Parking management system and computing device that accurately identify vehicles or utilize identifiers without causing stagnation.

Abstract

A parking management system capable of accurately identifying vehicles without the stagnation phenomenon is disclosed. The parking management system includes a first image capture device for recognizing the vehicle number of a vehicle at the entrance or inside of a parking lot, a second image capture device installed inside the parking lot, and a computing device communicatively connected to the first image capture device or the second image capture device. The computing device identifies a vehicle by considering detected vehicle information using RE-ID (re-identification) technology and vehicle information obtained by analyzing an image captured by at least one of the image capture devices, and the re-identification technology determines whether the vehicle is the same based on vehicle exterior information included in an image captured by the first image capture device and vehicle exterior information included in an image captured by the second image capture device.

Description

Technical Field

[0001] The present invention relates to a parking management system that accurately identifies a vehicle without a stagnation phenomenon or uses an identifier.

Background Art

[0002] In the case of a car-sharing service, when arriving at the site, one is often troubled by illegally parked vehicles. For this reason, various methods for checking whether a vehicle is on the parking surface have emerged. For example, as shown in FIG. 1, a method of checking the presence or absence of a vehicle through an IoT sensor in a 1:1 matching manner, a method of checking through a CCTV camera, a method of checking through the GPS reception signal of a mobile phone, etc. are used.

[0003] However, such parking control usually divides the external space and the controlled internal space through a barrier bar and prevents a vehicle from leaving without settling the fee. However, there are many inconveniences caused by such a barrier bar. There is a possibility that the barrier bar may malfunction, a high cost is incurred for installing the barrier bar, and vehicle stagnation may occur due to the barrier bar when a vehicle enters.

[0004] In a conventional parking management system as shown in FIG. 2, after recognizing a number through an ALPR camera, the barrier bar opens and parking is done in the parking lot. When leaving the vehicle, the ALPR camera recognizes the vehicle number again, and when settlement is made at the KIOSK, the barrier bar rises. In such a parking management system, as shown in FIG. 3, vehicle entry stagnation may occur due to the barrier bar, and since the barrier bar is expensive, the construction cost of the parking management system becomes high.

[0005] Also, as shown in FIG. 4, when an ALPR camera that recognizes a vehicle number takes a picture at an oblique angle, an error may occur in vehicle number recognition. Therefore, the access roads of most parking lots are designed to be narrow, and inexperienced vehicle drivers experience the inconvenience of the wheels or the body of the car being rubbed.

[0006] Furthermore, because the concrete structure for installing the ALPR camera occupies a certain amount of space in the entrance area, inexperienced drivers may be more likely to scrape against it. In other words, in order to take measurements with the ALPR camera, the road must be narrowed and the ALPR camera must be installed above the road, which could potentially cause damage to vehicles. [Overview of the Initiative] [Problems that the invention aims to solve]

[0007] This invention provides a parking management system that can accurately identify vehicles without causing congestion. Furthermore, the present invention provides RE-ID technology that can accurately identify vehicles. Furthermore, the present invention provides tracking technology that can determine whether a vehicle is parked or leaving the vehicle. Furthermore, the present invention provides a parking management system that can settle parking fees using an identifier. [Means for solving the problem]

[0008] To achieve the aforementioned objectives, a parking management system according to one embodiment of the present invention includes a first video acquisition device for recognizing the vehicle number of a vehicle at the entrance or inside of a parking lot, a second video acquisition device installed inside the parking lot, and a computing device that is communicated with the first or second video acquisition device. Here, the computing device identifies the vehicle by considering vehicle information detected using RE-ID (re-identification) technology and vehicle information obtained by analyzing images acquired by at least one of the first and second video acquisition devices, and the re-identification technology is a technology that determines whether or not the vehicle is the same vehicle via the external shape information of the vehicle included in the image acquired by the first video acquisition device and the external shape information of the vehicle included in the image acquired by the second video acquisition device.

[0009] A parking management system according to another embodiment of the present invention includes a first video acquisition device for recognizing the vehicle number of a vehicle, a second video acquisition device, an identifier, and a computing device that is communicated with the first or second video acquisition device. The computing device identifies the vehicle using at least one of the vehicle information obtained by analyzing an image acquired by at least one of the first and second video acquisition devices and vehicle information detected using RE-ID (re-identification) technology. When the user of the vehicle inputs user information via recognition of the identifier, the computing device either bills the user for parking fees based on the input user information or automatically settles the payment. The re-identification technology is a technology that determines whether the external shape information of the vehicle included in the image acquired by the first video acquisition device and the external shape information of the vehicle included in the image acquired by the second video acquisition device are the same vehicle.

[0010] A computing device used in a parking management system according to one embodiment of the present invention includes a communication unit which is a communication connection passage to a first video acquisition device or a second video acquisition device, and a re-identification unit which identifies a vehicle that has entered a parking lot by taking into consideration all of the vehicle information detected using RE-ID (re-identification) technology and vehicle information obtained by analyzing an image acquired by at least one of the first and second video acquisition devices. Here, the re-identification technology is a technology that determines whether or not the vehicle is the same vehicle based on the external shape information of the vehicle included in the image acquired by the first video acquisition device and the external shape information of the vehicle included in the image acquired by the second video acquisition device.

[0011] A computing device used in a parking management system according to another embodiment of the present invention includes a communication unit which is a communication connection path to a first video acquisition device or a second video acquisition device, an identifier unit which receives user information input via identifier recognition, a settlement unit, and a re-identification unit which identifies a vehicle that has entered the parking lot using vehicle information detected using RE-ID (re-identification) technology and vehicle information obtained by analyzing an image acquired by at least one of the first and second video acquisition devices. Here, when the user of the vehicle inputs user information via identifier recognition, the settlement unit either bills the user for parking fees based on the input user information or settles the payment automatically, and the re-identification technology is a technology that determines whether or not the external shape information of the vehicle included in the image acquired by the first video acquisition device and the external shape information of the vehicle included in the image acquired by the second video acquisition device are the same vehicle. [Effects of the Invention]

[0012] The parking management system according to the present invention identifies vehicles by using both the vehicle number recognized by the camera and RE-ID (re-identification) technology, so it can accurately identify vehicles regardless of the camera's position, recognition errors, etc. Furthermore, using the aforementioned parking management system eliminates the need for barrier bars, thus avoiding the high costs associated with installing them and preventing congestion. In addition, it significantly reduces the risk of vehicles scraping against each other when entering or exiting the parking lot. Furthermore, if parking fees are settled using an identifier in the aforementioned parking management system, it becomes possible to settle parking fees even for vehicles that leave without paying, eliminating the need to install payment machines and exit barriers. [Brief explanation of the drawing]

[0013] [Figure 1] This is a diagram illustrating a conventional method for checking parking spaces. [Figure 2] This is a diagram illustrating the configuration of a conventional parking management system. [Figure 3] It is a drawing illustrating an example where a stagnation occurs when a vehicle enters. [Figure 4] It is a drawing illustrating a case where a vehicle number cannot be recognized by diagonal shooting. [Figure 5] It is a block diagram schematically illustrating the configuration of a parking management system according to an embodiment of the present invention. [Figure 6] It is a drawing illustrating an example of Re-ID technology according to an embodiment of the present invention. [Figure 7] It is a drawing illustrating an example of a method for grasping differences for the same vehicle type according to an embodiment of the present invention. [Figure 8] It is a drawing illustrating the learning process of a RE-ID deep learning model using Triplet loss according to an embodiment of the present invention. [Figure 9] It is a block diagram schematically illustrating the configuration of a parking management system according to another embodiment of the present invention. [Figure 10] It is a block diagram schematically illustrating a parking management system according to another embodiment of the present invention. [Figure 11] It is a drawing illustrating a tracking process according to an embodiment of the present invention. [Figure 12] It is a drawing illustrating a tracking process according to an embodiment of the present invention. [Figure 13] It is a drawing illustrating a tracking process according to an embodiment of the present invention. [Figure 14] It is a drawing illustrating a tracking process according to an embodiment of the present invention. [Figure 15] It is a drawing schematically illustrating a QR code (registered trademark) recognition process according to an embodiment of the present invention. [Figure 16] It is a drawing illustrating a QR code (registered trademark) processing process according to an embodiment of the present invention. [Figure 17] It is a drawing illustrating a re-identification technology according to another embodiment of the present invention. [Figure 18] It is a drawing illustrating a re-identification technology according to another embodiment of the present invention. [Figure 19] The drawing illustrates the re-identification technology according to another embodiment of the present invention. [Figure 20] The drawing illustrates the re-identification technology according to another embodiment of the present invention. [Figure 21] The drawing illustrates the re-identification technology according to another embodiment of the present invention. [Figure 22] The drawing illustrates the re-identification technology according to another embodiment of the present invention. [Figure 23] The drawing illustrates the re-identification technology according to another embodiment of the present invention. [Figure 24] The drawing illustrates the re-identification technology according to another embodiment of the present invention. [Figure 25] The flowchart illustrates the guidance process of a parking space according to an embodiment of the present invention. [Figure 26] The block diagram illustrates a computing device according to an embodiment of the present invention.

Best Mode for Carrying Out the Invention

[0014] As used herein, the singular forms include plural forms unless the context clearly dictates otherwise. In this specification, terms such as "configured" or "including" should not necessarily be construed as including all of the multiple components or multiple steps described in the specification, and some of the components or steps may not be included, or additional components or steps may be further included. Also, terms such as "... part" and "module" described in the specification mean units that process at least one function or operation, and this can be implemented by hardware or software, or can be implemented by the combination of hardware and software.

[0015] The present invention relates to a parking management system that can have a structure that prevents congestion. In particular, the parking management system can be operated efficiently and is considerably cheaper than conventional systems in terms of cost.

[0016] Conventional parking management systems primarily use barrier bars. This is because, in order to accurately recognize vehicle numbers, it was necessary to photograph the vehicle with a camera after it had stopped. In particular, to accurately identify vehicles, it was necessary to narrow the vehicle entry lane and use barrier bars. Nevertheless, a problem arose where vehicle numbers could not be accurately identified if the camera photographed the vehicle at an oblique angle.

[0017] To solve these problems, the parking control system of the present invention uses a uniquely developed RE-ID (Re-identification) technology. That is, even if the vehicle number cannot be accurately recognized when the parking control system enters, it can accurately identify the vehicle by utilizing the RE-ID technology. Therefore, it is not necessary to narrow the vehicle entry lane, and it is not necessary to use a barrier bar when a vehicle enters. However, the barrier bar may be used if it is necessary for the safety of the vehicle.

[0018] Of course, even when using the barrier bar, if vehicle congestion is expected, or if the barrier bar malfunctions and is left open, the vehicle can be accurately identified, so the parking management system can charge the vehicle the correct fee (parking fee).

[0019] According to another embodiment, the parking management system can register vehicles entering the parking area using an identifier (e.g., a QR code®) installed in the parking area. As a result, even if payment is not made via a payment machine when exiting the parking lot, payment can be automatically made via the information registered with the identifier. Therefore, payment of fees is possible even if there is no barrier bar in the exit area.

[0020] Hereinafter, various embodiments of the present invention will be described in detail with reference to the accompanying drawings. Figure 5 is a schematic block diagram illustrating the configuration of a parking management system according to one embodiment of the present invention, and Figure 6 is a diagram illustrating an example of Re-ID technology according to one embodiment of the present invention. Figure 7 is a diagram illustrating an example of a method for identifying differences for the same vehicle type according to one embodiment of the present invention, and Figure 8 is a diagram illustrating the learning process of a RE-ID deep learning model using Triplet loss according to one embodiment of the present invention.

[0021] Referring to Figure 5, the parking management system of this embodiment may include an entrance camera and barrier installed on the entrance to the parking lot, an exit camera installed at the exit of the parking lot, a payment machine and barrier, and a computing device (not shown, e.g., a server). Here, if entry and exit occur in the same passage, there may be only one barrier for entry and exit. Furthermore, a camera (for example, an ALPR camera) is just one example of an image acquisition device capable of recognizing vehicle numbers, and a variety of devices can be used as long as they can recognize vehicle numbers. However, for the sake of explanation, we will assume that the aforementioned image acquisition device is a camera.

[0022] According to one embodiment, the parking management system of this embodiment can identify a vehicle, that is, recognize the vehicle number, by using video footage captured by an entrance camera and RE-ID technology to recognize an entering vehicle.

[0023] Of course, the video footage captured by the camera and the RE-ID technology can be applied according to priority. For example, the video footage captured by the camera may be applied in priority over the RE-ID technology. In this case, the RE-ID technology may not be applied if the vehicle number is clearly displayed in the video footage captured by the camera, or the RE-ID technology may be applied only if the vehicle number is not clearly displayed in the video.

[0024] However, if a barrier bar is not used, the images acquired by the camera alone have limitations. Therefore, the parking management system can apply the images captured by the camera and the RE-ID technology to all vehicles entering the parking area to identify them. In this case, the vehicles can be clearly identified regardless of the camera's position, shooting angle, vehicle number recognition rate, etc.

[0025] The aforementioned RE-ID technology is a technique that recognizes an object captured by one camera as the same object when it is captured by another camera. According to one embodiment, the computing device analyzes the video captured by the entrance camera to detect the vehicle number, applies the RE-ID technology as shown in Figures 6 and 7 to determine whether or not it is the same vehicle, and identifies (recognizes) the vehicle by considering the detected vehicle number and the results of applying the RE-ID technology.

[0026] For example, as shown in Figure 6, the RE-ID technology (re-identification technology) can identify the vehicle type by comparing the image of a vehicle included in the video captured by the entrance camera with the image of a vehicle stored in an internal or external database, and then, based on the identified vehicle type, can compare the image of a vehicle included in the video captured by the entrance camera with the image of a vehicle included in the video captured by the parking lot's internal camera to determine whether or not they are the same vehicle.

[0027] In other words, the re-identification technology can determine whether or not the vehicles are the same by comparing the vehicle's external shape (external shape information) included in the image acquired by the entrance camera with the vehicle's external shape (external shape information) included in the image acquired by the interior camera.

[0028] In particular, the re-identification technology can determine whether or not they are the same vehicle by comparing detailed parts, as illustrated in Figure 7, rather than just looking at the overall shape of the vehicle. Here, the detailed parts can be a part of the vehicle itself, but they can also be equipment installed or positioned on the vehicle (e.g., a high-pass terminal or a drive recorder).

[0029] According to one embodiment, the re-identification technique can be learned through machine learning, such as a deep learning model. For example, the deep learning model can use a loss function called Triplet loss. Specifically, using Triplet loss, as shown in Figure 8, when multiple images containing vehicles are input, images of the same vehicle can be guided to be located close to each other in an N (an integer of 2 or more) dimensional feature space, while images of different vehicles can be guided to be located far apart in the feature space so that they are easily distinguishable. As a result, images of vehicles included in the video captured by the entrance camera and images of vehicles stored in the DB can be compared quickly and efficiently.

[0030] In the above description, the re-identification technology was explained as comparing the image of a vehicle included in the video captured by the entrance camera with the image of a vehicle in DB. However, it is also possible to compare the image of a vehicle included in the video captured by the parking lot interior camera with the image of a vehicle in DB.

[0031] When a vehicle exits, the exit camera recognizes at least one of the vehicle's license plate number and / or image, and can charge a parking fee only for the duration of the vehicle's parking. Therefore, the user can exit after paying the parking fee via the payment machine, and once the parking fee is paid, the exit barrier opens.

[0032] If the entrances are located together in a single passageway, there may be only one barrier bar for entry and exit. Of course, without the aforementioned barrier bar, there may only be an exit camera and a fare adjustment machine on the exit side. Furthermore, the settlement of the aforementioned charges may not be processed through the settlement machine, but may be billed to the user separately at a later date.

[0033] On the other hand, if an internal camera is present inside the parking lot, even if the vehicle number cannot be accurately recognized upon entry, the internal camera can be used to recognize the vehicle number. In this case, the vehicle can be identified by considering the time it reached the internal camera after entry and the type of vehicle, and the vehicle number of the identified vehicle can be obtained via the internal camera. In this case, payment for the parking fee can be settled using only the vehicle number when the vehicle leaves.

[0034] In summary, the parking management system of this embodiment can accurately identify an entered vehicle by using not only vehicle information detected at the entrance but also vehicle information detected via RE-ID technology (e.g., vehicle type recognition) in an intersection manner. Therefore, even if the vehicle number is not accurately recognized at the entrance, the vehicle can be accurately identified, making it possible to charge parking fees. In this case, a barrier bar is not required, but it may be present for safety reasons.

[0035] Figure 9 is a schematic block diagram illustrating the configuration of a parking management system according to another embodiment of the present invention, Figure 10 is a schematic block diagram illustrating a parking management system according to another embodiment of the present invention, and Figures 11 to 14 are diagrams illustrating the tracking process according to one embodiment of the present invention.

[0036] Referring to Figure 9, the parking management system of this embodiment may include at least one interior camera, an exit camera, a payment machine, and a computing device. In this case, a barrier bar may or may not be present.

[0037] In other words, the parking management system may include an interior camera installed inside the parking lot, without including an entrance camera. Here, the interior camera may be installed in a position that can photograph the parking area. The aforementioned internal camera recognizes the vehicle number when the vehicle parks within the parking lines (parking area) and counts the parking time from the time parking is completed.

[0038] According to one embodiment, the computing device can identify a vehicle by applying the intersection of vehicle information detected via the internal camera and vehicle information (e.g., vehicle type) detected using RE-ID technology. Therefore, even when the vehicle is photographed at an angle by the internal camera and the vehicle number is not 100% recognized, the computing device can identify the vehicle, that is, clearly distinguish it.

[0039] According to other embodiments, the vehicle number of a vehicle can be recognized using a first internal camera, and RE-ID technology using the first and second internal cameras can be applied to a vehicle that has entered the parking lot to clearly identify the vehicle.

[0040] In another embodiment, the computing device can use tracking technology to detect the parking and exit of a vehicle and recognize the vehicle number. That is, even if the vehicle number is not accurately recognized by the internal camera when the vehicle is parked, the computing device can use the tracking technology to clearly identify and track the vehicle. In this case, an exit camera may not be present, as shown in Figure 10.

[0041] The technology that tracks a vehicle using the same camera can determine whether a vehicle has entered or left a given area by comparing the vehicle's position information acquired in the previous frame with the vehicle's position information acquired in the current frame.

[0042] For example, if a vehicle that was present in the previous frame cannot be found in the current frame, the computing device can confirm that the vehicle has left the parking area monitored by the camera.

[0043] As another example, if a vehicle present in the current frame is not present in the previous frame, the computing device can confirm that the vehicle has newly entered or parked in the parking area. For example, as illustrated in Figure 11, if other vehicles are in the same position in the previous frame and the current frame, and vehicle number 4 is detected in the parking area only in the current frame, the computing device can confirm that vehicle number 4 has newly entered and parked in the parking area.

[0044] The computing device can apply this tracking method to an internal camera, thereby enabling it to track the movement of a vehicle. Furthermore, the computing device can obtain a vehicle number during this vehicle tracking process and assign a specific number to the vehicle.

[0045] Referring to Figure 12, the computing device can detect a vehicle by comparing the previous frame with the current frame (detection result), and then track the vehicle by performing IoU (Intersection over Union) matching as follows to determine whether the vehicle parked in or left the parking area.

[0046] In other words, the computing device can determine whether the vehicle is parked in the parking area or has left by detecting how much of the combined area of ​​the parking area and the vehicle area the vehicle area (e.g., 6435, the rectangular box of the vehicle) detected by the AI ​​model occupies. Specifically, whether or not the vehicle is parked can be determined by the ratio (IOU) of the overlapping area of ​​the parking area and the vehicle area to the combined area, as shown in the following formula.

[0047] In Figure 13, it can be confirmed that vehicle 6435, which was parked, has left the parking area in Figure 14. In Figure 14, it can be confirmed that the IoU for vehicle 6435 is 0, and the computing device can thus confirm that vehicle 6435 has left the parking area. On the other hand, the IoU criterion value for determining whether a vehicle is parked in a parking area depends on the settings, but if it is 0.7 or higher, it can be determined that the vehicle is parked. JPEG0007850478000001.jpg63128

[0048] On the other hand, if the camera is obstructed or the AI ​​engine generates an error indicating that the vehicle has not been detected, it may be incorrectly determined that the vehicle has left the vehicle. To prevent such phenomena, the system can also choose to determine that the vehicle has left only when it is being continuously tracked via the AI ​​engine, and even if the parking area and the vehicle area do not overlap, the system can choose not to determine that the vehicle has left for a predetermined period of time (for example, about 1 minute), and only determine that the vehicle has left if the parking area and the vehicle area still do not overlap after the predetermined period of time.

[0049] In another embodiment, the computing device may also determine whether the vehicle is parked in the parking area or has left the parking area based on whether the center of the vehicle area is within the parking area.

[0050] Next, the computing device can delete tracking results for which IoU matching was not performed.

[0051] On the other hand, the computing device can generate new tracking results for detection results for which IoU matching was not performed, estimate the tracking results using a Kalman filter, and then perform IoU matching again.

[0052] Furthermore, the computing device can correct the tracking results obtained via IoU matching using a Kalman filter, and then perform IoU matching again.

[0053] In another embodiment, if a vehicle that has left the first parking area enters the second parking area within a predetermined time, the computing device may determine that the vehicle that entered the second parking area is the same as the vehicle that left the first parking area and assign it the same or similar number. In this case, a comparison of the external appearance of the vehicles may not be performed.

[0054] In summary, the parking management system of the present invention can use re-identification technology, tracking technology, IoU matching, etc., to identify vehicles. As a result, the parking management system can accurately identify an entering vehicle and charge a parking fee even when the camera cannot accurately recognize the vehicle number.

[0055] Figure 15 is a schematic diagram illustrating the QR code (registered trademark) recognition process according to one embodiment of the present invention, and Figure 16 is a diagram illustrating the QR code (registered trademark) processing process according to one embodiment of the present invention.

[0056] Referring to Figure 15, in addition to the vehicle number recognition and re-identification technology via the camera described above, the parking management system of this embodiment can also additionally use an identifier (e.g., a QR code®) visible on the floor of the parking area or on a separate device.

[0057] For the sake of explanation, we will assume that the identifier is a QR code (registered trademark) in the following discussion. The use of identifiers is to ensure that payment is processed later in case a parked user leaves without paying the parking fee at the payment machine. In this case, an exit camera and exit barrier are not required. Of course, since payment of parking fees is possible even without using a payment machine, a payment machine is also not required.

[0058] In other words, the parking management system can identify vehicles using cameras and re-identification technology, and settle parking fees based on information entered via QR codes (registered trademark). In this case, additional payments for parking fees may be made based on the information entered via QR codes only for vehicles that have not been paid for by the payment machine, or the payment machine may be eliminated and parking fees may be settled based solely on the information entered via QR codes (registered trademark).

[0059] Specifically, as shown in Figure 15, when a user takes a picture of a QR code plate installed on the floor of the parking area using a camera, it connects to a specific application as shown in Figure 16, and the application can display information about the currently parked parking area.

[0060] The user can enter their phone number, vehicle number, and card number into the application. As a result, parking fees for the user's vehicle can be automatically settled.

[0061] Of course, there may be cases where the aforementioned card number is not entered, in which case the parking fee can be billed to the phone number entered.

[0062] In other embodiments, if parking fees are not settled through the process described above, the computing device of the parking management system can transmit billing information to a server such as a parking management entity, causing the parking management entity to collect the parking fees or to prohibit parking on the next visit to the parking lot. If parking is prohibited, the computing device can store information about the vehicle for which payment has not been made.

[0063] To calculate parking fees in this manner, it is necessary to track vehicle parking, departure, and movement. For this purpose, the aforementioned tracking technology can be used.

[0064] In summary, the parking management system of this embodiment can use re-identification technology, tracking technology, IoU matching, and information input via QR code (registered trademark) to identify vehicles. As a result, even when the camera cannot accurately recognize the vehicle number, the parking management system can accurately identify the entering vehicle and charge a parking fee, and in particular, even if the vehicle leaves without paying the parking fee, the parking fee can be automatically settled later.

[0065] The following section details the re-identification technology that can operate robustly under diverse external environmental conditions (illumination, weather, etc.).

[0066] Figures 17 to 24 illustrate a re-identification technique according to another embodiment of the present invention. The aforementioned parking management system can identify vehicles by considering the physical location of the camera or the geographical features of the shooting location, in order to ensure that the re-identification technology operates robustly even under diverse external environmental conditions.

[0067] Such re-identification technologies use the similarity of the vehicle's external shape to determine whether or not it is the same vehicle. However, due to diverse external environments, it can be difficult to accurately identify a vehicle by simply comparing its external shape information. Therefore, the parking management system of the present invention can identify a vehicle by considering the physical position of the camera or the geographical features of the shooting location.

[0068] According to one embodiment, when multiple cameras are installed as shown in Figure 17, a moving vehicle can be captured differently by the cameras. Therefore, the computing device can construct an adjacency matrix based on the physical positions of the cameras belonging to a single network, and adjust the weighting of the matching function used in the re-identification process via the constructed adjacency matrix.

[0069] figure 17When a camera is installed as shown in the figure, the adjacency matrix of the cameras is as follows. Examining these adjacency matrices in detail, it can be confirmed that Camera A is adjacent to Camera B but not to Camera C, Camera B is adjacent to both Camera A and Camera C, and Camera C is adjacent to Camera B but not to Camera A. Cameras that are adjacent can have a value of "1", and those that are not adjacent can have a value of "0". However, such value assignment can be variously deformed.

[0070] JPEG0007850478000002.jpg32128

[0071] Under such camera installation conditions, assume that at time t0, Vehicle 1 is detected by Camera C as shown in Figure 18, and at time t1, Vehicle 2 is detected by Camera B (where t0 < t1). Since Camera B and Camera C are physically adjacent to each other, they will have a value of 1 in the adjacency matrix. When calculating the distance (distance, D(f1, f2)) between the feature vectors extracted from Vehicle 1 and the feature vectors extracted from Vehicle 2 by the re-identification engine, a distance function or a matching function can be constructed as follows in the [function] below so as not to suffer a penalty. [Function] JPEG0007850478000003.jpg14128

[0072] Here, N represents the dimension of the extracted feature vectors, and V represents the similarity matrix. VBC can have a value of 1 if Camera B and Camera C are physically adjacent, and a value of 0 otherwise. λ is a parameter representing the degree of penalty and can have a value greater than 1.

[0073] As another example, assume that as shown in FIG. 19, vehicle No. 1 is detected by camera C at time t0 and vehicle No. 3 is detected by camera A at time t1 (where t0 < t1). Since cameras A and C are not physically adjacent to each other, they have a value of 0 on the adjacency matrix. As a result, when calculating the distance between the feature vectors of the two vehicles, a weighting value is applied only by the numerical value of, that is, it will be disadvantaged.

[0074] To summarize, the computing device can set different weighting values according to the degree of adjacency of the cameras when calculating the distance between the feature vectors of the vehicles. Specifically, the computing device can increase the probability that a vehicle detected by one camera is matched with a vehicle detected by an adjacent camera, thereby improving the accuracy of re-identification. That is, the computing device can improve the accuracy by imposing spatial constraints on the re-identification algorithm.

[0075] According to another embodiment, the computing device can compare the distances between vehicles based on the modeling result including the geographical information of the space photographed by the camera, thereby improving the accuracy of re-identification. This is to solve the case where it is difficult to perform accurate re-identification by only utilizing the external shape information of the vehicle when performing re-identification in one camera.

[0076] For example, when a two-lane road is photographed by one camera as shown in FIGS. 20 and 21, assume that one lane is blocked by a tree (obstacle) and an object traveling in that lane disappears from the camera and then reappears.

[0077] In such an environment, as shown in FIG. 22, there are vehicle No. 1 detected at time t0 before being blocked by the tree and vehicle No. 2 and vehicle No. 3 detected at time t1 after a certain period of time has passed. If the external shapes of vehicle No. 2 and vehicle No. 3 are similar, it is difficult to obtain an accurate re-identification result only by comparing the external shapes of the vehicles.

[0078] To resolve this, additional spatial constraints need to be utilized, but using the commonly used linear distance (Euclidean distance) for spatial constraints can lead to the erroneous result that vehicle 2 is more similar to vehicle 1.

[0079] To prevent such errors, the computing device sets up virtual lines A and B to model a two-lane road, as shown in Figure 23, and when calculating the spatial distance between two vehicles, it does not directly calculate the distance between the vehicles but can indirectly calculate the distance between them by utilizing lines A and B. For example, as shown in Figure 24, we can assume that cars 1 and 3 are located on line A and are 0 distance from line A, and that car 2 is located on line B and is 0 distance from line B.

[0080] Comparing car 1 and car 3 using lines A and B, both cars are located on line A, so their distance from line A is 0 and the distance from line B should be similar. On the other hand, comparing car 1 and car 2, car 1 has a distance of 0 from line A, while car 2 has a large value. Therefore, car 1 should have a large distance from line B, and car 2 should have a distance of 0. As a result, the computing device can determine that car 1 and car 3 are more similar.

[0081] In short, the computing device can indirectly compare the distance between vehicles based on modeling results that include geographical information of the space captured by the camera, thereby obtaining more accurate re-identification results.

[0082] Of course, while the re-identification process was explained above using lanes as an example, it is not limited to lanes and can be carried out by considering other geographical environments. For example, instead of lanes, it could be two paths on which vehicles travel.

[0083] Figure 25 is a flowchart illustrating the parking space guidance process according to one embodiment of the present invention. Referring to Figure 25, the computing device first executes an object detection algorithm on the video footage acquired from the camera to obtain the vehicle's position.

[0084] Next, the computing device can convert the preset parking area and the detected vehicle's position into a coordinate system based on camera-real-world relationship information. In the converted coordinate system, the distance between two points on the same plane and the measured distance in the real world can be the same within an error range.

[0085] Next, the computing device can calculate the degree of overlap between the detected vehicle and the parking area, thereby enabling the detection of available parking spaces. Specifically, if the degree of overlap between the detected vehicle and the parking area is high, it is determined that a vehicle is parked in the parking area; if the overlap between the detected vehicle and the parking area is 0 or small, it is determined that no vehicle is parked in the parking area.

[0086] Next, when calculating parking spaces, the computing device can provide a function to display how many vehicles of each vehicle type can be parked in the currently available parking spaces, based on the average and standard deviation of the overall width and length for each vehicle type (e.g., small SUV, medium-sized sedan).

[0087] Figure 26 is a block diagram illustrating a computing device according to one embodiment of the present invention. Referring to Figure 26, the computing device of this embodiment may include a control unit 2600, a communication unit 2602, a device management unit 2604, a re-identification unit 2606, a tracking unit 2608, a settlement unit 2610, and a storage unit 2612. Furthermore, the computing device may additionally include an identifier unit.

[0088] The communication unit 2602 is a communication connection passage to a camera, barrier, or payment machine. The device management unit 2604 can manage the operation of cameras, barriers, or payment machines. The re-identification unit 2606 can re-identify the vehicle via RE-ID technology and identify the vehicle by combining the video information acquired by the camera with the information detected through the re-identification process. The tracking unit 2608 can track vehicles using internal cameras within the parking lot and can also detect vehicle numbers during the tracking process. In particular, the tracking unit 2608 can use IoU technology for the vehicle tracking.

[0089] The identifier unit receives user information input via identifier recognition. The settlement unit 2610 can settle payments for a specified vehicle only for the parking time. In particular, the settlement unit 2610 can automatically settle parking fees without the use of a payment machine based on user information entered via the identifier recognition.

[0090] The storage unit 2612 can store various types of information, such as images acquired by the camera, re-identification information, and tracking results. The control unit 2600 can generally control the operation of the components of the computing device.

[0091] On the other hand, the components of the aforementioned embodiment can be easily understood from a process perspective. That is, each component can be understood as its own process. Furthermore, the process of the aforementioned embodiment can be easily understood from the perspective of the components of the apparatus.

[0092] Furthermore, the aforementioned technical content can be embodied in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., individually or in combination. The program instructions recorded on the medium may be specifically designed and configured for the embodiment, or may be publicly known and available to those skilled in the computer software art. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROMs, RAMs, and flash memory. Examples of program instructions include not only machine code, such as that produced by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. Hardware devices may be configured to operate as one or more software modules to perform the operations of the embodiment, and vice versa.

[0093] The embodiments of the present invention described above are disclosed for illustrative purposes only, and a person skilled in the art with ordinary skill in the invention will know that various modifications, changes, and additions are possible within the spirit and scope of the invention, and such modifications, changes, and additions should be considered to fall within the scope of the claims below. [Explanation of Symbols]

[0094] 2600 Control Unit 2602 Communications Department 2604 Equipment Management Department 2606 Re-identification unit 2608 Tracking Department 2610 Settlement Department 2612 Storage Unit

Claims

1. A first video acquisition device for recognizing the vehicle number of a vehicle at the entrance or inside of a parking lot, A second video acquisition device installed inside the aforementioned parking lot, The system includes a computing device that is communicated with the first video acquisition device or the second video acquisition device, The computing device identifies the vehicle by considering the vehicle information detected using RE-ID (re-identification) technology and the vehicle information obtained by analyzing the image acquired by at least one of the first and second video acquisition devices. The re-identification technology is a technology that determines whether or not the images are the same vehicle based on the external shape information of the vehicle included in the image acquired by the first image acquisition device and the external shape information of the vehicle included in the image acquired by the second image acquisition device. The computing device tracks the vehicle via its tracking function, The tracking function determines whether the vehicle has parked in or left the parking area by comparing the vehicle's position information acquired in the previous frame and the vehicle's position information acquired in the current frame, both of which are acquired by the second video acquisition device. The aforementioned tracking function determines whether the vehicle has parked in or left the parking area via IoU (Intersection over Union) matching. The aforementioned IoU matching is a technique for detecting the extent to which the vehicle area detected by the AI ​​model occupies the combined area of ​​the parking area and the vehicle area. A parking management system characterized in that it determines that the vehicle has parked in the parking area if the result value of the IoU matching is greater than or equal to a previously set value, and determines that the vehicle has left the parking area if the result value is less than the previously set value.

2. The first video acquisition device is an entrance camera, The re-identification technology detects the vehicle type by comparing the vehicle image included in the image acquired by the first video acquisition device with the vehicle image stored in the DB, and based on the detected vehicle type, compares the vehicle image included in the image acquired by the second video acquisition device with the vehicle image included in the image acquired by the first video acquisition device to determine whether or not they are the same vehicle. The parking management system according to claim 1, characterized in that the computing device links the vehicle number when it is determined to be the same vehicle via the re-identification technology.

3. The aforementioned re-identification technique is learned through a deep learning model that utilizes the Triplet loss function, The parking management system according to claim 1, characterized in that, when multiple images containing vehicles are input, the re-identification technology positions images of the same vehicle relatively close together in an N (2 or more integers) dimensional feature space via the Triplet loss, and positions images of different vehicles relatively far apart in the feature space.

4. The parking management system according to claim 1, characterized in that the tracking function determines whether the vehicle has parked in or left the parking area based on whether the center of the vehicle area detected by the AI ​​model is within the parking area.

5. A first video acquisition device for recognizing the vehicle number of a vehicle at the entrance or inside of a parking lot, A second video acquisition device installed inside the aforementioned parking lot, The system includes a computing device that is communicated with the first video acquisition device or the second video acquisition device, The computing device identifies the vehicle by considering the vehicle information detected using RE-ID (re-identification) technology and the vehicle information obtained by analyzing the image acquired by at least one of the first and second video acquisition devices. The aforementioned re-identification technology is a technology that determines whether or not the vehicle is the same vehicle based on the external shape information of the vehicle included in the image acquired by the first image acquisition device and the external shape information of the vehicle included in the image acquired by the second image acquisition device. The computing device constructs an adjacency matrix based on the physical location of the video acquisition device belonging to a single network, and uses the constructed adjacency matrix to adjust the weighting values ​​of the function used in the re-identification process. The process for adjusting the weighted value is characterized by applying the values ​​included in the adjacency matrix to the function when calculating the distance between the feature vectors of the vehicle acquired by the video acquisition device to assign a weighted value. [function] Here, N represents the dimension of the extracted feature vector, V represents the similarity matrix, and VBC has a value of 1 if the video acquisition devices are physically adjacent, and a value of 0 otherwise.

6. A first video acquisition device for recognizing the vehicle number of a vehicle at the entrance or inside of a parking lot, A second video acquisition device installed inside the aforementioned parking lot, The system includes a computing device that is communicated with the first video acquisition device or the second video acquisition device, The computing device identifies the vehicle by considering the vehicle information detected using RE-ID (re-identification) technology and the vehicle information obtained by analyzing the image acquired by at least one of the first and second video acquisition devices. The aforementioned re-identification technology is a technology that determines whether or not the vehicle is the same vehicle based on the external shape information of the vehicle included in the image acquired by the first image acquisition device and the external shape information of the vehicle included in the image acquired by the second image acquisition device. The parking management system is characterized in that the computing device indirectly compares the distance between vehicles based on modeling results that include geographical information of the space captured by at least one of the first and second video acquisition devices, and identifies the vehicles.

7. If a first lane and a second lane exist, the computing device sets virtual lines A and B in the first lane and the second lane, and indirectly calculates the distance between the two vehicles by utilizing lines A and B when calculating the spatial distance between the two vehicles. The distance between a vehicle located in the first lane and line A is set to 0, and the distance between a vehicle located in the second lane and line A is set to a value greater than 0. The parking management system according to claim 6, characterized in that the distance between a vehicle located in the second lane and line B is set to 0, and the distance between a vehicle located in the first lane and line B is set to a value greater than 0.

8. A first video acquisition device for recognizing the vehicle number of a vehicle at the entrance or inside of a parking lot, A second video acquisition device installed inside the aforementioned parking lot, The system includes a computing device that is communicated with the first video acquisition device or the second video acquisition device, The computing device identifies the vehicle by considering the vehicle information detected using RE-ID (re-identification) technology and the vehicle information obtained by analyzing the image acquired by at least one of the first and second video acquisition devices. The aforementioned re-identification technology is a technology that determines whether or not the vehicle is the same vehicle based on the external shape information of the vehicle included in the image acquired by the first image acquisition device and the external shape information of the vehicle included in the image acquired by the second image acquisition device. The parking management system is characterized in that the computing device applies an object detection algorithm to the video acquired by the second video acquisition device to detect the position of the vehicle, converts the preset parking area and the detected vehicle's position into a coordinate system based on camera-real-world relationship information, and calculates the degree of overlap between the detected vehicle and the parking area based on the conversion to detect a parking space where a vehicle can be parked.

Images