System and Method for Barcode-Based VIN Recognition and Operation of High-Speed Moving Objects

Abstract

A barcode VIN recognition and operation system for a high-speed moving object according to the present invention may include: a first camera that detects a vehicle object from an entry image of a vehicle and analyzes the vehicle's exterior information, movement speed, and direction of travel; an AI analysis edge device that determines a vehicle type based on the vehicle exterior information received from the first camera, calculates an expected location of an identification code on the vehicle using a pre-trained VIN location prediction model, and generates a shooting time and shooting conditions based on the expected location and the vehicle's movement speed; a plurality of VIN recognition cameras that capture an identification code area while the vehicle is moving, under the control of the AI ​​analysis edge device; and a display device that indicates whether the vehicle has passed or stopped according to the identification code recognition result transmitted from the AI ​​analysis edge device.

Description

Technology Field

[0001] The present invention relates to a barcode VIN recognition and operation system for a high-speed moving object and a method for executing the same. More specifically, the invention relates to a barcode VIN recognition and operation system for a high-speed moving object and a method for executing the same, which predicts the location of an identification code using a pre-trained model based on the external shape information of the vehicle, and pre-controls the shooting time and shooting conditions of a VIN recognition camera according to the prediction result, thereby enabling improved identification code recognition accuracy even in a high-speed moving environment. Background Technology

[0003] Recently, in the logistics industry and vehicle access control sectors, technology utilizing VINs or barcodes attached to vehicles or containers to track logistics flow and manage incoming and outgoing vehicles is being widely used. In particular, in environments where large volumes of vehicles and cargo move within a short period, such as airports, seaports, logistics terminals, and customs areas, identification code-based automatic recognition technology is essential.

[0004] However, most conventional barcode recognition technologies rely on a method where an operator directly scans barcodes using a barcode reader. In this method, operators must manually adjust the position, angle, and direction of the barcode reader depending on where the barcode is attached, and there are issues where environmental factors such as light reflection, shadows, and contamination must be resolved based on the operator's skill level. In particular, when barcodes are attached in various locations on a vehicle, the operator must repeatedly change positions to perform scanning, resulting in reduced work efficiency.

[0005] Furthermore, conventional technology generally assumes that the target object is stationary for barcode recognition. Therefore, to recognize barcodes attached to high-speed moving objects, such as vehicles or large containers, an operator must stop the object and approach it to perform the scan. This results in delays in logistics flow, increased vehicle waiting times, and a decrease in overall logistics processing speed.

[0006] In particular, national infrastructure logistics facilities such as airports, ports, and customs areas must handle multiple vehicles and cargo simultaneously within a limited space; consequently, vehicle stopping and manual recognition processes cause bottlenecks and reduce logistics operational efficiency. Furthermore, while continuous vehicle movement is required in such environments, conventional recognition methods fail to support this, necessitating additional waiting space and manpower.

[0007] Furthermore, in manual barcode recognition processes, input errors or omissions caused by operator mistakes occur frequently. For instance, barcodes may not be scanned accurately or incorrect codes may be entered; these errors lead to inconsistencies in logistics data and result in problems that incur additional time and costs by requiring re-verification or reprocessing.

[0008] In addition, the barcode recognition environment is significantly affected by lighting, weather, time of day, and vehicle conditions. In nighttime environments, the recognition rate may decrease due to insufficient lighting, and in backlit or strongly reflected light environments, there is a problem where parts of the barcode are overexposed, making recognition difficult. Furthermore, recognition accuracy is significantly reduced if the barcode is partially obscured or distorted due to contamination, damage to the vehicle surface, or deviation in the attachment position.

[0009] As such, conventional barcode recognition technology suffers from problems such as reliance on manual processes, the premise of recognition in a stationary state, vulnerability to environmental changes, and high error rates; in particular, it faces limitations in achieving automatic recognition in high-speed moving environments. Therefore, there is a need for a new technology that can accurately recognize identification codes while the vehicle is in motion without stopping, minimize operator intervention, and ensure stable recognition under various environmental conditions.

[0010] Accordingly, there is a need to develop technology that can predict the location of identification codes in advance based on vehicle exterior information, automatically control the shooting time and conditions, and accurately recognize the identification codes of high-speed moving vehicles using multiple cameras.

[0011] Korean Registered Patent No. 10-1822655 relates to an object recognition method using a camera and a camera system for the same, Korean Registered Patent No. 10-2331766 relates to a control method for a VIN input system, Korean Registered Patent No. 10-2393068 relates to an image-based part recognition system and method, and Korean Registered Patent No. 10-2060455 relates to a truck number recognition system and method, but no solution for solving the above-mentioned problems is disclosed. Prior art literature

[0013] Korean Registered Patent No. 10-1822655, Korean Registered Patent No. 10-2331766, Korean Registered Patent No. 10-2393068, Korean Registered Patent No. 10-2060455 The problem to be solved

[0014] The present invention aims to provide a barcode VIN recognition and operation system for a high-speed moving object and a method for implementing the same, which can accurately recognize a VIN or barcode attached to a vehicle without stopping the vehicle moving at high speed.

[0015] In addition, the present invention aims to provide a barcode VIN recognition and operation system for a high-speed moving object and a method for executing the same, which can improve the accuracy of identification code recognition even in a high-speed moving environment by predicting the location of an identification code using a pre-trained model based on the external shape information of the vehicle and controlling the shooting time and shooting conditions of a VIN recognition camera in advance according to the prediction result.

[0016] In addition, the present invention aims to provide a barcode VIN recognition and operation system for a high-speed moving object and a method for implementing the same, which provides a recognition structure robust against environmental factors such as reflected light, contamination, angle distortion, and partial occlusion by using a plurality of VIN recognition cameras to photograph identification code areas at different angles and positions and combining a plurality of reading results to derive a final identification code.

[0017] In addition, the present invention aims to provide a barcode VIN recognition and operation system for high-speed moving objects and a method for implementing the same, which can improve operational efficiency in logistics sites or vehicle management environments by determining in real time whether a vehicle passes, stops, or is re-inspected based on the identification code recognition result and outputting this to a display device. means of solving the problem

[0019] A barcode VIN recognition and operation system for a high-speed moving object to achieve this purpose may include: a first camera that detects a vehicle object from an entry video of the vehicle and analyzes the vehicle's exterior information, movement speed, and direction of travel; an AI analysis edge device that determines the vehicle type based on the vehicle exterior information received from the first camera, calculates the expected location of the identification code on the vehicle using a pre-trained VIN location prediction model, and generates a shooting time and shooting conditions based on the expected location and the vehicle's movement speed; a plurality of VIN recognition cameras that capture an identification code area while the vehicle is moving, under the control of the AI ​​analysis edge device; and a display device that indicates whether the vehicle has passed or stopped according to the identification code recognition result transmitted from the AI ​​analysis edge device.

[0020] Furthermore, a method for recognizing and operating a barcode or VIN of a high-speed moving object to achieve this purpose comprises the steps of: acquiring an entry image of a vehicle using a first camera, detecting a vehicle object from the image, and analyzing the vehicle's exterior information, movement speed, and direction of travel; an AI analysis edge device determining the vehicle type based on the vehicle exterior information and calculating the predicted location of an identification code on the vehicle using a pre-trained VIN location prediction model; the AI ​​analysis edge device generating a shooting time and shooting conditions for a plurality of VIN recognition cameras based on the predicted location and vehicle movement speed; capturing an identification code area using a plurality of VIN recognition cameras while the vehicle is moving, under the control of the AI ​​analysis edge device; detecting and reading an identification code from the captured image; and determining whether the vehicle passes or stops based on the reading result and outputting it through a display device. Effects of the invention

[0022] According to the present invention, there is an advantage in that a VIN or barcode attached to a vehicle can be accurately recognized without stopping a vehicle moving at high speed.

[0023] In addition, according to the present invention, by using a pre-trained model based on the exterior information of a vehicle to predict the location of an identification code and by pre-controlling the shooting time and shooting conditions of a VIN recognition camera according to the prediction result, there is an advantage in that the accuracy of identification code recognition can be improved even in a high-speed moving environment.

[0024] In addition, according to the present invention, by using a plurality of VIN recognition cameras to photograph identification code areas at different angles and positions and combining a plurality of reading results to derive a final identification code, there is an advantage of providing a recognition structure robust against environmental factors such as reflected light, contamination, angle distortion, and partial occlusion.

[0025] In addition, according to the present invention, there is an advantage in that operational efficiency in logistics sites or vehicle management environments can be improved by determining in real time whether a vehicle passes, stops, or is re-inspected based on the identification code recognition result and outputting this to a display device. Brief explanation of the drawing

[0027] FIG. 1 is a drawing for explaining a barcode VIN recognition and operation system for a high-speed moving body according to one embodiment of the present invention. FIG. 2 is a block diagram illustrating the internal structure of an AI analysis edge device according to an embodiment of the present invention. FIG. 3 is a flowchart illustrating an embodiment of the method for recognizing and operating a barcode VIN of a high-speed moving body according to the present invention. FIG. 4 is a diagram illustrating the identification code recognition and verification procedure of a high-speed moving vehicle according to one embodiment of the present invention based on actual images. Specific details for implementing the invention

[0028] Preferred embodiments of the present invention will be described below with reference to the attached drawings.

[0029] However, embodiments of the present invention may be modified in various different forms, and the scope of the present invention is not limited to the embodiments described below. Furthermore, embodiments of the present invention are provided to more fully explain the present invention to those with average knowledge in the relevant technical field.

[0030] That is, the aforementioned objectives, features, and advantages will be described in detail below with reference to the attached drawings, and accordingly, a person skilled in the art to which the present invention pertains will be able to easily implement the technical concept of the present invention. In describing the present invention, detailed descriptions of known technologies related to the present invention are omitted if it is determined that such descriptions may unnecessarily obscure the essence of the present invention. Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings. In the drawings, the same reference numerals are used to indicate the same or similar components.

[0031] Additionally, singular expressions used in this specification include plural expressions unless the context clearly indicates otherwise. In this application, terms such as "composed of" or "comprising" should not be interpreted as necessarily including all of the various components or steps described in the specification, and should be interpreted as meaning that some of the components or steps may not be included, or that additional components or steps may be included.

[0032] Various flowcharts are disclosed to explain embodiments of the present invention, but these are for the convenience of explaining each step and do not necessarily mean that each step must be performed in the order of the flowchart. That is, each step in the flowchart may be performed simultaneously, in the order according to the flowchart, or in the reverse order of the flowchart.

[0034] FIG. 1 is a drawing for explaining a barcode VIN recognition and operation system for a high-speed moving body according to one embodiment of the present invention.

[0035] Referring to FIG. 1, a barcode VIN recognition and operation system for a high-speed moving object includes a first camera (100), a VIN recognition camera (200), an AI analysis edge device (300), and a display device (400).

[0036] The first camera (100) is configured to detect in advance whether a vehicle has entered, the speed of movement, the type of vehicle, and the location of the vehicle so that the VIN recognition camera (200) can photograph the VIN or barcode area of ​​the vehicle at an accurate time.

[0037] The first camera (100) is installed to face the vehicle entry section, and acquires continuous video or frame-by-frame vehicle video when a vehicle enters the recognition section. At this time, the first camera (100) functions as a pre-detection camera for determining the shooting conditions of the VIN recognition camera (200), rather than as a camera for directly reading the VIN or barcode.

[0038] The first camera (100) detects a vehicle object in the acquired image and analyzes the position change of the vehicle object between frames to calculate the direction of movement and speed of movement of the vehicle. For example, if a front or side reference point of the vehicle exists at a first position in the first frame and subsequently moves to a second position in the second frame, the first camera (100) can calculate the speed of movement of the vehicle using the amount of change between the two positions and the time difference between frames. Accordingly, it is determined whether the vehicle is entering at a low speed or passing at a high speed, and the calculated speed information is used to set the shooting trigger time and exposure conditions of the VIN recognition camera (200).

[0039] Additionally, the first camera (100) analyzes the exterior shape information of the vehicle from the vehicle image to determine the type of vehicle. The vehicle type can be classified into passenger cars, sports utility vehicles, trucks, buses, special purpose vehicles, etc., and can be determined based on at least one of the vehicle's overall height, overall width, body length, front shape, window position, bonnet height, or chassis area position information. The first camera (100) transmits this vehicle type information to the AI ​​analysis edge device (300), and the AI ​​analysis edge device (300) sets the shooting direction, field of view, focal length, and area of ​​interest of the VIN recognition camera (200) by referring to the expected location of the VIN or barcode for each vehicle type.

[0040] The first camera (100) generates basic data to calculate the scheduled shooting time of the VIN recognition camera (200) based on the vehicle's current location and movement speed. That is, the first camera (100) extracts information on the time when the vehicle passes a specific reference line, the vehicle's reference point coordinates, the vehicle's direction of movement, and speed, and transmits this to the AI ​​analysis edge device (300). The AI ​​analysis edge device (300) uses the above information to predict the time when the vehicle's VIN or barcode area reaches the shooting area of ​​the VIN recognition camera (200), and can control the VIN recognition camera (200) to perform shooting at that time.

[0041] In one embodiment, the first camera (100) can calculate a change in the relative distance between the vehicle and the camera based on a change in the size or position of a vehicle object within the image. If the vehicle object gradually increases in size within the frame, it is determined that the vehicle is approaching the first camera (100), and if the vehicle object moves in a certain direction, the driving direction of the vehicle is determined. This distance and direction information is used as prediction information to secure shooting preparation time before the vehicle reaches the shooting section of the VIN recognition camera (200).

[0042] In addition, when the first camera (100) detects whether a vehicle has entered, it does not simply transmit a shooting start signal to the VIN recognition camera (200), but can also generate different shooting conditions depending on the speed and type of the vehicle.

[0043] For example, if the vehicle speed is faster than the reference speed, the exposure time of the VIN recognition camera (200) can be set to be short and the shutter timing can be controlled to be advanced. Conversely, if the vehicle speed is low, the shooting waiting time can be set to be long or multiple frames can be secured.

[0044] In addition, if the vehicle type is a truck or a bus, the area of ​​interest for shooting can be adjusted upward considering the possibility that the VIN or barcode may be located in a relatively high position, and if it is a passenger car, it can be set to prioritize shooting the downward direction or the area around the front windshield.

[0045] The first camera (100) may detect image quality degradation due to illumination, backlight, shadows, or nighttime environments. In this case, the first camera (100) analyzes the brightness, contrast, and noise levels of the input image to generate shooting environment information, and the AI ​​analysis edge device (300) can adjust the lighting, exposure, gain, or shutter speed of the VIN recognition camera (200) based on this. Accordingly, blurring, overexposure, or underexposure of the VIN or barcode area can be reduced even in high-speed moving vehicles.

[0046] The first camera (100) can generate tracking information of a vehicle object using multiple frames. Specifically, the first camera (100) assigns a tracking identifier to the vehicle object and continuously updates the vehicle position for each frame. At this time, even when multiple vehicles enter consecutively, tracking identifiers are generated separately for each vehicle, thereby clearly specifying which vehicle's VIN or barcode the subsequent VIN recognition camera (200) should photograph. Therefore, even when the distance between vehicles is short or multiple vehicles are moving consecutively in a logistics site, it is possible to prevent confusion in the recognition results for each vehicle.

[0047] When the first camera (100) determines that a vehicle object has entered a pre-set shooting preparation zone, it transmits a shooting preparation signal to the AI ​​analysis edge device (300). The shooting preparation signal may include a temporary tracking number for vehicle identification, a vehicle type, vehicle speed, vehicle direction of travel, an estimated shooting time, and an estimated VIN location information. Based on this, the AI ​​analysis edge device (300) sets the shooting parameters of the VIN recognition camera (200), and when the vehicle reaches the shooting reference position, it transmits a shooting trigger signal to the VIN recognition camera (200).

[0048] Consequently, the first camera (100) functions not as a simple surveillance camera, but as a sensor for prior recognition and shooting control for recognizing identification codes of high-speed moving vehicles. Since the entry, type, speed, and location of the vehicle are analyzed in advance by the first camera (100), the VIN recognition camera (200) can set optimal shooting conditions before the vehicle reaches the shooting area, and accordingly, the recognition rate of the VIN or barcode can be improved without stopping the vehicle.

[0049] A plurality of VIN recognition cameras (200) are configured to overlappingly photograph the VIN or barcode area of ​​a vehicle moving at high speed at different locations, angles, or shooting conditions based on vehicle information previously detected by the first camera (100). The plurality of VIN recognition cameras (200) are not configured such that a single camera simply photographs the vehicle, but rather as an adaptive shooting structure in which at least two cameras operate selectively or sequentially, taking into account the expected VIN location by vehicle type, vehicle speed, vehicle direction of travel, and the possibility of shooting failure.

[0050] In one embodiment, a plurality of VIN recognition cameras (200) may include a first VIN recognition camera (210), a second VIN recognition camera (220), and a third VIN recognition camera (230). The first VIN recognition camera (210) may be positioned to photograph the lower part of the front windshield or the area around the dashboard of a passenger car or a low-floor vehicle, the second VIN recognition camera (220) may be positioned to photograph the expected location of a VIN or barcode of a vehicle with a relatively high body height, such as a sports utility vehicle, a van, or a truck, and the third VIN recognition camera (230) may be positioned to photograph an auxiliary barcode or logistics identification code attached to the side, top, or rear of the vehicle. However, the number and placement location of the plurality of VIN recognition cameras (200) are not limited thereto and may be varied depending on the width of the vehicle passage line, the direction of travel of the vehicle, the VIN display location, and the on-site installation conditions.

[0051] Multiple VIN recognition cameras (200) can each be configured with a global shutter method. Unlike a rolling shutter method, in which each line of the frame is exposed sequentially when photographing a moving vehicle at high speed, the global shutter method ensures that all pixels are exposed at the same time, thereby reducing image distortion and motion blur caused by vehicle movement. Accordingly, the multiple VIN recognition cameras (200) can clearly acquire the characters, lines, spaces, and code patterns of the VIN or barcode even when the vehicle is moving without stopping.

[0052] Multiple VIN recognition cameras (200) are selectively activated based on vehicle type information generated from the first camera (100). For example, if the first camera (100) classifies the vehicle as a passenger car, the AI ​​analysis edge device (300) may prioritize activating the first VIN recognition camera (210) corresponding to the expected VIN location of the passenger car, and maintain the second VIN recognition camera (220) and the third VIN recognition camera (230) in an auxiliary shooting or standby state.

[0053] Conversely, if the vehicle is classified as a truck or a bus, the AI ​​analysis edge device (300) may first activate a second VIN recognition camera (220) suitable for capturing identification codes at a high position, and, if necessary, activate a third VIN recognition camera (230) together.

[0054] At this time, a plurality of VIN recognition cameras (200) set an area of ​​interest based on a VIN location prediction model for each vehicle type. The VIN location prediction model for each vehicle type predicts an area where a VIN or barcode is likely to exist based on at least one of the vehicle's overall height, overall width, body length, front windshield height, bonnet height, chassis location, license plate location, or vehicle exterior contour.

[0055] For example, in the case of a passenger car, the lower part of the windshield or the area around the dashboard on the driver's side can be set as the priority area of ​​interest, and in the case of a truck, the frame area on the side of the vehicle body or the area around the driver's door can be set as the priority area of ​​interest.

[0056] In this way, the multiple VIN recognition cameras (200) do not photograph all areas equally, but select different cameras and shooting areas depending on the type of vehicle, thereby reducing unnecessary image processing and increasing the recognition rate.

[0057] Additionally, the plurality of VIN recognition cameras (200) adjust the shooting trigger time and exposure conditions based on the vehicle speed calculated by the first camera (100). Since the higher the vehicle speed, the shorter the time it takes for the VIN or barcode area to pass through the shooting range, the AI ​​analysis edge device (300) calculates the shooting time of each VIN recognition camera (210, 220, 230) using the vehicle's current location, vehicle speed, and VIN predicted location.

[0058] Specifically, the AI ​​analysis edge device (300) uses the time when the vehicle reference point passes the detection reference line of the first camera (100) and the vehicle speed to predict the time when the expected VIN location reaches the shooting center area of ​​each VIN recognition camera, and outputs a shooting trigger signal at that time.

[0059] This shooting time can be corrected by taking into account the vehicle's direction of travel, the camera installation location, the offset of the expected VIN location, and the system delay time. For example, the AI ​​analysis edge device (300) determines the shooting trigger time based on the distance from the vehicle reference point to the expected VIN location, the distance from the detection reference line of the first camera (100) to the shooting reference line of the VIN recognition camera (200), the vehicle speed, and the device processing delay time. Accordingly, even if the vehicle is moving quickly, an image can be acquired at the moment when the VIN or barcode area is located in the optimal shooting area of ​​the camera.

[0060] Multiple VIN recognition cameras (200) also set shooting conditions differently depending on the vehicle speed. When the vehicle speed is above a reference speed, the multiple VIN recognition cameras (200) can set the exposure time to be short and increase the output of the gain or auxiliary light to suppress motion blur.

[0061] On the other hand, when the vehicle speed is low, the exposure time can be set relatively long or multiple frames can be secured to increase recognition stability. In addition, in environments with varying illumination, such as night, backlight, or tunnel entrances, the lighting output, shutter speed, gain, and gamma correction values ​​can be dynamically adjusted based on the shooting environment information transmitted by the first camera (100) or the image brightness information of the VIN recognition camera (200).

[0062] In one embodiment, a plurality of VIN recognition cameras (200) may operate in a sequential shooting manner. In the sequential shooting manner, a first VIN recognition camera (210) first captures a VIN or barcode expected area, and if the reliability of the shooting result is less than a reference value, a second VIN recognition camera (220) or a third VIN recognition camera (230) provides an auxiliary shooting image.

[0063] For example, if the VIN area in the image of the first VIN recognition camera (210) is partially obscured or the characters are unclear due to reflected light, the AI ​​analysis edge device (300) can re-detect the VIN area using a different angle image of the second VIN recognition camera (220) and determine the final identification code by comparing the reading results of both images.

[0064] In another embodiment, a plurality of VIN recognition cameras (200) may operate in a simultaneous shooting mode. In the simultaneous shooting mode, when the vehicle reaches a shooting reference position, a plurality of VIN recognition cameras (210, 220, 230) simultaneously acquire images, and an AI analysis edge device (300) calculates a reliability score for the recognition result of each camera and selects the recognition result with the highest reliability as the final result.

[0065] In this case, even if the image quality of a specific camera deteriorates due to vehicle shaking, deviation in the code attachment position, reflection on the driver's seat glass, or body contamination, the overall recognition success rate is improved because the identification code can be obtained from the image of another camera.

[0066] Multiple VIN recognition cameras (200) can perform re-shooting or correction logic in the event of recognition failure. Specifically, if the AI ​​analysis edge device (300) determines that the recognition reliability calculated from the image of the first VIN recognition camera (210) is lower than the reference reliability, it performs re-reading using the image of the second VIN recognition camera (220) or the image of the third VIN recognition camera (230).

[0067] In addition, if only some characters are read from multiple images, the final VIN can be restored by combining the partial reading results of each camera according to character position. For example, if the first 10 digits of a 17-digit VIN are clear in the image of the first VIN recognition camera (210) and the last 7 digits are clear in the image of the second VIN recognition camera (220), the AI ​​analysis edge device (300) generates a final VIN string by comparing the reliability of each character position.

[0068] Additionally, multiple VIN recognition cameras (200) can provide basic data for performing correction processing on captured images. Each VIN recognition camera (210, 220, 230) stores the shooting time, camera identification information, camera installation angle, focal length, exposure value, lighting value, and vehicle tracking identifier together as image metadata.

[0069] The AI ​​analysis edge device (300) can perform image tilt correction, perspective correction, brightness correction, barcode line width correction, and character area sharpening processing using the metadata. In particular, when a barcode or VIN is captured at an angle, perspective distortion is corrected using the camera installation angle and the direction of vehicle movement, and the identification code area is converted into a normalized rectangular area and then reading is performed.

[0070] Multiple VIN recognition cameras (200) operate in conjunction with a vehicle-specific tracking identifier. When a vehicle tracking identifier generated by the first camera (100) is transmitted to the multiple VIN recognition cameras (200), each VIN recognition camera (210, 220, 230) assigns a common identifier to the image data indicating that it is an image of the same vehicle. Accordingly, even when multiple vehicles pass in succession at short intervals, the VIN image of the first vehicle and the VIN image of the second vehicle are not confused, and the AI ​​analysis edge device (300) can accurately match multiple images and recognition results on a vehicle-by-vehicle basis.

[0071] Additionally, the multiple VIN recognition cameras (200) can respond even when the direction of travel of the vehicle to be recognized changes or when the vehicle is driven off to the left or right within the lane. For example, when the first camera (100) detects a lateral position deviation of the vehicle, the AI ​​analysis edge device (300) can prioritize activating the VIN recognition camera positioned on the left or the VIN recognition camera positioned on the right in response to the deviation. This allows the VIN or barcode area to be stably captured even if the vehicle does not pass exactly through the center of the shooting section.

[0072] Multiple VIN recognition cameras (200) can independently evaluate the quality of the captured results. For example, each camera calculates the clarity, brightness, contrast, degree of shake, code area size, and code area location of the captured image, and transmits the corresponding quality indicators to the AI ​​analysis edge device (300). The AI ​​analysis edge device (300) selects only images with quality indicators above a threshold value as priority reading targets, and can classify images with low quality indicators as auxiliary images or discarded images. Accordingly, the image processing load can be reduced and reading accuracy can be increased.

[0073] In one embodiment, a plurality of VIN recognition cameras (200) are maintained in a pre-waiting state before a vehicle passes, and when a vehicle entry detection signal from the first camera (100) is received, they are switched to a shooting ready state.

[0074] Subsequently, when the AI ​​analysis edge device (300) selects a specific camera based on the vehicle type, speed, and VIN estimated location, only the selected VIN recognition camera is switched to an active shooting state, and the remaining cameras can be maintained in an auxiliary shooting or low-power standby state. This selective activation structure reduces power consumption and provides the effect of reducing unnecessary video transmission volume in a wireless communication environment.

[0075] Finally, multiple VIN recognition cameras (200) perform vehicle type-specific location prediction, speed-based shooting time correction, global shutter-based motion blur reduction, multiple angle overlapping shooting, reshooting in case of failure, or combination of partial reading results for recognizing the identification code of a high-speed moving vehicle.

[0076] Accordingly, the present invention can improve the success rate of recognizing VIN or barcode without stopping the vehicle, and enables stable identification code reading through a plurality of VIN recognition cameras (200) even if a specific camera fails to shoot, reflected light, vehicle body contamination, vehicle height deviation, or driving position deviation occurs.

[0077] The AI ​​analysis edge device (300) is configured to be linked with the first camera (100) and a plurality of VIN recognition cameras (200) to analyze the vehicle's exterior information, movement speed, expected VIN location, and captured video in real time, control the plurality of VIN recognition cameras (200), and finally produce a recognition result of the VIN or barcode.

[0078] First, the AI ​​analysis edge device (300) receives vehicle entry video or vehicle detection information from the first camera (100). The AI ​​analysis edge device (300) detects a vehicle object in the received video and analyzes the position change, size change, and direction of movement of the vehicle object to determine whether the vehicle has entered the recognition section. At this time, the AI ​​analysis edge device (300) assigns a temporary vehicle tracking identifier to the vehicle object and subsequently generates vehicle-unit tracking information so that the same vehicle can be matched with the video obtained from the multiple VIN recognition cameras (200).

[0079] When a vehicle object is detected, the AI ​​analysis edge device (300) calculates the vehicle's speed of movement using the change in the position of the vehicle reference point between frames. For example, if the front reference point of the vehicle exists at a first position in a specific frame and moves to a second position in a subsequent frame, the AI ​​analysis edge device (300) calculates the vehicle speed using the change in distance between the two positions and the time difference between frames. The vehicle speed calculated in this way is used to adjust the shooting time, exposure time, lighting output, and number of shooting frames of a plurality of VIN recognition cameras (200).

[0080] Additionally, the AI ​​analysis edge device (300) analyzes the vehicle's exterior information based on the vehicle image received from the first camera (100). The exterior information may include at least one of the vehicle's overall height, overall width, body length, front shape, window position, bonnet height, chassis structure, vehicle contour, and license plate position. Based on this exterior information, the AI ​​analysis edge device (300) may classify the vehicle as a passenger car, a sports utility vehicle, a van, a truck, a bus, or a special purpose vehicle.

[0081] The AI ​​analysis edge device (300) calls a vehicle type-specific VIN location prediction model corresponding to the determined vehicle type. The vehicle type-specific VIN location prediction model is a model that has been pre-learned for locations where VIN or barcodes are likely to exist for each vehicle type, and receives vehicle exterior information and vehicle type information as input to output a VIN predicted location, predicted height, predicted direction, and area of ​​interest for shooting.

[0082] For example, in the case of a passenger car, the lower part of the windshield or the area around the dashboard on the driver's side can be output as the expected VIN location, and in the case of a truck, the area around the driver's door, the chassis frame area, or the side barcode attachment area can be output as the expected VIN location.

[0083] The AI ​​analysis edge device (300) selects at least one of a plurality of VIN recognition cameras (200) based on the output value of the VIN location prediction model for each vehicle type. For example, if the predicted VIN location corresponds to the lower part of the vehicle's front windshield, a first VIN recognition camera (210) that captures the area may be selected first, and if the predicted VIN location corresponds to the vehicle's side frame, a second VIN recognition camera (220) that captures the side may be selected first. Additionally, if the vehicle height is greater than or equal to a reference value, an upwardly positioned VIN recognition camera may be selected, and if the vehicle is driving off to the left or right of the lane, a VIN recognition camera in the direction corresponding to the deviation may be selected.

[0084] The AI ​​analysis edge device (300) sets shooting conditions for the selected VIN recognition camera (200). The shooting conditions may include at least one of a shooting angle, field of view, focal length, zoom magnification, region of interest, exposure time, shutter speed, gain, lighting output, and number of shooting frames. At this time, the AI ​​analysis edge device (300) may set the exposure time shorter as the vehicle speed increases, and may reduce motion blur by increasing the lighting output or gain if necessary. Additionally, in nighttime or backlit environments, the lighting conditions may be corrected by analyzing the brightness and contrast of the input image.

[0085] The AI ​​analysis edge device (300) calculates the shooting trigger time using the vehicle's current location, vehicle speed, VIN predicted location, and the installation location of the VIN recognition camera (200). That is, it predicts the time when the VIN predicted location reaches the shooting center area of ​​the selected VIN recognition camera (200) after the vehicle passes the detection reference line of the first camera (100), and outputs a shooting trigger signal at that time. At this time, the AI ​​analysis edge device (300) can correct the shooting trigger time by considering the device processing delay time, wireless communication delay time, and camera response delay time together.

[0086] For example, the AI ​​analysis edge device (300) determines the shooting time using the distance between the detection baseline and the shooting baseline of the VIN recognition camera, the offset distance between the vehicle reference point and the expected VIN location, the vehicle speed, and the system delay time. Accordingly, even if the vehicle is moving at high speed, an image can be acquired at the moment when the VIN or barcode area is located in the optimal shooting area of ​​the camera.

[0087] When images are received from multiple VIN recognition cameras (200), the AI ​​analysis edge device (300) performs preprocessing on each image. The preprocessing may include at least one of brightness correction, contrast correction, noise removal, shake correction, perspective correction, tilt correction, and identification code area normalization. In particular, when a barcode or VIN is captured at an angle, the AI ​​analysis edge device (300) can correct perspective distortion using camera installation angle and vehicle direction information, and convert the identification code area into a normalized rectangular area.

[0088] The AI ​​analysis edge device (300) detects a VIN or barcode region in a preprocessed image. At this time, the AI ​​analysis edge device (300) extracts candidate identification code regions within the image using an object detection model, a character region detection model, or a barcode pattern detection algorithm. In the case of a VIN, candidate regions can be detected based on character arrangement, character spacing, background contrast, and location information within the vehicle, and in the case of a barcode, candidate regions can be detected based on barcode line width, space pattern, boundary direction, and pattern repeatability.

[0089] The AI ​​analysis edge device (300) performs character recognition or barcode decoding on the detected identification code candidate area. In the case of VIN, the AI ​​analysis edge device (300) reads each character using a character recognition model, and in the case of barcode, extracts an internal unique number using a decoding algorithm according to the barcode standard. Additionally, if the VIN consists of a 17-character string, the AI ​​analysis edge device (300) can verify the validity of the reading result by using the number of digits, allowed characters, check rules, or consistency with pre-registered vehicle information.

[0090] The AI ​​analysis edge device (300) calculates recognition reliability for multiple images acquired from multiple VIN recognition cameras (200). Recognition reliability can be calculated based on at least one of image clarity, character contrast, code area size, focus, degree of motion blur, character recognition probability, success of barcode decoding, and degree of agreement with VIN location by vehicle type. The AI ​​analysis edge device (300) may select the image with the highest recognition reliability as the priority result, or calculate the final VIN or barcode value by combining the partial recognition results of multiple images.

[0091] In one embodiment, when the front part of the VIN is clearly read in the image of the first VIN recognition camera (210) and the back part of the VIN is clearly read in the image of the second VIN recognition camera (220), the AI ​​analysis edge device (300) compares the reliability of each character position, selects a character with high reliability at each position, and combines them to generate a final VIN string. Accordingly, even when it is difficult to read in a single image due to reflected light, contamination, angle difference, or partial occlusion, identification code recovery based on multiple images is possible.

[0092] The AI ​​analysis edge device (300) performs failure processing or re-recognition processing when the recognition reliability is below a threshold value. For example, if the result of the initially selected VIN recognition camera falls below the threshold reliability, the AI ​​analysis edge device (300) may switch the image of the auxiliary VIN recognition camera to the analysis target or output an additional shooting trigger to the camera placed in the auxiliary shooting section. Additionally, if the reliability is below the threshold value in all multiple images, the vehicle is classified as a re-inspection target and a re-inspection guidance signal is output to the display device (400).

[0093] The AI ​​analysis edge device (300) stores the finally read VIN or barcode value by matching it with a vehicle tracking identifier. At this time, the stored information may include the time of vehicle passage, vehicle type, vehicle speed, selected VIN recognition camera information, captured image identifier, reading result, recognition reliability, and success / failure status. This information can be used for operational history management, logistics vehicle control, shipment vehicle verification, entry / exit vehicle management, or post-verification data.

[0094] The AI ​​analysis edge device (300) transmits the reading result to the display device (400). If recognition is successful, it controls the display device (400) to output a passable, normal recognition, or green display signal, and if recognition fails, it controls the display device to output a stop, retake, re-examination, or red display signal. Additionally, in the recognition waiting state, it controls the display device to output a yellow display or a waiting message.

[0095] The display device (400) is configured to provide status information visually and intuitively to a vehicle driver or field worker based on VIN or barcode recognition results transmitted from an AI analysis edge device (300), and to provide real-time guidance on whether the vehicle is moving, stopped, or re-inspected.

[0096] The display device (400) may include at least one of an LED display, an electronic display board, a lamp in the form of a traffic light, or a warning light, and is installed in front of the vehicle's direction of travel or in a location easily recognizable by the driver. The display device (400) is not limited to merely outputting recognition results, but also performs the role of generating an operation signal to control the flow of vehicles.

[0097] First, the display device (400) receives recognition result data along with a vehicle tracking identifier from the AI ​​analysis edge device (300). At this time, the recognition result data may include a VIN or barcode reading value, recognition success status, recognition reliability, vehicle type, whether re-inspection is required, and vehicle status information. The display device (400) determines an output state based on the recognition success status and status information among the received data.

[0098] In one embodiment, the display device (400) outputs different visual signals depending on the recognition result. For example, if the VIN or barcode is successfully recognized, a green display signal may be output to indicate that the vehicle can continue to proceed, and if recognition fails or the recognition reliability is below a reference value, a red display signal may be output to induce the vehicle to stop or re-inspect. Additionally, if the vehicle enters a recognition section and analysis is in progress, a yellow display signal or a waiting message may be output to inform the driver of the current status.

[0099] The display device (400) can output not only simple color output but also text information or graphic information. For example, it can display messages such as “Recognition complete,” “Pass,” “Re-inspection required,” or “Stop and wait,” and, if necessary, display part of the vehicle number or part of the VIN so that the driver or operator can identify the vehicle to be recognized. In addition, it can guide the vehicle’s movement path through arrows, direction indicator icons, or lane guidance indicators.

[0100] The display device (400) can be configured to enable individual control for each vehicle. That is, by outputting display information corresponding to a specific vehicle based on a vehicle tracking identifier generated by the first camera (100) and the AI ​​analysis edge device (300), the vehicle recognition results are not confused even when multiple vehicles pass by in succession. For example, if multiple lanes exist, the display device for each lane can be controlled independently, or individual results can be displayed in an area corresponding to the vehicle location within a single display device.

[0101] Additionally, the display device (400) can generate a signal to control the subsequent operation of the vehicle according to the recognition result. For example, when recognition is successful, it can output a control signal to open the barrier or allow the vehicle to pass, and when recognition fails, it can keep the barrier closed or generate a warning sound to stop the vehicle. At this time, the display device (400) can be linked with a separate gate control device, barrier control system, or logistics operation system.

[0102] In one embodiment, the display device (400) can subdivide the output form based on the recognition reliability received from the AI ​​analysis edge device (300). For example, if the recognition reliability is very high, it may output a “normal pass” signal; if the reliability is at an intermediate level, it may output a “caution pass” or “speed reduction” message; and if the reliability is low, it may output a “stop and re-inspection” message. Accordingly, multi-stage operational control rather than simple binary judgment is possible.

[0103] Additionally, the display device (400) can perform a guidance function to guide vehicles that fail to be recognized to a re-shooting section or a waiting section. For example, when recognition fails, the vehicle flow can be efficiently controlled by outputting a direction indicator along with a message such as “Move to left waiting section” or “Enter re-shooting section.” In this way, the display device (400) goes beyond being a simple output device and serves as an interface that controls the flow of logistics or vehicle operations.

[0104] The display device (400) can automatically adjust the output brightness and display method in response to the external environment. For example, during the day, high-brightness LED output can be used to ensure visibility, and at night, the brightness can be lowered or the color contrast adjusted to prevent glare. Additionally, the color combination, flashing pattern, or warning signal can be changed to maintain a recognizable display even in environmental conditions such as rain, fog, or dust.

[0105] The display device (400) can be connected to the AI ​​analysis edge device (300) via wireless or wired communication and is configured to enable real-time data reception and output. In case of communication delay or data error, the display device (400) may be configured to maintain a default state or a safe state. For example, it may be configured to automatically maintain a “stop” or “standby” state if a recognition result is not received for a certain period of time or longer.

[0106] Additionally, the display device (400) can store output history. The output history may include a vehicle identifier, output time, output status, recognition result, and operation message, which can be used as data for post-analysis, operational improvement, or dispute response.

[0108] FIG. 2 is a block diagram illustrating the internal structure of an AI analysis edge device according to an embodiment of the present invention.

[0109] Referring to FIG. 2, the AI ​​analysis edge device (300) includes a data receiving unit (310), a vehicle analysis unit (320), a VIN location prediction unit (330), a shooting control unit (340), an image preprocessing unit (350), an identification code detection unit (360), a code reading unit (370), an AI learning unit (380), a result judgment unit (390), and an output control unit (395).

[0110] The data receiving unit (310) is configured to receive vehicle-related image data and control information from the first camera (100), a plurality of VIN recognition cameras (200), and an external control signal source, and to perform an input interface function for transmitting this to an internal processing module.

[0111] The data receiving unit (310) receives a vehicle detection image or a vehicle entry signal from the first camera (100) when the vehicle enters the recognition section. At this time, the first camera (100) transmits a continuous frame image of the vehicle or an event-based detection signal to the data receiving unit (310), and the data receiving unit (310) receives this in real time and transmits it to the vehicle analysis unit (320). The image data may include a frame image, time information, and camera identification information.

[0112] Additionally, the data receiving unit (310) receives image data captured from a plurality of VIN recognition cameras (200). The plurality of VIN recognition cameras (200) generate an image including a VIN or barcode area of ​​a vehicle according to a shooting trigger, and the data receiving unit (310) receives image data transmitted individually from each camera and manages it together with a camera identifier, a shooting time, and a vehicle tracking identifier. Accordingly, it is possible to distinguish whether the images received from the plurality of cameras correspond to the same vehicle.

[0113] The data receiving unit (310) can receive metadata along with the image data. The metadata may include at least one of a shooting time, camera position, camera angle, focal length, exposure time, shutter speed, lighting output, and vehicle tracking identifier. The data receiving unit (310) stores this metadata in conjunction with the image data and transmits it so that it can be utilized by the image preprocessing unit (350) and the identification code detection unit (360).

[0114] The data receiving unit (310) groups data received from the first camera (100) and a plurality of VIN recognition cameras (200) by vehicle unit. To this end, the data receiving unit (310) matches image data received from each VIN recognition camera (200) based on a vehicle tracking identifier generated by the first camera (100) to form a data set for the same vehicle. Such vehicle-unit data grouping contributes to preventing data interference between vehicles even in an environment where multiple vehicles enter continuously at short intervals.

[0115] The data receiving unit (310) can perform synchronization of the received data. That is, if the time reference between the first camera (100) and the plurality of VIN recognition cameras (200) is different, the data receiving unit (310) performs time alignment between frames based on the shooting time information so that the data is processed consistently according to the movement flow of the vehicle. In addition, if network delay or packet loss occurs, the data receiving unit (310) can maintain the continuity of the data through buffering or a retransmission request.

[0116] In one embodiment, the data receiving unit (310) may operate in a high-speed data receiving mode in consideration of cases where the vehicle's movement speed is fast. In this case, the data receiving unit (310) may apply a parallel receiving structure or a multiple buffer structure to process high-resolution images transmitted from a plurality of VIN recognition cameras (200) without delay. For example, image loss can be prevented by allocating an independent receiving buffer for each camera and sequentially loading the received image data into a processing queue.

[0117] Additionally, the data receiving unit (310) can verify the validity of the received data. For example, it can check the resolution, frame size, data corruption, and metadata inclusion of the video data, and if the criteria conditions are not met, it can discard the data or request re-reception. Accordingly, errors that may occur during subsequent processing can be prevented in advance.

[0118] The data receiving unit (310) may receive additional control information from an external system or a higher-level control system. The control information may include vehicle processing policies, recognition thresholds, shooting priorities, or operating mode settings, and the data receiving unit (310) may transmit this to an internal module to be used for controlling system operation.

[0119] The vehicle analysis unit (320) is configured to detect a vehicle object based on image data from the first camera (100) received from the data receiving unit (310), and to determine the vehicle type by analyzing the vehicle's exterior information, movement speed, and direction of travel. Additionally, the vehicle analysis result is generated to be used as an input value for the VIN location prediction unit (330) and the shooting control unit (340).

[0120] First, the vehicle analysis unit (320) detects vehicle objects in the input image. To do this, it identifies vehicle regions within the image using a pre-trained vehicle object detection model and generates bounding box or mask information for the vehicle regions. If multiple vehicles exist, the vehicle analysis unit (320) distinguishes each vehicle object individually and assigns a unique tracking identifier to each vehicle.

[0121] The vehicle analysis unit (320) extracts external shape information of the vehicle based on the detected vehicle object. The external shape information may include at least one of the vehicle's overall height, overall width, body length, front shape, window position, bonnet height, chassis structure, and vehicle contour. To extract such external shape information, the vehicle analysis unit (320) may use vehicle contour detection, feature point extraction, or deep learning-based feature vector extraction techniques.

[0122] The vehicle analysis unit (320) inputs the extracted exterior information into a pre-trained vehicle type classification model to determine the type of vehicle. The vehicle type classification model is a model trained to classify a vehicle into any one of a passenger car, a sports utility vehicle, a van, a truck, a bus, or a special purpose vehicle. The vehicle analysis unit (320) outputs the classification result as vehicle type information and transmits the information to the VIN location prediction unit (330).

[0123] Additionally, the vehicle analysis unit (320) calculates the vehicle's speed of movement using the change in position of the vehicle object between consecutive frames. Specifically, when the reference point of the vehicle object is at the first position in the first frame and moves to the second position in the second frame, the vehicle analysis unit (320) calculates the vehicle speed using the change in distance between the two positions and the time difference between frames. At this time, camera correction information or a scale conversion coefficient may be used so that the distance of movement in the image coordinate system can be converted into an actual distance.

[0124] The vehicle analysis unit (320) can determine the direction of movement of the vehicle. For example, if the vehicle object moves from left to right within the image, it can be determined as the forward direction, and if it moves in the opposite direction, it can be determined as the reverse direction. In addition, if the vehicle moves off to the left or right within the lane, the vehicle analysis unit (320) can calculate the lateral position deviation to estimate the actual driving path of the vehicle. This information on the direction of movement and position deviation is used for the selection and shooting area setting of the VIN recognition camera (200).

[0125] The vehicle analysis unit (320) can estimate relative distance information between the vehicle and the camera. For example, if the size of the vehicle object gradually increases within the frame, it can be determined that the vehicle is approaching the camera, and if the size of the vehicle object decreases, it can be determined that the vehicle is moving away. Additionally, changes in the vehicle's distance can be calculated more precisely by utilizing the vehicle's bounding box size, position, and perspective changes. This distance information is used to calculate the shooting trigger point.

[0126] In one embodiment, the vehicle analysis unit (320) can generate vehicle state information by combining the vehicle's external shape information and movement speed. For example, it can generate composite state information such as "a large truck traveling at high speed" or "a passenger car traveling at low speed," which is used to calculate a more precise VIN location in the VIN location prediction unit (330).

[0127] Additionally, the vehicle analysis unit (320) performs a vehicle tracking function. When the same vehicle is continuously detected across multiple frames, the vehicle analysis unit (320) continuously updates the vehicle's location, speed, and direction information while maintaining the same tracking identifier for the vehicle. This allows the information of each vehicle to be managed independently even in an environment where multiple vehicles enter continuously.

[0128] The vehicle analysis unit (320) can analyze image environment information together. For example, it can determine the shooting environment status by analyzing the brightness, contrast, shadows, backlighting, and noise level of the image. This environment information can be used to set exposure, gain, and lighting conditions in the shooting control unit (340).

[0129] Additionally, the vehicle analysis unit (320) can estimate the location of specific parts of the vehicle. For example, it can define major structural locations within the vehicle by estimating the location of the vehicle's front windshield, bonnet, license plate, or chassis. This location information is used as reference information to more accurately calculate the predicted VIN location in the VIN location prediction unit (330).

[0130] The VIN location prediction unit (330) takes vehicle type information, exterior information, and movement status information transmitted from the vehicle analysis unit (320) as input, calculates a location where there is a high probability that a VIN or barcode exists on the vehicle based on a pre-trained VIN location prediction model, and provides this to the shooting control unit (340).

[0131] First, the VIN location prediction unit (330) receives vehicle type, vehicle exterior features, vehicle direction of travel, vehicle speed, and vehicle location information from the vehicle analysis unit (320). The vehicle exterior features may include the vehicle's overall height, overall width, body length, front windshield position, bonnet height, chassis structure, vehicle contour, and specific structural reference point position. The VIN location prediction unit (330) organizes this information into a single feature vector and uses it as an input value for the VIN location prediction model.

[0132] The VIN location prediction unit (330) calculates the expected location of a VIN or barcode for each vehicle type using a pre-trained VIN location prediction model. The VIN location prediction model is a model that has learned the location patterns where VINs are attached for each type of vehicle, and outputs an area within the vehicle coordinate system where there is a high probability that a VIN or barcode exists according to the input vehicle exterior information and vehicle type.

[0133] For example, in the case of a passenger car, the lower part of the windshield or the dashboard area on the driver's side can be printed; in the case of a cargo truck, the area around the driver's door or the body frame area can be printed; and in the case of a bus or special purpose vehicle, the side or top area of ​​the vehicle can be printed.

[0134] The VIN location prediction unit (330) can calculate multiple candidate locations along with probability values, rather than outputting a single location. That is, a probability score indicating the possibility of a VIN or barcode existing is assigned to each candidate location, and priorities are set in order of highest probability. Through this, the shooting control unit (340) can control the VIN recognition camera (200) centered on the location with the highest priority, and can set it to shoot multiple locations sequentially if necessary.

[0135] Additionally, the VIN position prediction unit (330) dynamically corrects the VIN position by reflecting the vehicle's movement state. For example, if the vehicle moves to the left or right of the lane center, the VIN position prediction unit (330) can correct the predicted VIN position to the left or right by reflecting the vehicle's lateral position deviation. Additionally, if the vehicle is driving in a tilted state or the vehicle's attitude changes depending on the road surface condition, the VIN position can be corrected by reflecting changes in the vehicle contour and reference point.

[0136] The VIN position prediction unit (330) converts the VIN position into an actual shooting coordinate system by considering the relative positional relationship between the vehicle and the camera. That is, the VIN position calculated in the vehicle coordinate system is converted into a camera coordinate system and converted into an expected coordinate or region of interest on the image frame of each VIN recognition camera (200). In this process, the camera's installation position, installation angle, focal length, and perspective transformation parameters may be reflected.

[0137] Additionally, the VIN position prediction unit (330) performs position prediction including the time axis using vehicle speed information. That is, when the vehicle is moving at a certain speed, it predicts not only the VIN position at the current time but also the VIN position after a certain time, so that the shooting control unit (340) can prepare the shooting time in advance. For example, by using the vehicle speed and the current position, it can calculate the time when the predicted VIN position reaches the shooting center area of ​​a specific VIN recognition camera and support shooting to be performed at that time.

[0138] In one embodiment, the VIN location prediction unit (330) may use a correction model in addition to a VIN location prediction model for each vehicle type. For example, a correction model reflecting vehicle speed, vehicle location deviation, shooting environment conditions, and past recognition success / failure history may be applied to the basic VIN location prediction result to calculate the final VIN location. This allows for improved accuracy in VIN location prediction even under various environmental conditions.

[0139] Additionally, the VIN location prediction unit (330) can reflect shooting environment information. For example, if the recognition rate of a specific area is reduced in backlight, night, or shadow environments, the VIN location prediction unit (330) can adjust the output result by lowering the priority of that area and raising the priority of other candidate areas.

[0140] When the VIN position prediction unit (330) transmits the output result to the shooting control unit (340), it may provide not only simple coordinate information but also additional information necessary for shooting. The additional information may include the size of the region of interest, the expected code direction, the code tilt, the code size, the shooting priority, and the expected recognition difficulty. The shooting control unit (340) can use this to more precisely set the shooting conditions of the VIN recognition camera (200).

[0141] Additionally, the VIN position prediction unit (330) can generate an optimal shooting area for each of the multiple VIN recognition cameras (200). For example, it can provide an optimized shooting area for each camera by generating a region of interest centered on the front area for the first VIN recognition camera (210) and a region of interest centered on the side area for the second VIN recognition camera (220).

[0142] The shooting control unit (340) is configured to determine and control the shooting time and shooting conditions of a plurality of VIN recognition cameras (200) based on VIN predicted location information transmitted from the VIN location prediction unit (330) and vehicle speed, location, and direction of travel information calculated by the vehicle analysis unit (320). The shooting control unit (340) goes beyond simply transmitting a shooting trigger and integrally controls time, space, and camera parameters so that the VIN or barcode area is captured at an optimal location even in an environment where the vehicle is moving at high speed.

[0143] First, the shooting control unit (340) receives the predicted location of the VIN or barcode on the vehicle (coordinates of the area of ​​interest), shooting priority information, and candidate shooting area information per camera from the VIN location prediction unit (330). At the same time, it receives the vehicle speed, vehicle direction of travel, vehicle location, and vehicle tracking identifier from the vehicle analysis unit (320). The shooting control unit (340) combines these information to generate an optimal shooting strategy for each VIN recognition camera (200).

[0144] The shooting control unit (340) selects a target camera from among a plurality of VIN recognition cameras (200). At this time, considering the priority of candidate locations calculated by the VIN location prediction unit (330), the installation location, field of view, and current vehicle location of each camera, the camera with the highest recognition probability is selected as the primary shooting camera, and the remaining cameras are set as auxiliary shooting cameras. Additionally, if the vehicle type changes or the vehicle location deviates, the target camera can be dynamically changed.

[0145] The shooting control unit (340) calculates the shooting time by considering the vehicle's moving speed. When the vehicle is moving at high speed, the time it takes for the VIN area to pass through the camera's shooting area is very short, so accurate timing control is required. To this end, the shooting control unit (340) calculates the shooting time using the distance between the vehicle reference point and the expected VIN position, the distance from the camera shooting centerline, and the vehicle speed.

[0146] In one embodiment, the shooting control unit (340) can calculate the shooting trigger time using the following relationship.

[0147] The shooting time is defined as the time from when the vehicle reference point passes the detection reference line of the first camera (100) until the VIN estimated position reaches the shooting center area of ​​a specific VIN recognition camera (200). This can be expressed as a formula as [Equation 1].

[0149] [Mathematical Formula 1]

[0150]

[0151] Tshoot: Time to the shooting trigger,

[0152] D cam : Distance from the detection baseline to the shooting centerline of the VIN recognition camera,

[0153] D vin : Distance from the vehicle reference point to the estimated VIN location,

[0154] V vehicle : Vehicle's movement speed

[0156] The shooting control unit (340) generates a shooting trigger signal based on the time value calculated by the above formula and controls the VIN recognition camera (200) to perform shooting at that time. In addition, the actual shooting time can be corrected by reflecting the system delay time (communication delay, camera response delay, etc.) as a correction value.

[0157] The shooting control unit (340) sets not only the shooting time but also the shooting conditions. The shooting conditions include at least one of the angle of view, focal length, zoom magnification, exposure time, shutter speed, gain, and light output. For example, as the vehicle speed increases, the exposure time may be set shorter, and the light output or gain may be increased to compensate for the insufficient amount of light. Conversely, when the vehicle speed is low, the exposure time may be set relatively longer or a continuous shooting mode may be set to secure multiple frames.

[0158] Additionally, the shooting control unit (340) adjusts the shooting area by reflecting the position deviation of the vehicle. If the vehicle deviates from the center of the lane and drives to the left or right, the shooting control unit (340) can correct the actual shooting area of ​​the camera by moving the area of ​​interest calculated by the VIN position prediction unit (330) in that direction. Through this, the VIN or barcode area can be stably captured even if the vehicle does not accurately follow a certain path.

[0159] The shooting control unit (340) can control multiple VIN recognition cameras (200) sequentially or in parallel. In the sequential control method, the primary shooting camera performs shooting first, and if the recognition result falls short of the standard reliability, the auxiliary camera is activated sequentially. On the other hand, in the parallel control method, multiple cameras are controlled to perform shooting simultaneously at the same time or at close times, thereby enabling the acquisition of images from various angles. The shooting control unit (340) can select one of sequential control and parallel control or use a combination thereof depending on the vehicle speed, vehicle type, and recognition environment.

[0160] The shooting control unit (340) can perform a reshoot control function in preparation for a shooting failure situation. For example, if the initial shooting result is determined to be a failure, the shooting control unit (340) can activate an auxiliary camera by creating an additional shooting trigger while the vehicle is still within the shooting range. Additionally, if the vehicle has moved a certain distance or more, it can control the reshoot to be performed using a camera placed in the reshoot section or the subsequent shooting section.

[0161] The shooting control unit (340) performs synchronization of the shooting operation. That is, when multiple VIN recognition cameras (200) shoot the same vehicle, the shooting timing is aligned based on the vehicle tracking identifier, and the images of the same vehicle are controlled so that they can be compared on the same time axis. Through this, consistency between images can be ensured during the subsequent result fusion process.

[0162] Additionally, the shooting control unit (340) can dynamically change the control strategy according to the shooting environment. For example, in night or low-light environments, increasing the lighting output and adjusting the exposure time may be applied first, and in backlit environments, HDR shooting or multiple exposure shooting may be applied. Additionally, in rain or fog environments, shooting parameters for contrast correction and image quality improvement may be applied.

[0163] The image preprocessing unit (350) receives image data from a plurality of VIN recognition cameras (200) received from the data receiving unit (310) and performs the function of correcting the geometric structure of the image, removing noise, optimizing contrast and brightness, and converting it into a form suitable for input to an artificial intelligence model in order to improve recognition accuracy in the identification code detection unit (360) and the code reading unit (370).

[0164] The image preprocessing unit (350) first extracts an area in the input image where there is a high probability that a VIN or barcode exists based on the area of ​​interest information transmitted from the VIN location prediction unit (330). At this time, instead of using the entire frame as is, the image preprocessing unit (350) crops the image data around the area of ​​interest to remove unnecessary background areas and reduces the processing target area, thereby improving computational efficiency and minimizing noise ingress.

[0165] Subsequently, the image preprocessing unit (350) performs geometric correction of the image. Since multiple VIN recognition cameras (200) are installed at various angles and positions, the VIN area may be distorted when captured due to the tilt of the vehicle, the camera installation angle, and perspective effects. The image preprocessing unit (350) performs perspective transformation using pre-set camera correction parameters or vehicle exterior reference points, and converts the tilted VIN area into a form that facilitates character recognition by aligning it horizontally or in the forward direction.

[0166] Additionally, the image preprocessing unit (350) performs noise removal and contrast enhancement processing to improve image quality. Since the image may contain sensor noise, low-light noise, reflected light, and shadows depending on the shooting environment, the image preprocessing unit (350) removes high-frequency noise by applying a moving average filter, a Gaussian filter, or a low-pass filter. Additionally, histogram equalization or an adaptive contrast enhancement technique is applied to increase the contrast between the VIN character area and the background.

[0167] In one embodiment, the image preprocessing unit (350) can perform an operation such as [Equation 2] to normalize the pixel value distribution of the input image.

[0169] [Mathematical Formula 2]

[0170]

[0171] I(x,y) is the pixel value of the input image,

[0172] μ is the average brightness of the entire image,

[0173] σ is the standard deviation of the brightness value,

[0174] Inorm(x,y): Normalized output image,

[0176] The image preprocessing unit (350) can stably maintain the performance of the artificial intelligence-based identification code recognition model by removing brightness deviations of the image through the normalization process and maintaining a constant input distribution even in various shooting environments.

[0177] Additionally, the image preprocessing unit (350) can perform binarization processing and boundary enhancement processing to facilitate the detection of character or barcode patterns. Specifically, the image preprocessing unit (350) uses a threshold-based binarization or adaptive binarization technique to emphasize character regions and remove background regions, and enhances the linear structure of character outlines or barcodes through an edge detection algorithm.

[0178] The image preprocessing unit (350) performs resolution and scale normalization. Since the size of the images received from multiple VIN recognition cameras (200) may vary depending on the shooting distance and camera resolution, the image preprocessing unit (350) resizes the size of the images to a preset standard resolution and normalizes them to maintain the same input size. Through this, consistency of input data in the identification code detection unit (360) and the code reading unit (370) can be ensured.

[0179] Additionally, the image preprocessing unit (350) can correct image blur according to the vehicle's movement speed. Motion blur may occur when the vehicle moves at high speed, and the image preprocessing unit (350) corrects the image to a level where character recognition is possible by applying a blur correction or clarity enhancement algorithm using the difference between multiple frames.

[0180] The image preprocessing unit (350) converts the preprocessed image into an artificial intelligence model input form. For example, it normalizes pixel values ​​to a certain range, reconstructs channel information, converts it into a tensor form, and transmits it to the identification code detection unit (360) or the code reading unit (370).

[0181] The identification code detection unit (360) receives preprocessed image data transmitted from the image preprocessing unit (350), detects areas within the image where VIN or barcodes exist, and extracts the area as a candidate area to be transmitted to the code reading unit (370). The identification code detection unit (360) performs a function to improve detection accuracy by recognizing the characteristics of a character array structure or a barcode pattern based on a pre-trained identification code detection model, rather than simple pattern search.

[0182] First, the identification code detection unit (360) searches for identification code candidate regions based on the region of interest in the input image. At this time, by utilizing the region of interest information and geometric correction results transmitted from the image preprocessing unit (350), it prioritizes the analysis of regions where there is a high probability that a VIN or barcode exists. By performing detection within a limited area rather than the entire image, computational efficiency is improved and false detections are reduced.

[0183] The identification code detection unit (360) detects a character area or a barcode pattern area within an image using a pre-trained object detection model. The object detection model is a model that has learned the characteristics of a VIN string (a constant character spacing, a linear arrangement structure, a specific length pattern, etc.) or the linear pattern of a barcode (a parallel line structure, a repeating width pattern, etc.), and outputs a location in the input image where the identification code is likely to exist in the form of a bounding box.

[0184] Additionally, the identification code detection unit (360) can generate multiple candidate regions. That is, instead of outputting only a single candidate region, it detects multiple candidate regions of different locations or different sizes and calculates the detection reliability for each candidate region. Through this, the code reading unit (370) can select the optimal region among the multiple candidate regions or read multiple regions in parallel to derive a final result.

[0185] The identification code detection unit (360) can perform character-based VIN detection and barcode-based detection in parallel. That is, by simultaneously applying a character region detection model and a barcode pattern detection algorithm to the input image, it can handle both cases where a VIN string exists and cases where a barcode exists. For example, depending on the vehicle, if the VIN is displayed in the form of characters, a character detection model is used, and if a barcode is attached, such as in a logistics vehicle, a barcode pattern detection algorithm is used to generate respective candidate regions. The identification code detection unit (360) can evaluate the consistency of the detected candidate area. For example, in the case of a VIN string, it has a certain length (e.g., 17 digits) and a character array form, so it can be checked whether the detected candidate area satisfies these structural conditions. In addition, in the case of a barcode, the validity of the candidate area can be determined by analyzing the consistency of the line spacing, line width, and repeating pattern. Through this consistency evaluation, incorrectly detected areas can be removed, and only the area containing the actual identification code can be selected.

[0186] The identification code detection unit (360) can independently perform detection on images received from multiple VIN recognition cameras (200), and can compare and integrate candidate regions detected in multiple images of the same vehicle. For example, if the location and shape of a candidate region detected by the first VIN recognition camera and a candidate region detected by the second VIN recognition camera are compared and determined to be the same code region, this can be generated as a single integrated candidate region.

[0187] Additionally, the identification code detection unit (360) can utilize information between consecutive frames as the vehicle moves. When the same VIN area is captured across multiple frames while the vehicle is moving, the identification code detection unit (360) can improve detection speed and increase detection stability by setting the location of the candidate area detected in the previous frame as the initial search area of ​​the current frame.

[0188] In one embodiment, the identification code detection unit (360) can calculate a feature score for the detected candidate region. The feature score is calculated based on at least one of character contrast, clarity, pattern consistency, code region size, and background contrast, and the code reading unit (370) can determine the reading priority based on the feature score.

[0189] Additionally, the identification code detection unit (360) can also calculate direction information of the detected candidate region. For example, if the VIN string is tilted, the tilt angle of the character array is calculated and transmitted to the code reading unit (370), thereby supporting rotation correction to be performed during the reading process.

[0190] The identification code detection unit (360) converts the detected candidate region into a normalized form and transmits it to the code reading unit (370). That is, the detected region is resized into an image of a certain size, and rotation and geometric correction are applied as needed to convert it into a standardized form that the reading model can use as input.

[0191] The code reading unit (370) receives a plurality of identification code candidate regions transmitted from the identification code detection unit (360), reads the VIN or barcode included in each candidate region, and evaluates the validity and reliability of the reading result to determine the final identification code. The code reading unit (370) operates based on a pre-trained character recognition model or barcode decoding model and integrally processes the reading results obtained from a plurality of candidate regions and a plurality of camera images.

[0192] First, the code reading unit (370) receives a candidate region image transmitted from the identification code detection unit (360). At this time, each candidate region is transmitted in a preprocessed state having a normalized size, orientation, and resolution, and the code reading unit (370) converts it to fit the input format of an artificial intelligence-based reading model.

[0193] The code reading unit (370) reads the VIN string using a pre-trained character recognition model. The character recognition model may have a structure that predicts character candidates for each character position and outputs a probability value for each candidate. For example, if the VIN consists of 17 digits, the code reading unit (370) calculates a probability distribution for a set of possible characters for each digit and selects the character with the highest probability to generate the VIN string.

[0194] Additionally, the code reading unit (370) extracts internal data using a barcode decoding algorithm when a barcode is detected. In the case of a barcode, the line width and space pattern are analyzed and converted into a binary code, and then a decoding rule corresponding to the code system is applied to calculate an identification code value.

[0195] The code reading unit (370) calculates the reliability of the reading result. At this time, the reliability is calculated by comprehensively reflecting the prediction probability for each character position, the consistency of the entire string, whether the code length condition is satisfied, whether the allowed character rule is satisfied, and the consistency with the vehicle type information. In one embodiment, the code reading unit (370) can calculate the reliability of the entire VIN string as [Equation 3].

[0197] [Mathematical Formula 3]

[0198]

[0199] S is the confidence level of the entire VIN string,

[0200] N is the total number of characters in VIN,

[0201] p i : Predicted probability for the character selected at the i-th character position,

[0203] The code reading unit (370) determines the validity of the reading result by comparing the reliability S with a reference value. For example, if the reliability is greater than or equal to the reference value, it is determined to be a successful reading, and if it is less than the reference value, it can be classified as a reading failure or a subject for re-examination.

[0204] Additionally, the code reading unit (370) can compare and combine reading results obtained from multiple candidate regions or multiple camera images. For example, a final VIN string can be generated by combining some characters of the VIN read from the first candidate region and the remaining characters of the VIN read from the second candidate region. At this time, the combination is performed by comparing the reliability of each character position and selecting the character with the highest reliability.

[0205] The code reading unit (370) can verify the structural validity of the VIN string. For example, since the VIN has a specific length and character rules, the code reading unit (370) can determine whether the read string conforms to the specifications, review characters that may be errors, or select replacement candidates. Additionally, if necessary, the accuracy of the reading result can be verified by comparing it with a pre-registered vehicle database.

[0206] In one embodiment, the code reading unit (370) may perform re-reading when there is a high uncertainty in the reading result. For example, if the probability value at a specific character location is below a reference, re-reading may be performed for that area using another candidate area or other camera image data.

[0207] The AI ​​learning unit (380) is configured to perform the function of pre-training the artificial intelligence model used in the vehicle analysis unit (320), VIN location prediction unit (330), and code reading unit (370), or updating the model based on data collected during the operation process.

[0208] The AI ​​learning unit (380) collects learning data including vehicle image data, vehicle exterior information, VIN location information, and identification code reading results from the data receiving unit (310) and the code reading unit (370). At this time, the learning data may include vehicle type, vehicle exterior features, VIN location prediction results, captured images, read VIN values, and whether reading was successful.

[0209] The AI ​​learning unit (380) learns a vehicle type classification model, a VIN location prediction model, and an identification code recognition model based on the above learning data. Specifically, it creates a vehicle type classification model by learning the relationship between vehicle exterior information and actual vehicle type, and creates a VIN location prediction model by learning the relationship between vehicle type and exterior information and actual VIN location.

[0210] Additionally, the AI ​​learning unit (380) improves the performance of the recognition model by comparing the reading results performed by the identification code detection unit (360) and the code reading unit (370) with the actual correct answer data. At this time, if the reading result is accurate, the corresponding data is used as correct answer training data, and if the reading result is an error, the error pattern is analyzed and reflected in the model correction.

[0211] In one embodiment, the AI ​​learning unit (380) can update model parameters to minimize the error between the reading result and the actual VIN. For example, the AI ​​learning unit (380) can train the model using a loss function such as [Equation 4].

[0213] [Mathematical Formula 4]

[0214]

[0215] L is the learning loss value,

[0216] N is the number of training data,

[0217] yi is the actual identification code value or correct label,

[0218] is the identification code value or prediction result predicted by the model,

[0220] The AI ​​learning unit (380) can improve vehicle type classification accuracy, VIN location prediction accuracy, and identification code recognition accuracy by repeatedly updating model parameters to minimize the loss value L.

[0221] Additionally, the AI ​​learning unit (380) can perform learning by separating data by vehicle type, shooting environment, and camera. For example, it can distinguish between data captured in a night environment and data captured in a day environment to learn model parameters suitable for each, or create a correction model specialized for data collected from a specific VIN recognition camera.

[0222] The AI ​​learning unit (380) distributes the learned models to each module of the AI ​​analysis edge device (300). That is, the trained vehicle type classification model is provided to the vehicle analysis unit (320), the VIN location prediction model is provided to the VIN location prediction unit (330), and the identification code recognition model is provided to the code reading unit (370). Accordingly, the AI ​​analysis edge device (300) can perform vehicle analysis, location prediction, and identification code reading using the latest learned models.

[0223] In one embodiment, the AI ​​learning unit (380) may update the model only when certain conditions are met. For example, it may be configured to perform retraining only when a certain amount of new data is accumulated or when recognition accuracy drops below a threshold value. This reduces unnecessary training operations and maintains system stability.

[0224] The result judgment unit (390) is configured to determine the validity of the final recognition result and the processing status of the vehicle based on multiple identification code reading results transmitted from the code reading unit (370) and reliability information for each result. The result judgment unit (390) does not simply transmit reading results, but performs the function of comparing and analyzing multiple input results to select the optimal result and determining whether recognition is successful, whether re-shooting is necessary, and the operational control status.

[0225] First, the result judgment unit (390) receives a plurality of candidate reading results from the code reading unit (370). The candidate reading results may include VIN strings or barcode values ​​derived from a plurality of VIN recognition cameras (200) or a plurality of candidate regions, a reliability value for each result, probability information for each read character position, and reading status information.

[0226] The result judgment unit (390) performs a reliability-based evaluation on multiple received reading results. Specifically, the result judgment unit (390) calculates a priority by comprehensively considering the overall reliability value of each reading result, the probability distribution per character, whether the code length condition is satisfied, and the suitability of the character rule. At this time, the reading result with the highest reliability can be selected as the primary candidate, and the remaining results can be maintained as auxiliary candidates.

[0227] Additionally, the result judgment unit (390) determines whether there is a match between multiple reading results. For example, if the same VIN string is read from different cameras or candidate areas, the result judgment unit (390) may determine the result as a confirmed result with high reliability. On the other hand, if only some characters match between different reading results, the final result may be reconstructed by comparing the reliability of each character position. That is, the result may be corrected by generating the final VIN string by selecting the character with the highest probability for each character position.

[0228] The result judgment unit (390) verifies the structural validity of the reading result. For example, since VIN has specific length and character rules, it checks whether the read string satisfies the specifications. In addition, if a character that is not allowed is included in a specific location, it determines this as an error and can select a replacement candidate or request re-reading.

[0229] The result judgment unit (390) determines whether recognition is successful by comparing the final reliability of the reading result with a reference value. For example, if the final reliability is greater than or equal to the reference value, it is determined to be a successful recognition, and if it is less than the reference value, it is determined to be a failure to recognize or a subject for re-examination. At this time, if the reliability is close to the reference value, it may be classified into an intermediate state such as “caution passed” or “speed reduction”.

[0230] Additionally, the result judgment unit (390) determines whether to retake the shot in the event of a recognition failure. If the vehicle is still within the shooting range, a retake request signal may be transmitted to the retake control unit or the shooting control unit (340) to activate the auxiliary camera. On the other hand, if the vehicle has already passed the shooting range, the vehicle is classified as a re-inspection target, and a re-inspection guidance signal is output through the output control unit (395).

[0231] The result judgment unit (390) defines the processing status for each vehicle. For example, it classifies the vehicle status into one of “recognition successful,” “recognition failed,” “re-shooting in progress,” or “re-inspection required,” and generates the corresponding status information along with a vehicle tracking identifier. This status information is transmitted to the output control unit (395) and the display device (400) and provided to the vehicle driver or operator.

[0232] In one embodiment, the result determination unit (390) may additionally verify the consistency between the vehicle type information and the reading result. For example, by comparing the actual reading result with the expected VIN format or barcode format for a specific vehicle type, if the formats do not match, it may be determined as an error.

[0233] Additionally, the result judgment unit (390) can perform a comparison with past reading history. If there is a record of the same vehicle being previously recognized, the current reading result is compared with the previous result to verify consistency, and if a discrepancy occurs, a re-examination can be requested.

[0234] The result judgment unit (390) transmits the finally determined recognition result and vehicle status information to the output control unit (395). Based on the information, the output control unit (395) provides visual guidance to the display device (400) or controls whether the vehicle passes.

[0235] The output control unit (395) is configured to generate output signals for the display device (400) and external control device based on the final identification code, recognition reliability, and vehicle processing status information transmitted from the result judgment unit (390), and to control the vehicle's passage, stop, re-inspection, or guidance path. Beyond simply transmitting display signals, the output control unit (395) performs an interface function that controls the field operation flow in real time.

[0236] First, the output control unit (395) receives a vehicle tracking identifier, a final VIN or barcode value, whether recognition was successful, a reliability value, and vehicle status information from the result determination unit (390). The output control unit (395) determines an output control policy based on the vehicle status among the received information. For example, if the vehicle status is “recognition successful,” a pass signal is generated, and if the status is “recognition failed” or “re-inspection required,” a stop or guidance signal is generated.

[0237] The output control unit (395) generates visual information to be output to the display device (400). The visual information may include colors, text, icons, or flashing patterns, and is set differently depending on the vehicle status. For example, a green indicator and a “pass” message may be output when recognition is successful, a red indicator and a “stop” or “re-inspection” message may be output when recognition fails, and a yellow indicator and a “wait” message may be output when recognition is in progress. Additionally, a direction indicator icon or an arrow may be output together when vehicle guidance is required.

[0238] The output control unit (395) performs individual output for each vehicle when multiple vehicles pass through in succession. To this end, the output control unit (395) manages output information corresponding to each vehicle based on a vehicle tracking identifier and displays the status of the corresponding vehicle in a specific area of ​​the display device (400) or at a specific timing. For example, in a multi-lane environment, independent display may be performed for each lane, or individual status may be output in an area corresponding to the vehicle location within a single display device.

[0239] Additionally, the output control unit (395) can perform physical control in conjunction with an external control device. For example, it can be connected to a circuit breaker, gate, traffic light, or warning light and output a control signal that opens the circuit breaker upon successful recognition, or keeps the circuit breaker closed or emits a warning sound upon failure of recognition. At this time, the output control unit (395) can transmit and receive data with the external control device via wired or wireless communication.

[0240] The output control unit (395) can perform multi-stage control based on recognition reliability. For example, if the reliability is high, it outputs an immediate pass signal; if the reliability is at an intermediate level, it outputs a “caution pass” or “speed reduction” message; and if the reliability is low, it outputs a “stop and re-inspect” signal. Accordingly, flexible operational control according to the situation is possible, rather than a simple success / failure dichotomy.

[0241] Additionally, the output control unit (395) performs the function of inducing follow-up measures for vehicles that fail to be recognized. For example, if a re-shooting section exists, it outputs a “re-shooting section move” message, and if a separate waiting space is required, it can provide direction information along with a “waiting area move” message. This prevents congestion in the vehicle flow and improves operational efficiency.

[0242] The output control unit (395) performs time synchronization of the output signal. That is, it adjusts the output timing by reflecting the vehicle position and speed of movement to match the time when the vehicle reaches the front of the display device (400) with the time when the display information is output. This allows the driver to accurately perceive the results regarding their vehicle.

[0243] Additionally, the output control unit (395) can store and manage output history. The output history may include a vehicle identifier, output time, output status, recognition result, and control signal, and can be used as operational analysis, system performance evaluation, or post-verification data.

[0244] The output control unit (395) can switch to a safe mode when a communication failure or system error occurs. For example, if a recognition result is not received for more than a certain period of time, it can be controlled to automatically maintain a “stop” or “standby” state to prevent accidents caused by malfunction.

[0246] FIG. 3 is a flowchart illustrating an embodiment of the method for recognizing and operating a barcode VIN of a high-speed moving body according to the present invention.

[0247] Referring to FIG. 3, in step S310, an entry image of a vehicle is obtained using a first camera, a vehicle object is detected from the image, and the vehicle's external shape information, movement speed, and direction of travel are analyzed.

[0248] In step S320, the AI ​​analysis edge device determines the vehicle type based on the vehicle exterior information and calculates the predicted location of the identification code on the vehicle using a pre-trained VIN location prediction model.

[0249] In step S330, the AI ​​analysis edge device generates shooting times and shooting conditions for a plurality of VIN recognition cameras based on the predicted location and vehicle movement speed.

[0250] In step S340, under the control of the AI ​​analysis edge device, the identification code area is captured using a plurality of VIN recognition cameras while the vehicle is in motion.

[0251] In step S340, an identification code is detected and read from the captured image.

[0252] In step S350, whether the vehicle passes or stops is determined based on the reading result and output through the display device.

[0254] FIG. 4 is a diagram illustrating the identification code recognition and verification procedure of a high-speed moving vehicle according to one embodiment of the present invention based on actual images.

[0255] Referring to Fig. 4, Explanation of the symbols

[0257] 100: First camera, 200: VIN recognition camera, 300: AI Analysis Edge Device, 300: AI Analysis Edge Device, 310: Data receiver, 320: Vehicle Analysis Department, 330: VIN position prediction unit, 340: Shooting control unit, 350: Image preprocessing unit, 360: Identification code detection unit, 370: Code reader, 380: AI Learning Department, 390: Result judgment unit, 395: Output control unit, 400: Display device,

Claims

Claim 1 In a barcode VIN recognition and operation system for a high-speed moving object, a first camera detects a vehicle object from an entry image of the vehicle, analyzes the vehicle's movement speed and direction of travel by analyzing the position change of the vehicle object, and determines the vehicle type by analyzing the vehicle's external shape information; a plurality of VIN recognition cameras that capture an identification code area while the vehicle is in motion;The estimated location of an identification code on a vehicle is calculated using a pre-trained VIN location prediction model, and the shooting direction, field of view, focal length, and region of interest of a plurality of VIN recognition cameras are set by referencing the estimated VIN location corresponding to the vehicle type; the time at which the identification code area reaches the captureable area of ​​the plurality of VIN recognition cameras is predicted using the vehicle's current location, vehicle speed, vehicle direction of travel, and the estimated identification code location; the time to the shooting trigger (Tshoot) is calculated using the distance from the detection reference line to the shooting centerline of the VIN recognition camera (Dcam), the distance from the vehicle reference point to the estimated VIN location (Dvin), and the vehicle speed (Vvehicle); the time to the shooting trigger (Tshoot) is calculated according to [Equation 1] below; a shooting trigger signal is generated based on the time value calculated by [Equation 1]; the actual shooting time is corrected by reflecting the system delay time as a correction value; shooting conditions corresponding to the shooting time are generated; at least one of the plurality of VIN recognition cameras is selectively activated according to the vehicle type to control the identification code area to be captured while the vehicle is in motion; and the plurality An AI analysis edge device that detects and reads an identification code from an image captured by a VIN recognition camera, calculates the reliability (S) of the entire VIN string using a prediction probability (pi) for each character position, and calculates the reliability (S) of the entire VIN string according to [Equation 3] below, determines the validity of the reading result by comparing the reliability (S) with a reference value, and generates an identification code recognition result by comparing and combining reading results obtained from a plurality of candidate regions or a plurality of camera images; and a display device that determines whether a vehicle passes or stops according to the identification code recognition result transmitted from the AI ​​analysis edge device and displays the result, [Equation 1]; Tshoot refers to the time until the shooting trigger, and D cam represents the distance from the detection baseline to the shooting centerline of the VIN recognition camera, and D vin represents the distance from the vehicle reference point to the estimated VIN location, and V vehicle represents the vehicle's speed of movement, [Equation 3] S represents the confidence level of the entire VIN string, N represents the total number of characters in the VIN, and p i A barcode VIN recognition and operation system for a high-speed moving object characterized by meaning the predicted probability for a character selected at the i-th character position. Claim 2 A method for recognizing and operating a barcode VIN of a high-speed moving object, comprising: a step in which a first camera acquires an entry image of a vehicle, detects a vehicle object from the vehicle image, analyzes the position change of the vehicle object to analyze the vehicle's speed and direction of travel, and analyzes the vehicle's exterior information to determine the vehicle type; a step in which an AI analysis edge device calculates an identification code predicted location, a predicted direction, and a shooting area of ​​interest on the vehicle using a pre-trained VIN location prediction model for each vehicle type; a step in which an AI analysis edge device selects at least one of a plurality of VIN recognition cameras by referencing the VIN predicted location corresponding to the vehicle type, and sets the shooting direction, angle of view, focal length, and shooting conditions of the selected VIN recognition camera; and a step in which an AI analysis edge device predicts the time when the identification code area reaches the shooting center area of ​​the selected VIN recognition camera using the vehicle's current location, the vehicle's speed of travel, the vehicle's direction of travel, and the identification code predicted location, and uses the distance from the detection reference line to the shooting reference line of the VIN recognition camera (Dcam), the offset distance from the vehicle reference point to the VIN predicted location (Dvin), and the vehicle's speed of travel (Vvehicle) to trigger a shooting. A step of calculating a time (Tshoot), wherein the shooting trigger time (Tshoot) is calculated according to the following mathematical formula 1; a step in which an AI analysis edge device drives a selected VIN recognition camera according to the shooting trigger time (Tshoot) to photograph a VIN or barcode while the vehicle is moving; a step in which an AI analysis edge device detects and reads the VIN or barcode from the captured image, and calculates the reliability (S) of the entire VIN string using the prediction probability (pi) for each character position, wherein the reliability (S) of the entire VIN string is calculated according to the following mathematical formula 3;The AI ​​analysis edge device determines the validity of the reading result by comparing the reliability (S) with a reference value, and if the reliability is less than the reference value, performs re-reading on an image acquired from another candidate region or another VIN recognition camera, and if the reliability is greater than or equal to the reference value, generates a final identification code; and the AI ​​analysis edge device determines whether the vehicle passes or stops based on the final identification code and outputs it through a display device, [Equation 1]; Tshoot refers to the time until the shooting trigger, and D cam represents the distance from the detection baseline to the shooting centerline of the VIN recognition camera, and D vin represents the distance from the vehicle reference point to the estimated VIN location, and V vehicle represents the vehicle's speed of movement, [Equation 3] S represents the confidence level of the entire VIN string, N represents the total number of characters in the VIN, and p i A method for recognizing and operating a barcode VIN of a high-speed moving object, characterized in that it represents the predicted probability for a character selected at the i-th character position.

Images