Computer implemented method for detecting objects in digital images and associated system for detecting the presence of obstacles on road junctions
The computer-implemented method using AI models for obstacle detection in digital images addresses the limitations of existing systems by ensuring accurate and continuous detection of obstacles at road intersections, reducing false alarms and enhancing safety through precise object identification and dual-check verification, regardless of environmental conditions.
Patent Information
- Application Number
- PCT/ES2023/070282
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-04
- Publication Date
- 2024-11-07
Smart Images

Figure ES2023070282_07112024_PF_FP_ABST
Abstract
Description
[0001] COMPUTER-IMPLEMENTED METHOD FOR DETECTING OBJECTS IN DIGITAL IMAGES AND ASSOCIATED SYSTEM FOR DETECTING THE PRESENCE OF OBSTACLES AT ROAD INTERSECTIONS
[0002] TECHNICAL FIELD OF THE INVENTION
[0003] The present invention relates to a computer-implemented method and associated system by which it is possible to detect an obstacle on a road in order, for example, to alert vehicles traveling on said road to take, for example, the corresponding evasive actions. In particular, the invention focuses on the detection of obstacles at road intersections such as intersections or level crossings, where the implemented method enables continuous monitoring of the road or intersection independently of external factors such as, for example, lighting (day / night light changes, season of the year, etc.) or weather (rain, snow, fog, etc.).
[0004] STATE OF THE ART
[0005] As is widely known, when traveling, land vehicles travel on routes that intersect with each other. For example, trains or railway vehicles travel along rail tracks that may intersect with a car track, in what is known as a level crossing. In this sense, another vehicle (e.g., a car) may obstruct the level crossing by being in the intersection in such a way that it would result in a collision, run-over, derailment, etc., if the latter were to travel on the rails across the level crossing. For example, a car may get caught between the barriers when they have been lowered and do not rise in time, or the car may suffer a mechanical failure that makes it unable to exit the level crossing.
[0006] Some well-known systems use radar to detect the presence of a vehicle or obstacle, generally speaking at a level crossing or other intersection, and provide warnings to railway personnel. The biggest limitation of these systems is that they can only classify detected objects by size.
[0007] In other cases, the presence of an obstacle on the road is detected using ultrasonic or infrared transducers, and the corresponding warning signals are sent. Other examples of transducers are metal coils, where detection systems using this type of transducer can only detect metallic objects because they produce interference in the intersection's magnetic field. This is also their greatest limitation, since, in addition to the implementation and maintenance costs, not being able to detect non-metallic objects, which may also be present at the intersection, can lead to safety problems or accidents.
[0008] According to the above, in some cases, the detection of the obstacle on the road, intersection, level crossing, etc., gives erroneous readings or detections, indicating the presence of an obstacle that, in reality, is no longer on the road, that is, a "false positive" or as in the case of metal loops for non-metallic obstacles, being unable to detect the presence of obstacles, that is, a "false negative", which generates errors in the notification of the detection of an obstacle or the absence of it, or in the subsequent processing of the data obtained from said detection.
[0009] In other cases, and depending on the means used to detect the obstacle, its detection may be affected by changing environmental conditions, such as the transition from day to night. Therefore, for example, in low-light conditions, obstacle detection may be affected or even prevented, with the corresponding disadvantages this entails.
[0010] Therefore, there is a latent need to provide a procedure and system for detecting an obstacle on the road, particularly at a road intersection, and more particularly at a level crossing, where the procedure avoids false detections and at the same time operates continuously and efficiently regardless of lighting conditions or other environmental factors.
[0011] DESCRIPTION
[0012] To address the need found, the present invention provides the computer-implemented method for detecting the presence and classification of objects in at least one control region of a digital image.
[0013] To this end, upon receiving a digital image, the claimed computer-implemented method comprises the steps of: detecting the presence of at least one object in the control region of the digital image by feeding said digital image to an artificial intelligence model configured to detect the presence of the at least one object in an image; when the presence of the object in the digital image is not detected; detecting the absence of at least one element in the control region of the digital image by feeding the digital image to an artificial intelligence model configured to detect the absence of the at least one element in an image, and in the event that the non-absence of at least one element in the control region is detected, the presence of at least one object in the digital image is detected.
[0014] Within the context of the invention, a digital image should be understood as a two-dimensional representation of an image, where said image is suitable or adapted for processing on a computer or similar.
[0015] A digital image can be a two-dimensional representation of an image from a numerical matrix, often in binary, where, depending on the resolution, the image is static or dynamic, it can be a matrix image (or bitmap or pixels) or a vector graphic.
[0016] With respect to the bitmap or pixel map, a digital image comprises a plurality of pixels, preferably arranged in a pattern or matrix, obtained by image acquisition means; a digital image should be considered as such even if it has been subjected to processing, for example, filtering, rotation, masking, decimation, or the like, as long as the target information can be recovered from the unprocessed digital image. Preferably, the image acquisition means are arranged and directed toward an area or region of interest, such as, for example, a road intersection, a level crossing, etc., such that the execution of the claimed method entails the detection of an obstacle at said road intersection.
[0017] The image acquisition means are preferably arranged or are part of a detection device and / or a detection system, such that said acquisition means are configured to capture an image of an area of interest, where said area of interest is represented two-dimensionally in a digital image. In addition to the foregoing, when said digital image is captured or recorded over a road or road intersection by the image acquisition means, i.e., the control region contains a road or a road intersection, the detection of the object in the control region of the digital image is indicative of the presence of an obstacle on the road or at the road intersection.
[0018] In addition to the pixels that comprise the digital image, in some embodiments, the digital image includes associated data, such as the image's geolocation data, time, image capture conditions (exposure, etc.), image identification, or other information associated with the image. In other embodiments, the associated data includes image metadata.
[0019] According to the scope of the invention, a road intersection should be understood, but not limited to, those elements of the road and transportation infrastructure where two or more roads intersect. These infrastructures allow users to interchange between roads. The intersection can occur at a level crossing or an underpass. The term road intersection also includes elements of other transportation systems, such as railways or bicycle paths, highways, sea crossings, etc.
[0020] On the other hand, in one embodiment, the artificial intelligence models are part of or are hosted in an object and / or obstacle detection system, where such models may in turn be distributed in one or more computer systems provided with processing means that may either be part of a server of the computer system or may be independent units in data communication with the computer system, for example, one or more processing units configured to store and execute an artificial intelligence model, feed it with input data, such as digital images and retrieve data from them.
[0021] An artificial intelligence model is understood to be a computer implementation of an artificial intelligence algorithm trained with a training data set, where the training data set comprises relevant digital images labeled according to the intended purpose of the artificial intelligence algorithm model. The artificial intelligence algorithms used to obtain the artificial intelligence model can be of any type, and preferably an artificial neural network, such as a convolutional neural network, or an open-source neural network. According to the above, the detection of an object in the digital image should be understood as a process based on artificial vision to detect whether an object, such as, for example, an object selected from a list of detectable objects, such as a car, motorcycle, animal, pedestrian, etc., has been represented in the image as an object, where the said object may be an estimate of the object or the exact object.
[0022] In relation to the artificial intelligence model, thanks to training, it is possible to classify and discern detected objects through it, for example, it is possible to discern between a car, a person, a motorcycle, an animal, etc., and also to discern between objects of the same category, for example, between a truck and a car, between a motorcycle and a bicycle, between a dog and a cow, etc.
[0023] Also, as evidenced in the steps of the method object of the invention, in the event that this first artificial intelligence model does not detect the presence of an object in the digital image, an additional step is carried out consisting of the detection of the absence of an element in such digital image through an artificial intelligence model trained for this purpose.
[0024] Similar to how artificial intelligence has been trained to determine whether there is an object in the control region of a digital image, another artificial intelligence is trained to determine whether there is an element in the control region of the image. Thus, when an object is not detected with the method of the invention, this means that a double check has been made: first, that the object is not in the control region and, second, that said control region is empty or lacks a detectable element.
[0025] Going into more detail, the list of detectable objects includes, but is not limited to, trains, motorcycles, trucks, and cars. The artificial intelligence model results in a "no object detection" when it does not detect any of these objects in the control region of said digital image. Therefore, other objects, such as a bicycle or a pedestrian, may be present in the control region, but since they do not match those on the list of detectable objects (e.g., train, motorcycle, car, etc.), they are not detected by the artificial intelligence model for detecting objects in an image.
[0026] Next, the control region of the image is checked for certain elements that are not present, that is, that are absent, to ensure that there is actually no object in the control region of the digital image. In this sense, the artificial intelligence model configured to detect the absence of at least one element in an image is trained to detect the absence of an element from a list of detectable elements, for example, pedestrian, dog, bicycle, etc., such that, if this artificial intelligence model detects the absence of such elements in the control region of the digital image, the object is not located in the control region of the digital image.
[0027] According to the above, object detection in the digital image must also be understood as a process based on artificial vision to detect if an element such as, for example, an element selected from a list of detectable elements, such as a pedestrian, dog, bicycle, etc., without being limited to these, is not present in the control region of the digital image.
[0028] In this sense, this "second" artificial intelligence will detect the absence of at least one element in the digital image, and if the absence of at least one element is detected, the presence of at least one object in the digital image is detected.
[0029] This stage of the method is designed so that, even if an object, for example, a digital representation of a car, is not found / detected in the control region, an element, for example, a digital representation of a pedestrian, can be detected in the control region if this element is located in the control region. In this sense, in the digital image, the control region will contain an element, which will be detectable by the artificial intelligence model configured to detect the absence of at least one element in an image. Thus, if this element is located in the control region, the non-absence of said element will be detected by the artificial intelligence model configured for this purpose, such that the presence of an object in the digital image will have been detected.
[0030] If, on the other hand, none of the elements in the detectable element list are detected in the control region by the artificial intelligence model, the absence of that element in the digital image is detected and, therefore, it will have been detected that there is no object in the digital image.
[0031] The implementation of this absence detection stage using this "second" artificial intelligence model is designed to perform a double check to ensure that the digital image is free of any objects. It determines, first, that there is no object in the control region, and second, that an element is absent in the control region, i.e., that it is empty. This double check has significant advantages when the method is executed in an obstacle detection system, as it ensures the accurate detection of an obstacle or its absence, and, above all, guarantees that an obstacle is not present when one is not detected.
[0032] Obstacle detection at a road intersection should also be understood as a process based on artificial vision to detect whether an obstacle, such as a pedestrian, animal or small vehicle, is not represented in the image as an element, in which said element may be an estimate of it or an exact representation.
[0033] In an alternative embodiment of the method, a registered digital image has been provided, that is, an image already registered previously or captured at some previous time, for example, one second before, such that, when the presence of an object in the digital image (the current digital image) is not detected by the artificial intelligence model configured to detect the presence of an object in an image, the step of detecting the absence of at least one element in the image is also carried out in said registered digital image.
[0034] Preferably, all captured digital images are stored on the storage media, so that there will always be recorded digital images stored.
[0035] In preferred embodiments, digital images are captured and stored at 1-second intervals. Thus, if, for example, in the current digital image (in which the presence of an object is being detected), the absence of an element is detected in the control region, a check is made as to whether the absence of an element (the same or another) has also been detected in the control region of the recorded digital image, i.e., the image obtained exactly 1 second ago and stored in the storage media. Thus, if the absence of the element has also been detected in the recorded digital image, the absence of the object in the control region can be more accurately determined, i.e., whether the control region is empty.
[0036] At this point, it is worth clarifying that the object and the element to be detected are generally "non-stationary," so that in a series of subsequent digital images, the object and / or element may appear in different positions within the same series, and will be detected whenever at least a portion of it is located within the control region. In preferred embodiments, the digital images in which the absence of an element has been detected are stored, for example, in a storage unit for recorded elements, so that they can be compared according to the teachings described. In even more preferred embodiments, the absence of the element is checked in at least the two images recorded subsequent to the digital image, when said absence of an element has been detected in the digital image.
[0037] As stated, the recorded digital image may contain an object that no longer appears in the (current) digital image. Thus, if at least one object is not detected in the digital image, but it is verified that an object has been detected in the recorded digital image, then the absence of at least one element in said digital image is detected to determine that the digital image has no objects and no elements, i.e., the control region is "empty." According to the preceding paragraphs, after a series of iterations, for example, three, in which the absence of an element is detected in subsequent recorded digital images, the algorithm will conclude that there is no object in the digital image, i.e., that the control region is empty or free of objects.
[0038] On the other hand, in alternative embodiments the digital image comprises associated data, where the method additionally comprises the step of including the detected object or absence of element in the associated data.
[0039] Additionally, the method comprises a step of determining a set of geographic coordinates of the digital image from the data associated with it.
[0040] Furthermore, the digital image comprises an RGB (Red, Green, Black) format or an IR (InfraRed) format, such that they can be thereby obtained by the image acquisition means, wherein the artificial intelligence model configured to detect the presence of at least one object in an image is additionally configured to detect the object in the digital image in any of the RGB format and / or IR format, and / or in a combination of said formats.
[0041] This is advantageous because it allows object detection regardless of the digital image format used, or by merging them to obtain the image definition that best allows for the detection of the object and / or the absence of an element. In addition to the above, the method includes a stage in which it is verified that the image acquisition means are capturing images in the best possible conditions. To this end, parameters such as lighting, focus, exposure, etc. are checked, for example, using appropriate sensors.
[0042] Furthermore, digital images from different formats can be combined so that the respective artificial intelligence models can detect the object or the absence of an element in the combined digital image. Thus, in changing lighting situations, such as sunrise or sunset, the combination of formats, for example, RGB and IR, will result in an image that provides the necessary means for the corresponding artificial intelligence model to detect the presence of at least one object or the absence of at least one element.
[0043] Regarding the control region of a digital image, this corresponds to an area or portion of said image in which the presence of an object or the absence of an element is expected to be detected. This control region is framed within a frame that preferably remains constant in each digital image. The control region is modifiable and can range from a specific area of the digital image to the entire image, the control zone.
[0044] It is also important to note that the control region can be comprised of the fusion of several digital images, where the control region is the sum of each of the individual control regions of each digital image. This is especially appropriate when receiving multiple digital images captured of a point of interest from different perspectives, where the control region is the sum, juxtaposition, or composition of each of the control regions of the individual digital images, such that the presence or absence of an object is detected in this composite control region.
[0045] Alternatively, it is possible to detect “permanent” objects in the control region. The definition of a permanent object refers to an object that is always present in the control region and is detectable by the artificial intelligence model configured to detect an object and is also detectable by the artificial intelligence model configured to detect the absence of an element, but its detection as an object or element in the control region does not result in the detection of an object in the control region according to the teachings described so far, while the detection of its absence will result in the absence of this element, which can be used to trigger a warning that a permanent object is missing or lost in the digital image.In another embodiment, the invention proposes a system for detecting the presence of obstacles at road intersections, where the system has image acquisition means configured to obtain at least one digital image, and processing means or computing means configured to execute the computer-implemented method as described so far.
[0046] In this embodiment, the image acquisition means are arranged such that they capture digital images, or at least one digital image, of a road intersection, such as, for example, a level crossing, such that a physical obstacle (e.g., a car crossing at the level crossing) at the intersection will be represented as an object in the control area of the digital image and, through the execution of the method, the system will be able to detect the presence or absence of an obstacle at the intersection.
[0047] In alternative embodiments, the system comprises storage media configured to store captured digital images. Thus, the system is capable of detecting an obstacle at the intersection and generating some type of action in accordance with said detection when the presence of the object has been detected in the (current) digital image and it has been verified that an object has also been detected in at least one of a series of subsequent images. Thus, if the object appears in the (current) digital image and also appears in two subsequent images, the system detects that there is an obstacle at the intersection. The images are captured at intervals of, for example, one second.
[0048] Alternatively, the system comprises communication means configured to transmit and receive data and / or external instructions.
[0049] Preferably, the system comprises a set of electronic cards that, based on the information extracted from collecting data from the detection of objects or obstacles by means of the implemented method of the invention and the information provided by it, allows informing, by means of, for example, voltage-free contacts, an external system or terminal of the detection of said obstacle on the road, based on the detection of an object or absence of an element in the control region of the captured digital image of the road. It is also important to highlight that all the electronic part associated with the system is configurable according to the type of application and in accordance with the inputs and outputs of the system. Advantageously, this system allows a complete and exhaustive monitoring in real time of a road passable by vehicles, in particular at a road intersection, to detect if there is an obstacle at the intersection.Furthermore, through the use of machine vision technology, the system detects the type of obstacle present at the intersection. It also provides continuous, real-time feedback to the computer system, improving inspection reliability and reducing false positives continuously and indefinitely.
[0050] In addition to the above, since stationary objects on a road or intersection will always appear in the control region of the digital image, such as a traffic light or a traffic sign, the absence of any of these elements can be detected and the corresponding warning signals sent.
[0051] BRIEF DESCRIPTION OF THE FIGURES
[0052] The above and other advantages and characteristics will be more fully understood from the following detailed description of some embodiments with reference to the attached drawings, which should be considered as illustrative and not limiting, in which:
[0053] Fig. 1 is a plan view of a level crossing in which an obstacle detection system has been arranged at a road intersection, where the aforementioned system executes the computer implementation procedure to detect objects in a digital image.
[0054] Fig. 2 is a view of a digital image captured by the image detection means of the obstacle detection system, where an object to be detected appears in the control region of the digital image.
[0055] DETAILED DESCRIPTION OF AN EXAMPLE OF EMPLOYMENT
[0056] In the following detailed description, numerous specific details are presented in the form of examples to provide a thorough understanding of the relevant teachings. However, it will be apparent to those skilled in the art that the present teachings can be implemented without such details.
[0057] The present invention is focused on:
[0058] Computer-implemented method for detecting the presence of objects in at least one control region of a digital image, comprising the following steps: detecting the presence of at least one object in the control region of the digital image by feeding said digital image to an artificial intelligence model configured to detect the presence of the at least one object in an image; when the presence of the object in the digital image is not detected; detecting the absence of at least one element in the control region of the digital image by feeding the digital image to an artificial intelligence model configured to detect the absence of the at least one element in an image, and in the event that the non-absence of at least one element is detected, the presence of at least one object in the digital image is detected.where, a particular exemplary embodiment of this method is observed in figure 1, in which the computer-implemented method is executed in a road intersection obstacle detection system to detect obstacles present at a road intersection, in the particular case, at a level crossing as illustrated.
[0059] Thus, the system for detecting the presence of obstacles at road intersections comprises image detection means 101, which in this embodiment take the form of a digital camera 101 capable of capturing photographs of the road intersection in digital format. Although a single digital camera has been illustrated in this embodiment, it is possible to obtain digital images from various points of the road intersection using different digital cameras, where the control region 301 is the sum of the control regions of each of the captured digital images.The digital images of the road intersection are captured continuously and processed by the artificial intelligence model configured to detect the presence of at least one object in an image that is stored in a computer system that has processing and storage means (not shown) suitable for carrying out the described function, where the digital images are stored continuously.
[0060] Thus, in this embodiment, based on the digital images, the artificial intelligence model is detecting the presence of an object (obstacle at the intersection) in the control zone 301 of each digital image 300 image by image. Since, depending on the focal aperture of the digital camera 101 and its position in relation to the road intersection, or the arrangement of several digital cameras, as mentioned above, everything that is at the road intersection is captured in the digital image 300, the control region 301 is therefore defined as that area of the digital image 300 in which the object 200 is expected to be detected.
[0061] As can be seen in Figure 1, a car 201 is positioned in the middle of the intersection, so, as can be seen in Figure 2, in the control region 301, the car 201 is seen as an object 302 within it.
[0062] In this regard, the artificial intelligence model is configured (trained) to, based on the shape of the object detected in the control zone, discern what type of object it is from a list of detectable objects. The list may contain, but is not limited to, a car, motorcycle, train, truck, bicycle, etc. Thus, if object 302 within control region 301 corresponds to one on the list, the artificial intelligence model will have detected an object 302 in digital image 300, and therefore, consequently, the system will have detected the presence of an obstacle at the road intersection.
[0063] Since the obstacle detection system at road intersections is designed to operate continuously, that is, 24 hours a day, every day, it is possible that, when the object 302, or at least a part of it, appears in the control region 301, it will be the first time that said object 302 has been detected, a single detection of the object 302 being sufficient for the system to have detected the presence of an obstacle 200 at the road intersection and, consequently, proceed to execute the actions corresponding to said detection, for example, sending signals to the vehicles approaching the intersection to warn that it is obstructed.On the other hand, when the artificial intelligence model does not detect an object in the control region 301 of the digital image 300, that is, there is no obstacle 200 at the road intersection, for example, the car 201 represented as object 200 in the digital image 300 has not yet entered the control region 301 or has left it, the system proceeds to detect that the road intersection is empty, that is, free of an obstacle 200, which is carried out by detecting the absence of an element 400 in the digital image 300 by means of an artificial intelligence model configured to detect the absence of at least one element 400 in an image.
[0064] According to the above, this artificial intelligence model is configured (trained) to detect at least one element 400 from the shape of said element 400 detected in the control region 301, and discern what type of element it is from a list of detectable elements. Said list may contain, but is not limited to, car, train, pedestrian, bicycle, dog, scooter, etc. So that, if the element 400 within the control region 301 corresponds to one of the list, the artificial intelligence model will have detected the non-absence of an element 400 in the digital image 300, so that, consequently, the system will have detected the presence of an obstacle at the road intersection.
[0065] The above is so that, even if, for example, a car 200 is not found / detected in the middle of the road intersection, a pedestrian 201 can be located if it is in said road intersection, so that in the digital image 300, the control region 301 will have a representation of said pedestrian 201 as an element 400, which will be detectable by the artificial intelligence model configured to detect the absence of at least one element in an image. Thus, if this element 400 is located in the control region 300, the non-absence of said element 400 will be detected by the artificial intelligence model configured for this purpose, such that the system will have detected the presence of an obstacle at the road intersection.
[0066] If, on the other hand, none of the elements 400 of the detectable element list are detected in the control region 300 by the dedicated artificial intelligence model, the absence of said element 400 is detected in the digital image and, therefore, the system will have detected that there is no obstacle at the road intersection. Following the example, there is no car 201 , nor pedestrian 202, etc., at the road intersection.
[0067] By means of this “second” artificial intelligence, a double check is made to ensure that the road intersection is free of any obstacles 200, whether an object 302 is present in the digital image 300, or if an element 400 is absent.
[0068] The system will consider the presence of an obstacle 200 as not detected when the object 302 has not been detected in a single digital image, i.e., in the first capture. However, in the preferred embodiment, when the object 302 is not detected, by means of double checking, the absence of the element 400 in the control region 301 is verified, for which purpose it is checked in the subsequent stored images, for example, in the two subsequent stored digital images 300, if the absence of the element 400 has been detected in said stored images to establish the non-presence of an obstacle 302 and the non-presence of an element 400 at the road intersection, i.e., that the road intersection is empty.
[0069] The number of iterations is configurable. In the preferred embodiment, three iterations are performed for the system to establish the absence of an object 400, i.e., for the system to determine that the road intersection is clear of any obstacles. However, this can be done, as stated, with a single iteration (a single image capture) up to "n" iterations. The number of iterations will be related to the time interval between the digital images captured. In this example embodiment, the digital images 300 are captured every second, so after three seconds the system will have detected that the road intersection is clear of obstacles and safe.
[0070] In a preferred embodiment, the list of detectable objects and the list of detectable elements are equal, such that, for example, a car 201 is not detected, and the absence of said car 201 at the road intersection is detected (with the second artificial intelligence). Therefore, the road intersection is free.
[0071] Now, when the presence of the object 302 is not detected in the digital image 300 by the artificial intelligence model configured to detect the presence of an object in an image, the absence of the element 400 in the digital image 300 is detected, if the object 302 has been detected in at least one registered digital image 300.In this way, if object 302 had been detected in the previous iteration, since it is not detected in the current iteration, it is verified that said object is no longer in the control region 301 and that the latter is in turn free of elements 400 by means of several iterations to guarantee that the road intersection is free. In addition to the above, the camera can capture digital images in different formats such as IR, RGB, among others, being able to detect objects or the absence of elements in one or another type of format, or in a combination of said formats that allows detection to be carried out independently of the lighting conditions of the intersection.
[0072] Thanks to this, the obstacle detection system works in real time, that is, 24 hours a day, 7 days a week, monitoring the intersection. This means it works at all times, regardless of the state of the intersection, the digital entrances / exits, or whether or not there are obstacles.
[0073] On the other hand, in a preferred embodiment, the system performs update checks every 5-10 minutes of the camera monitoring, with an image capture to determine if they are being performed optimally and there are no errors.
[0074] The different camera checks that are carried out are:
[0075] Verification of the required use of the IR format. An image is captured, and its darkness level is verified based on a preset brightness threshold. An update is performed every 5-10 minutes.
[0076] Exposure time check, "image freeze," where the exposure time is so short that nothing in the frame can move or shift from the camera's perspective. Image freeze is checked with the IR function on (black and white image) and off.
[0077] Checking camera orientation. The digital image frame area is checked to ensure that it contains control region 301.
[0078] Checking camera focus. Verifying automatic lens settings to achieve maximum resolution, sharpness, and contrast for the digital image and, therefore, for the control region.
Claims
CLAIMS 1. Computer-implemented method for detecting the presence of objects in at least a control region 301 of a digital image 300, comprising the following steps: detecting the presence of at least one object 302 in the digital image 300 by feeding said digital image 300 to an artificial intelligence model configured to detect the presence of the at least one object in an image; when the presence of the object 302 in the digital image 300 is not detected; detecting the absence of at least one element 400 in the digital image 300 by feeding the digital image 300 to an artificial intelligence model configured to detect the absence of the at least one element in an image, and in the event that the non-absence of at least one element is detected, the presence of at least one object 302 in the digital image is detected.
2. The method according to claim 1, wherein, when the presence of an object 302 in the digital image 300 is not detected by the artificial intelligence model configured to detect the presence of an object in an image, the step of detecting the absence of at least one element 400 in the digital image 300 in a recorded digital image is carried out.
3. The method according to claim 2, wherein the recorded digital image corresponds to a digital image 300 in which the artificial intelligence model configured to detect the absence of the at least one element in an image has detected the absence of at least one element 400.
4. The method according to any of the preceding claims, wherein the artificial intelligence model configured to detect the presence of the at least one object in an image and / or the artificial intelligence model configured to detect the absence of the at least one element in an image is configured to detect the at least one object 302 or the at least one element 400 from predefined lists of objects and elements.
5. The method of the preceding claim wherein the predefined lists of objects and elements comprise, among others, person, animal, vehicle.
6. The method according to any of the preceding claims wherein the control region 301 of the digital image 300 comprises a road intersection.
7. The method according to any of the preceding claims, comprising the step of receiving the digital image 300 from a detection device 100 comprising image acquisition means 101 configured to capture a digital image of an area of interest.
8. The method of the preceding claim, wherein the image acquisition means 101 of the detection device 100 is configured to capture a digital image of a road intersection.
9. The method of any one of the preceding claims wherein the digital image 300 comprises associated data, and wherein the method further comprises the step of including the detected object 302 or absence of element 400 to the associated data.
10. The method of any of the preceding claims, wherein the method further comprises the step of determining a set of geographic coordinates of the digital image 300 from the associated data.
11. The method according to any of the preceding claims, wherein the digital image 300 comprises an RGB format or an IR format, and wherein the artificial intelligence model configured to detect the presence of the at least one object in an image is further configured to detect the object 302 in the digital image 300 in RGB format and / or in IR format, and / or in a combination of said formats.
12. The method according to any one of claims 7 to 11 comprising a step of checking the images acquired by the image acquisition means 101 of the detection device 100.
13. A system for detecting the presence of obstacles at road intersections by executing a method according to any of claims 1 to 12, wherein the system comprises: at least one detection device 100 provided with image acquisition means 101 configured to obtain digital images of a road intersection, at least one processing unit configured to detect the presence of an object or the absence of an element in the digital image, where when the presence of the object or the absence of an element is detected in the digital image, the presence of an obstacle has been detected at the road intersection; and communication means configured to allow data communication with a data network.
14. A detection device 100 for detecting an obstacle at a road intersection belonging to a system according to claim 14, the device 100 comprising image acquisition means 101 and data processing means.
Citation Information
Patent Citations
Computer Vision Based Driver Assistance Devices, Systems, Methods and Associated Computer Executable Code
US20180012085A1
Accelerating convolutional neural network computation throughput
US20190102640A1
