Apparatus and method for detecting potential vehicle intruding into driving path

Abstract

An apparatus for detecting a potential vehicle intruding into a driving path, according to the present embodiment, comprises: an image obtainment unit that obtains a forward image from a camera mounted on a vehicle and preprocesses the forward image; a forward vehicle detection unit that inputs the preprocessed forward image into a pre-prepared object detection model, to detect a forward vehicle; a vehicle wheel detection unit that, if the detected vehicle is located within a preset region of interest, detects each wheel region of the vehicle by using the object detection model; an ellipse parameter extraction unit that performs ellipse fitting on each detected wheel region to extract an ellipse parameter; a change tracking unit that calculates an ellipse feature value by using the extracted ellipse parameter and tracks the degree of change in the ellipse feature value; and a dangerous vehicle detection unit that determines the presence of a potential risk on the basis of the degree of change in the ellipse feature value and detects a dangerous vehicle on the basis of a determination result.

Description

Device and method for detecting potential path-invading vehicles

[0001] The present invention relates to a potential driving path intrusion detection device and method thereof, and more particularly, to a potential driving path intrusion detection device and method thereof that detects a potential driving path intrusion vehicle by estimating a change in the wheel direction of a surrounding vehicle.

[0002] An autonomous driving system is a system capable of autonomously driving to a pre-determined destination by recognizing the surrounding environment and vehicle status without the driver's input. These autonomous driving systems consist of perception, judgment, path generation, and vehicle control stages. Path generation must detect changes in obstacles, generate an avoidance path in real time, and generate a path that reflects the vehicle's kinematic behavioral characteristics.

[0003] In particular, various dangerous situations may arise when autonomous vehicles attempt to change lanes in real-time changing urban areas, congested areas, and highways. Therefore, the vehicle is controlled to ensure safe driving by generating a stable driving path while considering collisions with surrounding dynamic obstacles.

[0004] In conventional technology, artificial intelligence and deep learning techniques were widely used to determine the driving plan of a vehicle by understanding the behavior of surrounding vehicles.

[0005] However, this method can produce results that are different from the developer's intention, and there is a problem in that it is difficult to identify the cause and correct and supplement the problem when an accident occurs.

[0006] The technology underlying the invention is disclosed in Korean Patent Publication No. 10-2360817 (published on February 10, 2022).

[0007] The present invention has been devised to solve the above problems, and the technical task to be achieved by the present invention is to provide a potential driving path intrusion detection device and method for detecting a potential driving path intrusion vehicle by estimating a change in the wheel direction of a surrounding vehicle.

[0008] In addition, the present invention provides a potential driving path intrusion vehicle detection device and method for detecting sudden wheel direction changes by calculating and tracking elliptical feature values ​​for vehicle wheels.

[0009] According to an embodiment of the present invention for achieving such a technical task, a potential driving path intrusion vehicle detection device includes: an image acquisition unit for acquiring a front image from a camera mounted on a vehicle and preprocessing it; a front vehicle detection unit for inputting the preprocessed image into a pre-prepared object detection model to detect a front vehicle; a vehicle wheel detection unit for detecting each wheel area of ​​the vehicle using the object detection model when the detected vehicle is located within a preset region of interest; an ellipse parameter extraction unit for performing ellipse fitting on each detected wheel area to extract ellipse parameters; a change tracking unit for calculating ellipse feature values ​​using the extracted ellipse parameters and tracking the degree of change in the ellipse feature values; and a dangerous vehicle detection unit for determining whether there is a potential danger based on the degree of change in the ellipse feature values ​​and detecting a dangerous vehicle based on the determination result.

[0010] At this time, the front vehicle detection unit may use the object detection model that has been learned in advance to enable detection of vehicles and wheels within an image to detect a vehicle within the preprocessed image, generate a bounding box for the vehicle, and extract an area for the vehicle. The vehicle wheel detection unit may use the object detection model to create a bounding box for each wheel of the vehicle, and extract an area for each wheel, if the vehicle is located within an area of ​​interest that has been preset for detection of a potential dangerous vehicle, with reference to the extracted area.

[0011] In addition, the ellipse parameter extraction unit can perform ellipse fitting on the area of ​​each wheel using any one of RANSAC (Random Sample Consensus), Least Squares, and Hough Transform, and extract ellipse parameters including a semi-major axis and a semi-minor axis.

[0012] In addition, the change tracking unit can track the degree of change in the ellipse feature value by calculating the ellipse feature value including the flattening rate and circularity to estimate the degree of flatness of the ellipse fitted to each detected wheel using the extracted ellipse parameter.

[0013] In addition, if the change in the elliptical feature value exceeds a threshold value set in consideration of the relative distance and direction with respect to another vehicle, the dangerous vehicle detection unit determines that a change has occurred in the wheel direction of the other vehicle, and can detect the vehicle as a potential dangerous vehicle that may invade the driving path.

[0014] In addition, a method for detecting a potential driving path intrusion vehicle according to another embodiment of the present invention includes the steps of: acquiring a front image from a camera mounted on a vehicle and preprocessing it; inputting the preprocessed image into a pre-prepared object detection model to detect a front vehicle; detecting each wheel area of ​​the vehicle using the object detection model when the detected vehicle is located within a pre-set region of interest; performing ellipse fitting on each detected wheel area to extract ellipse parameters; calculating ellipse feature values ​​using the extracted ellipse parameters and tracking a degree of change in the ellipse feature values; and determining whether there is a potential danger based on the degree of change in the ellipse feature values, and detecting a dangerous vehicle based on the determination result.

[0015] At this time, the step of detecting the front vehicle uses the object detection model that has been learned in advance to enable detection of vehicles and wheels within the image, and when a vehicle is detected within the preprocessed image, a bounding box for the vehicle is generated to extract an area for the vehicle, and the step of detecting the area of ​​each wheel of the vehicle uses the object detection model to generate a bounding box for each wheel of the vehicle and extract an area for each wheel if the vehicle is located within an area of ​​interest that has been preset for detection of a potential dangerous vehicle with reference to the extracted area.

[0016] In addition, the step of extracting the ellipse parameters may perform ellipse fitting for the area of ​​each wheel using any one of the methods of RANSAC (Random Sample Consensus), Least Squares, and Hough Transform, and extract ellipse parameters including the semi-major axis and the semi-minor axis.

[0017] In addition, the step of tracking the degree of change may track the degree of change in the ellipse feature values ​​by calculating the ellipse feature values ​​including the flattening and circularity to estimate the degree of flatness of the ellipse fitted to each detected wheel using the extracted ellipse parameters.

[0018] In addition, in the step of detecting the above-mentioned dangerous vehicle, if the change in the above-mentioned elliptical feature value exceeds a threshold value set in consideration of the relative distance and direction with respect to the other vehicle, it is determined that a change has occurred in the direction of the wheels of the other vehicle, and the vehicle can be detected as a potential dangerous vehicle that may invade the driving path.

[0019] In this way, according to the present invention, by estimating the change in wheel direction of surrounding vehicles and detecting potential vehicles that invade the driving path, there is an effect of minimizing damage loss due to a collision accident in a situation where a parked vehicle or a vehicle in an adjacent lane makes a close cut-in when driving on an irregular road with unclear lane lines or in a low-speed traffic jam.

[0020] In addition, according to the present invention, by securing computational efficiency by calculating and tracking elliptical feature values ​​for vehicle wheels and detecting sudden wheel direction changes, it is possible to complement the limitations of existing autonomous driving and driving safety systems by detecting potential path-invading vehicles early.

[0021] In addition, according to the present invention, there is an advantage in that functions can be implemented without installing additional sensors and equipment by utilizing cameras that are basically installed in autonomous vehicles and mobile devices.

[0022] In addition, according to the present invention, early detection of vehicles invading the driving path is possible, so that the overall performance and safety of autonomous driving and advanced driver assistance systems (ADAS) functions can be expected to be improved.

[0023] In addition, according to the present invention, in the case of a mobile vehicle such as a delivery robot that has a lower height than a general vehicle and needs to drive on an irregular road with many parked vehicles, there is an effect of minimizing damage loss due to a collision accident.

[0024] Additionally, according to the present invention, an appropriate avoidance path can be created and provided when necessary to prevent a violation of the driving path.

[0025] FIG. 1 is a block diagram showing a potential driving path intrusion vehicle detection device according to an embodiment of the present invention.

[0026] FIG. 2 is a drawing illustrating a process for detecting a front vehicle according to an embodiment of the present invention.

[0027] FIG. 3 is a drawing illustrating a process for detecting a wheel area of ​​a front vehicle according to an embodiment of the present invention.

[0028] Figure 4 is an example showing the results of ellipse fitting and ellipse parameter extraction for the wheel area detected in Figure 3.

[0029] FIG. 5 is a flowchart illustrating the operation flow of a potential driving path intrusion vehicle detection method according to an embodiment of the present invention.

[0030] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. In this process, the thickness of lines and the sizes of components depicted in the drawings may be exaggerated for clarity and convenience of explanation.

[0031] Furthermore, the terms described below are defined based on their functions within the present invention, and may vary depending on the intent or custom of the user or operator. Therefore, the definitions of these terms should be based on the overall content of this specification.

[0032] Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

[0033] First, a potential driving path intrusion vehicle detection device according to an embodiment of the present invention will be described with reference to FIGS. 1 to 4.

[0034] FIG. 1 is a block diagram showing a potential driving path intrusion vehicle detection device according to an embodiment of the present invention.

[0035] As shown in FIG. 1, a potential driving path intrusion vehicle detection device (100) according to an embodiment of the present invention includes an image acquisition unit (110), a front vehicle detection unit (120), a vehicle wheel detection unit (130), an ellipse parameter extraction unit (140), a change tracking unit (150), and a dangerous vehicle detection unit (160).

[0036] First, the image acquisition unit (110) acquires a front image from a camera (not shown) mounted on the vehicle and preprocesses it.

[0037] In detail, after acquiring an image of the driving direction forward from the camera, a preprocessing process is performed to improve image quality, such as distortion correction, normalization, smoothing, resizing, and color adjustment, on the acquired image.

[0038] And the front vehicle detection unit (120) detects the front vehicle by inputting the image preprocessed by the image acquisition unit (110) into a pre-prepared object detection model.

[0039] At this time, the front vehicle detection unit (120) uses an object detection model that has been trained in advance to enable detection of vehicles and wheels within an image, and when a vehicle is detected within a preprocessed image, a bounding box for the vehicle is created to extract the area for the vehicle.

[0040] At this time, the object detection model may be a deep learning-based model that detects the main object in the image and displays a bounding box centered on the object to detect the object.

[0041] In detail, artificial intelligence algorithms such as Fast R-CNN and Faster R-CNN, which are object detection models that improve YOLO (You Only Look Once) or R-CNN (Regions with Convolutional Neural Network) that are suitable for real-time applications by detecting objects in a single step, can be applied to provide faster learning and detection speed.

[0042] FIG. 2 is a drawing illustrating a process for detecting a front vehicle according to an embodiment of the present invention.

[0043] As illustrated in Fig. 2, the front vehicle detection unit (120) inputs a preprocessed image into an object detection model that has been pre-trained to be suitable for vehicle detection, and when a bounding box for a vehicle in the image is generated, the area for the vehicle is extracted using the generated bounding box.

[0044] At this time, the object detection model can detect multiple vehicles included in the input image at once and generate multiple bounding boxes for each vehicle.

[0045] And, when the vehicle detected by the front vehicle detection unit (120) is located within a preset area of ​​interest, the vehicle wheel detection unit (130) detects each wheel area of ​​the vehicle using an object detection model.

[0046] At this time, the vehicle wheel detection unit (130) refers to the area of ​​the vehicle detected by the front vehicle detection unit (120), and if the vehicle is located within the area of ​​interest preset for detection of a potential dangerous vehicle, it uses an object detection model to create a bounding box for each wheel of the vehicle and extracts the area for each wheel.

[0047] FIG. 3 is a drawing illustrating a process for detecting a wheel area of ​​a front vehicle according to an embodiment of the present invention.

[0048] As illustrated in FIG. 3, the vehicle wheel detection unit (130) extracts an area for the vehicle when the detected vehicle is located in an area of ​​interest (e.g., close range, left and right front sides) for detecting a potential vehicle invading the driving path, and then creates a bounding box for the vehicle wheel in the image using a pre-trained object detection model for vehicle wheel detection. At this time, an additional preprocessing step for making the image size and illuminance uniform, etc. may be applied to the image from which the detected vehicle area is extracted, and then used as input for the object detection model for vehicle wheel detection.

[0049] The ellipse parameter extraction unit (140) performs ellipse fitting on each wheel area detected by the vehicle wheel detection unit (130) to extract ellipse parameters.

[0050] At this time, the ellipse parameter extraction unit (140) performs ellipse fitting on the area of ​​each wheel detected by the vehicle wheel detection unit (130) using a method such as RANSAC (Random Sample Consensus), Least Squares, or Hough Transform, and extracts ellipse parameters including the semi-major axis and the semi-minor axis.

[0051] The change tracking unit (150) calculates elliptical feature values ​​using the elliptical parameters extracted from the elliptical parameter extraction unit (140) and tracks the degree of change in the elliptical feature values.

[0052] At this time, the change tracking unit (150) calculates ellipse feature values ​​including flattening and circularity to estimate the degree of flatness of the ellipse fitted to each wheel detected by the vehicle wheel detection unit (130) using the ellipse parameters extracted by the ellipse parameter extraction unit (140), and tracks the degree of change in the ellipse feature values.

[0053] Figure 4 is an example showing the results of ellipse fitting and ellipse parameter extraction for the wheel area detected in Figure 3.

[0054] As illustrated in FIG. 4, the change tracking unit (150) calculates ellipse feature values ​​for estimating the flatness of an ellipse fitted to each vehicle wheel detected by the vehicle wheel detection unit (130) and tracks the degree of change. The flatness and circularity are calculated by the following mathematical expression 1.

[0055] [Mathematical Formula 1]

[0056]

[0057] Here, f is the flatness, a is the semi-major axis of the ellipse, b is the semi-minor axis of the ellipse, C is the circularity, A is the area of ​​the ellipse, and P is the perimeter of the ellipse.

[0058] Finally, the dangerous vehicle detection unit (160) determines whether there is a potential danger based on the degree of change in the elliptical feature value, and detects a dangerous vehicle based on the determination result.

[0059] At this time, if the change in the elliptical feature value calculated by the change tracking unit (150) exceeds a threshold value set in consideration of the relative distance and direction with respect to other vehicles, the dangerous vehicle detection unit (160) determines that a change has occurred in the wheel direction of the other vehicle and detects the vehicle as a potential dangerous vehicle that may invade the driving path.

[0060] That is, if the change in the elliptical feature value of the wheel of the vehicle in front being tracked exceeds a threshold that takes into account changes in the relative distance and relative direction between the vehicle and other vehicles, the vehicle is judged as a potential dangerous vehicle that may invade the driving path of other vehicles.

[0061] At this time, time series prediction can be performed using Kalman Filter, LSTM (Long Short-Term Memory) model, etc. to track elliptical feature values ​​and detect potential dangerous vehicles, or anomaly detection techniques such as CUSUM (Cumulative Sum) and EWMA (Exponentially Weighted Moving Average) can be applied.

[0062] In addition, when a potential dangerous vehicle is detected by the dangerous vehicle detection unit (160), the driving path intrusion vehicle detection device (100) may execute an avoidance path generation algorithm to generate an appropriate avoidance path in preparation for a driving path intrusion.

[0063] That is, when the avoidance path generation algorithm is executed, it determines whether there is an adjacent lane opposite the dangerous vehicle that allows lane changes. If there is an adjacent lane and a lane change is possible (provided there are no hazards such as collisions), an avoidance path can be generated by changing lanes to the adjacent lane.

[0064] However, if there is no adjacent lane opposite the hazardous vehicle that can be changed to, or if changing to an existing adjacent lane is not possible, an avoidance route may be created to attempt to drive in a deviated lane away from the hazardous vehicle.

[0065] Hereinafter, a method for detecting a potential driving path intrusion vehicle according to an embodiment of the present invention will be described through FIG. 5.

[0066] FIG. 5 is a flowchart illustrating the operation flow of a potential driving path intrusion vehicle detection method according to an embodiment of the present invention, and specific operations of the present invention will be described with reference to this.

[0067] According to an embodiment of the present invention, first, the image acquisition unit (110) of the driving path intrusion vehicle detection device (100) acquires a front image from a camera mounted on the vehicle and preprocesses it (S10).

[0068] In detail, after acquiring an image of the driving direction forward from the camera, a preprocessing process is performed to improve image quality, such as distortion correction, normalization, smoothing, resizing, and color adjustment, on the acquired image.

[0069] Next, the front vehicle detection unit (120) inputs the preprocessed image in step S10 into a pre-prepared object detection model to detect the front vehicle (S20).

[0070] In detail, when a vehicle is detected in a preprocessed image using an object detection model that has been trained in advance to enable vehicle and wheel detection in the image, a bounding box for the vehicle is created to extract the area for the vehicle.

[0071] At this time, the object detection model may be a deep learning-based model that detects the main object in the image and displays a bounding box centered on the object to detect the object.

[0072] In detail, artificial intelligence algorithms such as Fast R-CNN and Faster R-CNN, which are object detection models that improve YOLO (You Only Look Once) or R-CNN (Regions with Convolutional Neural Network) that are suitable for real-time applications by detecting objects in a single step, can be applied to provide faster learning and detection speed.

[0073] Next, the vehicle wheel detection unit (130) determines whether the vehicle detected in step S20 is located within a preset area of ​​interest (S30).

[0074] As a result of the judgment at step S30, if the vehicle detected at step S20 is located within a preset area of ​​interest, each wheel area of ​​the vehicle detected at step S20 is detected using an object detection model (S40).

[0075] At this time, step S30 refers to the area for the vehicle detected in step S20, and if the vehicle is located within a preset area of ​​interest for detecting a potential dangerous vehicle, the object detection model is used to create a bounding box for each wheel of the vehicle and extract the area for each wheel.

[0076] That is, if the vehicle detected in step S20 is located in the area of ​​interest (e.g., near-field, left, right, and front) for potential path-invading vehicle detection, the area for the corresponding vehicle is extracted, and then a pre-trained object detection model is used to detect vehicle wheels to generate a bounding box for the vehicle wheels in the image. At this time, an additional preprocessing step to make the image size and illumination uniform, etc. may be applied to the image from which the detected vehicle area is extracted, and then used as input for the object detection model for vehicle wheel detection.

[0077] Next, the ellipse parameter extraction unit (140) performs ellipse fitting for each wheel area detected in step S40 to extract ellipse parameters (S50).

[0078] At this time, the S50 step performs ellipse fitting on the area of ​​each wheel detected by the vehicle wheel detection unit (130) using methods such as RANSAC (Random Sample Consensus), Least Squares, and Hough Transform, and extracts ellipse parameters including the semi-major axis and the semi-minor axis.

[0079] Next, the change tracking unit (150) calculates the ellipse feature value using the ellipse parameter extracted in step S50 and tracks the degree of change in the ellipse feature value (S60).

[0080] In detail, step S60 calculates ellipse features including flattening and circularity to estimate the degree of flatness of the ellipse fitted to each wheel detected in step S40 using the ellipse parameters extracted in step S50, and tracks the degree of change in the ellipse features.

[0081] At this time, the flatness and circularity are calculated by mathematical formula 1 as described above.

[0082] Finally, the dangerous vehicle detection unit (160) determines whether there is a potential danger based on the degree of change in the elliptical feature value calculated in step S60 (S70).

[0083] As a result of the judgment at step S70, if the change in the elliptical feature value calculated at step S60 exceeds the threshold value set in consideration of the relative distance and direction with respect to other vehicles, it is determined that a change has occurred in the wheel direction of the other vehicle, and the vehicle is detected as a potential dangerous vehicle that may invade the driving path (S80).

[0084] That is, if the change in the elliptical feature value of the wheel of the vehicle in front being tracked exceeds a threshold that takes into account changes in the relative distance and relative direction between the vehicle and other vehicles, the vehicle is judged as a potential dangerous vehicle that may invade the driving path of other vehicles.

[0085] At this time, time series prediction can be performed using Kalman Filter, LSTM (Long Short-Term Memory) model, etc. to track elliptical feature values ​​and detect potential dangerous vehicles, or anomaly detection techniques such as CUSUM (Cumulative Sum) and EWMA (Exponentially Weighted Moving Average) can be applied.

[0086] Additionally, when a potential dangerous vehicle is detected at step S80, an avoidance path generation algorithm can be executed to generate an appropriate avoidance path in preparation for a path violation.

[0087] That is, when the avoidance path generation algorithm is executed, it determines whether there is an adjacent lane opposite the dangerous vehicle that allows lane changes. If there is an adjacent lane and a lane change is possible (provided there are no hazards such as collisions), an avoidance path can be generated by changing lanes to the adjacent lane.

[0088] However, if there is no adjacent lane opposite the hazardous vehicle that can be changed to, or if changing to an existing adjacent lane is not possible, an avoidance route may be created to attempt to drive in a deviated lane away from the hazardous vehicle.

[0089] If the change in the elliptical feature value calculated at step S60 is below the threshold, the vehicle is not detected as a potential dangerous vehicle.

[0090] The method for detecting a potential driving path intrusion vehicle according to an embodiment of the present invention has been described above.

[0091] The method for detecting a potential driving path intrusion vehicle according to an embodiment of the present invention as described above may be provided in the form of a computer-readable medium suitable for storing computer program commands and data.

[0092] In particular, the computer program of the present invention can execute the steps of acquiring a front image from a camera mounted on a vehicle and preprocessing it, the step of inputting the preprocessed image into a pre-prepared object detection model to detect a front vehicle, the step of detecting each wheel area of ​​the vehicle using the object detection model when the detected vehicle is located within a pre-set region of interest, the step of performing ellipse fitting on each detected wheel area to extract ellipse parameters, the step of calculating ellipse feature values ​​using the extracted ellipse parameters and tracking the degree of change in the ellipse feature values, and the step of determining whether there is a potential danger based on the degree of change in the ellipse feature values, and detecting a dangerous vehicle based on the determination result.

[0093] Such computer-readable recording media may contain program commands, data files, data structures, etc. alone or in combination, and include all types of recording devices that store data that can be read by a computer system. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs (Compact Disk Read Only Memory) and DVDs (Digital Video Disks), magneto-optical media such as floptical disks, and hardware devices specifically configured to store and execute program commands such as ROMs (Read Only Memory), RAMs (Random Access Memory), and flash memory.

[0094] Additionally, computer-readable recording media can be distributed across network-connected computer systems, allowing computer-readable code to be stored and executed in a distributed manner. Furthermore, functional programs, codes, and code segments for implementing the present invention can be readily inferred by programmers skilled in the art to which the present invention pertains.

[0095] According to the present invention as described above, by estimating the change in wheel direction of surrounding vehicles and detecting potential vehicles that invade the driving path, it is possible to minimize damage loss due to a collision accident in a situation where a parked vehicle or a vehicle in an adjacent lane makes a close cut-in when driving on an irregular road with unclear lane lines or in a low-speed traffic jam.

[0096] In addition, by securing computational efficiency by calculating and tracking elliptical features for vehicle wheels and detecting sudden wheel direction changes, it is possible to complement the limitations of existing autonomous driving and driving safety systems by detecting potential path-invading vehicles early.

[0097] Additionally, by utilizing cameras that are basically installed in autonomous vehicles and mobile devices, there is an advantage in that functions can be implemented without installing additional sensors and equipment.

[0098] Additionally, early detection of vehicles encroaching on the driving path is possible, which is expected to improve the performance and safety of overall autonomous driving and advanced driver assistance systems (ADAS) functions.

[0099] Additionally, for mobile vehicles such as delivery robots, which have a lower overall height than regular vehicles and need to drive on irregular roads with many parked vehicles, it is possible to minimize damage losses due to collision accidents.

[0100] Additionally, in case of a violation of the driving path, an appropriate avoidance route can be created and provided when necessary.

[0101] While the present invention has been described with reference to the embodiments illustrated in the drawings, these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent embodiments are possible. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the following claims.

[0102] [Explanation of symbols]

[0103] 100: Vehicle detection device for invading driving path

[0104] 110: Image acquisition unit 120: Front vehicle detection unit

[0105] 130: Vehicle wheel detection unit 140: Ellipse parameter extraction unit

[0106] 150: Change Tracking Unit 160: Dangerous Vehicle Detection Unit

Claims

1. An image acquisition unit that acquires and preprocesses a front image from a camera mounted on a vehicle; A front vehicle detection unit that inputs the above preprocessed image into a pre-prepared object detection model to detect a front vehicle; A vehicle wheel detection unit that detects each wheel area of ​​the vehicle using the object detection model when the detected vehicle is located within a preset area of ​​interest; An ellipse parameter extraction unit that extracts ellipse parameters by performing ellipse fitting for each of the detected wheel areas; A change tracking unit that calculates elliptical feature values ​​using the elliptical parameters extracted above and tracks the degree of change in the elliptical feature values; and A potential driving path intrusion vehicle detection device including a dangerous vehicle detection unit that determines whether there is a potential danger based on the degree of change in the above elliptical feature value and detects a dangerous vehicle based on the determination result.

2. In paragraph 1, The above front vehicle detection unit, When a vehicle is detected in the preprocessed image using the object detection model that has been pre-trained to enable vehicle and wheel detection in the image, a bounding box for the vehicle is created to extract the area for the vehicle. The above vehicle wheel detection unit, A potential driving path intrusion vehicle detection device that extracts an area for each wheel by generating a bounding box for each wheel of the vehicle using the object detection model when the vehicle is located within a preset area of ​​interest for detecting a potential dangerous vehicle by referring to the extracted area.

3. In paragraph 1, The above ellipse parameter extraction unit is, A potential driving path intrusion vehicle detection device that performs ellipse fitting for the area of ​​each wheel using any one of RANSAC (Random Sample Consensus), Least Squares, and Hough Transform, and extracts ellipse parameters including a semi-major axis and a semi-minor axis.

4. In paragraph 1, The above change tracking unit, A potential driving path intrusion vehicle detection device that calculates ellipse feature values ​​including flattening and circularity to estimate the degree of flatness of an ellipse fitted to each detected wheel using the ellipse parameters extracted above, and tracks the degree of change in the ellipse feature values.

5. In paragraph 1, The above dangerous vehicle detection unit, A potential driving path intrusion detection device that detects a vehicle as a potential dangerous vehicle that may intrude on a driving path by determining that a change in the wheel direction of another vehicle has occurred when the change in the above elliptical feature value exceeds a threshold value set in consideration of the relative distance and direction from another vehicle.

6. In a method for detecting a potential driving path intrusion vehicle using a potential driving path intrusion vehicle detection device, A step of acquiring and preprocessing a front image from a camera mounted on a vehicle; A step of inputting the above preprocessed image into a pre-prepared object detection model to detect a front vehicle; A step of detecting each wheel area of ​​the vehicle using the object detection model when the detected vehicle is located within a preset area of ​​interest; A step of extracting ellipse parameters by performing ellipse fitting for each of the detected wheel areas; A step of calculating an elliptical feature value using the elliptical parameter extracted above and tracking the degree of change in the elliptical feature value; and A method for detecting a potential driving path intrusion vehicle, comprising a step of determining whether there is a potential danger based on the degree of change in the above elliptical feature value, and detecting a dangerous vehicle based on the determination result.

7. In paragraph 6, The step of detecting the above-mentioned front vehicle is: When a vehicle is detected in the preprocessed image using the object detection model that has been pre-trained to enable vehicle and wheel detection in the image, a bounding box for the vehicle is created to extract the area for the vehicle. The step of detecting each wheel area of ​​the above vehicle is: A method for detecting a potential driving path intrusion vehicle, wherein, with reference to the above extracted area, if the vehicle is located within a preset area of ​​interest for detecting a potential dangerous vehicle, a bounding box for each wheel of the vehicle is generated using the object detection model, and an area for each wheel is extracted.

8. In paragraph 6, The step of extracting the above ellipse parameters is: A method for detecting a potential driving path intrusion vehicle, wherein ellipse fitting is performed on the area of ​​each wheel using any one of RANSAC (Random Sample Consensus), Least Squares, and Hough Transform, and ellipse parameters including a semi-major axis and a semi-minor axis are extracted.

9. In paragraph 6, The steps for tracking the above change level are: A method for detecting a potential driving path intrusion vehicle, which calculates elliptical feature values ​​including flattening and circularity to estimate the degree of flatness of an ellipse fitted to each detected wheel using the elliptical parameters extracted above, and tracks the degree of change in the elliptical feature values.

10. In paragraph 6, The step of detecting the above dangerous vehicle is: A method for detecting a potential driving path intrusion vehicle, which detects a potential dangerous vehicle that may intrude on a driving path by determining that a change in the wheel direction of another vehicle has occurred when the change in the above elliptical feature value exceeds a threshold value set in consideration of the relative distance and direction with respect to another vehicle.

Images