System and method for recognizing object and evaluating confidence level based on artificial intelligence and autonomous driving sensor

Abstract

The present invention proposes a system and method for recognizing objects by an AI module and an autonomous driving sensor. Specifically, the system according to the present invention comprises: a processor having an object recognition module as part of the AI ​​module that recognizes objects existing in the surrounding environment from data input from the autonomous driving sensor; and a memory that stores the point cloud data. The AI ​​module is characterized by executing the steps of: clustering a plurality of points corresponding to objects in the surrounding environment acquired from the autonomous driving sensor to calculate a natural number N predicted objects; and assigning a reliability score to each of the N predicted objects, wherein the reliability score is assigned differentially to each of the N predicted objects based on the Region of Interest (ROI) of the vehicle in autonomous driving.

Description

Technology Field

[0001] The present invention relates to a system and method for recognizing an object and evaluating a confidence score (Confidence Level) by AI (Artificial Intelligence) and autonomous driving sensors, and more specifically, to an object recognition and confidence evaluation system and method that can provide greater robustness and reliability to object recognition in autonomous driving by differentially evaluating confidence according to ROI (Region of Interest). Background Technology

[0002] With the recent development and commercialization of autonomous vehicles, there is an increasing number of cases utilizing various sensors and artificial intelligence (AI) technologies to support autonomous driving functions. For instance, research is continuing on what objects exist in front of a moving vehicle, how far the distance is between the object and the vehicle, and what algorithms the vehicle must respond to specific situations to ensure safety.

[0003] Accordingly, vehicle sensor technology is becoming more advanced, and there is a trend of installing high-performance sensors in vehicles, such as LiDAR (Light Detection and Ranging) which uses a laser beam to perceive the surrounding environment, RADAR (Radio Detection and Ranging) which uses radio waves, ultrasonic sensors, fisheye cameras capable of capturing 360-degree images, multifocal lenses, and GPS (Global Positioning System).

[0004] As described above, the realization of so-called Super Sensor Vehicles has become possible by aggregating measurement results obtained from multiple sensors. In the context of self-driving and autonomous driving, the concept of a super sensor refers to a technology that aims to more accurately perceive the surrounding environment by combining measurements from various sensors rather than relying on individual sensors, for the convenience and safety of driving. With the addition of ICT (Information and Communications Technology) and cloud technologies, the sensors and related AI algorithms required for autonomous driving are becoming unprecedentedly sophisticated. This involves not only targeting a single vehicle but also remotely accumulating data across a fleet of vehicles and training AI servers and databases to enhance the reliability of vehicle sensor judgments.

[0005] In particular, autonomous driving sensors such as radar and LiDAR, which are used to perceive the external environment, recognize various objects present on the road during autonomous driving by emitting radio waves or lasers and measuring the reflection time and intensity of the waves or lasers reflected from surrounding objects.

[0006] However, object recognition using radar, LiDAR, and AI algorithms is not perfect. Despite the advancement of autonomous driving sensors such as radar and the development of AI technology, technical requirements for accurate object recognition remain; in particular, there is a demand for reliability evaluations related to object recognition to reflect more realistic driving situations. The problem to be solved

[0007] The present invention relates to a system and method for recognizing an object and evaluating its confidence score (Confidence Level) using AI (Artificial Intelligence) and autonomous driving sensors. Specifically, the main technical objective of the present invention is to implement an object recognition and confidence evaluation system and method that can provide greater robustness and reliability to object recognition in autonomous driving by differentially evaluating confidence based on the Region of Interest (ROI).

[0008] Of course, the technical problems of the present invention are not limited to those mentioned above, and other technical problems not explicitly mentioned will be clearly understood by those skilled in the art from the detailed description of the present invention and the accompanying drawings. means of solving the problem

[0009] To solve all or at least part of the aforementioned technical problems, the present invention may be implemented in various aspects as follows.

[0010] That is, the first aspect of the present invention relates to a system for recognizing objects by an AI module and an autonomous driving sensor, wherein the system comprises: a processor having an object recognition module as part of the AI ​​module for recognizing objects existing in a surrounding environment from data input from the autonomous driving sensor; and a memory for storing the point cloud data, wherein the AI ​​module performs the steps of: clustering a plurality of points corresponding to objects in the surrounding environment acquired from the autonomous driving sensor to calculate a natural number N predicted objects; and assigning a reliability score to each of the N predicted objects, wherein the reliability score is assigned differentially to each of the N predicted objects based on the Region of Interest (ROI) of the vehicle in autonomous driving. An AI object recognition system is presented as the first aspect of the present invention.

[0011] In the case of a system according to the second aspect of the present invention, the reliability score includes a track of interest reliability score calculated within the ROI based on a first weight applied to the current lane of the vehicle or a second weight applied to the left and right lanes of the current lane.

[0012] In the case of a system according to the third aspect of the present invention, the interest track reliability score for an object existing in the current lane is calculated according to the following mathematical formula 1, and the interest track reliability score for objects existing in the left and right lanes It is characterized by being calculated according to the following mathematical formula 2.

[0013] [Mathematical Formula 1]

[0014]

[0015] [Mathematical Formula 2]

[0016]

[0017] Here, is the above-mentioned first weight, and is the above-mentioned second weight, and is the maximum confidence score.

[0018] In the case of a system according to the fourth aspect of the present invention, the reliability score is characterized by including a non-interest track reliability score calculated for an object existing in a non-interest region outside the ROI.

[0019] In the case of a system according to the fifth aspect of the present invention, the non-interest area is characterized by being set according to a first non-interest condition that the area is outside the road boundary.

[0020] In the case of a system according to the sixth aspect of the present invention, the non-interest area is characterized by being set according to a second non-interest condition, which is an area that is further in longitudinal distance than a preceding vehicle, which is an object existing in the current lane of the vehicle or in the left or right lane of the current lane.

[0021] In the case of a system according to the seventh aspect of the present invention, even if an object exists in an area that is closer in length than the preceding vehicle, if it is a stationary object or an unidentified object, it is determined to satisfy a third non-interest condition in which it is considered an object belonging to the non-interest area.

[0022] In the case of a system according to the eighth aspect of the present invention, the non-interest track reliability score is characterized by the maximum reliability score for an object existing in the non-interest area being determined by an artificially limited upper limit as a predetermined uncontrollable level.

[0023] In the case of a system according to the ninth aspect of the present invention, a band-pass filter is applied to objects having a reliability level below the uncontrollable level.

[0024] In the case of a system according to the 10th aspect of the present invention, for an object having a reliability level exceeding the uncontrollable level, the reliability score assigned to the object is finally recognized as the reliability score in a differential manner.

[0025] Meanwhile, the eleventh aspect of the present invention relates to a method for recognizing an object by an AI module and an autonomous driving sensor, wherein the AI ​​module clusters a plurality of points corresponding to an object in the surrounding environment acquired from the autonomous driving sensor to calculate a natural number of predicted objects N; and the method comprises the step of assigning a reliability score to each of the N predicted objects, wherein the reliability score is assigned differentially to each of the N predicted objects based on the Region of Interest (ROI) of the vehicle in autonomous driving. Effects of the invention

[0026] The present invention relates to a system and method for recognizing an object and evaluating its confidence score (Confidence Level) using AI (Artificial Intelligence) and autonomous driving sensors. Specifically, the main technical objective of the present invention is to implement an object recognition and confidence evaluation system and method that can provide greater robustness and reliability to object recognition in autonomous driving by differentially evaluating confidence based on the Region of Interest (ROI).

[0027] To this end, the system according to the present invention comprises: a processor having an object recognition module as part of the AI ​​module that recognizes objects existing in the surrounding environment from data input from the autonomous driving sensor; and a memory that stores the point cloud data, wherein the AI ​​module performs the steps of: clustering a plurality of points corresponding to objects in the surrounding environment acquired from the autonomous driving sensor to calculate a natural number N predicted objects; and assigning a reliability score to each of the N predicted objects, wherein the reliability score is assigned differentially to each of the N predicted objects based on the Region of Interest (ROI) of the vehicle in autonomous driving.

[0028] In particular, in the case of a system according to the second aspect of the present invention, the reliability score includes a track of interest reliability score calculated within the ROI based on a first weight applied to the current lane of the vehicle or a second weight applied to the left and right lanes of the current lane, and in the case of a system according to the fourth aspect of the present invention mentioned above, the reliability score includes a track of non-interest reliability score calculated for an object existing in a non-interest area outside the ROI.

[0029] In other words, the present invention rapidly increases the reliability score to raise the control priority in the case of a track that enters the region of interest, and introduces the concept of first and second weights to the maximum value of the reliability score according to the importance in autonomous driving.

[0030] In addition, for tracks located in a non-interest area, even if the reliability score based on autonomous driving sensor information is recognized as a control target (e.g., greater than 40), the maximum reliability score is artificially limited to, for example, 40, thereby forcibly excluding them from the control target. In other words, by lowering the priority of objects that are relatively far away or outside the interest area (especially objects located outside the road boundary, etc.), the effect of reducing the influence of control on tracks that are not in the interest area can be expected.

[0031] In addition to the above, through the detailed description of the present invention and the accompanying drawings, those skilled in the art will be able to understand various effects other than those described above from the present invention. Brief explanation of the drawing

[0032] FIG. 1 is a block diagram showing the overall system for automatically recognizing objects and controlling a vehicle for purposes such as autonomous driving, according to one embodiment of the present invention. FIG. 2 is a flowchart showing an algorithm for recognizing an object and evaluating reliability by an AI module and an autonomous driving sensor according to an embodiment of the present invention. FIG. 3 is a drawing for explaining the components of a reliability score newly defined according to an embodiment of the present invention. FIGS. 4a and 4b are drawings illustrating exemplary driving scenes to explain a differential reliability calculation method for a region of interest and a non-region of interest according to an embodiment of the present invention. FIG. 5 is a diagram showing actual experimental results to which an object recognition and reliability evaluation technique according to an embodiment of the present invention is applied. FIG. 6 is a block diagram showing a computing system for autonomous driving vehicle control and object recognition operations according to an embodiment of the present invention. Specific details for implementing the invention

[0033] Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. It should be noted that in assigning reference numerals to the components of each drawing, the same components are given the same reference numeral whenever possible, even if they are shown in different drawings. Furthermore, in describing the embodiments of the present invention, if it is determined that a detailed description of related known components or functions would hinder understanding of the embodiments of the present invention, such detailed description is omitted.

[0034] In describing the components of the embodiments of the present invention, terms such as first, second, A, B, (a), (b), etc., may be used. These terms are intended merely to distinguish the components from other components, and the essence, order, or sequence of the components is not limited by these terms. Furthermore, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application. For example, in the present invention, "object" has virtually the same meaning as "object," and the expressions "object" and "object" are to be used interchangeably within the present invention.

[0035] FIG. 1 is a block diagram showing the overall system for automatically recognizing objects and controlling a vehicle for purposes such as autonomous driving, according to one embodiment of the present invention.

[0036] Referring to FIG. 1, a vehicle control device (100) according to one embodiment of the present invention may be implemented inside or outside a vehicle, and some of the components included in the vehicle control device (100) may be implemented inside or outside the vehicle. In this case, the vehicle control device (100) may be formed integrally with the internal control units of the vehicle, or it may be implemented as a separate device and connected to the control units of the vehicle by a separate connection means. For example, the vehicle control device (100) may further include components not shown in FIG. 1.

[0037] A vehicle control device (100) according to one embodiment may include a processor (110), a LiDAR (120, LiDAR), and a memory (130). The processor (110), the LiDAR (120), or the memory (130) may be electrically and / or operably coupled with each other by an electronic component including a communication bus.

[0038] In the following, the hardware being operatively coupled may include direct and / or indirect connections between the hardware being established via wired and / or wireless connections so that the second hardware is controlled by the first hardware among the hardware.

[0039] Although illustrated in different blocks, the embodiments are not limited thereto. For example, some of the hardware of FIG. 1 may be included in a single integrated circuit including a system on a chip (SoC). The type and / or number of hardware included in the vehicle control unit (100) are not limited to those illustrated in FIG. 1. For example, the vehicle control unit (100) may include only some of the hardware illustrated in FIG. 1.

[0040] A vehicle control device (100) according to one embodiment may include hardware for processing data based on one or more instructions. For example, the hardware for processing data may include a processor (110). For example, the hardware for processing data may include an arithmetic and logic unit (ALU), a floating point unit (FPU), a field programmable gate array (FPGA), a central processing unit (CPU), and / or an application processor (AP). The processor (110) may have a structure of a single-core processor or a structure of a multi-core processor including a dual core, a quad core, a hexa-core, or an octa core.

[0041] According to an embodiment, the processor (110) may include at least one of a GPU (graphic processing unit), an NPU (neural processing unit), or any combination thereof. For example, the GPU may be referred to as a VPU (visual processing unit). For example, the NPU may be referred to as a neural network processing unit.

[0042] A vehicle control device (100) according to one embodiment may include a depth sensor for detecting external objects. For example, the depth sensor for detecting external objects may include at least one of a time of flight (ToF) sensor, a light detection and ranging (LiDAR) sensor (120), a structured light sensor, an ultrasonic sensor, an infrared sensor, a radio detection and ranging (RADAR) sensor, an optical distance sensor, or any combination thereof. For convenience of explanation, the following description will focus on the LiDAR sensor.

[0043] A vehicle control device (100) according to one embodiment may include a lidar (120) that acquires a plurality of points based on a pulsed laser signal. For example, the lidar (120) may acquire data sets that identify surrounding objects of the vehicle control device (100) (or a vehicle including the vehicle control device (100)). For example, the lidar (120) may identify the position, direction of motion, speed, or at least one of any combination thereof of surrounding objects based on the pulsed laser signal emitted from the lidar (120) being reflected back from surrounding objects.

[0044] For example, the lidar (120) can acquire data sets representing external objects in a space formed by the x-axis, y-axis, and z-axis based on pulsed laser signals reflected from surrounding objects. For example, the lidar (120) can acquire data sets including multiple points in a space formed by the x-axis, y-axis, and z-axis based on receiving pulsed laser signals at specified intervals. For example, the multiple points may include points representing external objects in a three-dimensional virtual coordinate system. The three-dimensional virtual coordinate system may include at least one of a vehicle coordinate system, a lidar coordinate system, or any combination thereof. However, examples of the three-dimensional virtual coordinate system are not limited to those described above.

[0045] A memory (130) of a vehicle control device (100) according to one embodiment may include a hardware component for storing data and / or instructions that are input and / or output to a processor (110) of the vehicle control device (100). For example, the memory (130) may include a volatile memory including a random-access memory (RAM) and / or a non-volatile memory including a read-only memory (ROM).

[0046] For example, volatile memory may include at least one of DRAM (dynamic RAM), SRAM (static RAM), Cache RAM, PSRAM (pseudo SRAM), or any combination thereof. For example, non-volatile memory may include at least one of PROM (programmable ROM), EPROM (erasable PROM), EEPROM (electrically erasable PROM), flash memory, hard disk, compact disk, SSD (solid state drive), eMMC (embedded multi-media card), or any combination thereof.

[0047] In the memory (130) of the vehicle control unit (100), one or more instructions (or commands) representing operations and / or actions to be performed by the processor (110) of the vehicle control unit (100) based on data may be stored. A set of one or more instructions may be referred to as a program, firmware, operating system, process, routine, sub-routine, and / or application.

[0048] In the following, the statement that an application is installed in the vehicle control unit (100) means that one or more instructions provided in the form of an application are stored in memory (130), and that one or more applications are stored in an executable format (e.g., a file having an extension specified by the operating system of the vehicle control unit (100)) by the processor (110) of the vehicle control unit (100).

[0049] For example, the memory (130) may include a first neural network model for detecting objects. For example, the memory (130) may include a second neural network model for outputting the type of a plurality of points acquired by the lidar (120) and / or the score of the plurality of points.

[0050] In one embodiment, the processor (110) may obtain at least one of a first virtual box for representing a target object, a first class representing the type of a target object, or any combination thereof, based on a plurality of points obtained through the lidar (120) and a first neural network model stored in memory (130).

[0051] For example, the processor (110) may obtain at least one of a first virtual box for representing a target object, a first class representing the type of the target object, or any combination thereof, based on inputting a plurality of points into a first neural network model. For example, the first neural network model may include an object detection model. For example, the target object may include an external object located within a specified distance from the vehicle control device (100) (or a self-vehicle including the vehicle control device (100)). For example, the target object may include an object identified by the vehicle control device (100) and continuously tracked. For example, the type of the target object may include a plurality of types for classifying the target object. For example, the type of the target object may include at least one of a first type representing the ground, a second type representing a type different from the ground, or any combination thereof. However, the type of the target object is not limited to the above. For example, the type of the target object may include at least one of a third type representing a person, a fourth type representing a vehicle, or any combination thereof, but is not limited thereto.

[0052] In one embodiment, the processor (110) may obtain, based on a plurality of points and a second neural network model, at least one of first partial points corresponding to at least a part of a target object among the plurality of points, a second class identified through the first partial points and indicating the type of the target object, or any combination thereof. For example, the second neural network model may include a segmentation model.

[0053] For example, the second neural network model may include a neural network model for obtaining the types of multiple points and the scores of multiple points.

[0054] For example, the processor (110) can obtain first partial points corresponding to at least a part of the target object among the multiple points based on inputting multiple points into the second neural network model. For example, the processor (110) can identify the types of the multiple points based on inputting multiple points into the second neural network model. For example, the processor (110) can obtain first partial points corresponding to at least a part of the target object among the multiple points based on the type of each of the multiple points.

[0055] In one embodiment, the processor (110) may perform a first designated algorithm on a plurality of points. For example, the processor (110) may perform a first designated algorithm for classifying the type of each of the plurality of points. For example, the processor (110) may classify second part points corresponding to a designated type among the plurality of points. For example, the designated type may include a type representing a surface.

[0056] For example, the processor (110) can classify second part points corresponding to a specified type based on performing a first specified algorithm on a plurality of points, and obtain (or identify) first part points by excluding the second part points from among the plurality of points.

[0057] In one embodiment, the processor (110) may obtain at least one of a subclass for obtaining a second class, a score for each of the multiple points, or any combination thereof, based on inputting a plurality of points into a second neural network model. For example, the processor (110) may obtain a subclass and a score for each of the multiple points based on inputting a plurality of points into a second neural network model. For example, the subclass may include classifying each of the multiple points into an arbitrary type.

[0058] For example, the processor (110) can fuse a partial class, the score of each of a plurality of points, and a second partial point. For example, the processor (110) can perform clustering based on the fusion of the partial class, the score of each of a plurality of points, and the second partial point. For example, clustering may include grouping first partial points corresponding to at least a part of a target object.

[0059] For example, the processor (110) can obtain a point cloud for generating a second virtual box based on first partial points. For example, the processor (110) can obtain a point cloud based on grouping the first partial points.

[0060] For example, the processor (110) may generate a second virtual box for representing a target object, which is different from the first virtual box, based on a point cloud. For example, the second virtual box may include a box that includes at least some of the first partial points.

[0061] For example, the processor (110) can identify a heading direction indicating the direction of progress of a target object based on at least one of first partial points, a point cloud, or any combination thereof.

[0062] For example, the processor (110) can identify the location of the second virtual box in a virtual coordinate system based on at least one of the first partial points, a point cloud, or any combination thereof. For example, the processor (110) can identify the size of the second virtual box based on at least one of the first partial points, a point cloud, or any combination thereof. For example, the processor (110) can identify the second class based on at least one of the first partial points, a point cloud, or any combination thereof. For example, the processor (110) can identify the heading direction indicating the direction of travel of the target object, the location of the second virtual box in a virtual coordinate system, the size of the second virtual box, the second class, or at least one combination thereof based on at least one of the first partial points, a point cloud, or any combination thereof. For example, the processor (110) can identify the heading direction of the bounding box based on at least one of the first virtual box, the first class, the heading direction of the second virtual box, the location of the second virtual box, the size of the second virtual box, the second class, or any combination thereof. For example, the processor (110) can identify the location of the bounding box in a virtual coordinate system based on at least one of the first virtual box, the first class, the heading direction of the second virtual box, the location of the second virtual box, the size of the second virtual box, the second class, or any combination thereof. For example, the processor (110) can obtain a third class representing the type of the target object corresponding to the bounding box based on at least one of the first virtual box, the first class, the heading direction of the second virtual box, the location of the second virtual box, the size of the second virtual box, the second class, or any combination thereof.For example, the processor (110) can obtain the heading direction of the bounding box, the position of the bounding box in a virtual coordinate system, a third class representing the type of the target object corresponding to the bounding box, or at least one combination thereof, based on at least one of the first virtual box, the first class, the heading direction of the second virtual box, the position of the second virtual box, the size of the second virtual box, the second class, or any combination thereof.

[0063] For example, the processor (110) may assign a first identifier to the second virtual box for tracking the second virtual box. For example, the processor (110) may assign a second identifier corresponding to the first identifier to the bounding box.

[0064] For example, the processor (110) can track bounding boxes using a second identifier. For example, the processor (110) can track a target object based on identifying a plurality of bounding boxes, including a bounding box to which the second identifier is assigned, in a plurality of frames. For example, since the second identifier is an identifier assigned to a bounding box corresponding to the target object, the processor (110) can track a target object by identifying a plurality of bounding boxes to which the second identifier is assigned in a plurality of frames.

[0065] In one embodiment, the processor (110) may output a bounding box corresponding to a target object based on at least one of a first virtual box, a first class, first partial points, a second class, or any combination thereof. For example, the bounding box may include an example in which the target object is represented in a virtual coordinate system in the form of a hexahedron.

[0066] Hereinafter, the operations performed by the CPU, GPU, and / or NPU included in the processor (110) are briefly described.

[0067] In one embodiment, the processor (110) may include at least one of a CPU, a GPU, an NPU, or any combination thereof. For example, at least one of a GPU, an NPU, or any combination thereof may obtain a first virtual box and a first class based on a first neural network model. For example, at least one of a GPU and an NPU may obtain a first virtual box and a first class. For example, at least one of a GPU, an NPU, or any combination thereof may obtain a subclass and a score for each of a plurality of points for obtaining a second class based on a second neural network model. For example, at least one of a GPU and an NPU may obtain a subclass and a score for each of a plurality of points for obtaining a second class based on a second neural network model. For example, the CPU may classify second subpoints corresponding to a specified type among a plurality of points based on performing a first specified algorithm for classifying the type of each of a plurality of points for a plurality of points.

[0068] As described above, a vehicle control device (100) according to one embodiment may include at least one processor (110). The vehicle control device (100) can accurately detect a target object by detecting the target object using at least one processor (110). In addition, by performing parallel processes, the vehicle control device (100) can reduce the load on each processor.

[0069] To add one important point regarding FIG. 1, the autonomous driving sensor (120) applied to the vehicle control device (100) according to one embodiment of the present invention is not limited to LiDAR but may be radar (120). As previously mentioned, at the point where object recognition technology fused with various autonomous driving sensors is introduced, there is no need to limit the autonomous driving sensor (120) to necessarily being only a type of LiDAR. Therefore, for the autonomous driving sensor (120), the reference number will be used interchangeably as either LiDAR (120) or radar (120) as needed. In fact, whether it is LiDAR or radar, it is merely one of many sensors that can be used for autonomous driving.

[0070] Meanwhile, prior to examining Figure 2, to gain an overall understanding of AI object recognition according to the present invention, an explanation regarding object recognition based on autonomous driving sensors is added as follows.

[0071] The object recognition process by the autonomous driving sensor (120) and AI module goes through three stages: preprocessing, segmentation, and tracking.

[0072] The object recognition system (1000, see FIG. 6) according to the present invention undergoes a pre-processing process before executing an object recognition function. The pre-processing may include, for example, removing points forming the ground based on laser / radio wave (or other) sensing data (i.e., unprocessed raw data) input from an autonomous driving sensor (120). Since lasers / radio waves reflected from the ground may cause a misperception that an object exists on the ground, a process of distinguishing between the ground and non-ground is performed in the pre-processing stage, and if necessary, can also be performed in the segmentation stage.

[0073] In short, preprocessing in AI object recognition can be understood as a process in which an image processing tool of an AI module within a processor (110) removes noise from a point cloud image and reduces the total number of points present in the point cloud image through techniques such as voxel downsampling to improve computational efficiency.

[0074] For reference, point cloud images can also be displayed using the BEV (Bird's Eye View) method. When an autonomous driving sensor map is created to resemble the scene seen through a bird's eyes as it looks at a city while flying, it is referred to as a BEV image.

[0075] That is, the autonomous driving sensor (120) transmits lasers / radio waves, etc., to the surrounding environment as described above, and by recording the time when the lasers / radio waves are reflected back from external objects, it generates a point (i.e., a point) for each of the numerous signals and calculates the distance to that point. By repeating the emission of numerous lasers / radio waves in this way, the processor (110) can generate a real-time map of the surrounding environment as a BEV-type 3D map, and if necessary, can also generate a 2D map.

[0076] Lines or surfaces that appear black and white in a point cloud map are actually composed of countless points (each of these points is generated by a LiDAR / radar (120)), and for this reason, autonomous driving sensing images are also called point cloud images. Of course, for example, if an RGB-D (Red, Green, Blue - Depth) sensor and a LiDAR sensor are combined, the LiDAR point cloud image can be recreated in color.

[0077] Although it is difficult for humans to recognize objects based on just one of the numerous points within a point cloud image, if the point cloud is viewed comprehensively—for example, from the perspective of a BEV or a 2D plan—one can roughly estimate what the surrounding environment of a vehicle currently driving autonomously looks like. Furthermore, it is possible to recognize objects such as cars, buses, pedestrians, street trees, and traffic signs present within the point cloud image. In AI image recognition technology, these objects or people are referred to as "Objects," and each Object can be classified into a Class belonging to a group of specific properties, such as a car class or a bus class.

[0078] Distinguishing whether an object within a point cloud image belongs to the car or bus class requires the assistance of a deep AI neural network. To locate objects within the point cloud image and identify their classes using the AI ​​neural network, AI training must be performed beforehand.

[0079] For example, PANDASET TM The dataset (Dataset, source: https: / / pandaset.org / #data-collection) contains over 48,000 camera images (mostly taken in the Silicon Valley area of ​​the United States) and over 16,000 LiDAR scan images, with a total of 28 classes, including pedestrians, passenger cars, bicycles, construction site signs, and traffic signs, annotated in the form of annotations.

[0080] In addition, LiDAR point cloud images are Open3D TM(Source: https: / / www.open3d.org / ) and point cloud working tools can be used to visualize the options desired by the user. Since LiDAR (120) is capable of distance detection, for example, Open3D TM 3D LiDAR images can be reproduced more realistically by visualizing distant objects in dark blue and close objects in light blue during visual processing.

[0081] In addition, as mentioned above, the original image (i.e., Raw Data) can be preprocessed by applying a technique called voxel downsampling to the point cloud image. Here, a voxel refers to a cube-shaped 3D pixel, and voxel downsampling is a technique that reduces the number of points without requiring excessive AI computation while maintaining the structure of various objects included in the point cloud.

[0082] Meanwhile, the autonomous driving sensor (120) emits, for example, m lasers / radio waves n times during one scan cycle, and in this case, the laser / radio wave scan values ​​that return after colliding with an external object form an (mxn) matrix, and this (mxn) matrix data is called a range image. Each point constituting the point cloud image contains depth (i.e., range) information, and in addition to this, it includes the intensity, azimuth, inclination, and other additional information of the returned laser pulse. Regarding the range image, for example, WOD (Waymo TM AI training is possible using large datasets such as Open Datasets.

[0083] Range View (RV) refers to a technique that converts a 3D point cloud into a 2D or, in short, 2.5D scene, representing it as a 3D map based on autonomous driving sensors that humans can intuitively understand, much like an analog drawing, instead of numerous points. In a range view image, the 3D point cloud image acquires 2D coordinates, but the 3D laser-related information (angle, tilt, intensity, etc.) recorded when obtaining the range image is not discarded. By applying a variable called width to the (x, y) coordinate among the (x, y, z) coordinate values ​​of the 3D image to obtain the coordinate of one axis in 2D, and by applying range image information representing the range (depth) and a variable called height to the (z) coordinate to obtain the coordinate of another axis in 2D, a 2D range view image is created.

[0084] To add further details, the AI ​​algorithm according to the present invention may include a Convolutional Neural Network (CNN), which is an AI training module frequently used to extract features (or feature points) from image data. For this purpose, there is a dataset consisting of tens of thousands of commercially available images, and currently, there are versions of CNN capable of processing images ranging from 1D to 3D. That is, the result of the range view image processing tool is trained by the CNN neural network to perform the function of helping the AI ​​accurately recognize objects within the image.

[0085] Since objects surrounding an autonomous vehicle are ultimately recognized by the machine, generating Ground Truth (GT) bounding boxes on the aforementioned autonomous driving sensor map is also a crucial process in object recognition. In machine learning, Ground Truth (GT) is a term used to represent the original or actual values ​​of the data that the AI ​​intends to learn. Generally, it can be viewed as a type of image annotation applied to a point cloud image, appearing as a bounding box with a box-shaped boundary.

[0086] In other words, for the AI ​​module to recognize objects, it retrieves labels and groups various objects. Of course, it also sets the spacing of 3D data points used to output GT bounding boxes, and can set it so that about 50 to 1000 cloud points are included within a single GT bounding box.

[0087] Of course, such GT annotations do not exist in the raw data captured by sensors such as radar (120, or LiDAR) while the vehicle is in motion. The processor (110) must recognize objects belonging to various classes, such as road signs, crosswalks, pedestrians, other vehicles, and center lanes, as objects. GT annotations are a means used to measure object recognition errors and sometimes evaluate AI performance by comparing the object determination results recognized by the AI ​​algorithm of the processor (110) with the actual results. GT bounding boxes applied to the original image in the form of annotations can be set manually by the user, but there are also commercially available GT calculation tools, such as Grid-striding.

[0088] Meanwhile, when the AI ​​object recognition module is operated, the predicted bounding box can also be verified. The result of the processor (110) recognizing an object of a specific class from the original image data obtained from the sensor (120), etc., appears in the form of another bounding box similar to the GT bounding box. Unlike the GT bounding box, the predicted bounding boxes are the result of the autonomous driving AI's calculations. The predicted bounding box may match the GT bounding box, but it may not match the GT bounding box or there may be no overlapping area at all.

[0089] For reference, since predicted bounding boxes alone cannot definitively determine that an object of a specific class actually exists at a certain location, they are generally referred to as P-Boxes (Probability Boxes) or Predicted Bounding Boxes.

[0090] Segmentation processing, which is performed after preprocessing, refers to marking specific parts of a road (e.g., traffic lights) in red and the rest (asphalt roads) in blue. Clustering point clouds into specific groups and generating P-Boxes can also be done during the segmentation stage. For reference, there is also a technique called cluster expansion, where the expansion target can include all points within an epsilon distance (the minimum distance for constituting a cluster) from a seed point.

[0091] During segmentation processing, clustering based on point clouds and P-Box generation are performed. In short, the AI ​​network performing the segmentation is used to obtain point labels from the autonomous driving sensor (120).

[0092] In the case of the present invention, a "Rule-based" road surface recognition and label fusion technique is proposed to solve the problem of ground recognition errors that occur during segmentation. For this road surface recognition algorithm, any of the various techniques, such as slope-based road surface recognition, grid-based road surface recognition, or other non-planar-based road surface recognition, may be applied.

[0093] For reference, semantic segmentation is the process of assigning a unique class label to each point within a point cloud generated by an autonomous driving sensor (120). In image processing technology composed of points, semantic segmentation is a technique that finds and uses meaningful information from autonomous driving sensor data for object recognition or scene reconstruction necessary for implementing autonomous driving; various semantic segmentation AI models already exist, such as projection-based methods, point-based methods, and sparse convolution-based methods. For example, the semantic segmentation result is that the processor (110) NVIDIA DRIVE 쪠 It may be the result of AI computations performed with the AGX system, and through this configuration, various colors can be added to, for example, a point cloud image.

[0094] Sensor images undergo a process called post-processing following the segmentation process described above. Post-processing refers to the conversion of point cloud data into 3D maps or models, which are meaningful information for autonomous driving. Post-processing includes the final removal of noise or errors from point cloud images, the recognition of objects such as vehicles or pedestrians from the point clouds, and, if necessary, the registration of these objects by attaching unique identifiers to the point cloud information. Assigning confidence scores to each object or P-Box recognized by AI may also be performed during the post-processing stage.

[0095] Now, with reference to FIG. 2, the core algorithm of the present invention will be described. FIG. 2 is a flowchart showing, in its entirety, an algorithm (200) that recognizes an object and evaluates reliability by an AI module and an autonomous driving sensor according to an embodiment of the present invention. The algorithm (200) of FIG. 2 can be implemented and executed, for example, as an AI module mounted on the processor (100) of FIG. 1, and the system (1000) that recognizes an object and evaluates reliability by an AI module and an autonomous driving sensor according to the present invention can correspond to the computing system (1000) of FIG. 6 including the algorithm (200) of FIG. 2.

[0096] However, in the process of explaining FIG. 2, FIG. 3 to FIG. 5 will be additionally referred to for convenience of explanation as necessary. FIG. 3 is a drawing for explaining the components of a reliability score newly defined according to an embodiment of the present invention. FIG. 4a and FIG. 4b are drawings showing exemplary driving scenes to explain a differential reliability calculation method for a region of interest and a region of non-interest according to an embodiment of the present invention. FIG. 5 is a drawing showing actual experimental results to which an object recognition and reliability evaluation technique according to an embodiment of the present invention is applied.

[0097] In step S100 of FIG. 2, the AI ​​module according to the present invention performs a preprocessing operation to recognize an object using the autonomous driving sensor (120, radar, etc.) described above in FIG. 1 and a prediction operation to generate a P-Box.

[0098] In step S200, it is determined whether track information (including lanes and driving path) regarding the vehicle currently driving (450, see FIG. 4a and FIG. 4b) has been updated during the object tracking process after step S100. If the result of the determination is NO, the reliability of the object is evaluated according to the conventional method in step S300 and the procedure is terminated. For reference, when calculating the reliability score in step S300, the reliability can be calculated by setting, for example, the maximum reliability score to 100 and then subtracting the value obtained by multiplying it by a certain weight (e.g., 0.1, etc.) from the reliability value calculated by the AI.

[0099] If the result of the judgment in Step S200 is YES, the process moves to Step S400 to calculate the overall reliability score. Referring to Fig. 3, the overall reliability calculated in Step S400 is actually calculated by a process of summing or weighting scores excluding the interest track score and non-interest track score, which are the core of the present invention, namely the Age score, Heading stability score, combination score (of various sensors), and other association scores (referring to a process of selecting a cluster with attributes most similar to oneself for an individual prediction object and creating a new object using a cluster that has no similarity to any prediction object). The remaining interest / non-interest track scores are calculated in subsequent procedures.

[0100] In other words, regarding the "temporary reliability score" calculated in Step S400, a determination of whether it is an autonomous vehicle track is made according to Step S500. That is, the present invention is configured to pre-set a Region of Interest (ROI) and rapidly increase the reliability score for tracks that fall within this ROI in order to increase the priority of AI computation.

[0101] That is, if it is determined in the judgment of step S400 that the object exists in the same lane as the self-vehicle lane and is located within the ROI area, then the judgment is made YES and the process moves to step S600 to perform the calculation of a confidence score with the introduction of the "first weight."

[0102] Even if a NO judgment is made in step S400, if the object (other vehicle) being analyzed exists in the left or right lane of the track of the vehicle (450) and is within the ROI area, a YES judgment is made in step S700 and the process proceeds to step S800 to perform a reliability score calculation using the "second weight" according to the present invention.

[0103] In short, for the algorithm (200) according to the present invention, the first type of interest track reliability score for an object existing in the current lane of a driving vehicle (450) is, for example Defined as such and calculated according to the following mathematical formula 1, for objects existing in the left and right lanes of the driving vehicle (450), the second type of interest track reliability score Define and then calculate according to the following mathematical formula 2.

[0104] <Mathematical Formula 1>

[0105]

[0106] <Mathematical Formula 2>

[0107]

[0108] Here, is the above-mentioned first weight, and is the above-mentioned second weight, and is the maximum confidence score.

[0109] That is, in steps S600 and S800, for objects located within the ROI area and in their own lane or the lanes to the left or right, the maximum confidence score (e.g., 100) is multiplied by a first weight (e.g., 0.1) and a second weight (e.g., 0.05), and the result is added to the original confidence score to calculate Type 1 and Type 2 confidence scores, respectively. This will affect the priority of vehicle object recognition and autonomous driving control following the above steps, and ultimately result in increasing the importance in AI computation for "track of interest" objects that are within the ROI and satisfy the conditions of step S500 or step S700. This will be of great help in preventing situations where the object to be controlled is not controlled or is controlled incorrectly.

[0110] In fact, conventionally, when calculating the reliability score of an object generated and tracked based on radar, for example, the same standard algorithm was applied to all detection areas; however, the present invention introduces a reliability judgment algorithm different from the conventional one through steps S500 to S800.

[0111] If a NO result is received in Step S700, the process proceeds to the "Non-Interest Track Reliability Score" calculation algorithm step starting from Step S900.

[0112] That is, in step S900, it is determined whether an object is located outside the road boundary (i.e., reference number 430 in FIG. 4a and 4b). For example, vehicles (403, 406, 409) in FIG. 4a and vehicles (413, 417, 420) in FIG. 4b will receive a YES judgment in step S900. In the present invention, as in step S900, the non-interest area is defined as an area outside the road boundary, and whether an object belongs there is defined as the first non-interest condition.

[0113] Even if a NO judgment is received in Step S900, the process moves to Step S1100 to determine whether the object is greater than a certain vertical distance from the preceding vehicle. That is, in Step S1100, a second non-interest condition is defined based on whether the object belongs to an area that is vertically farther than the preceding vehicle, which is an object located in the current lane of the current vehicle (450) or in the left or right lanes of the current lane.

[0114] Even if a NO judgment is received in Step S1100, it is determined whether the object is stationary or an unidentified object. That is, even if a NO judgment is received in both Step S900 and Step S1100, if the object is stationary (e.g., movement can be identified by the radar's Doppler effect) or unidentified (e.g., an object with an unclear class classification), it is determined that the third non-interest condition is satisfied, which considers the object to belong to a non-interest area.

[0115] If a YES judgment is received in steps S900, S1100, and S1200, it is considered an object that may interfere with AI computation, and a band filter is applied in step S1000, after which tracking of these objects is terminated. For reference, the band filter applied in step S1000 is such that, for example, even if the confidence score calculated in step S400 for an object that received a YES judgment in steps S900, S1100, and S1200 is 70 points, an upper limit of the score is artificially applied to an uncontrollable level of 40 points, and consequently, the object that received a YES judgment in steps S900, S1100, and S1200 becomes an uncontrollable object due to the artificially applied upper limit and becomes a target for filtering by the band filter.

[0116] Of course, if a NO judgment is received in all of steps S900, S1100, and S1200, the process moves to step S1300 to calculate the standard reliability judgment as the final reliability score and terminates the procedure. That is, in the preceding steps S600 and S800, for objects existing within the ROI area and located in their own lane or the lanes to the left and right, the maximum reliability score (e.g., 100) is multiplied by a first weight (e.g., 0.1) and a second weight (e.g., 0.05), and the result is "summed" to the original reliability score to calculate the Type 1 and Type 2 reliability scores, respectively, whereas in step S1300, the above summing process does not exist.

[0117] FIG. 4a is a diagram relating to the process of calculating the reliability score of the track of interest in steps S500 to S800. In the driving situation (400a) exemplified in FIG. 4a, a total of 10 vehicles (401, 402, 403, 404, 405, 406, 407, 408, 409, 410) exist on the road in addition to the self-vehicle (450). As illustrated in FIG. 4a, the vehicle (407), which is the closest vehicle in the same lane as the self-vehicle (450), becomes the first priority object of interest to which the first weight is applied, and the vehicles closest to the self-vehicle (408) in the left and right lanes (i.e., vehicle (402) and vehicle (408)) are selected as the second priority objects of interest to which the second weight is applied.

[0118] FIG. 4b is a diagram relating to the process of calculating the reliability score of a non-interest track according to steps S900 to S1200. In the driving situation (400b) exemplified in FIG. 4b, the ROI (440) can be identified, and the objects satisfying the first non-interest condition described above, that is, objects existing outside the road boundary (430), would be three vehicles (413, 417, 420). These are first-priority non-interest targets. For reference, all other areas shown in FIG. 4b, excluding the ROI (440) box, can be considered as non-interest areas.

[0119] For reference, in the case of vehicle (414), it is longer in longitudinal distance than the vehicle closest to the left lane relative to its own vehicle (450) (i.e., 412), but is shorter in longitudinal distance than the first priority object of interest (vehicle (418) in FIG. 4b), so it is excluded from the non-object of interest. The longitudinal distance in the vertical direction can be measured, for example, based on the point at the rear of the vehicle shown in FIG. 4b. If excluded from the non-object of interest, there is a possibility of a cut-in.

[0120] For reference, in Fig. 4b, the second-ranked vehicles of interest are two vehicles (412, 419).

[0121] Meanwhile, objects that are longer in the vertical direction than the object of interest are second-ranked objects of non-interest. In FIG. 4b, two vehicles (415, 416) would fall into this category. Although not shown in FIG. 4b, stationary or unidentified objects are third-ranked objects of non-interest.

[0122] As discussed above, the present invention proposes a strategy that differentiates reliability calculation methods based on the level of interest, rather than applying absolute standards to all detection areas. Through this, when autonomous driving sensors, such as radar, detect objects, those objects gain priority in AI computation and control.

[0123] In addition, during the post-processing process described above, the present invention deletes objects falsely recognized by autonomous driving sensors (120), such as radar, and manages the lifecycle, movement attributes, and class of radar-tracked objects. Of course, the present invention also performs the role of estimating optimal current location / speed information using objects predicted as objects and clusters associated therewith.

[0124] The algorithm (200) applied in the present invention aims to calculate the reliability of the final generated, for example, radar track differentially according to the degree of interest of the area, so it will be implemented in the reliability evaluation module, which can be said to be the last module after all states and positions of the radar track are estimated.

[0125] Finally, referring to Fig. 5, the reliability score of the track with a longer longitudinal distance than the track of the right leading vehicle, which has a high cut-in probability and is likely to be selected as a control target, is calculated to be much higher in the left (a) diagram of Fig. 5.

[0126] On the other hand, looking at the right side (b) of Fig. 5, the reliability score for each area of ​​interest is calculated by increasing the reliability score for the track that is the right leading vehicle selected as the track of interest, and suppressing the reliability score for the track that is farther from the track of interest to a maximum value, for example, 40 (a value where control cannot be used).

[0127] The reliability score of the tracking object of the autonomous driving sensor (120, radar, etc.) is used in a strategy to determine the output priority of the radar track to be used during multi-sensor fusion, and helps in selecting the control track by increasing the priority of the target track that requires control during autonomous driving.

[0128] Accordingly, according to the present invention, by lowering the priority of objects that are relatively far away or outside the area of ​​interest (such as objects outside the road boundary), the effect of reducing the influence of control on tracks that are not in the area of ​​interest can be expected.

[0129] FIG. 6 is a block diagram showing a computing system (1000) for autonomous driving vehicle control and object recognition computation according to one embodiment of the present invention.

[0130] Referring to FIG. 6, the computing system (1000) may include at least one processor (1100), memory (1300), user interface input device (1400), user interface output device (1500), storage (1600), and network interface (1700) connected via a bus (1200).

[0131] The processor (1100) may be a semiconductor device that executes processing on instructions stored in a central processing unit (CPU) or in memory (1300) and / or storage (1600). The memory (1300) and storage (1600) may include various types of volatile or non-volatile storage media. For example, the memory (1300) may include ROM (Read Only Memory) and RAM (Random Access Memory).

[0132] Accordingly, the steps of the method or algorithm described in connection with the embodiments disclosed in this specification may be directly implemented in hardware, software modules, or a combination of both, executed by the processor (1100). The software modules may reside in storage media such as RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs (i.e., memory (1300) and / or storage (1600)).

[0133] An exemplary storage medium is coupled to a processor (1100), and the processor (1100) can read information from the storage medium and write information to the storage medium. Alternatively, the storage medium may be integrated with the processor (1100). The processor and the storage medium may reside within an application-specific integrated circuit (ASIC). The ASIC may reside within a user terminal. Alternatively, the processor and the storage medium may reside as separate components within the user terminal.

[0134] The above description is merely an illustrative explanation of the technical concept of the present invention, and those skilled in the art to which the present invention pertains will be able to make various modifications and variations within the scope of the essential characteristics of the present invention.

[0135] Accordingly, the embodiments disclosed in this invention are intended to illustrate, not limit, the technical concept of the invention, and the scope of the technical concept of the invention is not limited by these embodiments. The scope of protection of this invention shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of this invention.

Claims

Claim 1 An AI object recognition system comprising: a processor having an object recognition module as part of the AI ​​module for recognizing objects existing in a surrounding environment from data input from the autonomous driving sensor; and a memory for storing point cloud data, wherein the AI ​​module performs the steps of: clustering a plurality of points corresponding to objects in the surrounding environment acquired from the autonomous driving sensor to calculate a natural number N predicted objects; and assigning a reliability score to each of the N predicted objects, wherein the reliability score is assigned differentially to each of the N predicted objects based on the Region of Interest (ROI) of the vehicle in autonomous driving. Claim 2 An AI object recognition system according to claim 1, wherein the reliability score includes a track of interest reliability score calculated within the ROI according to a first weight applied to the current lane of the vehicle or a second weight applied to the left and right lanes of the current lane. Claim 3 In claim 2, the interest track reliability score for the object existing in the current lane is calculated according to the following mathematical formula 1, and the interest track reliability score for objects existing in the left and right lanes An AI object recognition system characterized by each being calculated according to the following mathematical formula 2.[Mathematical Formula 1] [Mathematical Formula 2] (Here, is the above-mentioned first weight, and is the above-mentioned second weight, and is the maximum confidence score) Claim 4 An AI object recognition system according to claim 1, characterized in that the reliability score includes a non-interest track reliability score calculated for an object existing in a non-interest region outside the ROI. Claim 5 An AI object recognition system characterized in that, in claim 4, the above-mentioned non-interest area is set according to a first non-interest condition that the area is outside the road boundary. Claim 6 An AI object recognition system according to claim 5, wherein the non-interest area is set according to a second non-interest condition that the non-interest area is an area that is further in vertical distance than a preceding vehicle, which is an object existing in the current lane of the vehicle or in the left and right lanes of the current lane. Claim 7 An AI object recognition system according to claim 6, characterized in that even if an object exists in an area with a vertical distance closer than the preceding vehicle, if it is a stationary object or an unidentified object, it is determined to satisfy a third non-interest condition in which it is considered an object belonging to the non-interest area. Claim 8 An AI object recognition system characterized in that, in any one of claims 4 to 7, the non-interest track reliability score is determined by an artificially limited upper limit as a predetermined uncontrollable level for the maximum reliability score for an object existing in the non-interest area. Claim 9 An AI object recognition system characterized by applying a band-pass filter to objects having a reliability level below the uncontrollable level in claim 8. Claim 10 An AI object recognition system according to claim 8, characterized in that for an object having a reliability level exceeding the uncontrollable level, the reliability score assigned to the object is finally recognized as the reliability score in the differential manner. Claim 11 A method for recognizing an object by an AI module and an autonomous driving sensor, comprising: a step in which the AI ​​module clusters a plurality of points corresponding to objects in the surrounding environment acquired from the autonomous driving sensor to produce a natural number of predicted objects N; and a step of assigning a confidence score to each of the N predicted objects, wherein the confidence score is assigned differentially to each of the N predicted objects based on the Region of Interest (ROI) of the autonomous driving vehicle. Claim 12 An AI object recognition method according to claim 11, wherein the reliability score includes a track of interest reliability score calculated within the ROI based on a first weight applied to the current lane of the vehicle or a second weight applied to the left and right lanes of the current lane. Claim 13 In claim 12, the interest track reliability score for the object existing in the current lane is calculated according to the following mathematical formula 1, and the interest track reliability score for objects existing in the left and right lanes An AI object recognition method characterized by each being calculated according to the following mathematical formula 2.[Mathematical Formula 1] [Mathematical Formula 2] (Here, is the above-mentioned first weight, and is the above-mentioned second weight, and is the maximum confidence score) Claim 14 An AI object recognition method according to claim 11, characterized in that the reliability score includes a non-interest track reliability score calculated for an object existing in a non-interest region outside the ROI. Claim 15 An AI object recognition method according to claim 14, characterized in that the non-interest area is set according to a first non-interest condition that the non-interest area is an area outside the road boundary. Claim 16 An AI object recognition method according to claim 15, wherein the non-interest area is set according to a second non-interest condition that the non-interest area is an area that is further in vertical distance than a preceding vehicle, which is an object existing in the current lane of the vehicle or in the left and right lanes of the current lane. Claim 17 An AI object recognition method according to claim 16, characterized in that even if an object exists in an area with a vertical distance closer than the preceding vehicle, if it is a stationary object or an unidentified object, it is determined to satisfy a third non-interest condition in which it is considered an object belonging to the non-interest area. Claim 18 An AI object recognition method according to any one of claims 14 to 17, wherein the non-interest track reliability score is determined by an artificially limited upper limit as a predetermined uncontrollable level for the maximum reliability score for an object existing in the non-interest region. Claim 19 An AI object recognition method according to claim 18, characterized by applying a band-pass filter to objects having a reliability level below the uncontrollable level. Claim 20 In claim 18, an AI object recognition method characterized by finally recognizing the reliability score assigned to an object as the reliability score in a differential manner for objects having a reliability level exceeding the uncontrollable level.

Images