Object recognition data processing device for autonomous vehicles and method for implementing the same

Abstract

This invention provides an object recognition data processing device for autonomous vehicles and a method for implementing the same, which can enhance real-time performance and increase the efficiency of autonomous driving systems. [Solution] The object recognition data processing device for autonomous vehicles according to the present invention includes a lidar sensor that generates distance information by measuring the time it takes for a laser pulse to be emitted and reflected back from surrounding objects, and converts the distance information into raw point cloud data; an autonomous driving controller that provides the raw point cloud data; and a computing module that is mounted on an autonomous vehicle in the form of an embedded board, and when it receives the raw point cloud data from the autonomous driving controller, it processes the raw point cloud data to generate and store object recognition data, and provides the object recognition data to a control center server.

Description

[Technical Field]

[0001] The present invention relates to an object recognition data processing device for an autonomous vehicle and a method for implementing the same, and more specifically, to an object recognition data processing device for an autonomous vehicle and a method for implementing the same, which enables object recognition from data acquired by a lidar sensor by including a separate device in the form of an embedded board for storing raw data acquired by a lidar sensor.

[0002] This deliverable is the result of an initiative supported by the 2025 Gyeongsangbuk-do Regional Innovation-Centered University Support System (RISE) - (Regional Growth Innovation LAB). (Project Management Number OOO) [Background technology]

[0003] Autonomous driving technology is a technology that allows vehicles to recognize and judge the road environment on their own and control their driving. It must recognize and respond to objects and obstacles around the vehicle in real time through a variety of sensors and algorithms.

[0004] In particular, LiDAR (Light Detection and Ranging) sensors are core sensors for autonomous vehicles, providing high-resolution 3D spatial information about the environment around the vehicle. These LiDAR sensors primarily emit laser pulses, receive reflected signals, calculate distances, and based on this, represent the position and shape of surrounding objects in the form of 3D point cloud data.

[0005] However, the raw point cloud data obtained from numerous lidar sensors is enormous in volume, and processing this data in real time places a significant burden on the system. The raw point cloud data consists of millions of points, each containing not only spatial coordinates but also reflectivity, temporal information, and more, resulting in a very large amount of data. Such large volumes of data overload the autonomous driving control system in the vehicle and significantly impact its real-time data processing performance.

[0006] Existing autonomous driving systems primarily employ a method where all data collected from the lidar sensor is transmitted to the autonomous driving controller, which then filters out unnecessary data to recognize the object. However, this method leads to the following problems.

[0007] Firstly, autonomous driving controllers must process a large amount of data within the vehicle, and raw point cloud data provided by lidar sensors accounts for a significant portion of this data. Processing such large volumes of data in real time requires high-performance hardware, and slow processing speeds can negatively impact the safety and driving performance of autonomous vehicles.

[0008] Secondly, the raw point cloud data collected from the lidar sensor contains a lot of unnecessary information (e.g., the sky, small stones on the road surface, distant objects unrelated to the vehicle) in addition to the information that is absolutely necessary for object recognition. Processing all of this unnecessary information would waste computational resources and make it difficult to quickly extract only the necessary information.

[0009] Thirdly, a structure in which the autonomous driving controller processes all data to perform object recognition increases system complexity and raises hardware and software development costs. Furthermore, such complexity increases the likelihood of system errors and exacerbates maintenance difficulties.

[0010] Fourth, autonomous vehicles are systems that require high responsiveness and must react quickly to their surroundings. However, if there is a delay in the process of processing all the LiDAR data, problems can arise with real-time responsiveness. This increases the likelihood that the autonomous vehicle may not be able to respond appropriately to unexpected situations, potentially leading to an accident.

[0011] As mentioned above, in order to improve the sophistication of autonomous vehicles, it is extremely important that the vehicle accurately perceives its surrounding environment and makes accurate decisions and controls regarding its driving based on that perception.

[0012] For this purpose, advanced recognition technologies that utilize surround sensors (such as cameras, radars, lidars, etc.) are essential, and such recognition technologies highly depend on the quality of the artificial intelligence learning data used in the autonomous driving system. In particular, lidar sensors provide high-resolution 3D point cloud raw data to assist the autonomous driving vehicle in accurately recognizing surrounding objects.

[0013] However, in current autonomous driving systems, it is difficult to process the large volume of raw data collected from lidar sensors in real time. Since lidar sensors generate approximately 70 MB or more of data per second, it is computationally burdensome for the autonomous driving controller to process all such data.

[0014] Therefore, existing systems integrate lidar data with other sensors (e.g., cameras, radars) to perform sensor fusion, and then delete the data unnecessary for driving, leaving only the important data. In this process, the lidar raw data is deleted, resulting in the problem that the accurate object recognition data using this data does not remain.

[0015] For the commercialization of autonomous driving at level 3 or higher, it is necessary to be able to integrally manage the object information and traffic situation data around the vehicle collected in real time in cooperation with the control center. However, the information currently provided to the control center is limited to the vehicle's status data (PVD) and video data in snapshot form, and in fact, the provision of data that accurately reflects the surrounding objects and traffic situation is insufficient through this. Summary of the Invention Problems to be Solved by the Invention

[0016] The present invention aims to provide an object recognition data processing device for an autonomous driving vehicle and an execution method thereof, which can significantly reduce the computational burden of the autonomous driving controller by extracting and storing only the necessary object recognition information in a separate preprocessing system instead of the autonomous driving controller processing all the large-capacity data collected from the rider sensor, thereby enhancing real-time performance and increasing the efficiency of the autonomous driving system.

[0017] In addition, the present invention aims to provide an object recognition data processing device for an autonomous driving vehicle and an execution method thereof, which can contribute to improving the accuracy of object recognition and more precisely recognizing the surrounding environment of the autonomous driving vehicle by extracting and storing only the object information from the rider data in real time.

[0018] In addition, the present invention aims to provide an object recognition data processing device for an autonomous driving vehicle and an execution method thereof, which can manage traffic road conditions, traffic flow, event situations such as accidents, etc. in real time by providing the preprocessed object recognition data to the control center in real time, thereby strengthening the safety and response ability of the autonomous driving system.

[0019] In addition, the present invention aims to provide an object recognition data processing device for an autonomous driving vehicle and an execution method thereof, which can provide a basis for further accurately learning the recognition algorithm of the autonomous driving system by utilizing the stored object recognition data as artificial intelligence learning data for enhancing autonomous driving, and further developing the autonomous driving technology.

[0020] In addition, the present invention aims to provide an object recognition data processing device for an autonomous driving vehicle and an execution method thereof, which can strengthen the connection between vehicle data and road infrastructure and contribute to the commercialization of autonomous driving because it can store and provide not only rider data but also GPS RTK data and vehicle state data (PVD).

[0021] The objectives of the present invention are not limited to those mentioned above, and other objectives and advantages of the present invention not mentioned can be understood from the following description and will be more clearly understood from the embodiments of the present invention. Furthermore, it will be readily apparent that the objectives and advantages of the present invention can be embodied by the means and combinations thereof set forth in the claims. [Means for solving the problem]

[0022] To achieve this objective, an object recognition data processing device for autonomous vehicles includes a lidar sensor that generates distance information by measuring the time it takes for a laser pulse to reflect back from surrounding objects and convert the distance information into raw point cloud data; an autonomous driving controller that provides the raw point cloud data; and a computing module that is mounted on the autonomous vehicle in the form of an embedded board, and upon receiving the raw point cloud data from the autonomous driving controller, processes the raw point cloud data to generate and store object recognition data, and provides the object recognition data to a control center server.

[0023] In one embodiment, the computing module can group points in the point cloud raw data to form objects, extract features of the objects, identify the type of object using a deep learning model, and then store it as object recognition data.

[0024] In one embodiment, the computing module can apply a density-based clustering algorithm to the point cloud raw data to define the contour of an object and store the object's shape, size, and relative position as object recognition data.

[0025] Furthermore, a method for processing object recognition data for an autonomous vehicle to achieve this objective includes the steps of: generating distance information by measuring the time it takes for a laser pulse emitted by a lidar sensor to reflect back from surrounding objects and converting the distance information into raw point cloud data; an autonomous driving controller receiving the raw point cloud data from the lidar sensor and providing it to a computing module mounted on the autonomous vehicle in the form of an embedded board; when the computing module receives the raw point cloud data from the autonomous driving controller, processing the raw point cloud data to generate and store object recognition data; and the computing module providing the object recognition data to a control center server.

[0026] In one embodiment, the step of processing the point cloud raw data to generate and store object recognition data includes the step of grouping points in the point cloud raw data to form objects, and the step of extracting the features of the objects, identifying the type of object using a deep learning model, and then storing it as object recognition data.

[0027] In one embodiment, the step of processing the point cloud raw data to generate and store object recognition data may include the step of applying a density-based clustering algorithm to the point cloud raw data to define the contour of an object and storing the object's shape, size, and relative position as object recognition data. [Effects of the Invention]

[0028] As described above, according to the present invention, instead of the autonomous driving controller processing all of the large amount of data collected from the lidar sensor, a separate preprocessing system extracts and stores only the necessary object recognition information, thereby significantly reducing the computational burden on the autonomous driving controller. This has the advantage of improving real-time performance and increasing the efficiency of the autonomous driving system.

[0029] Furthermore, according to the present invention, by extracting only object information from the LiDAR data, storing it in real time, and providing it, it has the advantage of improving the accuracy of object recognition and contributing to more precise recognition of the environment surrounding the autonomous vehicle.

[0030] Furthermore, according to the present invention, by providing pre-processed object recognition data to the control center in real time, road traffic conditions, traffic flow, and event situations such as accidents can be managed in real time, which has the advantage of enhancing the safety and responsiveness of the autonomous driving system.

[0031] Furthermore, the present invention has the advantage of providing a foundation for further developing autonomous driving technology by utilizing the stored object recognition data as artificial intelligence learning data for improving autonomous driving, thereby allowing the recognition algorithm of the autonomous driving system to learn more accurately.

[0032] Furthermore, according to the present invention, not only rider data but also GPS RTK data and vehicle status data (PVD) can be stored and provided together, which has the advantage of strengthening the linkage between vehicle data and road infrastructure and contributing to the commercialization of autonomous driving. [Brief explanation of the drawing]

[0033] [Figure 1] This is a network configuration diagram illustrating an object recognition data processing system for an autonomous vehicle according to one embodiment of the present invention. [Figure 2] This is a network configuration diagram illustrating the internal structure of an object recognition data processing device for an autonomous vehicle according to one embodiment of the present invention. [Figure 3] This is a block diagram illustrating the internal structure of an object recognition data processing device for autonomous vehicles. [Figure 4] This is a flowchart illustrating one embodiment of the object recognition data processing method for autonomous vehicles according to the present invention. [Modes for carrying out the invention]

[0034] The aforementioned objectives, signatures, and advantages will be described in detail below with reference to the attached drawings, and accordingly, a person with ordinary skill in the art to which the present invention pertains will be able to readily implement the technical idea of ​​the present invention. In describing the present invention, detailed explanations of prior art related to the present invention will be omitted if it is judged that such explanations may unnecessarily obscure the gist of the present invention. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. In the drawings, the same reference numerals are used to indicate the same or similar components.

[0035] Figure 1 is a network configuration diagram illustrating an object recognition data processing system for an autonomous vehicle according to one embodiment of the present invention.

[0036] Referring to Figure 1, the object recognition data processing system for autonomous vehicles includes autonomous vehicles 100_1 to 100_N, a base station 200, and a control center server 300.

[0037] The autonomous vehicle 100_1 processes raw data through a separate device in the form of an embedded board, and recognizes and classifies objects in its surroundings in real time.

[0038] The autonomous vehicle 100_1 collects raw data in the form of point cloud raw data through LiDAR sensor data and provides it to a computing module in the form of an embedded board formed on the autonomous vehicle 100_1. The computing module then preprocesses the raw data to remove noise and extracts object features (size, position, reflectivity, etc.).

[0039] The computing module recognizes all objects around the autonomous vehicle 100_1 based on pre-processed data, and classifies the characteristics of each object by obtaining information such as its position, size, direction of movement, and speed. At this time, the computing module stores the information about the objects on an embedded board so that the object recognition data can be used by the autonomous driving system.

[0040] In one embodiment, the computing module provides pre-processed object information to the base station 200 via the OBU (On-Board Unit). This communication is conducted via C-V2X (Cellular Vehicle-to-Everything) or WAVE (Wireless Access in Vehicular Environment) technology, and the base station 200 transmits the object recognition data transmitted by the autonomous vehicle 100_1 to the control center server 300 in real time.

[0041] The base station 200 receives object recognition data, traffic condition information, and vehicle location information transmitted by the autonomous vehicle 100_1 through bidirectional communication with the On-Board Unit (OBU) of the autonomous vehicle 100_1, and provides this information to the control center server 300.

[0042] The base station 200 transmits traffic information and hazardous factors analyzed by the control center to other autonomous vehicles in the vicinity, helping them make safe driving decisions. For example, the base station 200 can transmit accident information that occurred in a specific section to other vehicles, helping them to slow down or change their route.

[0043] Furthermore, the base station 200 can collect and transmit real-time traffic information in conjunction with road infrastructure such as traffic lights, road signs, and traffic cameras. Through this, the road environment can be managed more safely and efficiently.

[0044] Furthermore, the base station 200 provides location-based services based on vehicle location data. For example, it collects GPS information from vehicles and uses this to analyze and share traffic conditions occurring at specific locations in real time. Through this, if traffic congestion occurs in a specific area, the base station 200 can detect it and send warning messages to nearby autonomous vehicles 100_2 to 100_N or suggest alternative routes.

[0045] Furthermore, the base station 200 can recognize hazardous elements based on information received from the autonomous vehicle 100_1 and the control center server 300, and transmit safety warning messages to surrounding autonomous vehicles 100_2 to 100_N. For example, the base station 200 can contribute to accident prevention by transmitting information such as vehicles suddenly stopping ahead, obstacles on the road, and pedestrian intrusions in real time.

[0046] The control center server 300 can analyze real-time traffic conditions by integrating various object recognition data, traffic condition information, and location information of autonomous vehicles 100_1 collected through the base station 200. For example, the control center server 300 can optimize traffic flow by collecting information on obstacles and accidents on the road, and can respond immediately to emergency situations.

[0047] The control center server 300 collects traffic information transmitted by the base station 200 and autonomous vehicles 100_1 in real time to monitor the overall road conditions. For example, the control center server 300 analyzes information such as congested areas, accident locations, vehicle movement routes, and obstacles to understand the real-time traffic conditions. This monitoring function helps ensure that vehicles on the road can travel safely and enables the rapid detection of sudden situations occurring in specific areas.

[0048] Furthermore, the control center server 300 analyzes the data collected in real time to identify hazardous elements such as traffic accidents, road obstacles, and emergency situations, and immediately notifies the autonomous vehicle 100_1 of this to prevent accidents. For example, when an obstacle or accident suddenly occurs on the road, the control center server 300 transmits this information to the nearby autonomous vehicle 100_1 and the base station 200, helping surrounding autonomous vehicles 100_2 to 100_N to react quickly.

[0049] Furthermore, the control center server 300 analyzes road congestion and presents the optimal route to vehicles to optimize traffic flow. If congestion is expected in sections with heavy traffic or during specific time periods, the control center server 300 will suggest alternative routes or adjust vehicle speeds to alleviate traffic congestion. Through this, it is possible to support the smooth flow of traffic across the entire road network and reduce travel time.

[0050] Furthermore, the control center server 300 manages object recognition data transmitted from the base station 200 and autonomous vehicle 100_1, and distributes necessary object information to surrounding autonomous vehicles 100_2 to 100_N to support cooperative driving among the autonomous vehicles 100_1. For example, the control center server 300 provides object recognition data such as pedestrians, other vehicles, and obstacles on the road to each autonomous vehicle 100_1 to 100_N, enabling the autonomous vehicles 100_1 to 100_N to drive while maintaining a safe distance. Through this, surrounding autonomous vehicles 100_2 to 100_N can predict the situation on the road and prevent accident risks in advance.

[0051] Furthermore, the control center server 300 expands its traffic management capabilities by integrating with road infrastructure such as traffic signal systems, road signs, and CCTV. If traffic volume surges at a particular intersection, the control center server 300 can adjust signals to smooth vehicle flow and enhance pedestrian safety. Additional visual data can be analyzed through CCTV and traffic camera footage on the road to identify situations that are difficult to grasp from vehicle sensor information alone.

[0052] Figure 2 is a network diagram illustrating the internal structure of an object recognition data processing device for an autonomous vehicle according to one embodiment of the present invention. Figure 3 is a block diagram illustrating the internal structure of an object recognition data processing device for an autonomous vehicle.

[0053] Referring to Figures 2 and 3, autonomous vehicles 100_1 to 100_N include a lidar sensor 110, an autonomous driving controller 120, and a computing device 130.

[0054] The lidar sensor 110 emits a laser pulse and measures the time it takes for the light to reflect off surrounding objects and return, thereby calculating the distance to the object. This is called ToF (Time of Flight), and because the speed of light is known, the distance can be accurately measured by the time it takes for the pulse to travel and return.

[0055] In one embodiment, the lidar sensor 110 rapidly emits laser light in various directions, and when it receives the laser light that has been reflected back from an object, it measures the time it takes for the laser to reach the object and for the reflected light to return to the sensor, and calculates the distance to the object.

[0056] In the above embodiment, the lidar sensor 110 converts distance information collected through numerous laser pulses and their reflections into three-dimensional data called point cloud raw data. This data enables the autonomous vehicle to perceive the surrounding environment in 3D. The point cloud raw data is very high-resolution raw data directly collected by the lidar sensor and is a collection of numerous points containing information about objects around the vehicle, their distances, and their positions.

[0057] The aforementioned raw data is extremely large in volume, and processing all of it in real time places a heavy computational burden on the autonomous driving controller 120. The amount of data generated by the lidar can exceed 70 MB per second, which causes problems with processing and storing large amounts of data.

[0058] When the autonomous driving controller 120 receives raw data from the lidar sensor 110, it transmits the raw data to a separate device in the form of an embedded board. In this process, in addition to lidar data, GPS RTK data and vehicle status data (PVD) may be included as needed.

[0059] The aforementioned GPS RTK (Real-Time Kinematic) data is a technology that provides extremely accurate location information and is used to track the position of autonomous vehicles in real time. RTK GPS typically provides positional accuracy within a few centimeters, making it essential for autonomous driving systems.

[0060] The aforementioned vehicle status data represents the physical state of the vehicle, including its speed, acceleration, steering angle, and braking status. This data helps to understand the vehicle's current driving condition and supports safe and efficient operation based on this information.

[0061] A separate device in the form of an embedded board is a dedicated computing module 130 installed inside the autonomous vehicle. The computing module 130 preprocesses the raw data of the LiDAR received from the autonomous driving controller in real time.

[0062] The computing module 130 uses the point cloud raw data from the autonomous driving controller 120 to determine whether something is an object and classifies the object.

[0063] First, the computing module 130 performs filtering on the raw point cloud data. That is, the raw point cloud data may contain noise generated by external factors or sensor errors. Therefore, the computing module 130 uses a filtering algorithm to remove such noise.

[0064] In one embodiment, the computing module 130 also removes points that are more than a certain distance away from the point cloud raw data (e.g., objects in the distant background) and anomalous points (e.g., abnormal values ​​due to sensor errors).

[0065] The computing module 130 subdivides the point cloud raw data into individual objects. This is the process of grouping closely located points to form each object. For example, the DBSCAN (Density-Based Clustering) algorithm can be used.

[0066] In one embodiment, the computing module 130 applies a density-based clustering algorithm to point cloud raw data to analyze the shape based on the cluster points and define the contour of the object. Through this, the computing module 130 can determine the shape, size, and relative position of the object.

[0067] In another embodiment, the computing module 130 performs downsampling by applying a voxel grid filter after separating the object from the background. Point cloud raw data consists of millions of 3D points, such as LiDAR data collected by autonomous vehicles. However, processing such a massive amount of data in real time places a heavy computational burden on the system, so a voxel grid filter is used to simplify the data and retain only the necessary information.

[0068] The aforementioned downsampling is not simply about reducing the amount of data, but rather a process that removes unnecessary or redundant data while preserving important information. For example, an object that is far away from a vehicle and for which detailed information is no longer needed can be represented by a small number of points. In this process, techniques such as voxel grid filtering are used to divide the data into a grid structure, and then only representative points are selected within each grid to reduce the size of the data.

[0069] First, the computing module 130 performs a process of dividing the 3D space where the cloud data exists into small 3D cubes (lattices). At this time, the 3D cube is a voxel, which is a regular hexahedron with a certain size, and the three-dimensional space is divided according to the resolution (the size of the voxel) specified by the user. At this time, the smaller the size of the voxel, the finer the data is maintained, and the larger the size, the more data is removed.

[0070] After the 3D space is divided into voxels by the computing module 130, it checks all the points existing in each voxel, analyzes all the points inside the voxel, and selects a representative point.

[0071] In one embodiment, the computing module 130 determines a specific point as the representative point based on the reflection intensity of the lidar point cloud raw data within the voxel. At this time, the computing module 130 can select the representative point according to [Equation 1].

[0072] [Equation 1] JPEG2026085897000002.jpg43127

[0073] P i、selected : The point with the maximum reflection intensity within Vi,

[0074] V i : Voxel,

[0075] P i :{P i1 、P i2 、…、P in}

[0076] I ij : The reflection intensity of each point Pij,

[0077] The computing module 130, as in [Equation 1], after finding the point (P i ) with the highest reflection intensity within each voxel (V ij ), for each voxel (V iThe point selected for (P) i、selected Only the points with high reflectivity remain, and the remaining points are removed. In this way, only the points with high reflectivity are selected as representative points within each voxel.

[0078] For these reasons, the data collected by the lidar sensor includes reflectivity information, and points with high reflectivity mainly represent the surface of an object, while points with low reflectivity are likely to be the background. Therefore, the computing module 130 can use a method to preferentially select points with high reflectivity as representative points.

[0079] In another embodiment, the computing module 130 determines a specific point as a representative point based on the height of the ridapoint cloud raw data within the voxel. In this case, the computing module 130 can select the representative point using [Mathematical Formula 2].

[0080] [Mathematical formula 2] JPEG2026085897000003.jpg17127

[0081] P i、filtered : The set of points whose height satisfies the minimum criteria,

[0082] V i : Voxel,

[0083] H ij :P ij Height (i.e., Z coordinate)

[0084] P ij :{P i1 , P i2 ..., P in}

[0085] In other words, the computing module 130, as shown in [Mathematical Equation 2], has each voxel (V i After extracting the set of points whose height satisfies the minimum criterion within ), the set of points (P i、filtered) allows us to select the average coordinates of points that satisfy the height criteria as a representative point, as shown in [Mathematical Formula 3].

[0086] [Mathematical formula 3] JPEG2026085897000004.jpg34127

[0087] P i、selected : The average coordinates for the point whose height satisfies the minimum criterion.

[0088] P ij :{P i1 , P i2 ..., P in}

[0089] P i、filtered : The set of points whose height satisfies the minimum criteria,

[0090] The reason for this is that if we can distinguish between the background (e.g., roads) and the object (e.g., vehicles, people) primarily based on the difference in height from the ground, then we can select only points within each voxel that are above a certain height as representative points. This is useful for recognizing and removing parts parallel to the ground as part of the background.

[0091] Subsequently, the computing module 130 converts the collected point cloud raw data into the vehicle's coordinate system. This is because the lidar sensor data is measured relative to the location where the sensor is installed, and therefore must be converted to match the autonomous vehicle's world coordinate system or map coordinate system.

[0092] Therefore, the computing module 130 applies rotation and translation to convert the coordinates of each point collected by the lidar sensor into the vehicle's reference coordinate system. Through this, the lidar data is combined with data collected by other sensors (camera, radar, etc.) to enable integrated recognition.

[0093] The computing module 130 distinguishes different objects in the point cloud raw data.

[0094] First, the computing module 130 uses clustering techniques to group points that are spatially close together into a single object. Typical techniques include Euclidean clustering and the RANSAC algorithm. Euclidean clustering groups points within a specific radius into a single cluster based on the distance between them, while RANSAC is useful for identifying and removing planes such as roads and walls.

[0095] Subsequently, the computing module 130 extracts the features of the object in each cluster.

[0096] In one embodiment, the computing module 130 calculates the difference between the minimum and maximum coordinates of an object using a bounding box as shown in [Mathematical Equation 4] in order to determine the size of the object from the ridapoint cloud raw data.

[0097] [Mathematical formula 4] JPEG2026085897000005.jpg52127

[0098] L: Length of the object,

[0099] W: width of the object,

[0100] H: Height of the object,

[0101] x max : The maximum value of the x-coordinate relative to the object,

[0102] x min : The minimum x-coordinate relative to the object,

[0103] y max : The maximum value of the y-coordinate relative to the object,

[0104] y min : The minimum value of the y-coordinate relative to the object.

[0105] zmax : The maximum value of the z-coordinate relative to the object,

[0106] z min : Minimum value of the z-coordinate relative to the object,

[0107] In the above embodiment, the computing module 130 can enclose the space occupied by each object's points in the form of a bounding box, and calculate the size of each object through the bounding box.

[0108] Furthermore, the computing module 130 can determine the center point by calculating the average of each coordinate, as shown in [Mathematical Equation 5].

[0109] [Mathematical formula 5] JPEG2026085897000006.jpg8389

[0110] n: Number of points in the point cloud's original data set.

[0111] (x c , y c , z c ): center point,

[0112] Furthermore, the computing module 130 estimates the object's velocity and direction using raw point cloud data from multiple frames. At this time, the computing module 130 calculates the object's velocity and direction vector based on the movement of the center point of the raw point cloud data from multiple frames, according to [Mathematical Equations 5] to [Mathematical Equation 7].

[0113] JPEG2026085897000007.jpg23166

[0114] [Mathematical formula 6] JPEG2026085897000008.jpg19127△c: translation vector,

[0115] JPEG2026085897000009.jpg18140

[0116] JPEG2026085897000010.jpg17166

[0117] [Mathematical formula 7] JPEG2026085897000011.jpg20127

[0118] V: speed, △c: movement vector, △t: time interval,

[0119] JPEG2026085897000012.jpg15140

[0120] JPEG2026085897000013.jpg14166

[0121] [Mathematical formula 8] JPEG2026085897000014.jpg5777

[0122] d: Movement direction, △c: movement vector,

[0123] Subsequently, the computing module 130 calculates the reflection intensity of the i-th point inside the object using [mathematical formula 9].

[0124] [Mathematical formula 9] JPEG2026085897000015.jpg5089

[0125] I avg : Average value of reflectance

[0126] As described above, after the computing module 130 extracts the features of the objects, it uses a machine learning or deep learning model to identify what kind of object each cluster is, and then stores only the object information.

[0127] Figure 4 is a flowchart illustrating one embodiment of the object recognition data processing method for autonomous vehicles according to the present invention.

[0128] Referring to Figure 4, the LiDAR sensor emits a laser pulse, measures the time it takes for the pulse to reflect back from surrounding objects, generates distance information, and converts this distance information into raw point cloud data (step S410).

[0129] The autonomous driving controller receives the point cloud raw data from the lidar sensor and provides it to a computing module mounted on the autonomous vehicle in the form of an embedded board (step S420).

[0130] When the computing module receives the raw point cloud data from the autonomous driving controller, it processes the raw point cloud data to generate and store object recognition data (step S430).

[0131] In one embodiment of step S430, the computing module groups points in the point cloud raw data to form objects, extracts features of the objects, identifies the type of object using a deep learning model, and then stores it as object recognition data.

[0132] In another embodiment for step S430, the computing module applies a density-based clustering algorithm to the point cloud raw data to define the contours of objects and stores the shape, size, and relative position of the objects as object recognition data.

[0133] The computing module provides the object recognition data to the control center server (step S440).

[0134] As described above with reference to the embodiments and drawings, the present invention is not limited to the embodiments described above, and various modifications and variations can be made from such descriptions by those with ordinary skill in the art to which the present invention pertains. Therefore, the spirit of the present invention should be understood solely by the claims described below, and any equivalent or comparable variations can be said to fall within the scope of the spirit of the present invention. [Explanation of symbols]

[0135] 100_1~100_N: Autonomous vehicles 200:Base station 300: Control Center Server 110: Lida Sensor 120: Autonomous driving controller 130: Computing equipment

Claims

1. A lidar sensor that generates distance information by measuring the time it takes for a laser pulse to reflect back from surrounding objects after being emitted, and converts this distance information into raw point cloud data; An autonomous driving controller that provides the aforementioned point cloud source data; and An object recognition data processing device for an autonomous vehicle, characterized in that it is mounted on an autonomous vehicle in the form of an embedded board, and includes a computing module that, upon receiving raw point cloud data from the autonomous driving controller, processes the raw point cloud data to generate and store object recognition data, and provides the object recognition data to a control center server.

2. The computing module The object recognition data processing device for an autonomous vehicle according to claim 1, characterized in that it groups points in the raw point cloud data to form objects, extracts features of the objects, identifies the type of object using a deep learning model, and then stores it as object recognition data.

3. The computing module The object recognition data processing device for an autonomous vehicle according to claim 2, characterized in that it applies a density-based clustering algorithm to the aforementioned point cloud raw data to define the contour of an object and stores the shape, size, and relative position of the object as object recognition data.

4. The LiDAR sensor emits a laser pulse, measures the time it takes for the pulse to reflect back from surrounding objects, generates distance information, and converts this distance information into raw point cloud data; The steps include: the autonomous driving controller receiving the point cloud raw data from the lidar sensor and providing it to a computing module mounted on the autonomous vehicle in the form of an embedded board; and When the computing module receives point cloud raw data from the autonomous driving controller, it processes the point cloud raw data to generate and store object recognition data; A method for processing object recognition data for an autonomous vehicle, characterized by including the step of the computing module providing the object recognition data to a control center server.

5. The step of processing the aforementioned point cloud raw data to generate and store object recognition data is The step of grouping points in the aforementioned point cloud raw data to form an object; and The method for processing object recognition data for an autonomous vehicle according to claim 4, characterized in that it includes the step of extracting the features of the object, identifying the type of the object using a deep learning model, and then storing it as object recognition data.

6. The step of processing the aforementioned point cloud raw data to generate and store object recognition data is The method for processing object recognition data for an autonomous vehicle according to claim 4, characterized by including the step of applying a density-based clustering algorithm to the point cloud raw data to define the contour of an object and saving the shape, size, and relative position of the object as object recognition data.

Images