Search

KR-102963140-B1 - Edge AI-based Parking Control System for Pedestrian Safety and Prevention of Unauthorized U-turns

KR102963140B1KR 102963140 B1KR102963140 B1KR 102963140B1KR-102963140-B1

Abstract

A parking control system for pedestrian safety and prevention of unauthorized turning based on Edge AI according to an embodiment of the present invention may include: a display blocking bar installed at the entrance and exit sides of a parking lot, respectively, and equipped with a display for outputting visual information to indicate whether an event related to safety accidents and violations has occurred; an operating PC installed at the entrance and exit sides of the parking lot, respectively, and equipped with an electronic display board for outputting the vehicle number, entry time, and exit time of a vehicle entering or exiting, and visual information to indicate whether the event has occurred; and an Edge AI device that analyzes camera images captured by cameras installed at the entrance and exit sides of the parking lot to output metadata of the vehicle, and transmits a control signal to control the operation of the display blocking bar and the operating PC based on the analysis result of the camera images, and an output signal to output the visual information on the display and the electronic display board to the display blocking bar and the operating PC.

Inventors

  • 김태경

Assignees

  • 주식회사 넥스파시스템

Dates

Publication Date
20260511
Application Date
20260121

Claims (12)

  1. A display barrier bar installed at the entrance and exit sides of the parking lot, respectively, and equipped with a display for outputting visual information to indicate whether an event related to safety accidents and violations has occurred; An operating PC installed at the entrance and exit sides of the above-mentioned parking lot, respectively, and equipped with an electronic display board for displaying the vehicle number, entry time, and exit time of vehicles entering and exiting, and time information for indicating whether the above-mentioned event has occurred; and An Edge AI device that analyzes camera images captured by cameras installed at the entrance and exit sides of the parking lot to output metadata of the vehicle, and transmits a control signal to control the operation of the display blocking bar and the operating PC based on the analysis result of the camera images, and an output signal to output the time information from the display blocking bar and the operating PC to the display blocking bar and the operating PC; The metadata of the above vehicle is, Includes the vehicle number, license plate, and entry time of the above vehicle, The above Edge AI device is, An Edge AI module that detects an object from the camera image, analyzes the attributes of the detected object to output metadata of the vehicle, and determines whether an event has occurred within the parking lot based on object detection and attribute analysis; and The above Edge AI module is, Generates a control signal and an output signal based on the result of determining whether the above event has occurred, and transmits the control signal and the output signal to the display blocker and the operating PC. The above Edge AI module is, An object tracker that assigns a unique identification number (ID) to each object in each image frame of the camera image before detecting objects and analyzing attributes from the image of the camera; and An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning around, characterized by including: a Region of Interest (ROI) Selector that tracks a unique identification number held by the object tracker in real time and selects an image frame at the best shot point in which the entire shape of the vehicle is most exposed among image frames containing an object with the same unique identification number.
  2. delete
  3. delete
  4. In Article 1, The above Edge AI module is, An object detection model for detecting the object, which is a lightweight model for object detection and attribute analysis, and an object analysis model for analyzing the attributes of the object detected by the object detection model are applied. An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning, characterized by being designed with a sequential 2-Stage structure in which the object detection model detects the object and the object analysis model analyzes the attributes of the object.
  5. In Article 4, The above object detection model is, An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning, characterized by utilizing the lightweight processing results of the YOLO-Tiny model for object detection.
  6. In Article 4, The above object analysis model is, An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning, characterized by being an artificial intelligence model that combines a MobileNet model and an EfficientNet-Lite model for analyzing the attributes of the above-mentioned object.
  7. In Article 6, The above object analysis model is, An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning, characterized by adopting a multi-structure that shares a backbone network during the inference operation process of analyzing the attributes of the above-mentioned object, but branches into multiple heads that perform metadata determination of the vehicle at the output end.
  8. In Article 7, The above object analysis model is, An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning around, characterized by having a multi-structure including a vehicle entry time head that incorporates time information of the camera image acquisition time into the calculation, configured with multiple inference heads behind a single MobileNet backbone to extract the vehicle number and license plate of the vehicle through a single inference based on a multi-structure.
  9. In Article 8, The above object analysis model is, The entry time of the above vehicle is applied as a weight for an inference operation to analyze the attributes of the above object, and The entry time of the above vehicle is, An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning, characterized by being combined with movement data of a vehicle detected upon exiting the parking lot and utilized as a threshold to determine whether the vehicle has made an unauthorized turn.
  10. delete
  11. In Article 1, The above Edge AI module is, After storing the metadata of a vehicle corresponding to an object assigned a unique identification number through the object tracker, re-recognize whether it is the same vehicle when the vehicle with the stored metadata exits the parking lot, and The above Edge AI device is, An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning, characterized by transmitting control signals and output signals to the display blocking bar and the operating PC to block the violation of a vehicle when the vehicle re-recognized in the exit section attempts to evade the exit without exiting, based on the re-recognition of the same vehicle by the Edge AI module.
  12. In Article 1, An Edge AI-based parking control system for pedestrian safety and prevention of unauthorized turning around, further comprising: a parking control server that monitors the entry and exit status of the vehicle and transmits an alarm to a terminal equipped by the manager to notify the manager of the event when the event occurs.

Description

Edge AI-based Parking Control System for Pedestrian Safety and Prevention of Unauthorized U-turns The present invention relates to a parking control system for pedestrian safety and prevention of unauthorized turning around based on Edge AI. Recently, urban environments are changing due to new modes of transportation and the utilization of edge-based traffic information, and the parking environment is also undergoing significant changes as unmanned parking technology can affect parking lot capacity and parking patterns. Furthermore, urban and residential areas face different parking challenges. With the development of the mobility industry, simplified modes of transportation, and dedicated road environments, complaints have arisen regarding inconveniences caused by safety issues among pedestrians; consequently, the ownership and operation of parking infrastructure require improvement from the perspective of various stakeholders. Generally, a conventional parking control system refers to a system that operates or drives an operating PC after recognizing the license plate number of a passing vehicle by embedding loop coils in the road surface within the parking lot to manage vehicle entry and exit, such as in public institution parking lots, department stores, apartments, and paid parking lots. In order to reduce labor costs associated with the use of personnel for vehicle management, the operation and management of basic parking control systems involves the development of systems that recognize vehicles by installing cameras at the entrances and exits of parking lots to detect license plates. These systems utilize a method in which loop coils are embedded in the vehicle entry lanes and cameras are installed at the front of the lanes; when a vehicle's entry is detected by the loop coils, the camera captures the license plate, and the license plate number of the vehicle that entered the lane is extracted from the captured image. Here, burying loop coils in the road surface is time-consuming and costly, and there are often problems where they fail to function properly even after installation due to reasons such as electrical leakage or disconnection. In addition, regarding vehicles entering and exiting, there is a problem where they are judged to be the same vehicle due to the function of the loop coil installed on the road surface and the structural design during the burial stage, or there is a concern about loss of toll collection. In addition, the method using loop coils has limitations in environmental structure and increased costs, as it can only detect whether a vehicle is passing through the location, or requires burying two loop coils side by side to detect the direction of travel of a vehicle based on license plate recognition by an operating PC. Furthermore, the configuration of existing parking systems introduced for parking lot management adopts a centralized control method for overall management. Consequently, in environments influenced by the size of the parking lot and the status of moving vehicles, there are operational difficulties due to limitations in statistics, operational losses, and equipment management (status information management including camera video signals, loop coil disconnection, operation PC operation, and data transmission). To address these issues, problems related to control technology have been resolved to enhance the efficiency of local system configurations and server operations. However, since applicable technologies vary depending on the system installation environment and there are unique characteristics of vehicles—ranging from large trucks to motorcycles—regarding movement paths for entry and exit, system configurations capable of satisfying these requirements are increasingly becoming localized (dependent on the terminal itself). Representative examples include edge devices such as embedded systems or Raspberry Pi products, and these devices are being expanded and distributed through the development of lightweight products. Against this backdrop, for vehicles using limited parking spaces, restrictions such as boarding and alighting areas, movement passages, and height considerations for the driver are unnecessary, allowing for a revolutionary change in parking spaces, and the availability of parking spaces is gradually expanding and increasing from manned to unmanned parking to levels ranging from 20% to 87%. In particular, due to changes in the domestic social environment, technical automation is urgently needed as limitations in workforce management and automation are becoming apparent in the management and operation of production sites—including a rapidly aging society and even for drivers—as well as in handling safety issues, system operations, malicious complaints, and unexpected situations such as equipment failures. In addition, there are operational risks and revenue leakage due to the limitations of the existing central server method. In the case of existin