Ai-based system and method for sharing idle parking spaces in small-scale residential complexes

Abstract

The present invention relates to an AI-based control method for sharing idle parking spaces in small-scale multi-unit housing, comprising: a step of collecting parking lot image data from a shooting device installed in each of a plurality of small-scale multi-unit housings; a step of analyzing the collected parking lot image data through an artificial intelligence prediction model to generate predicted idle parking space information including the idle state of individual parking spaces by future time periods; a step of providing the generated predicted idle parking space information to a user terminal; a step of receiving a reservation request for a specific parking space and a specific idle time from the user terminal; and a step of processing the reservation of the corresponding parking space based on the received reservation request.

Description

Technology Field

[0001] The present invention relates to an AI-based idle parking space sharing control system for small-scale multi-unit housing and a method of operating the same. Background Technology

[0002] The present invention may fall within the technical field of vehicle access control systems and parking space sharing platforms.

[0003] Due to the intensification of urbanization and the rapid increase in the number of vehicles owned, parking shortages are emerging as a serious social problem, particularly in areas densely populated with small-scale multi-unit housing such as row houses, multi-family houses, and small officetels.

[0004] Recently, in addition to the shortage of parking spaces, unauthorized parking by external vehicles within apartment complexes and parking conflicts among residents have emerged as serious social issues. In fact, the severity of these conflicts is escalating, with the number of related complaints skyrocketing by more than 150 times over the past decade or so. The root cause of these problems lies in the fact that vehicle access routes and parking lots within apartment complexes do not qualify as "roads" under the current Road Traffic Act, making it difficult to implement coercive administrative measures, such as imposing fines or towing vehicles, against illegal or double parking. Consequently, residents are forced to rely on the voluntary efforts of residents or internal management regulations; however, this approach shows clear limitations in resolving issues regarding unauthorized occupation by external vehicles or long-term abandoned vehicles.

[0005] Even if attempts are made to technically resolve these management gaps, existing parking management systems require the installation of expensive dedicated hardware, such as license plate recognition cameras and vehicle barriers. Due to the high burden of initial setup costs and ongoing maintenance expenses, it is currently difficult to apply these systems economically to cost-sensitive small-scale apartment complexes.

[0006] Therefore, there is a very high technical demand for a low-cost, high-efficiency integrated management solution that can effectively control unauthorized parking by external vehicles while simultaneously efficiently sharing residents' idle parking resources. The problem to be solved

[0007] This invention was developed to solve the aforementioned problems and aims to provide a new parking sharing control system and a method of operation thereof that can alleviate parking shortages in small-scale multi-unit housing and maximize the utilization of idle resources. In particular, this invention focuses on technically preventing and resolving parking conflicts among residents and the problem of unauthorized parking by external vehicles.

[0008] Specifically, the present invention aims to collect parking status data by utilizing various existing CCTV cameras already installed in most buildings in order to minimize the burden of installing expensive dedicated equipment.

[0009] Furthermore, by analyzing collected video data using artificial intelligence (AI) technology to go beyond simply identifying current occupancy status and predict future idle parking times, we aim to activate legal sharing while effectively identifying abnormal occupancy (unauthorized parking).

[0010] Furthermore, another objective of this invention is to fundamentally prevent the possibility of unauthorized parking by external vehicles by establishing a system that allows only residents and pre-authorized visitors to access parking spaces via a mobile application. Through this, the invention seeks to establish parking order by supplementing the limitations of legal enforcement measures with technical access control.

[0011] The problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned problems will be clearly understood by a person skilled in the art from the description below. means of solving the problem

[0012] The present invention relates to an AI-based control method for sharing idle parking spaces in small-scale multi-unit housing, comprising: a step of collecting parking lot image data from a shooting device installed in each of a plurality of small-scale multi-unit housings; a step of analyzing the collected parking lot image data through an artificial intelligence prediction model to generate predicted idle parking space information including the idle state of individual parking spaces by future time periods; a step of providing the generated predicted idle parking space information to a user terminal; a step of receiving a reservation request for a specific parking space and a specific idle time from the user terminal; and a step of processing the reservation of the corresponding parking space based on the received reservation request. Effects of the invention

[0013] According to the present invention, existing CCTVs are utilized without the need to install new expensive dedicated sensors or cameras, thereby significantly reducing the initial investment costs and maintenance burden associated with the introduction of the system. This provides a foundation for distributing parking sharing systems even to small-scale multi-unit housing with limited economic conditions.

[0014] In addition, by notifying users of future idle parking times in advance through an AI-based prediction model, users can reliably secure parking spaces according to their schedules, and parking space providers can generate additional revenue during idle time, thereby maximizing the efficiency of resource utilization.

[0015] In addition, by integrating and sharing parking spaces from various buildings within the region into a single platform, a wider range of parking spaces is secured, thereby making a substantial contribution to alleviating the city's parking shortage and significantly improving user convenience.

[0016] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art from the description below. Brief explanation of the drawing

[0017] Various embodiments are described with reference to the drawings, wherein similar reference numbers are used to collectively refer to similar components. In the following embodiments, for illustrative purposes, a number of specific details are presented to provide a comprehensive understanding of one or more embodiments. However, it will be apparent that such embodiment(s) may be practiced without these specific details. FIG. 1 is a diagram illustrating, in an exemplary manner, the operating environment of an AI-based idle parking space sharing system according to one embodiment of the present invention. FIG. 2 is a block diagram schematically illustrating the internal configuration of a shared control system according to one embodiment of the present invention. FIG. 3 is a flowchart illustrating the basic processing steps of an AI-based idle parking space prediction method according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating the main processing steps of the idle parking space search and reservation process through a user app according to an embodiment of the present invention. FIG. 5 is a diagram illustrating an exemplary screen configuration of a user application according to one embodiment of the present invention. FIG. 6 is a flowchart illustrating the processing process of a method for improving analysis efficiency using edge computing according to another embodiment of the present invention. FIG. 7 is a flowchart illustrating the processing process of an external sensor linkage and event-triggered parking status identification method according to another embodiment of the present invention. To provide a clearer understanding of the technical concept related to the present invention, the following explanation is provided with reference to the attached drawings. However, the present invention is not limited to the attached drawings. Specific details for implementing the invention

[0018] FIG. 1 illustrates an exemplary operating environment (10) of a densely populated area of ​​small-scale multi-unit housing where an AI-based idle parking space sharing system according to an embodiment of the present invention can be applied. The operating environment (10) illustrated in FIG. 1 may be in a form where a number of small-scale multi-unit housing units (12a, 12b, 12c) (e.g., row houses, multi-family houses, small officetels, villas, etc.) are located adjacent to each other, and each multi-unit housing unit possesses its own parking space (14a, 14b, 14c). These parking spaces (14a, 14b, 14c) are, in principle, used for priority by residents, but may remain idle during specific time periods.

[0019] Existing CCTV cameras (20a, 20b, 20c) for monitoring the parking spaces (14a, 14b, 14c) of each apartment building (12a, 12b, 12c) or their surroundings may already be installed. The present invention may be characterized by collecting and utilizing video data from these various types of existing CCTV cameras (20a, 20b, 20c). Accordingly, additional costs associated with the installation of separate dedicated cameras and limitations on installation space can be minimized.

[0020] The collected video data can be transmitted to a central AI analysis center (40) via a communication network (30) (e.g., wired internet, wireless internet, mobile communication network, etc.). The AI ​​analysis center (40) can be implemented as a cloud-based server system and can perform the function of predicting the idle status of individual parking spaces within each parking space (14a, 14b, 14c) by time period by comprehensively analyzing the received videos.

[0021] Information on idle parking spaces predicted by the AI ​​analysis center (40) can be provided to a user terminal (50) via a communication network (30). The user terminal (50) may be a smartphone, a tablet PC, etc., and can perform functions such as checking information on idle parking spaces through a dedicated application (hereinafter "user app"), and reserving or paying for the use of the corresponding parking space if necessary. The user terminal (50) may be a terminal of an external visitor who needs a parking space, or a terminal of a resident who checks their own parking space information or manages sharing settings.

[0022] In this operating environment (10), the present invention can contribute to alleviating urban parking shortages and maximizing resource utilization by utilizing existing infrastructure to the fullest extent and by using AI technology to discover hidden idle parking resources within small-scale apartment complexes and connecting them to people who need them. In particular, the scalability and utility of the service can be enhanced by providing a model that integrates and manages multiple apartment parking spaces in a specific area or neighborhood unit rather than on an individual building unit basis.

[0023] However, existing CCTVs (20a, 20b, 20c) may vary greatly in terms of manufacturer, model, installation angle, resolution, and video output format, and some CCTVs may provide low-quality images or images with poor identifiability during specific times (e.g., night). In addition, the network environment of each apartment building may also differ, which may make it difficult to transmit video data stably. The present invention may include technical means for effectively analyzing idle parking spaces and providing sharing services even in such heterogeneous CCTV environments and various network conditions, which will be explained in more detail through the drawings described below.

[0024] FIG. 2 may be a block diagram schematically illustrating the internal configuration of an AI-based idle parking space sharing control system (hereinafter referred to as the "sharing control system (100)") according to an embodiment of the present invention. Referring to FIG. 2, the sharing control system (100) according to an embodiment of the present invention may be connected to communicate with a plurality of existing CCTVs (20a, 20b, 20c in FIG. 1), a plurality of user terminals (50 in FIG. 1), and optionally an IoT smart barrier (not shown), an external sensor (not shown), etc. through a communication network (30 in FIG. 1). The sharing control system (100) may be implemented as a central server or a distributed cloud platform that provides an AI-based idle parking space prediction and sharing solution.

[0025] The shared control system (100) may include a communication interface unit (110), an image data collection and preprocessing unit (120), an AI idle screen prediction unit (130), a data storage unit (140), a policy and reservation management unit (150), a user authentication and blocker control unit (160), and a revenue settlement unit (170).

[0026] The communication interface unit (110) can provide a communication interface for transmitting and receiving data between the shared control system (100) and external systems and devices. The communication interface unit (110) can support various wired and wireless communication protocols (e.g., TCP / IP, HTTP / HTTPS, MQTT, RTSP, WebRTC, etc.).

[0027] In one embodiment, the communication interface unit (110) may receive video streams or snapshot images in real-time or periodically from a plurality of existing CCTVs (20a, 20b, 20c of FIG. 1). Additionally, it may receive user account-related information, parking space search requests, reservation requests, payment information, location information, etc. from a user terminal (50 of FIG. 1), and transmit predicted idle space information, reservation confirmation results, notification messages, barrier control signals (if applicable). Optionally, it may be linked with an IoT smart barrier or external sensors (e.g., IR sensors, loop sensors) to receive status information of the devices or transmit control commands. Furthermore, it may be linked with external information systems, such as a weather information server, a holiday information database, and a map service API, to collect additional data for improving AI prediction accuracy and user services.

[0028] The video data collection and preprocessing unit (120) can perform the important function of collecting, normalizing, and preprocessing heterogeneous CCTV video data with various characteristics received through the communication interface unit (110) into a form that the AI ​​idle area prediction unit (130) can effectively process.

[0029] In one embodiment, the video data collection and preprocessing unit (120) can recognize that the codecs (e.g., H.264, H.265, MJPEG), resolution, frame rate, transmission protocol (e.g., RTSP, HTTP Live Streaming), etc., of video streams transmitted from CCTVs installed in different apartment buildings are different, and can perform the task of converting them into a standardized format suitable for AI analysis. For example, consistency of the AI ​​model input data can be ensured by transcoding with a specific codec, downscaling videos of excessively high resolution, and performing limited upscaling or interpolation processing on videos of too low resolution.

[0030] In another embodiment, the image data collection and preprocessing unit (120) can perform preprocessing for image quality improvement, such as image noise removal, shake correction (when the CCTV is installed outdoors and affected by wind, etc.), and brightness improvement in low-light environments. Additionally, by receiving information on the installation location, viewing angle, and surveillance area of ​​each CCTV in advance or by the system automatically inferring it, a Region of Interest (ROI) corresponding to the actual parking space can be set, and only that area can be extracted and analyzed, thereby reducing unnecessary computations and increasing analysis efficiency.

[0031] Since these collection and preprocessing processes have a direct impact on the accuracy and stability of AI predictions, the function of this part can be considered very important in the present invention.

[0032] The AI ​​idle space prediction unit (130) can perform the core function of the present invention by analyzing normalized parking lot image data provided from the image data collection and preprocessing unit (120) to predict the idle state of each individual parking space or a specific parking area at future times. The AI ​​idle space prediction unit (130) may be composed of a plurality of sub-modules including various algorithms based on deep learning and machine learning.

[0033] For example, the AI ​​idle space prediction unit (130) includes a vehicle detection and parking space occupancy analysis module (131) and can quickly and accurately detect vehicles in input video frames using deep learning-based object detection algorithms such as YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), and Faster R-CNN, and determine the occupancy status (presence or absence of a vehicle) of each parking space in real time or periodically.

[0034] Additionally, by including a temporal pattern learning module (132), it is possible to learn and model occupancy patterns by time of day, day of the week, and season (e.g., idle patterns due to weekday commuting times of a specific vehicle, long weekend outing patterns, etc.) based on past occupancy data of a specific parking space or a specific vehicle (if linked with a vehicle license plate recognition function). This learning can be performed by accumulating data over a long period (e.g., several weeks or several months), and through this, the possibility of periodically repeating idle periods can be predicted.

[0035] Optionally, by including an external factor integration analysis module (133), weather information (e.g., changes in parking demand due to reduced outdoor activities during heavy rain or snow), information on holidays and specific events (e.g., surge in parking demand during nearby festivals), and surrounding traffic conditions information can be collected via an external API, and the prediction accuracy can be further improved by analyzing the combined impact of these factors on parking demand and the occurrence of idle spaces and reflecting this in the prediction model.

[0036] Finally, the idle space prediction model generation and inference module (134) can synthesize the analyzed information to generate a model that predicts the probability that each individual parking space will be idle at a specific future point in time (e.g., after the next hour, at a specific time of day the next day) and the expected duration of idleness (e.g., "85% probability that parking space No. 2 at Villa A will be empty from 9 AM to 5 PM tomorrow"), and through this, infer and provide actual predicted values.

[0037] The data storage unit (140) can perform the function of permanently or temporarily storing and managing various data required for the operation of the shared control system (100). The data storage unit (140) can be configured by combining various types of storage, such as a relational database (RDBMS), a NoSQL database, a file system, and an in-memory data store.

[0038] In one embodiment, the data storage unit (140) may store key frames or metadata of collected and preprocessed image data, historical information on the past and present occupancy status of each parking space, learned AI prediction model parameters and weights, predicted idle space information (parking space ID, expected idle start / end time, probability, etc.), user account information (ID, encrypted password, contact information, etc.), user registered vehicle information (vehicle number, vehicle type, etc.), user reservation history, payment information and history, sharing policy information for each building / parking space (allowed sharing time, unit price, resident-only time, etc.), system logs, etc.

[0039] The policy and reservation management unit (150) can manage policies regarding parking space sharing set by a building manager (or an authorized resident representative, etc.), process parking space reservation requests from a user terminal (50 in FIG. 1), and perform the function of managing the reservation status in real time.

[0040] In one embodiment, the policy and reservation management unit (150) can set and store, through an administrator interface (in the form of a web or app, not shown), time slots available for sharing for each row house or individual parking space, hourly or per-case usage fee policy, rules for priority use by residents during specific time periods, external sharing restriction conditions, etc., in the data storage unit (140).

[0041] In another embodiment, the policy and reservation management unit (150) may provide the user with a list of parking spaces available for reservation at a specific time in the present or future and an estimated time of availability based on real-time prediction information provided by the AI ​​idle space prediction unit (130). When a user's reservation request is received, the availability of the parking space (e.g., prevention of duplicate reservations, compliance with sharing policies) is quickly determined, the user is notified of the result of the reservation confirmation or rejection, and the confirmed reservation information is recorded in the data storage unit (140).

[0042] The user authentication and barrier control unit (160) can perform the function of authenticating a user approaching to use a parking space and, if necessary, remotely controlling the opening and closing of the IoT smart barrier of the apartment complex (if installed). If there is no barrier, this part can operate mainly as a usage guidance function through an app after user authentication and access permission verification.

[0043] In one embodiment, the user authentication and barrier control unit (160) may grant access rights by verifying the reserved user's vehicle number (if registered in advance and linked with the LPR camera) or an authentication request through a user app (e.g., clicking a button within the app, scanning a QR code, detecting a BLE signal, etc.).

[0044] In another embodiment, when valid access rights are confirmed, the user authentication and barrier control unit (160) can allow the entry of a vehicle by transmitting an opening signal to the IoT smart barrier of the parking lot through the communication interface unit (110). A similar authentication and control process may be performed when exiting. In the absence of a barrier, the user app may be clearly notified of the “parking available” status and guided to the exact location of the parking space (e.g., photo, map) to encourage autonomous parking by the user.

[0045] The revenue settlement unit (170) can perform a settlement function that calculates revenue generated from the actual use of the shared parking space and transparently distributes it among stakeholders, such as building owners and platform operators, according to established criteria (e.g., fee policy, distribution ratio).

[0046] In one embodiment, the revenue settlement unit (170) can automatically calculate the usage fee based on the user's actual parking usage time (based on the entry and exit time records) and the fee policy set in the policy and reservation management unit (150), charge the user, and process the payment through a linked payment system.

[0047] In another embodiment, the profit settlement unit (170) can automatically separate and settle the profit attributable to each row house (building owner) and platform operation fees, etc., according to a pre-agreed distribution ratio (e.g., building owner 70%, platform operator 30%) for the total profit generated during a specific period (e.g., every month), and generate a related detailed report and provide it to the manager.

[0048] These individual component modules are organically interconnected to automate and streamline the entire process, ranging from AI-based idle parking space prediction using existing CCTVs to user sharing, reservation, payment, and settlement. However, in actual implementation, the functions of some modules may be integrated or further subdivided, which may be considered as design variations that do not deviate from the scope of the present invention.

[0049] With reference to FIG. 3 below, the basic processing steps of the AI-based idle parking space prediction method of the present invention will be explained.

[0050] FIG. 3 may be a flowchart illustrating the basic processing steps of an AI-based idle parking space prediction method according to an embodiment of the present invention. For example, the exemplary process (S300) illustrated in FIG. 3 may be performed through the organic operation of various hardware and software components of the shared control system (100) illustrated in FIG. 2.

[0051] First, in step S310, the video data collection and preprocessing unit (120) of the shared control system (100) can periodically collect and normalize video data related to parking lots from existing CCTVs installed in each of the multiple small-scale apartment buildings.

[0052] In one embodiment, a scheduler (e.g., a cron job or a dedicated scheduling service) running on the main processor (CPU) of the shared control system (100) can trigger a video acquisition task at a predefined time interval (e.g., every 5 minutes). In response to this trigger, the NIC of the communication interface unit (110) can connect to the IP address and streaming path (e.g., an RTSP URL) of each CCTV to request and receive video frames or short video clips. The received raw video data (e.g., H.264 compressed data) can be loaded into system memory and then converted into raw pixel data (e.g., a bitmap in RGB or YUV format) through decoding software (e.g., a call to an FFmpeg library function) running on the CPU or VPU of the video data acquisition and preprocessing unit (120).

[0053] Afterward, the converted pixel data may undergo preprocessing steps such as resizing (e.g., resizing to 608x608 pixels), normalization (e.g., scaling pixel values ​​to a 0-1 range), and changing the channel order (e.g., BGR to RGB) through an image processing routine (e.g., OpenCV function) running on the same processor or GPU to fit the input format of the AI ​​idle surface prediction unit (130). The preprocessed final data (e.g., a multidimensional array consisting of floating-point values) may be prepared in a specific area of ​​system memory or GPU memory for AI analysis.

[0054] Next, in step S320, the AI ​​idle space prediction unit (130) of the shared control system (100) can predict future idle parking spaces by receiving the preprocessed image data prepared in step S310 as input and performing AI analysis.

[0055] In one embodiment, preprocessed image data (multidimensional array) can be transmitted to the GPU of the AI ​​idle surface prediction unit (130). An inference engine of a deep learning object detection model (e.g., YOLOv5) loaded on the GPU can receive this input data and perform a forward pass to generate an output tensor containing the location (bounding box coordinates), class (whether it is a vehicle), and confidence score of each vehicle object in the image. This output tensor is then transmitted to the CPU, and post-processing logic executed on the CPU can generate a digital value that determines the current occupancy status of each parking space (e.g., occupancy=1, non-occupancy=0) by analyzing the spatial relationship (e.g., calculation of IoU value) between each parking space area and the detected vehicle bounding box.

[0056] Next, a time series prediction model (e.g., an LSTM-based model) running on the CPU or GPU of the AI ​​idle space prediction unit (130) can receive as input a currently determined occupancy state value, a past occupancy history sequence retrieved from the data storage unit (140) (e.g., an array of occupancy states in 5-minute intervals over the last 7 days), and external factor data (e.g., a weather code quantified from a weather API, a flag indicating whether it is a public holiday), and generate a prediction result including a probability value (e.g., a real number between 0.0 and 1.0) that each parking space will be idle for a specific period of time in the future and an expected idle start / end time (timestamp).

[0057] Subsequently, in step S330, the data management logic executed on the CPU of the shared control system (100) can record the prediction results (ID of each parking space, predicted idle time slot, probability value, etc.) generated in step S320 by executing SQL or NoSQL commands to insert them as new records or update existing records in a related table (e.g., 'Idle_Space_Prediction_Information' table) within the data storage unit (140). During this process, database transaction management is applied to ensure data consistency and integrity.

[0058] Finally, in step S340, the communication interface unit (110) of the shared control system (100) can provide the latest idle parking space information updated in the data storage unit (140) to the application of the user terminal (50 in FIG. 1).

[0059] In one embodiment, when a user app is launched or switched to a specific screen, the app may request idle information for a specific region or time zone from the server via an HTTPS GET request. Web application logic running on the server's CPU may receive this request, query the data storage unit (140), and construct the queried result into JSON formatted data and send it to the user app as an HTTP response. The user app may parse the received JSON data and display it visually on the screen. Alternatively, the server may asynchronously transmit important idle information changes to the user app via a push notification service (e.g., FCM, APNS).

[0060] The method of FIG. 3 described above may represent the basic idle parking space prediction and information provision method of the present invention, and may include a series of processes in which various hardware and software modules of the system interact in a predetermined order and manner to process data and generate results.

[0061] FIG. 4 may be a flowchart illustrating the main processing steps of the process of searching for and reserving idle parking spaces through a user app according to an embodiment of the present invention. This process may include data exchange and processing through network communication between a user app running on the processor and memory of a user terminal (50 in FIG. 1) according to user input, and the server hardware and software of the shared control system (100) of FIG. 2.

[0062] First, in step S410, the user can input the desired parking area (e.g., tapping a specific point on a map or entering an address string) and time zone (e.g., manipulating a date and time selection widget) into the user app (see FIG. 5) through the touchscreen interface of their terminal (e.g., smartphone). The input information may be temporarily stored in the app's internal memory.

[0063] Next, in step S420, the user app can use the terminal’s CPU and communication module (e.g., Wi-Fi or LTE modem) to generate an HTTPS request message (e.g., JSON payload) containing search conditions (local coordinates, time range, etc.) entered in step S410, and transmit it to the communication interface unit (110) of the shared control system (100) via the communication network (30).

[0064] Subsequently, at step S430, the application logic of the policy and reservation management unit (150) running on the CPU of the shared control system (100) can parse the received search request and execute an SQL or NoSQL query on the idle parking space information table stored in the data storage unit (140). This query can search for records by filtering them according to the regional range (e.g., radius search based on latitude / longitude) and time range (e.g., parking spaces that are idle between the requested start time and end time). The searched results can be loaded into system memory.

[0065] Subsequently, at step S440, the CPU of the shared control system (100) serializes the list of parking spaces (e.g., parking space ID, building name, idle time, fee, etc.) retrieved at step S430 into a standard data format such as a JSON array, and can send it back to the user terminal through the communication interface unit (110) by including it in an HTTPS response message. The user app receives this response through the communication module of the terminal, converts the JSON data into an internal object through parsing logic executed on the CPU, and then passes it to a UI rendering engine (e.g., Android View System, iOS UIKit) for screen display to visually present it to the user in the form of a map or a list (see FIG. 5 (a) and (b)).

[0066] When a user selects a specific parking space from the displayed search results via the touchscreen, in step S450, the user app can generate a reservation request message (HTTPS POST) including the unique identifier of the selected parking space, the confirmed usage time, and, if necessary, the vehicle number (keyboard input), and send it back to the shared control system (100).

[0067] Next, in step S460, the policy and reservation management unit (150) running on the CPU of the shared control system (100) can process the received reservation request. This may include checking whether there are simultaneous reservation attempts for the corresponding parking space and time in the reservation table of the data storage unit (140) (e.g., atomic operation using database locking or transactions), and if there are no issues, inserting new reservation information into the table. If payment is required, the process of sending a payment approval request via HTTPS communication with an external payment gateway (e.g., PG company API) and receiving and processing the result may be performed on the CPU.

[0068] Finally, in step S470, the shared control system (100) may send an HTTPS response containing the result of the reservation processing (success or failure code, reservation ID, etc.) to the user terminal. The user app may receive this response and display a notification of reservation completion or failure (e.g., pop-up message, screen transition) to the user, and if successful, store the reservation details in internal memory or local storage so that the user can check them at any time.

[0069] This S400 process can provide a user experience of efficiently searching for and reserving physically distributed idle parking resources through active user input and the system's automated data processing and communication processes.

[0070] FIG. 6 may be a flowchart illustrating a process of a method to increase the efficiency of idle parking space analysis and reduce the load of a central shared control system (100) by utilizing edge computing as another embodiment of the present invention. For example, the exemplary process (S600) illustrated in FIG. 6 may represent a distributed cooperation model between the processor, memory, storage, and specialized AI accelerator (if present) of an edge computing device (hereinafter “edge device”) installed near each apartment building or CCTV and the central shared control system (100). The edge device may be implemented in the form of, for example, an NVIDIA Jetson series, a Google Coral Dev Board, or a Raspberry Pi combined with an AI acceleration module (e.g., Movidius Neural Compute Stick).

[0071] In particular, in one embodiment of the present invention, a single edge device may have a distributed structure in which it manages only the CCTVs belonging to a small group of physically adjacent apartment buildings or the apartment building in which it is installed, rather than managing all CCTVs of the entire system. For example, an edge device installed in Villa A is responsible only for the CCTVs of Villa A, and an edge device in Villa B is responsible for the CCTVs of Villa B. This local clustering method ensures the overall scalability of the system, optimizes the local network load between a specific edge device and a CCTV, and contributes to improving the accuracy and responsiveness of analysis by allowing the edge device to focus only on video analysis of the adjacent environment it can best understand.

[0072] First, in step S610, video data (e.g., RTSP stream) captured from the existing CCTV of each apartment building can be preferentially input into the network interface of an edge device directly connected to the CCTV via a local network (e.g., Ethernet or Wi-Fi). The CPU of the edge device can decode the received video stream and store it in internal memory frame by frame.

[0073] Next, in step S620, the main processor (CPU) or built-in AI accelerator (e.g., GPU, NPU) of the edge device can perform primary analysis or lightweight AI processing on the input video frame.

[0074] In one embodiment, a lightweight object detection model (e.g., MobileNet-SSD, TinyYOLO) or motion detection algorithm (e.g., background subtraction, optical flow analysis) mounted on the edge device may be executed. This model / algorithm is optimized for the limited computational resources (CPU clock, memory capacity) of the edge device to detect the appearance or disappearance of vehicles within a parking area, or changes in the occupancy status of a specific parking space (e.g., a vehicle parking in an empty space, a parked vehicle exiting), in real-time or near real-time, and can generate results (e.g., event type, time of occurrence, related parking space ID) in the form of text or numeric data. For example, the edge device may monitor a threshold for pixel change in a specific parking space area and generate a “high probability of occupancy status change” event if the change exceeds the threshold.

[0075] Next, in step S630, the data transmission logic executed in the CPU of the edge device can determine the data to be transmitted to the central shared control system (100) based on the primary analysis result of step S620.

[0076] In one embodiment, the edge device may selectively package a key video frame (low-resolution or compressed image file) or a short video clip (a video file several seconds in length) at the time the change occurred, along with corresponding event information (text / numeric data), only when a significant change in the parking status is detected, and transmit it (e.g., an HTTPS POST request or an MQTT message issuance) to the communication interface unit (110) of the central shared control system (100) via the Internet through its own communication module (e.g., Ethernet, Wi-Fi, or LTE modem) via the Internet. If the edge device does not detect any significant change for a certain period of time, it may transmit only periodic “good status” signals or minimal summary information (e.g., current total parking space occupancy) to the central server, or omit data transmission.

[0077] After that, in step S640, the AI ​​idle screen prediction unit (130) of the central shared control system (100) can receive limited data (selected event information, low-resolution images / clips) transmitted from the edge device as input, perform a more precise secondary analysis, and update the idle screen prediction model.

[0078] In one embodiment, a high-performance GPU of a central server can apply a precise deep learning model (e.g., high-resolution YOLO, time-series forecasting model) described in FIG. 2 to an image / clip delivered by an edge device to generate results that more reliably predict the future idle time of the corresponding parking space, accurately identify the type of vehicle, recognize license plates (if possible), and predict the exact type of vehicle. Since primary filtering has already been performed at the edge, the central server can concentrate computational resources only on data that is actually of high analytical value.

[0079] Finally, in step S650, the central shared control system (100) can perform the process of storing the results of the secondary analysis and prediction update in the data storage unit (140) and providing the relevant information to the user app if necessary (similar to S330 and S340 of FIG. 3). Since event information with enhanced real-time capabilities is reflected through edge computing, the timeliness of the idle information provided to the user can be further increased.

[0080] The method of Fig. 6 has the potential to compensate for the disadvantages of centralized analysis and provide reduced data transmission volume, improved response speed, central server load distribution, and even limited autonomous operation capabilities at edge devices (e.g., simple entry / exit recording) even when the network is disconnected.

[0081] FIG. 7 is another embodiment of the present invention and may be a flowchart illustrating a process for identifying and responding to parking status selectively or preferentially at a necessary time by utilizing a detection signal from an external physical sensor (e.g., IR sensor) or a logical event within the system, such as a user reservation, as a trigger. For example, the exemplary process (S700) illustrated in FIG. 7 may be performed through interaction between various modules of the shared control system (100) of FIG. 2, external sensor hardware installed at the parking site, and a user terminal (50 of FIG. 1).

[0082] First, in step S710, external sensors installed at the entrance of the apartment building parking lot or at individual parking spaces, such as an infrared (IR) beam sensor (composed of a transmitter and a receiver to detect beam blockage when a vehicle passes) or a floor-embedded magnetic loop detector (detecting changes in the magnetic field when a vehicle passes through a metal body) can physically detect the presence or movement of a vehicle.

[0083] When a detection event occurs (e.g., when the IR beam is blocked for more than a certain amount of time, or when the inductance value of the loop coil exceeds a threshold), in step S712, a microcontroller unit (MCU, e.g., ESP32, Arduino) or edge device connected to the sensor via wired (e.g., GPIO connection) or wireless (e.g., Zigbee, LoRaWAN) recognizes this detection signal (e.g., electrical pulse or digital data packet) as a trigger event and can immediately transmit it to a central shared control system (100) or a nearby edge device (Fig. 6). Transmission may use MQTT messages, UDP packets, or low-power wide area network (LPWAN) protocols.

[0084] The system that receives the trigger signal (the communication interface unit (110) of the central server or the CPU of the edge device) can, in step S714, immediately check the CCTV video of the location where the trigger occurred (identified by the sensor ID) or perform priority AI analysis on the video (the AI ​​idle space prediction unit (130) of FIG. 2 or the execution of the AI ​​model of the edge device). For example, when the IR sensor detects a vehicle passing through the entrance gate, the system can obtain the video of the gate CCTV for the last few seconds to activate the license plate recognition (LPR) module or perform parking space occupancy analysis to track which parking space the vehicle is parked in. This result is immediately reflected in the data storage unit (140) to increase the real-time nature of the parking status.

[0085] In step S720, an event in which a new reservation for a specific parking space occurs or existing reservation information is changed through an app of a user terminal (50 in FIG. 1) can be processed by the policy and reservation management unit (150) of the shared control system (100) (see S450-S470 in FIG. 4). This reservation information (reservation ID, parking space ID, reservation start / end time, user ID) can be recorded in the data storage unit (140).

[0086] In step S722, the scheduler or event management logic running on the CPU of the shared control system (100) may periodically scan the reservation information of the data storage unit (140) or register specific time-based notifications related to the reservation (e.g., 15 minutes before the reservation starts, 5 minutes before the reservation ends) in an internal timer queue or message queue at the time of creating or changing the reservation.

[0087] In step S724, when a set reservation-related event (e.g., timer expiration) occurs, the AI ​​idle space prediction unit (130) of the system may optionally perform intensive monitoring or detailed analysis of the CCTV footage capturing the reserved parking space. For example, when the reservation start time is imminent, the system may perform vehicle detection and occupancy analysis on the latest video frame to determine whether the parking space is actually empty or if it is being occupied without authorization by a vehicle other than the reserved user. Additionally, when a user presses a button in the app indicating a status change, such as “parking complete” or “expiration scheduled,” this is also considered an event and may trigger the relevant CCTV video analysis. Here, when the reservation time is imminent, it may mean a preset time (e.g., 20 minutes) based on the reservation time.

[0088] Parking status information identified based on events in step S714 or S724 (e.g., the status of a specific parking space changing from 'reserved idle' to 'occupied') can be immediately updated in step S730 by the CPU of the shared control system (100) in the idle space status table and the reservation status table within the data storage unit (140). If a significant discrepancy (e.g., an unauthorized vehicle found in a reserved parking space) is detected, the system can take countermeasures, such as automatically sending a warning notification (e.g., email, SMS) to the administrator or suggesting another available parking space to the reservation user.

[0089] Finally, in step S740, the real-time parking status information based on events is integrated with the periodic sampling-based information of Fig. 3 and provided to a user app through the communication interface unit (110), so that the user can always receive the latest and more reliable parking space information.

[0090] The method of Fig. 7 can go beyond the system passively collecting and analyzing data and intelligently respond to active events, such as actual physical changes within the parking environment or user behavior, thereby providing a means to significantly improve the accuracy of the service and user satisfaction along with the efficient use of system resources.

Claims

Claim 1 An AI-based control method for sharing idle parking spaces in small-scale multi-unit housing, comprising: a step in which a control system collects parking lot image data from a plurality of heterogeneous shooting devices pre-installed in each of the small-scale multi-unit housings and converts at least one of the codec, resolution, and transmission protocol of the received image data to generate normalized image data suitable for input to an AI prediction model; a step in which the control system determines the current occupancy status of an individual parking space from the normalized image data using a deep learning-based object detection model, inputs the determined current occupancy status and the time-series past occupancy history information of the individual parking space pre-stored in a data storage unit into a time-series prediction model to generate predicted idle parking space information including the time-series future idle status and expected idle duration of the individual parking space; a step in which the control system provides the generated predicted idle parking space information to a user terminal; a step in which the control system receives a reservation request for a specific parking space and a specific idle time from the user terminal; and a step in which the control system processes the reservation of the corresponding parking space based on the received reservation request. Claim 2 The AI-based idle parking space sharing control method according to claim 1, wherein the step of generating the predicted idle parking space information comprises additionally inputting external environmental factor information, including at least one of weather information and holiday information received from an external information system, into the time series prediction model to predict the idle state and expected idle duration for each future time period. Claim 3 The AI-based idle parking space sharing control method according to claim 1, wherein the steps of generating normalized image data and generating predicted idle parking space information comprise: a step in which an edge computing device deployed at the site of the multi-unit housing directly receives an image stream from one or more local shooting devices deployed at a certain distance adjacent to the edge computing device among the plurality of shooting devices and performs a primary analysis to detect a change in parking status; a step in which the edge computing device transmits core event data and related image frames to a central control system only when a change is detected as a result of the primary analysis; and a step in which the central control system applies a high-performance AI prediction model to the core event data and related image frames to perform a precise secondary analysis. Claim 4 The AI-based idle parking space sharing control method according to claim 1, wherein the step of generating normalized image data comprises removing image noise, improving brightness in low-light environments, and extracting a Region of Interest (ROI) corresponding to an actual parking space for image data received from the heterogeneous imaging device, and wherein the time series prediction model learns periodic occupancy patterns by time of day, day of the week, and season from the past occupancy history information and reflects this in the prediction of the future idle state and expected idle duration. Claim 5 The AI-based idle parking space sharing control method according to claim 1 further comprises: a step of analyzing the current occupancy status of a reserved parking space using a deep learning-based object detection model when an event occurs indicating that the reservation start time is imminent based on reservation information received from the user terminal; and a step of transmitting a warning notification to an administrator terminal and providing alternative available parking space information to the reservation user terminal when unauthorized occupancy by a vehicle other than the reserved user's vehicle is detected as a result of analyzing the current occupancy status of the reserved parking space. Claim 6 The AI-based idle parking space sharing control method according to claim 1, wherein the step of collecting parking lot video data further includes the step of calculating the actual usage time of the parking space based on the result of processing the reservation of the parking space, calculating the usage fee according to the calculated usage time and a preset fee policy, and settling the calculated usage fee by automatically separating it into profit attributable to the owner of the multi-unit housing complex to which the parking space belongs and a platform operation fee according to a preset distribution ratio. Claim 7 As an AI-based control system for sharing idle parking spaces in small-scale multi-unit housing, the system comprises: a communication interface unit that is connected to communicate with heterogeneous imaging devices and user terminals pre-installed in each of multiple small-scale multi-unit housing; and a processor. A memory connected to the processor and storing executable commands; wherein, when the commands are executed by the processor, the system collects parking lot image data from the heterogeneous shooting device through the communication interface unit and converts at least one of the codec, resolution, and transmission protocol of the received image data to generate normalized image data suitable for input to an AI prediction model; determines the current occupancy status of an individual parking space from the normalized image data using a deep learning-based object detection model, inputs the determined current occupancy status and the time-series past occupancy history information of the individual parking space stored in the data storage unit into a time-series prediction model to generate predicted idle parking space information including the future time-series idle status and expected idle duration of the individual parking space; provides the generated predicted idle parking space information to the user terminal; and is configured to receive a reservation request for a specific parking space and a specific idle time from the user terminal and process the reservation.

Images