interface.

[0031] Furthermore, since the processing medium and the neural network are in the cloud, the system can simultaneously analyze information from multiple intraoral devices and return the corresponding three-dimensional models in real time. With this functionality, it is possible to further reduce the cost of the system, since it is not necessary to equip each intraoral device in operation with a processing medium, unlike in the usual solutions.

[0032] According to another embodiment of the application, the at least one set of cameras comprises at least one first camera and at least one second camera. This allows the acquisition of pairs of synchronized images necessary for the algorithm and the neural network to correctly estimate the depth of the scanned surface.

[0033] According to another embodiment of the application, the at least one intraoral device communicates wirelessly with the at least one processing medium.

[0034] According to another embodiment of the application, the at least one intraoral device communicates with the at least one processing medium through a communication cable.

[0035] According to another embodiment of the application, the at least one intraoral device communicates wirelessly with the at least one receiving device.

[0036] According to another embodiment of the application, the at least one intraoral device communicates with the at least one receiving device through a communication cable.

[0037] According to another embodiment of the application, the at least one receiving device communicates wirelessly with the at least one cloud.

[0038] According to another embodiment of the application, the at least one intraoral device further comprises at least one battery that enables the intraoral device to operate without being directly connected to a power source, such as a plug.

[0039] According to another aspect, a method for building a three-dimensional model of the dentition of at least one user, according to a second preferred embodiment of the application, is also described, comprising the following steps: a) capturing at least one stereoscopic image by means of at least one set of cameras arranged in at least one intraoral device; b) receiving at least one stereoscopic image by means of at least one processing medium; c) analyzing at least one stereoscopic image by means of at least one trained neural network comprised in at least one processing medium, to estimate at least one depth map; and d) sequentially integrating said at least one stereoscopic image and said at least one depth map into a generated three-dimensional model by means of at least one localization and mapping module comprised in said at least one processing medium.

[0040] As mentioned above, while capturing the images, the intraoral device starts sending the stereoscopic images in real time to a linked equipment, such as a computer, which then sends said information to a cloud server where the processing medium is located, or directly to the processing medium arranged in the computer. Once the images are available for analysis by the processing medium, a neural network specifically trained for this task analyzes each stereoscopic image sent and estimates a depth map from them.

[0041] The depth maps are integrated together with the stereoscopic images into the reconstruction generated at the current instant by means of at least one localization and mapping module that compares the information it receives with the partial reconstruction of the scene and predicts the camera poses of the stereoscopic images.

[0042] According to another embodiment of the application, the method further comprises removing at least one noisy depth value from the three-dimensional model and recalculating the poses of the at least one set of cameras by means of at least one post-processing module comprised in at least one processing medium.

[0043] According to another embodiment of the application, the method further comprises, before step c), receiving said at least one stereoscopic image from said at least one set of cameras by means of at least one receiving device, which then sends it to said at least one processing medium.

[0044] According to another embodiment of the application, the method further comprises displaying the information sent and received from at least one processing medium by means of at least one display interface provided on at least one receiving device.

[0045] According to another embodiment of the application, the method also comprises generating in real time a three-dimensional model of the dentition of the at least one user and sending it to at least one display interface.

[0046] Finally, according to a third preferred embodiment of the application, a computer-readable storage medium is also described, comprising instructions which, when executed by at least one processor, cause the at least one processor to perform the method for constructing a three-dimensional model of the dentition of at least one user.

[0047] As can be seen from the foregoing, the important difference between the present application and the prior art solutions lies in the fact that the present system does not employ active stereo, structured light or confocal microscopy techniques, but rather uses a powerful neural model which allows it to estimate depth values using only passive stereo techniques. This means that the sensors of the intraoral device are considerably less complex and expensive, since they only comprise at least one ordinary camera and a circuit which synchronizes them.

[0048] Furthermore, the depth estimation from the stereo images, as well as the reconstruction process and post-processing, can be performed on a server set up in the cloud, where the processing medium is located together with the trained neural network. In this case, the dentist's computer only sends the information obtained through the intraoral device to this data cloud and allows the status of the reconstruction to be viewed in a real-time display interface, so that the dentist can control the process. By contrast, traditional scanners perform the reconstruction and post-processing on the dentist's own computer, which needs to have sufficiently powerful hardware to accommodate the processing algorithms used, where the algorithms used by these solutions must usually be adapted to this type of computer, which directly impairs the quality of the three-dimensional modeling obtained, whereas this does not occur in the present application, where the quality of the modeling is improved due to the fact that there is no such limitation in terms of processing capacity.

[0049] Finally, none of the prior art solutions is capable of providing the possibility of using a single processing medium to analyze the data sent by multiple intraoral devices, thus optimizing the use of resources, reducing the costs associated with the implementation of the system and making it easier for dentists and patients to access this type of technology. BRIEF DESCRIPTION OF DRAWINGS

[0050] As part of the present application, the following representative diagram is shown, which shows the preferred configuration of the present application and therefore should not be considered as a limitation of the claimed subject matter.

[0051] Figure 1 A block diagram of an intraoral scanning process according to the prior art is shown.

[0052] Figure 2 A general scheme of the passive stereo technique according to the preferred configuration of the present application is shown.

[0053] Figure 3 a block diagram of an intraoral scanning process according to a preferred configuration of the present application is shown. DETAILED DESCRIPTION

[0054] With reference to the drawings, Figure 1 A block diagram of a traditional intraoral scanning process to acquire a three-dimensional image of a patient's dentition is shown. In particular, a first phase of data capture and reconstruction consisting of two sub-phases (1) is observed. The first sub-phase (la) refers to the actual scanning performed by an expert, such as a dentist, in the patient's mouth. This is done by inserting a physical device into the user's mouth, which includes at least one data capture device. During the scanning process, the scanner sends the data it captures in real time to the computer of the dentist, where these technical scanners usually work under the theory of confocal microscopy or structured light (active stereo), which are usually measurements of images, depth and inertial measurement units (IMU) including accelerometers and gyroscopes to measure angular velocity and acceleration. Once the information obtained by the scanner is sent to the computer of the expert, the computer sequentially integrates the images, depth values and other measurements to estimate the pose of the camera in each image, thus constructing a three-dimensional model (sub-phase (lb)). In this way, a first three-dimensional model of the patient's dentition is obtained, which is still inaccurate and unclean (block (2)).

[0055] In view of the three-dimensional model obtained in the first phase, which is not suitable for dental treatment, a post-processing phase (3) is required, in which the first three-dimensional model is cleaned by removing noise points and non-dentition corresponding points (sub-phase (3a)). Thereafter, the computer of the dentist recalculates the reconstruction using all the information received during the scan (sub-phase (3b)). Finally, the camera poses and depth values are optimized to minimize the reprojection error (sub-phase (3c)), which corresponds to the difference between the captured images and the images generated from the three-dimensional model reconstruction, after which a corrected and cleaned three-dimensional model is obtained (block (4)), which allows the expert to use it to develop a specific treatment for the patient (block (5)), where we can mention invisible aligners, relaxed occlusion planes, crowns, etc.

[0056] It should be noted that, given the number of images to be processed, these prior art require the expert to be equipped with a computer (10) with high resource processing capacity and capable of executing the processing algorithms necessary to generate the three-dimensional model, where at least the phases described in blocks (1), (2) and (3) must be executed.

[0057] On the other hand, Figure 2 a general scheme of the passive stereo technology used by the present application is shown, where it can be understood how the scanner or intraoral device (30) of this technology works. In this sense, in Figure 2In the illustrated embodiment, the intraoral device (30) comprises two cameras (12a, 12b) on the baseline (11), where the cameras (12a, 12b) in turn comprise left and right lenses (13a, 13b), respectively. The left and right cameras (12a, 12b) can be placed parallel to the baseline (11) or at an angle thereto.

[0058] Each camera (12a, 12b) of the intraoral device (30) forms an image plane (14a, 14b) through which the real points (15, 16) are displayed in each image obtained by the cameras (12a, 12b). As mentioned earlier, the intraoral device (30) uses passive stereo technology to estimate the depth values (17) of the scanned surface, which is achieved by an algorithm that processes pairs of synchronized images without the need to project anything on said images, as occurs in other prior art solutions that address this problem.

[0059] This is extremely relevant in scenarios with little texture and / or reflections, such as the patient's mouth, where traditional passive stereo algorithms tend to work poorly because they search for key points in each image and then try to associate these key points of one image with the key points of the other image (stereo matching). Since the surfaces inside the mouth are mostly almost textureless and / or reflective, the key points are very blurry and difficult to locate, so these traditional algorithms make many mistakes and give inaccurate results.

[0060] The present invention improves the accuracy in low-texture and / or reflection situations, such as in the patient's mouth, by using a neural network instead of a traditional stereo matching algorithm, without the need to locate explicit key points for the calculation of the image depth values. To achieve this accuracy, the neural network must be trained with highly realistic synthetic images and depth values.

[0061] Finally, Figure 3 A block diagram of the intraoral scanning process according to the preferred embodiment of the present invention is shown. Specifically, a first data capture phase (100) is shown, which is divided into two sub-phases. The first sub-phase aims to capture pairs of synchronized images (stereo images) by an expert through a scanner or intraoral device, which are then transmitted wirelessly or through a data cable to his computer (sub-phase (100a)). Then, the computer transmits the stereo images in real time to a server or cloud (sub-step (100b)). This intraoral device preferably consists of two cameras and a circuit that synchronizes the capture of the two cameras, where the depth values are calculated from these images by a neural network, which is preferably located in the same cloud that receives the images from the intraoral device. As a result of the first data capture phase (100), a stream of stereo images (block (200)) is obtained from the intraoral device, which is then transmitted to the expert's computer and then to the cloud.

[0062] This is an important difference between the state of the art and what is described in the present application, since it can be seen that the system of the present application only requires the use of the expert's computer (10) in this data capture phase (100), after which the information is processed in the cloud, where the processing algorithms are deposited together with the neural networks. This makes it unnecessary for the expert to have a computer with high processing capacity, since he even only has an electronic device, such as a tablet or a smartphone, capable of connecting to the cloud.

[0063] After this first phase (100) of data capture and sending information to the cloud, where the stereoscopic image stream (200) is obtained, a reconstruction phase (300) is carried out in the cloud, which includes a depth map calculation sub-phase (300a) based on the use of neural networks for each pair of images, and a sequential integration sub-phase (300b) of the depth maps and the three-dimensional reconstruction of the images. The neural networks compare two images (for example, RGB) and estimate the depth value of each pixel based on the relative movement of the objects between the two images. In contrast, conventional scanners use confocal microscopes or structured light to estimate the depth value, which is less accurate and requires the surface to have a significant texture. As a result of this reconstruction phase (300) in the cloud, an inaccurate and unclean three-dimensional model is obtained (block (400)).

[0064] Once the first three-dimensional model (400) is obtained, a post-processing phase (500) must be carried out in the cloud, which includes three sub-phases. The first sub-phase aims to clean the three-dimensional reconstruction by removing noise points or points that do not correspond to the dentition (500a), after which the processing medium uses all the information received during the scanning and data capture phase (100) to recalculate the reconstruction (500b). Finally, the camera poses and depth values are optimized (500c) to minimize the reprojection error, which, as mentioned above, corresponds to the difference between the captured images and the images generated from the reconstruction.

[0065] As a product of these two reconstruction and post-processing phases in the cloud, a corrected and cleaned three-dimensional model is obtained (block (600)) with greater accuracy with respect to the state of the art, where the expert can download the three-dimensional model generated in the cloud to carry out dentition treatments (700), such as invisible aligners, relaxation bite planes, crowns, etc.

[0066] In this sense, it is important to emphasize the importance of carrying out the reconstruction phase (300) and the post-processing phase (500) in the cloud (20), where the processing medium is located, including the trained neural networks and algorithms that make up the localization and mapping module that sequentially integrates the at least one stereoscopic image and the at least one depth map into the generated three-dimensional model. As mentioned above, this not only avoids the need for the expert to have a device that includes high processing capacity, but also allows having, in the cloud, processing tools with higher capacity than those used in the prior art, which are limited by the processing capacity of the intraoral device or the expert's computer. This generates a solution in which the expert receives the result with the corrected and cleaned three-dimensional model in real time, thanks to the high processing capacity and the neural networks that allow obtaining more accurate results in a shorter time.

[0067] Finally, it is also highlighted the possibility of the system of the present invention to operate with a single cloud with a plurality of intraoral devices, which can significantly reduce the costs associated with the implementation of the system, making it available to most experts, requiring only one processing medium, while in the solutions of the prior art, the expert must pay for the entire processing system each time he purchases the product.

[0068] Reference signs 1 Data capture and reconstruction 1a Scanning by the expert and sending the data to the computer 1b Reconstruction of the three-dimensional model 2 Inaccurate and unclean three-dimensional model 3 Post-processing 3a Cleaning of the inaccurate and unclean three-dimensional model 3b Recalculation of the reconstruction 3c Optimization of the camera poses and depth values 4 Corrected and cleaned three-dimensional model 5 Processing 10 Computer of the expert 11 Baseline 12a Left camera 12b Right camera 13a Left lens 13b Right lens 14a Left image plane 14b Right image plane 15, 16 Real points 17 Depth value 20 Cloud 30 Intraoral device 100 Data capture 100a Scanning by the expert and sending the data to the computer 100b Transmission of the stereoscopic images to the cloud in real time 200 stereo image stream 300 cloud reconstruction 300a depth map computation 300b sequential integration into three-dimensional reconstruction 400 inaccurate and dirty three-dimensional model 500 post-processing in the cloud 500a cleaning of inaccurate and dirty three-dimensional model 500b recomputation of reconstruction 500c optimization of camera poses and depth values 600 corrected and cleaned three-dimensional model 700 processing

Claims

1. A system for constructing a three-dimensional model of at least one user's dentition, characterized in that, The system comprises: - at least one intraoral device arranged in the oral cavity of at least one user; - at least one set of cameras arranged in the at least one intraoral device for capturing at least one stereoscopic image of the oral cavity of the at least one user; and - at least one processing medium receiving the at least one stereoscopic image; wherein the at least one processing medium comprises at least one trained neural network analyzing the at least one stereoscopic image to estimate at least one depth map; and wherein the at least one processing medium further comprises at least one localization and mapping module sequentially integrating the at least one stereoscopic image and the at least one depth map into a generated three-dimensional model.

2. The system of claim 1, wherein, The at least one processing medium further comprises at least one post-processing module to eliminate at least one noisy depth value of the three-dimensional model and to recalculate the pose of the at least one set of cameras.

3. The system of any of claims 1-2, wherein, The system further comprises at least one receiving device receiving at least one stereoscopic image from the at least one set of cameras and sending the at least one stereoscopic image to the at least one processing medium.

4. The system of claim 3, wherein, The at least one receiving device is at least one of a computer, a notebook, a tablet and a smartphone.

5. The system of any of claims 3-4, wherein, The at least one processing medium is arranged in the at least one receiving device.

6. The system of any one of claims 3-5, wherein, The at least one receiving device comprises at least one display interface.

7. The system of any one of claims 1-2, wherein, The at least one processing medium is arranged in the cloud.

8. The system of any one of claims 1-7, wherein, The at least one set of cameras comprises at least one first camera and at least one second camera.

9. The system of any one of claims 1-8, wherein, The at least one intraoral device is in wireless communication with the at least one processing medium.

10. The system of any one of claims 1-9, wherein, The at least one intraoral device is in communication with the at least one processing medium through a communication cable.

11. The system of any one of claims 3-10, wherein, The at least one intraoral device is in wireless communication with the at least one receiving device.

12. The system of any one of claims 3-10, wherein, The at least one intraoral device is in communication with the at least one receiving device through a communication cable.

13. The system of any of claims 7-12, wherein, The at least one receiving device is in wireless communication with the at least one cloud.

14. The system of any one of claims 1-13, wherein, The at least one intraoral device further comprises at least one battery.

15. A method of constructing a three-dimensional model of at least one user's dentition using the system of any one of claims 1-14, wherein, The method comprises the following stages: a) capturing at least one stereoscopic image by at least one set of cameras arranged in at least one intraoral device; b) receiving the at least one stereoscopic image from at least one processing medium; c) analyzing the at least one stereoscopic image by at least one trained neural network comprised in the at least one processing medium to estimate at least one depth map; and d) sequentially integrating the at least one stereoscopic image and the at least one depth map into a generated three-dimensional model by at least one localization and mapping module comprised in the at least one processing medium. The method further comprises eliminating at least one noisy depth value of the three-dimensional model and recalculating the pose of the at least one set of cameras by at least one post-processing module comprised in the at least one processing medium.

16. The method of claim 15, wherein, The method further comprises, prior to stage c), receiving at least one stereoscopic image from at least one set of cameras by at least one receiving device and subsequently sending the at least one stereoscopic image to at least one processing medium.

17. The method of any one of claims 15-16, wherein, ​ 18. The method of claim 17, wherein, The method further comprises visualizing the information transmitted and received from the at least one treatment medium via at least one display interface provided in the at least one receiving device.

19. The method according to any one of claims 15-18, characterized by, The method further comprises generating a three-dimensional model of the at least one user's dentition in real time and transmitting it to at least one display interface.

20. A computer-readable storage medium, characterized in that, It comprises instructions which, when executed by at least one processor, cause the at least one processor to perform the method of any one of claims 15-19.

Images