Apparatus, method and computer program for estimating velocity of object using radar sensor

Abstract

A device for estimating the speed of an object using a radar sensor includes a driving information derivation unit that derives driving information of an object located on a target road through a radar sensor installed on a target road, an initial value acquisition unit that selects one cell corresponding to the driving information of the object based on an initial value learning model including a plurality of cells pre-learned for the target road and acquires at least one parameter information corresponding to the one cell, and a speed calculation unit that calculates the actual driving speed of the object based on the at least one parameter information.

Description

Technology Field

[0001] The present invention relates to an apparatus, method, and computer program for estimating the speed of an object using a radar sensor. Background Technology

[0002] A traffic control system refers to a system that maintains optimal traffic conditions by intensively managing traffic flow, such as increases or decreases in traffic volume, and automatically controlling the flashing times of traffic lights.

[0003] Such traffic control systems primarily utilize radar sensors, which are excellent for determining vehicle speed, distance, angle, etc. Regarding technology for controlling roads and traffic using radar sensors, the prior art Korean Registered Patent No. 10-0987177 discloses a road control method and apparatus using radar.

[0004] However, while it is possible to measure the vehicle's position and radial velocity using signal processing technology and the Doppler effect via radar sensors, it has the disadvantage that it is not possible to determine the longitudinal and lateral velocity components constituting the vehicle's actual speed from a single frame.

[0005] To improve this, a technology is being developed to estimate the actual speed of a vehicle detected by a radar sensor using a Kalman filter. A Kalman filter refers to an algorithm that more accurately extracts the actual signal by estimating the noise included in the measured value from the measured target. The Kalman filter operates in two stages. The first stage is the prediction step, which calculates the predicted value of the next state and the prediction error covariance using the system's current state and the error covariance. The second stage is the update step, which corrects the predicted state using new measurements and updates the error covariance to improve the accuracy of the estimation. Through this process, the actual speed of the vehicle can be estimated by more accurately updating the system state using the Kalman filter. The technology for estimating the actual speed of a vehicle using a Kalman filter will be explained in detail through Figure 1.

[0006] FIG. 1 is a diagram illustrating the process of estimating the speed of an object in a conventional speed estimation device. Referring to figure (a) of FIG. 1, a radar sensor (100) transmits a radar signal and receives a reflected signal reflected from a vehicle (110), thereby obtaining two-dimensional spatial information and one-dimensional speed information (radial velocity) of the vehicle (110). Here, radial velocity refers to the speed when an object moves in the direction of the line of sight.

[0007] The radar sensor (100) is the position of the vehicle in two-dimensional space and one-dimensional line-of-sight velocity ( It can provide ), but this provides only limited information about the actual 2D velocity (here, If information obtained using a radar sensor is used directly as the initial value for the prediction step of the Kalman filter, the radar sensor provides only limited information about the actual two-dimensional velocity, which leads to a degradation in the performance of the Kalman filter.

[0008] To improve this, when the speed of the vehicle (110) is estimated based on information from multiple initial frames, information loss occurs for the duration of the number of frames. Consequently, if the initial value of the Kalman filter is inaccurate, the speed estimation does not converge to the actual speed value of the vehicle (110) but diverges.

[0009] This is because the Kalman filter has a limitation in that it relies heavily on initial values; therefore, if the initial values ​​are inaccurate, performance degradation occurs, such as increased convergence time, decreased accuracy of estimates, reduced responsiveness and stability, and the persistence of incorrect estimates.

[0010] In other words, to predict the next position of a vehicle moving in a two-dimensional space, two-dimensional space and two-dimensional velocity information are required, but there is a limitation of the Kalman filter in that performance degradation occurs if the initial value information of the Kalman filter is inaccurate.

[0011] To improve this, when using pseudo-sensors such as cameras and Lidar, the pseudo-sensors have the disadvantage of not being able to collect sufficient information to estimate the actual speed of the vehicle. Specifically, since pseudo-sensors such as cameras and Lidar do not have information about the speed of the vehicle, speed information must be estimated by utilizing the vehicle's position information for more than two frames, which requires more than double the amount of information and processing speed to be processed in a single frame.

[0012] Referring to Figure (b), when there are multiple signals reflected from a single vehicle (110), a technique has been developed to estimate the actual speed of the vehicle using only the information from one frame by solving the positional difference and line of sight velocity of the multiple signals simultaneously.

[0013] However, referring to Figure (c), when the actual speed of a vehicle is estimated using only the information from one frame by solving the positional difference and line-of-sight velocity of multiple signals as in Figure (b), a disadvantage arises in that there is a large loss of information due to an error in the vehicle (110) moving perpendicular to the radiation direction of the radar sensor (100). Therefore, when using a Kalman filter, performance degradation occurs if the information of the initial value is inaccurate, so a method is required to estimate the initial value of the two-dimensional speed more accurately. The problem to be solved

[0014] The present invention provides a speed estimation device, method, and computer program that derives driving information of an object located on a target road through a radar sensor installed on the target road, selects one cell corresponding to the driving information of the object based on an initial value learning model including a plurality of cells pre-learned for the target road, and obtains at least one parameter information corresponding to one cell.

[0015] The present invention aims to provide a speed estimation device, method, and computer program that calculates the actual driving speed of an object based on at least one parameter information.

[0016] However, the technical problems that this embodiment aims to solve are not limited to the technical problems described above, and other technical problems may exist. means of solving the problem

[0017] As a means to achieve the technical problem described above, one embodiment of the present invention may provide a speed estimation device comprising: a driving information derivation unit that derives driving information of an object located on a target road through a radar sensor installed on the target road; an initial value acquisition unit that selects one cell corresponding to the driving information of the object based on an initial value learning model including a plurality of cells that have been learned for the target road and acquires at least one parameter information corresponding to one cell; and a speed calculation unit that calculates the actual driving speed of the object based on the at least one parameter information.

[0018] Another embodiment of the present invention may provide a speed estimation method comprising the steps of: deriving driving information of an object located on a target road through a radar sensor installed on the target road; selecting one cell corresponding to the driving information of the object based on an initial value learning model comprising a plurality of cells pre-learned for the target road; obtaining at least one parameter information corresponding to the one cell; and calculating the actual driving speed of the object based on the at least one parameter information.

[0019] According to another embodiment of the present invention, a computer program stored in a medium comprising a sequence of instructions that, when executed by a computing device, derive driving information of an object located on a target road through a radar sensor installed on the target road, select one cell corresponding to the driving information of the object based on an initial value learning model comprising a plurality of cells pre-learned for the target road, obtain at least one parameter information corresponding to the one cell, and calculate the actual driving speed of the object based on the at least one parameter information.

[0020] The above-described means for solving the problem are merely exemplary and should not be interpreted as intended to limit the invention. In addition to the exemplary embodiments described above, additional embodiments described in the drawings and the detailed description of the invention may exist. Effects of the invention

[0021] According to any one of the means for solving the problem of the present invention described above, a speed estimation device, method, and computer program can be provided, which derive driving information of an object located on a target road through a radar sensor installed on a target road, select one cell corresponding to the driving information of the object based on an initial value learning model including a plurality of cells pre-learned for the target road, and obtain at least one parameter information corresponding to one cell.

[0022] Conventionally, only the position and line-of-sight velocity of an object could be determined through a Kalman filter, but the present invention can provide a speed estimation device, method, and computer program capable of calculating the actual driving speed of an object based on at least one parameter information.

[0023] Conventionally, when attempting to estimate the velocity of an object using a Kalman filter, it was difficult to accurately estimate the actual velocity due to the limitation that the Kalman filter relies on initial values. However, a velocity estimation device, method, and computer program can be provided that enable the accurate estimation of the actual velocity of an object through the learning of an initial value learning model.

[0024] A velocity estimation device, method, and computer program can be provided that can be universally utilized in fixed-installation pseudo-sensors that detect dynamic objects such as cameras and lidar, and can complement and improve the performance of radar sensors by allowing the pseudo-sensor to estimate precise initial values ​​using a Kalman filter or a modified Kalman filter. Brief explanation of the drawing

[0025] Figure 1 is a diagram illustrating the process of estimating the speed of an object in a conventional speed estimation device. FIG. 2 is a configuration diagram of a speed estimation device according to one embodiment of the present invention. FIG. 3 is an exemplary diagram illustrating the process of generating an initial value learning model according to an embodiment of the present invention. FIG. 4 is an exemplary diagram illustrating the process of performing learning on any one cell when the first parameter is false according to one embodiment of the present invention. FIG. 5 is an exemplary drawing illustrating a histogram of an angle-weighted average value obtained through the completion of learning of an initial value learning model according to one embodiment of the present invention. FIG. 6 is an exemplary drawing illustrating a histogram in which the availability of data is determined based on an angle-weighted average according to an embodiment of the present invention. FIG. 7 is an exemplary drawing for explaining the process of calculating the radius-weighted average and the angle-weighted average for any one cell, respectively, according to an embodiment of the present invention. FIG. 8 is an exemplary drawing illustrating a preset radius-weighted average value according to one embodiment of the present invention. FIG. 9 is an exemplary drawing comparing the line of sight speed of an object and the actual speed of an object according to one embodiment of the present invention. FIG. 10 is a flowchart of a method for estimating the speed of an object using a radar sensor according to an embodiment of the present invention. FIG. 11 is a flowchart of a method for calculating the actual driving speed of an object based on a first parameter and a second parameter corresponding to any one cell according to an embodiment of the present invention. Specific details for implementing the invention

[0026] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0027] Throughout the specification, when a part is described as being "connected" to another part, this includes not only cases where they are "directly connected" but also cases where they are "electrically connected" with other elements interposed between them. Furthermore, when a part is described as "including" a component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components, and it should be understood that this does not preclude the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0028] In this specification, the term "part" includes a unit realized by hardware, a unit realized by software, and a unit realized using both. Additionally, one unit may be realized using two or more hardware, and two or more units may be realized by one hardware.

[0029] Some of the operations or functions described in this specification as being performed by a terminal or device may instead be performed by a server connected to said terminal or device. Likewise, some of the operations or functions described as being performed by a server may also be performed by a terminal or device connected to said server.

[0030] The functions realized by the components described herein may be realized in a general-purpose processor, a specific-purpose processor, an integrated circuit, an Application Specific Integrated Circuit (ASIC), a Central Processing Unit (CPU), a circuit, and / or a combination thereof, which are programmed to realize the described functions. A processor may include transistors or other circuits and is considered to be a circuit or a processing circuit. A processor may be a programmed processor that executes a program stored in memory.

[0031] In this specification, circuits, parts, units, and means are hardware programmed to perform or execute the described functions. Such hardware may be any hardware disclosed in this specification or any hardware known to be programmed or execute the described functions.

[0032] If the hardware is a processor considered to be a circuit type, the circuit, the part, means, or unit is a combination of the hardware and the software used to constitute the hardware and / or processor.

[0033] An embodiment of the present invention will be described in detail below with reference to the attached drawings.

[0034] FIG. 2 is a configuration diagram of a speed estimation device according to an embodiment of the present invention. Referring to FIG. 2, the speed estimation device (200) may include a driving information derivation unit (210), an initial value acquisition unit (220), a speed calculation unit (230), and a learning unit (240).

[0035] The learning unit (240) can generate an initial value learning model based on driving information from at least one learning target object driving on the target road collected using a radar sensor installed on the target road, and can train the generated initial value learning model. Here, the initial value learning model can be configured to be trained such that the speed error between the current frame and the next frame measured through the radar sensor is reduced based on at least one of the position, initial direction of travel angle, and line of sight speed of at least one learning target object driving on the target road. The process of training the initial value learning model will be explained in detail through FIG. 3.

[0036] FIG. 3 is an exemplary diagram illustrating the process of training an initial value learning model according to an embodiment of the present invention. Referring to figure (a) of FIG. 3, a radar sensor can collect driving information from a learning target object driving on a target road.

[0037] Here, the radar sensor installed on the target road has the advantage of being fixed to facilities, equipment, etc., making it easy to collect driving information for a predetermined period of time. Through this, the radar sensor transmits a signal in a specific direction and then receives a signal reflected from the object to be learned, thereby enabling the prediction of the direction of movement (300) of an object generated at a specific coordinate.

[0038] The initial value learning model collects driving information reflected from the target object through a radar sensor installed on the target road for a predetermined period of time, and thus the direction of travel of the target object occurring in the section where subsequent driving information is collected can be used as the initial value of the Kalman filter.

[0039] Referring to Figure (b), the learning unit (240) can learn the optimal initial speed for a learning target object detected at a specific location using a radar sensor installed on the target road.

[0040] For example, the target object detected within the green area is characterized by a high probability of moving to the left. Additionally, while the line-of-sight velocity provided by the radar sensor is limited, it is characterized by providing some information regarding the actual velocity of the target object.

[0041] Accordingly, by utilizing the 2D spatial information and 1D velocity of the learning target object detected by the radar sensor as parameters, the learning unit (240) can construct an initial value learning model in the form of a 3D grid map composed of multiple cells. At this time, the learning unit (240) can assign an initial value to each cell.

[0042] Referring to Figure (c), the learning unit (240) can divide the area of ​​the target road detected by the radar sensor into a grid of a reference size (e.g., 5m x 5m) corresponding to a section where both longitudinal and transverse components must be considered, and then divide the line-of-sight speed section between -6m / s and 6m / s into 3m / s intervals to generate a 3D grid cell. Here, the 3D grid cell may include a histogram of the initial direction of travel angle based on a Kalman filter for the learning target object detected by the radar sensor.

[0043] The learning unit (240) may additionally configure sections lower than -6 m / s and sections higher than 6 m / s. At this time, the learning unit (240) can generate an initial value learning model (310) based on a 3D grid cell by adding a z-axis for each line of sight speed section to the x-axis and y-axis corresponding to the longitudinal and transverse components of the area of ​​the target road. For example, by adding a z-axis, the learning unit (240) can generate an initial value learning model (310) composed of a grid of [-infinity, -6 m / s], [-6 m / s, -3 m / s], [-3 m / s, 0 m / s], [0 m / s, 3 m / s], [3 m / s, 6 m / s], [6 m / s, infinity].

[0044] The total number of cells included in this initial value learning model can be composed of W*H*R. Here, W represents the detection width of the radar sensor, H represents the length of the detection range, and R represents the total number of segments of the line-of-sight velocity.

[0045] Returning to FIG. 2, the driving information derivation unit (210) can derive driving information of an object located on the target road through a radar sensor installed on the target road. Here, the driving information of the object may include the location where the radar sensor detected the object and the line-of-sight velocity.

[0046] The initial value acquisition unit (220) can select one cell corresponding to the driving information of an object based on an initial value learning model that includes a plurality of cells that have been learned for a target road, and can acquire at least one parameter information corresponding to one cell. For example, the initial value acquisition unit (220) can acquire a first parameter indicating the completion of learning for one cell and a second parameter indicating whether data for one cell can be utilized.

[0047] Here, the pre-trained initial value learning model can be configured to be trained such that the speed error between the current frame and the next frame measured by the radar sensor is reduced based on at least one of the position, initial direction of travel angle, and line of sight velocity of at least one learning target object traveling on the target road.

[0048] The speed calculation unit (230) can calculate the actual driving speed of an object based on at least one parameter information. Here, the speed calculation unit (230) can calculate the actual driving speed considering the angle of the object's direction of travel based on the first parameter and the second parameter.

[0049] Below, we will explain the case where the first parameter is true.

[0050] For example, the speed calculation unit (230) can calculate the actual driving speed of an object based on the initial driving direction angle stored in one cell when the first parameter and the second parameter are true (e.g., indicating that learning for one cell is complete and data is available). For example, the speed calculation unit (230) can calculate the actual driving speed of an object by applying sine and cosine functions to the initial driving direction angle stored in one cell.

[0051] In another example, the speed calculation unit (230) can calculate the actual driving speed of an object by setting one of the azimuth angles among the plurality of azimuth angles as the initial driving direction angle of the object when the first parameter is true and the second parameter is false (e.g., indicating that learning is complete and data cannot be utilized for any one cell). For example, the speed calculation unit (230) can calculate the actual driving speed of an object by setting any one of the azimuth angles among the 8 azimuth angles divided into 45-degree intervals between 0 degrees and 360 degrees as the initial driving direction angle based on a Kalman filter. At this time, one cell corresponds to a cell where data cannot be utilized as the second parameter is false, but no additional learning is performed for that cell. This is because, in the case of that cell, learning is not completed because the condition for learning completion cannot be satisfied, as an arbitrary value is assigned to the object but the count of the optimal value according to the preset condition is not recorded.

[0052] Below, we will explain the case where the first parameter is false.

[0053] When the first parameter is false, the learning unit (240) sets one of the orientation angles among the multiple orientations as the initial direction of movement angle of the object and can perform learning on one of the cells based on the initial direction of movement angle. The process of performing learning on one of the cells will be explained in detail through FIG. 4.

[0054] FIG. 4 is an exemplary diagram illustrating the process of performing learning on any one cell when the first parameter according to an embodiment of the present invention is false while learning in relation to the target road (400). Referring to FIG. 4, when the first parameter is false, the learning unit (240) sets any one of the eight azimuth angles (410) divided into 45-degree intervals between 0 degrees and 360 degrees as the initial direction of travel angle based on a Kalman filter, and can perform learning on any one cell (420) based on the set initial direction of travel angle.

[0055] At this time, the learning unit (240) records a count (430) when the initial direction of movement of an object satisfies a preset condition for one cell (420), and based on the recorded count (430), when the completion count for the initial direction of movement angle (e.g., set to 6 times, but is a parameter that can be tuned according to the radar sensor) is reached, the learning for one cell (420) can be completed.

[0056] The preset conditions may include cases where the initial score is maintained above a preset number of frames, cases where the assigned score above a reference score within a preset number of frames, and cases where the initial direction angle of movement of the object after a preset number of frames is within the allowable angle error range of the preset conditions.

[0057] The learning unit (240) assigns a score based on the error range between frame-by-frame prediction information and observation information in relation to a single cell, and can determine whether the assigned score maintains an initial score or higher within a preset number of frames. For example, the learning unit (240) assigns an initial score of 2 points to an object, and then, for each frame, assigns +2 points if it falls within the error range between prediction information and observation information (an error in x and y direction distance of up to 2m, and an error in line speed of up to 1.5m), and assigns -1 point if it does not fall within the error range, thereby determining whether the score is maintained so as not to fall below 0 points within 25 frames (e.g., 1.2 seconds). At this time, if the score is maintained so as not to fall below 0 points, it is considered an optimal value and a count can be recorded. Here, the position, line speed, and direction of travel information of the current frame can be substituted into the following mathematical formulas 1 and 2 to predict the position and line speed of the next frame. This is expressed using the most basic equations of motion, but prediction algorithms utilizing various filters may be used. In this invention, Kalman filters and alpha filters are used, but are not limited thereto.

[0058]

[0059] Referring to Equation 1, x is the x-coordinate, y is the y-coordinate, the k-index is frame information, and the prediction index is the predicted value. is the length of a single frame (e.g., 0.05 seconds), is line of sight speed, can represent the direction of movement. That is, by substituting the position in the current frame into Equation 1, the position in the next frame can be predicted.

[0060]

[0061] Referring to Equation 2, the line of sight velocity of the current frame can be predicted by substituting the line of sight velocity of the current frame into Equation 2. Here, if the error between the information predicted in the current frame based on the initial direction of travel angle and the information actually observed in the next frame is large, it may be determined that it is not the optimal value. Meanwhile, since the line of sight velocity error is not dependent on the initial direction of travel, it can act as a filter to prevent cases where the prediction is incorrect and is mistaken for an observation of a completely different target (such as a vehicle other than the one currently intended to be detected) when the error in the initial direction of travel angle—which directly determines whether the initial direction of travel angle is the optimal value—is large.

[0062] The learning unit (240) can determine whether, with respect to any one cell, the case is satisfied where the score assigned within a preset number of frames is greater than or equal to the reference score. For example, the learning unit (240) can record a count by considering the score assigned within 25 frames to be 7 points or more as the optimal value.

[0063] The learning unit (240) can determine whether, with respect to any one cell, the initial direction of movement of an object after a preset number of frames satisfies the condition that the angle of movement of the object is within the allowable angle error range of a preset condition. For example, the learning unit (240) can record a count by considering the initial direction of movement after 25 frames as an optimal value if the maximum allowable angle error range is within 25 degrees with any one of the initial azimuth angles assigned.

[0064] Through this process, the learning unit (240) can complete the learning for any one cell (420) when it reaches the completion count for the initial direction angle of movement as any one direction angle is considered as the optimal value and recorded as a count, and then set the first parameter for any one cell as completed learning and proceed with the process of determining whether the data can be utilized. Here, the time parameter for completing the learning can be composed of a positive integer that increases by 1 every 10 minutes.

[0065] Through this process, when learning is complete, the learned value for an object generated in a single cell can be used as the initial value for the Kalman filter.

[0066] Returning to Fig. 2, the learning unit (240) can convert the initial direction of movement for one cell into a polar coordinate system to derive the radius and angle, and calculate the radius-weighted average and angle-weighted average for one cell, respectively.

[0067] The learning unit (240) can set the second parameter to true for any cell and store the calculated angle weighted average value in any cell when the calculated radius weighted average value is greater than or equal to the preset radius weighted average value. The process of setting the second parameter to true for any cell based on the radius weighted average value and the angle weighted average value will be explained through FIGS. 5 to 8.

[0068] FIG. 5 is an exemplary diagram illustrating a histogram of an angle-weighted average obtained through the completion of training of an initial value learning model according to an embodiment of the present invention. Referring to FIG. 5, Figures (a) and (b) illustrate a histogram obtained by the completion of training for a single cell, wherein an angle-weighted average is calculated without ignoring angle information other than the peak value and is set to be used as the value of a single cell. The angle-weighted average can be calculated, for example, through the following mathematical formulas 3 to 5.

[0069]

[0070] Referring to mathematical formula 3, the learning unit (240) can calculate the weights of the sine and cosine components of the optimal values ​​for the 8 directional angles.

[0071]

[0072] Referring to mathematical formula 4, the learning unit (240) can calculate the average of the angles that repeat in a 360-degree cycle using trigonometric functions. For example, the learning unit (240) can calculate the average of the sine and cosine components derived through mathematical formula 3.

[0073]

[0074] Referring to mathematical formula 5, the learning unit (240) can calculate an angle-weighted average by taking the inverse tangent to convert the average of the calculated sine and cosine components back into an angle.

[0075] Here, the reason for using trigonometric functions in the angle-weighted average is that, in the case of 0 degrees and 270 degrees, the actual average value is 315 degrees at 90-degree intervals, but if trigonometric functions are not used, an incorrect average of 135 degrees may be calculated, so the purpose is to calculate an accurate angle-weighted average by using trigonometric functions.

[0076] FIG. 6 is an exemplary drawing illustrating a histogram in which the availability of data is determined based on an angle-weighted average value according to an embodiment of the present invention. The learning unit (240) can determine whether data is available for use for a cell when a first parameter for a cell is set to true. At this time, the learning unit (240) can determine whether data is available or unavailable based on whether the conditions for data availability are satisfied for a cell.

[0077] Conditions for data usability may include, for example, distinguishing trends between usable and unusable data with high accuracy, facilitating optimization by using few parameters, requiring less computer processing time and memory, and not being constrained by the installation location of various types of radar sensors.

[0078] Referring to Fig. 6, Figure (a) shows a histogram of available data, and Figure (b) shows a histogram of unavailable data.

[0079] Figure (a) shows that in a histogram of available data, if the peak value occurs at a single angle or at multiple angles, the data can be determined to be available data if the angle between the peaks is within 90 degrees.

[0080] Figure (b) shows that in the histogram of unusable data, if peak values ​​are identified at multiple angles or if the angle between peaks exceeds 90 degrees, the data can be determined to be unusable data. In this case, even if the first parameter is set to true as training is completed for any cell, the value of the corresponding angle data may not be used.

[0081] FIG. 7 is an exemplary diagram illustrating the process of calculating the radius-weighted average and the angle-weighted average for any one cell, respectively, according to an embodiment of the present invention. Referring to FIG. 7, the learning unit (240) can derive the radius and the angle by converting the initial direction of movement angle (700) for any one cell into a polar coordinate system (710).

[0082] For example, the learning unit (240) considers the characteristics of angle data that repeats in a 360-degree cycle and histogram (θ, N) of the initial direction of movement angle (700). θ ) can be converted into a plot of (ρ, θ) in polar coordinates (710).

[0083] At this time, in the plot of the polar coordinate system (710), θ is maintained as is, and N is the count of the histogram of the initial direction of travel angle (700). θ can be expressed as the radius ρ, which is the distance from the center.

[0084] Afterwards, the learning unit (240) can perform normalization on the polar coordinate system (710) converted from the histogram of the initial direction of travel angle (700) through the following mathematical formula 6.

[0085]

[0086] Referring to mathematical formula 6, the learning unit (240) can perform normalization on the polar coordinate system (710) so that it can be expressed as a plot of the normalized polar coordinate system (720).

[0087] Afterwards, the learning section (240) is the radius-weighted average and the angle-weighted average ( , ) can be represented as an identifier (e.g., X) in a normalized polar coordinate system (720).

[0088] Here, the radius-weighted average (730) of the available data It can have a significantly larger value compared to the radius-weighted average (740) of unusable data. In addition, since there are no other tuning parameters other than the radius, the logic for determining the validity of the data is easily optimized.

[0089] FIG. 8 is an exemplary drawing illustrating a preset radius-weighted average value according to an embodiment of the present invention. Referring to FIG. 8, the learning unit (240) is μ, which is a radius-weighted average. ρ The threshold parameter must satisfy a preset threshold condition. For example, the preset threshold condition may limit the False Negative Rate (FNR), which determines available data as unavailable, to 1%, and the False Positive Rate (FPR), which determines unavailable data, to 5%.

[0090] Here, for example, μ of the initial training data including approximately 1800 available and unavailable data. ρ By normalizing the distribution and setting the threshold to 0.05, it can be confirmed that the conditions are satisfied with FNR=0.098% and FPR=4.18%.

[0091] Figure (b) shows μ, the radius-weighted average corresponding to the available data. ρ This threshold parameter is 0.05 or higher (μ ρ This is an exemplary drawing illustrating a polar coordinate system corresponding to ≥0.05).

[0092] Figure (c) shows μ, the radius-weighted average corresponding to the unusable data. ρ This threshold parameter is less than 0.05 (μ ρ This is an exemplary drawing illustrating a polar coordinate system corresponding to <0.05).

[0093] Through this process, the learning unit (240) can set the second parameter to true for one cell and store the calculated angle weighted average value in one cell if the calculated radius weighted average value is greater than or equal to the preset radius weighted average value.

[0094] The present invention can not only derive speed based on the initial direction of travel angle, but also derive various additional values, thereby utilizing information on traffic volume in a specific direction that frequently occurs based on the learned initial direction of travel angle to identify the signal system of a target road or to infer various unexpected situations, such as determining that an object is driving in the wrong direction when a direction of travel significantly different from the learned information is detected.

[0095] Furthermore, in addition to deriving the initial direction of travel of an object based on its position and line-of-sight velocity, the present invention can derive other values ​​based on various parameters. In this case, the z-axis must be changed from the line-of-sight velocity to the corresponding parameter, and the criteria for the optimal value can be modified according to the situation. For example, in the case of a radar sensor, signal strength can be measured, allowing for the filtering of radar ghost signals based on this. As another example, in the case of a camera sensor, additional results can be derived based on RGB values ​​in the image and object information detected by an object detection algorithm. As yet another example, by collecting data such as hourly vehicle traffic volume and road usage by lane and using them in an initial value learning model, the derived results can contribute to road safety, such as signal optimization and determining road maintenance timing.

[0096] FIG. 9 is an exemplary diagram comparing the line-of-sight velocity of an object and the actual velocity of an object according to an embodiment of the present invention. Referring to FIG. 9, Figure (a) shows the line-of-sight velocity of a vehicle using a Kalman filter, and Figure (b) shows the actual velocity of a vehicle using an initial value learning model based on a Kalman filter. According to Figure (b), by learning the initial value through the initial value learning model, an effect of improved accuracy in estimating the actual velocity of a vehicle can be obtained even when using a Kalman filter that depends on the initial value.

[0097] FIG. 10 is a flowchart of a method for estimating the speed of an object using a radar sensor according to an embodiment of the present invention. Referring to FIG. 10, the method for estimating the speed of an object using a radar sensor performed by a speed estimation device (200) includes steps processed chronologically according to the embodiments illustrated in FIG. 2 to 9. Therefore, even if the following is omitted, it is also applicable to the method for estimating the speed of an object using a radar sensor performed by a speed estimation device (200) according to the embodiments illustrated in FIG. 2 to 9.

[0098] In step S1010, the speed estimation device (200) can derive driving information of an object located on the target road through a radar sensor installed on the target road.

[0099] In step S1020, the speed estimation device (200) can select any one cell corresponding to the driving information of the object based on an initial value learning model that includes a plurality of cells that have been learned for the target road.

[0100] In step S1030, the speed estimation device (200) can obtain at least one parameter information corresponding to any one cell.

[0101] In step S1040, the speed estimation device (200) can calculate the actual driving speed of the object based on at least one parameter information.

[0102] In the description above, steps S1010 to S1040 may be further divided into additional steps or combined into fewer steps, depending on an embodiment of the present invention. Additionally, some steps may be omitted as necessary, and the order between steps may be changed.

[0103] FIG. 11 is a flowchart of a method for calculating the actual driving speed of an object based on a first parameter and a second parameter corresponding to any one cell according to an embodiment of the present invention. Referring to FIG. 11, a speed estimation device (200) can derive driving information of an object located on a target road through a radar sensor installed on the target road (S1100).

[0104] The speed estimation device (200) selects one cell corresponding to the driving information of an object based on an initial value learning model that includes a plurality of cells learned for a target road (S1101), and can obtain a first parameter indicating the completion of learning for one cell and a second parameter indicating whether data for one cell can be utilized (S1102).

[0105] The speed estimation device (200) can determine whether the first parameter is true (S1103). Below, the case where the first parameter is true will be described.

[0106] The speed estimation device (200) can further determine whether the second parameter is true when the first parameter is true (S1105).

[0107] For example, when the first parameter is true and the second parameter is false (S1106), the speed estimation device (200) sets one of the plurality of azimuth angles as the initial direction of travel of the object (S1107), and can calculate the actual driving speed of the object based on the set initial direction of travel of the object (S1108).

[0108] In another example, the speed estimation device (200) can retrieve the initial driving direction angle stored in one cell when the first parameter and the second parameter are true (S1109), and calculate the actual driving speed of the object based on the retrieved initial driving direction angle stored in one cell (S1111).

[0109] Below, we will explain the case where the first parameter is false.

[0110] When the first parameter is false (S1112), the velocity estimation device (200) can set one of the multiple orientations as the initial direction of travel angle of the object and perform learning on one of the cells based on the set initial direction of travel angle (S1113).

[0111] The speed estimation device (200) records a count when the initial direction of movement of an object satisfies a preset condition for any one cell, and when the completion count for the initial direction of movement angle is reached based on the recorded count, the learning for any one cell can be completed (S1114).

[0112] The speed estimation device (200) calculates the radius-weighted average and the angle-weighted average for each cell (S1115), and can determine whether the radius-weighted average is greater than or equal to a preset radius-weighted average (S1116).

[0113] For example, if the radius-weighted average value is greater than or equal to a preset radius-weighted average value (S1117), the speed estimation device (200) can set a second parameter to true for one cell and store an angle-weighted average value in one cell (S1118).

[0114] As another example, the speed estimation device (200) can set the possibility of using the data to be impossible if the radius-weighted average value is less than the preset radius-weighted average value (S1119) (S1120).

[0115] The method of estimating the speed of an object using a radar sensor and the method of calculating the actual driving speed of an object based on a first parameter and a second parameter corresponding to any one cell, performed in a speed estimation device described through FIGS. 2 to 11, may also be implemented in the form of a computer program stored on a medium executed by a computer or a recording medium containing instructions executable by a computer. In addition, the method of estimating the speed of an object using a radar sensor and the method of calculating the actual driving speed of an object based on a first parameter and a second parameter corresponding to any one cell, performed in a speed estimation device described through FIGS. 2 to 11, may also be implemented in the form of a computer program stored on a medium executed by a computer.

[0116] A computer-readable medium may be any available medium accessible by a computer and includes both volatile and non-volatile media, and both removable and inremovable media. Additionally, a computer-readable medium may include a computer storage medium. A computer storage medium includes both volatile and non-volatile, removable and inremovable media implemented by any method or technique for storing information such as computer-readable instructions, data structures, program modules, or other data.

[0117] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0118] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention. Explanation of the symbols

[0119] 200: Speed ​​Estimation Device 210: Driving Information Derivation Unit 220: Initial value acquisition section 230: Speed ​​calculation unit 240: Learning Department

Claims

Claim 1 A speed estimation device comprising: a driving information derivation unit that derives driving information of an object located on a target road through a radar sensor installed on a target road; an initial value acquisition unit that selects one cell corresponding to the driving information of the object based on an initial value learning model including a plurality of cells pre-learned for the target road, and acquires at least one parameter information corresponding to the one cell; and a speed calculation unit that calculates the actual driving speed of the object based on the at least one parameter information, wherein the initial value acquisition unit acquires a first parameter indicating the completion of learning for the one cell and a second parameter indicating whether data for the one cell is available for use, and the speed calculation unit calculates the actual driving speed considering the angle of the object's direction of travel based on the first parameter and the second parameter, and further comprises a learning unit that performs learning for the one cell based on one of a plurality of azimuths when the first parameter is false. Claim 2 A speed estimation device according to claim 1, wherein the pre-learned initial value learning model is configured to be learned such that the speed error between the current frame and the next frame measured through the radar sensor is reduced based on at least one of the position, initial direction of travel angle, and line of sight speed of at least one learning target object traveling on the target road. Claim 3 delete Claim 4 A speed estimation device according to claim 1, wherein the speed calculation unit calculates the actual driving speed of the object based on the initial driving direction angle stored in any one of the cells when the first parameter and the second parameter are true. Claim 5 A speed estimation device according to claim 1, wherein the speed calculation unit calculates the actual driving speed of the object by setting one of a plurality of azimuth angles as the initial driving direction angle of the object when the first parameter is true and the second parameter is false. Claim 6 delete Claim 7 A velocity estimation device according to claim 1, wherein the learning unit sets one of the plurality of orientations as the initial direction of movement angle of the object, records a count when the initial direction of movement angle of the object satisfies a preset condition for one of the cells, and completes the learning for one of the cells when a completion count for the initial direction of movement angle is reached based on the recorded count. Claim 8 A speed estimation device according to claim 7, wherein the learning unit assigns a score based on the error range between frame-by-frame prediction information and observation information in relation to any one of the cells, and determines whether the assigned score maintains an initial score or higher within a preset number of frames. Claim 9 A speed estimation device according to claim 8, wherein the preset condition includes at least one of the following: maintaining an initial score or higher within the preset number of frames; having an assigned score or higher than a reference score within the preset number of frames; and having an initial direction of movement angle of the object after the preset number of frames within the allowable angle error range of the preset condition. Claim 10 A speed estimation device according to claim 7, wherein the learning unit converts the initial direction of travel angle for any one cell into a polar coordinate system to derive a radius and an angle, and calculates a radius-weighted average and an angle-weighted average for any one cell, respectively. Claim 11 A velocity estimation device according to claim 9, wherein the learning unit sets the second parameter to true for any one cell and stores the calculated angle-weighted average value in any one cell when the calculated radius-weighted average value is greater than or equal to a preset radius-weighted average value. Claim 12 A speed estimation method for estimating the speed of an object using a radar sensor in a speed estimation device, comprising: a step of deriving driving information of an object located on a target road through a radar sensor installed on a target road; a step of selecting one cell corresponding to the driving information of the object based on an initial value learning model including a plurality of cells that have been learned for the target road; a step of obtaining at least one parameter information corresponding to one cell; and a step of calculating the actual driving speed of the object based on the at least one parameter information, wherein the step of obtaining at least one parameter information includes a step of obtaining a first parameter indicating the completion of learning for one cell and a second parameter indicating whether data for one cell is available for use, and the step of calculating the actual driving speed of the object includes a step of calculating the actual driving speed considering the angle of the object's direction of travel based on the first parameter and the second parameter, and further comprising a step of performing learning for one cell based on one of a plurality of azimuths when the first parameter is false. Claim 13 A speed estimation method according to claim 12, wherein the pre-learned initial value learning model is configured to be learned such that the speed error between the current frame and the next frame measured through the radar sensor is reduced based on at least one of the position, initial direction of travel angle, and line of sight speed of at least one learning target object traveling on the target road. Claim 14 delete Claim 15 A speed estimation method according to claim 12, wherein the step of calculating the actual driving speed of the object includes the step of calculating the actual driving speed of the object based on the initial driving direction angle stored in any one of the cells when the first parameter and the second parameter are true. Claim 16 A speed estimation method according to claim 12, wherein the step of calculating the actual driving speed of the object includes the step of calculating the actual driving speed of the object by setting one of a plurality of azimuth angles as the initial driving direction angle of the object when the first parameter is true and the second parameter is false. Claim 17 delete Claim 18 A speed estimation method according to claim 12, wherein the step of performing learning for any one cell comprises: setting an angle of one of the plurality of directions as the initial direction of movement angle of the object, and recording a count when the initial direction of movement angle of the object satisfies a preset condition for any one cell; and completing the learning for any one cell when a completion count for the initial direction of movement angle is reached based on the recorded count. Claim 19 A speed estimation method according to claim 18, wherein the step of performing learning on any one cell comprises: a step of assigning a score according to the error range between frame-by-frame prediction information and observation information in relation to any one cell; and a step of determining whether the assigned score maintains an initial score or higher within a preset number of frames. Claim 20 A speed estimation method according to claim 19, wherein the preset condition includes at least one of the following: maintaining an initial score or higher within the preset number of frames; having an assigned score or higher than a reference score within the preset number of frames; and having an initial direction of movement angle of the object after the preset number of frames within the allowable angle error range of the preset condition. Claim 21 A speed estimation method according to claim 18, wherein the step of performing learning for any one cell comprises: a step of deriving a radius and an angle by converting the initial direction of travel for any one cell into a polar coordinate system; and a step of calculating a radius-weighted average and an angle-weighted average for any one cell, respectively. Claim 22 A computer program stored on a computer-readable recording medium comprising a sequence of instructions for estimating the speed of an object using a radar sensor, wherein, when the computer program is executed by a computing device, the computer program comprises a sequence of instructions for deriving driving information of an object located on a target road through a radar sensor installed on the target road, selecting one cell corresponding to the driving information of the object based on an initial value learning model comprising a plurality of cells previously learned for the target road, obtaining a first parameter indicating the completion of learning for the one cell and a second parameter indicating whether data for the one cell is available for use, calculating the actual driving speed of the object considering the angle of the object's direction of travel based on the first parameter and the second parameter, and, if the first parameter is false, performing learning for the one cell based on the angle of one of a plurality of azimuths. Claim 23 A device for estimating the speed of an object using a radar sensor, comprising: at least one processor; and at least one memory including computer program code, wherein the device, through the at least one processor, derives driving information of an object located on a target road through a radar sensor installed on a target road, selects one cell corresponding to the driving information of the object based on an initial value learning model including a plurality of cells previously learned for the target road, obtains a first parameter indicating the completion of learning for the one cell and a second parameter indicating whether data for the one cell is available for use, calculates the actual driving speed of the object considering the angle of the object's direction of travel based on the first parameter and the second parameter, and, if the first parameter is false, performs learning for the one cell based on the angle of one of a plurality of azimuths.

Images