Portable device for analyzing a video stream

Abstract

Portable Video Stream Analysis Device The invention relates to a portable device (2) for analyzing a video stream, the device (2) comprising: - an input (6) for receiving said video stream, for example, a video stream in the visible or invisible spectrum, in particular infrared, - a processing unit (8), receiving said video stream as input and being configured to provide as output information relating to said video stream, for example, identification or recognition information for elements in said video stream, and - a display (10) for displaying said information provided by the processing unit (8), optionally simultaneously with said video stream. The processing unit (8) includes an analysis means (16), for example, an image analysis means, based on artificial intelligence and configured to analyze the video stream and determine information relating to said video stream.The invention also relates to a video analysis system (1), and a corresponding video stream analysis method. Figure for the abstract: Fig. 1.

Description

Title of the invention: Portable device for analyzing a video stream Technical field

[0001] The present invention relates to the field of image processing, in particular video processing. The invention relates more specifically to devices for analyzing a video stream by an artificial intelligence model. Prior art

[0002] It is known to use artificial intelligence models to analyze images. Artificial intelligence models can enable, in particular, better decision-making in a shorter period of time.

[0003] However, artificial intelligence models are most often stored on remote servers, thus requiring an internet connection to transfer the images to be analyzed to the server, before receiving the results, always via the internet connection.

[0004] Online storage of the artificial intelligence model(s) may be due to several factors: significant memory resources incompatible with mobile devices in particular, protection against copying of the model(s), modifications to the model(s), etc.

[0005] Thus, the use of such artificial intelligence models is made restrictive by the quality of internet access, and by the duration of transmission and processing of information, which makes such uses uncomfortable and poorly suited to certain applications, particularly mobile applications. In particular, the use of such remote artificial intelligence models is poorly suited to the analysis of videos for which the number of images per second requires a very high and stable data transfer rate over time. Statement of the invention

[0006] The present invention aims to solve the various technical problems stated above. In particular, the present invention aims to propose a device for analyzing a video stream with more flexible and more efficient use. More specifically, the present invention aims to propose a device for analyzing a video stream making it possible to exploit the advantages of artificial intelligence models more easily and universally.

[0007] Thus, according to one aspect, a portable device for analyzing a video stream is proposed, the device comprising: - an input for receiving said video stream, for example a video stream in the visible or invisible range, in particular infrared, - a processing unit, receiving said video stream as input and being configured to provide as output information relating to said video stream, for example information for identifying or recognizing elements in said video stream, and - a display for displaying said information provided by the processing unit, possibly at the same time as said video stream.

[0008] The processing unit comprises an analysis means, for example an image analysis means, based on artificial intelligence and configured to analyze the video stream and determine the information relating to said video stream.

[0009] Thus, the device according to the invention makes it possible to use, in a portable and autonomous manner, one or more artificial intelligence models for the analysis of any type of video stream. It is then no longer necessary to have an internet connection to analyze a video, but just a transmission of data between the video recording means, or camera, and the portable device according to the invention.

[0010] In addition, the portable device also makes it possible to provide the results of the video analysis in real time, thanks to its display, as the data is provided by the analysis means. The portable device according to the invention thus makes it possible to obtain information from the video stream more reliably and more quickly, without requiring a network or a connection to a network.

[0011] Preferably, the input for receiving said video stream comprises a connection means, for example a socket, in particular HDMI, for connecting, in a wired manner, the portable device to a video acquisition means, preferably separate from the portable device.

[0012] In such a case, the video acquisition means is in proximity to the portable device according to the present invention, and can be connected wired, with a cable, to the portable device according to the present invention to transmit the video data to be analyzed. Such a configuration makes it possible in particular to obtain high transfer rates between the video acquisition means and the portable analysis device.

[0013] Preferably, the input for receiving said video stream comprises a wireless connection means, for example WIFI, for connecting, wirelessly, the portable device to a video acquisition means, preferably separate and remote from the portable device.

[0014] In such a case, the video acquisition means may be remote or sufficiently mobile relative to the portable analysis device to make a wired connection between the two impractical. The video acquisition means then communicates with the portable analysis device via waves, for example WIFI, to enable real-time transmission of video data between the video acquisition means and the portable analysis device, despite the constraints of distance or movement of the acquisition means relative to the portable analysis device. Furthermore, and Unlike an internet connection, the video acquisition device and the portable analysis device always remain at a certain distance from each other, so as to guarantee data transmission between the two.

[0015] Preferably, the analysis means based on artificial intelligence comprises a model based on machine learning, in particular trained or automatic, and possibly the data used for training said model.

[0016] The analysis means is based on artificial intelligence. It can thus comprise an artificial intelligence model, in particular a trained one, but also the data used for training said model. In particular, when the training data of the model are recorded in the analysis means, it is then possible to modify, correct and / or enrich them from the data collected by the video acquisition means, to continue to improve the artificial intelligence model of the portable analysis device. In other words, the model can use new data to improve its analyses, such data being in particular derived from the video streams received by the portable analysis device.

[0017] Preferably, the analysis means based on artificial intelligence comprises an image analysis model, for example a neural network system such as a convolutional neural network.

[0018] Neural network systems, in particular convolutional neural networks, are particularly suitable for analyzing images or videos, and make it possible to provide relevant information on such data.

[0019] Preferably, the analysis means based on artificial intelligence is configured to carry out the analyses of said video stream entirely and autonomously.

[0020] The portable analysis device is designed to enable analysis of the video stream locally, without requiring external resources: the analysis means based on artificial intelligence is designed to be able to analyze, alone, the video stream and provide information on it. This thus makes the analysis of video streams more flexible and more mobile.

[0021] Preferably, the portable video stream analysis device also comprises a learning means configured to record one or more images of said video stream, and possibly the corresponding information, when the user indicates that said information provided by the analysis means is erroneous or absent.

[0022] The portable device may also comprise a learning means intended to enrich learning data, or to create learning data intended to improve the information provided by the analysis model. For this purpose, the learning means is configured to record one or more images of said video stream, and corresponding information. The recorded information may be based on that provided by the analysis means, when these are erroneous, or may be indicated by the user, when the analysis means has not provided information. Such a learning method therefore makes it possible to develop the analysis means, even if it is disconnected from the internet and / or updates, in order to maintain reliable analyses.

[0023] Preferably, the analysis means is modified as a function of the data recorded by the learning means, for example by sending the data recorded by the learning means to a remote server for modification of the analysis means and then by updating the analysis means of said portable device by a remote server.

[0024] According to this other embodiment, the improvement of the artificial intelligence model is carried out via a remote server: the analysis means may only include the artificial intelligence model, without the model's training data. The training is then carried out by updating the model, via a connection to a server when internet access is possible. During such a connection of the portable device to the internet, it is then possible to transmit the data recorded by the training means, so that they can be added to the model's training data and therefore can be taken into account. Thus, even if the model based on artificial intelligence can operate and analyze the video stream autonomously, it can nevertheless be improved by updates, in order to keep it up to date.

[0025] Preferably, the portable device is devoid of video acquisition means, for example devoid of a camera.

[0026] As indicated previously, in order to be able to exploit an artificial intelligence model in a universal and autonomous manner, it is designed in the form of a portable analysis device configured to receive any type of video stream, but does not include a video acquisition means.

[0027] According to another aspect, there is also provided a video analysis system, comprising a portable device as described previously, and a video acquisition means, for example a camera, preferably separate from the portable device and providing said video stream to the portable analysis device, in which the video acquisition means is mounted on a drone or on an endoscope, for example medical, and possibly in which the analysis means comprises an analysis model adapted to the video stream provided by the video acquisition means.

[0028] The video analysis system is thus formed by the assembly: portable analysis device and video acquisition means. The video analysis system therefore makes it possible to acquire and analyze images directly, locally, without requiring external resources. The video acquisition means can thus be an endoscope or a drone, whose functions are specific to their uses, while the portable analysis device remains universal and allows the corresponding video streams to be analyzed.

[0029] According to another aspect, there is also provided a method for analyzing a video stream using a portable device, preferably without video acquisition means, in which said video stream is received, said video stream is analyzed and information relating to said video stream is determined with a processing unit mounted in said portable device, then the information obtained by analyzing said video stream is displayed on a display of said portable device, possibly at the same time as said video stream, in which the processing unit comprises the analysis means is based on artificial intelligence and configured to analyze the video stream and determine the information relating to said video stream. Brief description of the drawings

[0030] [Fig-1] [Fig.l] represents a schematic view of a first embodiment of a video analysis system comprising a portable device according to the invention; and

[0031] [Fig.2] [Fig.2] represents an example of a flowchart of a mode of implementation implementation of a method according to the invention. Description of the embodiments

[0032] [Fig.l] schematically illustrates an exemplary embodiment of a video analysis system 1 comprising a portable device 2 for analyzing a video stream according to the invention.

[0033] In this exemplary embodiment, the portable device 2 comprises a frame 4 with an input 6 for receiving a video stream to be analyzed, a processing unit 8 making it possible in particular to analyze said video stream and to obtain information therefrom, and a display 10 for displaying said information on the video stream and possibly the video stream. The analysis device 2 may also comprise a battery (not shown) and a handle 12 making it autonomous and portable, for easy and rapid use.

[0034] The video analysis system 1 comprises, in addition to the portable device 2 according to the present invention, a video acquisition means 14 providing said portable device 2 with the video stream to be analyzed. The acquisition means 14 is preferably distinct from the portable device 2, that is to say that the portable device 2 is not designed to operate with only one given acquisition means, but can on the contrary be connected, by a cable or wirelessly, to different acquisition means 14, different or identical, in order to provide an analysis of the video stream provided. The portable device 2 is a device independent of the acquisition means 14.

[0035] The acquisition means 14 may be, for example, a camera capable of detecting radiation in the visible range and / or in the invisible range, such as infrared. The acquisition means 14 is in particular adapted to the conditions of use in which it is used: thus, the acquisition means 14 may be mounted on an endoscope, in the case of a medical application or a fault-finding application or other, or may be mounted on a drone, in the case of an application at height. The acquisition means 14 therefore relies on existing equipment specific to the intended use: the portable device 2 according to the present invention makes it possible in particular to analyze the video stream of the acquisition means 14 without requiring a change, modification or adaptation of said acquisition means 14.It is therefore possible to keep the acquisition means 14 which are the most advantageous and most suited to the intended use, while obtaining an analysis of the video stream using the portable device 2 according to the present invention.

[0036] The acquisition means 14 may also be a camera detecting infrared, for example a thermal camera. In this case, the portable analysis device 2 makes it possible to facilitate the analysis of the video stream provided by the thermal camera and which may be difficult to interpret, at first glance, by the user.

[0037] The portable device 2 thus comprises the input 6 for receiving said video stream from the acquisition means 14. The input 6 comprises, in the example shown in [Fig.l], a wired connection means, such as an HDMI socket, configured to connect a cable for transmitting the video data provided by the acquisition means 14. The input 6 in the form of a wired connection means makes it possible in particular to overcome any transmission hazards that may occur in the context of a wireless transmission. The wired connection means allows faster and more stable transmission of the data between the acquisition means 14 and the portable device 2.

[0038] Alternatively, or in addition to the wired connection means described above, the input 6 may also comprise a wireless connection means, for example WIFI or other. The wireless connection means may then comprise one or more antennas configured to communicate with the acquisition means 14. In this case, it is then possible to analyze the video stream provided by the acquisition means 14 while maintaining the latter's freedom of movement. The input 6 in the form of a wireless connection means makes it possible to benefit from a video analysis of the stream provided by the acquisition means 14, without limiting its movements and / or use.

[0039] Input 6 is therefore chosen according to the use of the acquisition means 14, and makes it possible to adapt to different uses or situations, to provide a functionality for analyzing the video stream without modifying or constraining the acquisition means. 14.

[0040] The video stream received by the input 6 is then transmitted to a processing unit 8 in order to be able to analyze it in particular, and possibly to the display 10 to allow it to be viewed by the user. The purpose of the processing unit 8 is to provide information relating to the video stream. The processing unit 8 can in particular allow the identification of elements, such as anomalies in the variations of contrasts or others, which can indicate the possible presence of hidden or buried elements. The processing unit 8 can also allow the recognition of elements, such as objects or people, present on the video. This can thus involve the recognition of the different elements constituting the image, in order to allow a complementary analysis thereof.

[0041] For this purpose, the processing unit comprises an analysis means 16 based on an artificial intelligence model and making it possible to analyze said video stream to provide said information. The analysis model 16 may in particular comprise one or more artificial intelligence models, depending on the type of information desired and / or the type of video stream to be analyzed. Thus, depending on the application and the video acquisition means 14 used, the user will be able to choose and select the artificial intelligence model to be used to perform the analysis of the video stream.

[0042] In general, the artificial intelligence model(s) are image analysis models, configured to be able to analyze computer data corresponding to images or videos. Such image analysis models are known, for example models based on a neural network system, and more particularly models based on a convolutional neural network.

[0043] Such artificial intelligence models are specifically designed and intended for the analysis of images and / or videos.

[0044] The artificial intelligence model may also be based on learning. In other words, and as for the examples of image analysis models mentioned previously, the artificial intelligence model(s) of the analysis means 16 may carry out their image and / or video analyses based on the examples and learning that have been previously integrated into them.

[0045] Thus, the analysis means 16 of the portable device 2 may comprise one or more artificial intelligence models based on learning, the learning data of which are stored remotely, for example on a server. In this case, the artificial intelligence model(s) operate autonomously, on the basis of the learning data used during the last improvement, or update, of the model. The memory required for storing and using the artificial intelligence model is then limited, while allowing regular improvements by connection to the server with the learning data, when the internet connection of portable device 2 is possible.

[0046] Alternatively, or in addition, the analysis means 16 of the portable device 2 may comprise one or more artificial intelligence models based on learning, but also the learning data of the artificial intelligence model(s). In the latter case, all the elements on which the artificial intelligence relies to carry out its analysis are integrated into the portable device 2 and can therefore, in particular, be modified, improved, supplemented, etc.

[0047] In particular, the integration of the learning data of the artificial intelligence model into the analysis means 16 possibly makes it possible, depending on the results obtained with the portable device 2, to adapt or even modify the analysis results provided by the analysis means 16, in particular by correcting the analyses provided when they are erroneous. By thus correcting the learning data, it is then possible to directly and locally modify the artificial intelligence model so as to obtain results consistent with those expected.

[0048] The portable device 2 may thus comprise a learning means 18, preferably included in the processing unit 8. The learning means 18 is configured to record the images or videos corresponding to the erroneous or missing analyses of the analysis means 16. Thus, it is possible to enrich the learning data of the artificial intelligence model by recording, via the learning means 18, the images or videos giving rise to erroneous or missing information, and by indicating the expected information. By adding such elements to the learning data of the artificial intelligence model, it is then possible to allow said artificial intelligence model to correct itself and to limit the erroneous or missing information during subsequent analyses.

[0049] In the case of an analysis means 16 comprising the artificial intelligence model and the learning data, the learning means 18 can be configured to directly and locally modify the learning data of the artificial intelligence model, in order to improve the results thereof. Thus, even modifications and improvements to the artificial intelligence model can be carried out without requiring an internet connection.

[0050] Alternatively, or in addition, the learning means 18 may be intended to transmit the recorded data to a remote server, for example a server comprising learning data for the artificial intelligence model, in order to allow, in a second stage, a modification and improvement of the artificial intelligence model, during a subsequent connection of the portable device 2 to the internet.

[0051] The information obtained by the analysis means 16 is then provided, for example at the same time as the video stream or separately, to the display. 10. The display 10 may be a standard screen, configured to display both the video stream and, superimposed, the information provided by the analysis means 16.

[0052] In order to enable the portable device 2 to be controlled, the display 10 may also comprise, on its display surface, a touch-sensitive selection means. It is then possible for the user to order modifications or to carry out actions on the portable device 2 via the touch-sensitive selection means. For example, the processing unit 8 may be configured to display, on the display 10, menus 20 for entering commands. The menus 20 may include in particular the choice of the artificial intelligence model to be used by the analysis means 16. The menus 20 may also include a possibility of indicating erroneous or missing information, leading to the recording, in the learning means 18, of the corresponding image or video with the erroneous or missing information.

[0053] Thanks to the menus 20, the user can directly interact with the operation of the portable device 2 to adapt it to the use made of it, in an intuitive manner.

[0054] Alternatively, the portable device 2 may comprise physical buttons, for example mounted on the frame 4, making it possible to control the various functions of the processing unit 8 in particular.

[0055] Thus, with the portable device 2 according to the present invention, it becomes possible and easy to analyze, with artificial intelligence, any type of video stream, without requiring an internet connection.

[0056] [Fig.2] illustrates an example of a flowchart of an embodiment of a method 30 according to the invention.

[0057] According to a first step 32 of the method 30, a video stream to be analyzed originating from any acquisition means is received on a portable device 2 as described previously.

[0058] In a second step 34, said video stream is analyzed with the artificial intelligence model of the portable device 2, in order to obtain information relating to the video stream.

[0059] Then, in a third step 36, the information obtained by the artificial intelligence model is displayed, preferably at the same time as said video stream.

[0060] Thus, thanks to the portable device according to the present invention, it becomes easier to analyze any type of video stream, in real time and locally, that is to say without an internet connection, with an artificial intelligence model, in order to obtain information on said video stream.

Claims

1.

2.

3. Claims Video analysis system (1), comprising a portable device (2) for analyzing a video stream and a video acquisition means (14) mounted on a drone and providing said video stream to the portable analysis device (2), the device comprising: - an input (6) for receiving said video stream, for example a video stream in the visible or invisible range, in particular infrared, - a processing unit (8), receiving said video stream as input and being configured to provide as output information relating to said video stream, for example information for identifying or recognizing elements in said video stream, and - a display (10) for displaying said information provided by the processing unit (8), possibly at the same time as said video stream, wherein the processing unit (8) comprises an analysis means (16), for example an image analysis means, based on artificial intelligence and configured to analyze the video stream and determine the information relating to said video stream, wherein the portable device (2) also comprises a learning means (18) configured to record one or more images of said video stream when the user indicates that said information provided by the analysis means (16) is erroneous or absent, and wherein the analysis means (16) is modified according to the data recorded by the learning means (18) by sending the data recorded by the learning means (18) to a remote server for modification of the analysis means and then by an update of the analysis means (16) of said portable device by a remote server.Video analysis system (1) according to claim 1, wherein the input (6) for receiving said video stream comprises a connection means, for example a socket, in particular HDMI, for connecting, in a wired manner, the portable device (2) to the video acquisition means (14) separate from the portable device. Video analysis system (1) according to claim 1 or 2, wherein the input (6) for receiving said video stream comprises a wireless connection means, for example WIFI, for connecting, wirelessly, the portable device (2) to the video acquisition means (14) separate and remote from the portable device.

4. Video analysis system (1) according to any one of the preceding claims, wherein the analysis means (16) based on artificial intelligence comprises a model based on machine learning, in particular trained or automatic, and possibly the data used for training said model.

5. A video analysis system (1) according to any preceding claim, wherein the artificial intelligence-based analysis means (16) comprises an image analysis model, for example a neural network system such as a convolutional neural network.

6. Video analysis system (1) according to any one of the preceding claims, wherein the analysis means (16) based on artificial intelligence is configured to fully and autonomously perform the analyses of said video stream.

7. Video analysis system (1) according to any one of the preceding claims, wherein the learning means (18) is configured to record the corresponding information of the image(s) of said video stream, when the user indicates that said information provided by the analysis means (16) is erroneous or absent.

8. Video analysis system (1) according to any one of the preceding claims, in which the portable device (2) is devoid of video acquisition means, for example devoid of camera.

9. Video analysis system (1) according to any one of the preceding claims in which the analysis means (16) comprises an analysis model adapted to the video stream provided by the video acquisition means (14).

Images