Information processing device, control method, program, and storage medium

Abstract

To accurately determine a noise point generated by noise while suppressing a calculation amount.SOLUTION: An information processor 1 functions as an information processor having acquisition means and evaluation value calculation means. The acquisition means acquires point group data, which is a set of data representing a reflection intensity value and a measured distance for each point measured by a measurement device. The evaluation value calculation means calculates a noise evaluation value for a target point on the basis of noise evaluation function evaluating whether the target point corresponding to each data is an object point that is a point to be measured of an object or a noise point generated by noise, on the basis of data of a reference point existing within a search range set using the target point as a reference. In this case, the evaluation value calculation means determines the search range on the basis of at least one of the reflection intensity value and the measured distance corresponding to the target point.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present disclosure relates to a technique for processing measured data. [Background technology]

[0002] A distance measuring device has been known that irradiates a measurement object with light, detects the light reflected from the measurement object, and calculates the distance to the measurement object based on the time difference between the time when the light is irradiated to the measurement object and the time when the light reflected from the measurement object is detected. For example, Patent Document 1 discloses a noise determination method that determines noise points generated by noise at each measurement point indicated by point cloud data obtained by a LIDAR based on temporal or spatial stability. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-43838 Summary of the Invention [Problem to be solved by the invention]

[0004] Noise points generated by noise are measured with a roughly constant probability, while object points representing objects tend to have a lower probability of being measured as the distance increases. When an object is far away or when the object originally has low reflection intensity, the probability of measuring the object point decreases, making it difficult to distinguish between noise points and object points.

[0005] The above is an example of a problem to be solved by the present invention. The main object of the present disclosure is to provide an information processing device, a control method, a program, and a storage medium storing the program, which are capable of accurately determining noise points generated by noise while reducing the amount of calculation. [Means for solving the problem]

[0006] The claimed invention is an acquisition means for acquiring point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by the measurement device; Whether the target point corresponding to each of the data is an object point that is a measured point of the object or a noise point generated by noise is determined based on the target point. , including a spatial range related to the measurement direction and a temporal range related to the measurement time measured by the measurement device. Data of reference points within the search area a distance index, a time index, and a space index, which respectively represent the difference in the measured distance, the difference in time, and the difference in space between the target point and the target point; and an evaluation value calculation means for calculating an evaluation value for the target point based on an evaluation function that evaluates the target point based on the evaluation value calculation means determines the search range based on at least one of the reflection intensity value or the measurement distance corresponding to the target point; The information processing device has:

[0007] The claimed invention also includes: Acquire point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by the measurement device; Whether the target point corresponding to each of the data is an object point that is a measured point of the object or a noise point generated by noise is determined based on the target point. , including a spatial range related to the measurement direction and a temporal range related to the measurement time measured by the measurement device. Data of reference points within the search area a distance index, a time index, and a space index, which respectively represent the difference in the measured distance, the difference in time, and the difference in space between the target point and the target point; Calculating an evaluation value for the target point based on an evaluation function that evaluates based on determining the search range based on at least one of the reflection intensity value or the measurement distance corresponding to the target point; It is a control method.

[0008] The claimed invention also includes: Acquire point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by the measurement device; Whether the target point corresponding to each of the data is an object point that is a measured point of the object or a noise point generated by noise is determined based on the target point. , including a spatial range related to the measurement direction and a temporal range related to the measurement time measured by the measurement device. Data of reference points within the search area a distance index, a time index, and a space index, which respectively represent the difference in the measured distance, the difference in time, and the difference in space between the target point and the target point;Calculating an evaluation value for the target point based on an evaluation function that evaluates based on The program causes a computer to execute a process of determining the search range based on at least one of the reflection intensity value or the measured distance corresponding to the target point. [Brief explanation of the drawings]

[0009] [Figure 1] 1 shows a schematic configuration of a lidar according to a first embodiment. [Figure 2] 10 is a graph showing a probability density function of a noise evaluation function. [Figure 3] 10 is an example of a flowchart of point cloud information processing according to the first embodiment. [Figure 4] (A) A diagram showing, on a virtual plane, the target point of the current frame and its surrounding measurement points where the reflection intensity value is less than the threshold. (B) A diagram showing, on a virtual plane, the target point of the current frame and its surrounding measurement points where the reflection intensity value is greater than or equal to the threshold. [Figure 5] 10 is an example of a flowchart of a noise point determination process according to a second embodiment. [Figure 6] 10 is another example of a flowchart of the noise point determination process according to the second embodiment. [Figure 7] FIG. 10 is a diagram showing the arrangement of sample points of past target points determined based on the first method. [Figure 8] 10A and 10B are diagrams showing the arrangement of sample points of past target points determined based on the second and third methods, respectively; [Figure 9] 10 is an example of a flowchart showing a procedure of a noise point determination process according to a third embodiment. [Figure 10] FIG. 10 is a configuration diagram of a lidar system according to a fourth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] In a preferred embodiment of the present invention, an information processing device includes an acquisition means for acquiring point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by a measurement device, and an evaluation value calculation means for calculating an evaluation value for each target point based on an evaluation function that evaluates whether the target point corresponding to each of the data is an object point that is a measured point of an object or a noise point generated by noise, based on data of reference points present within a search range set with the target point as a reference, and the evaluation value calculation means determines the search range based on at least one of the reflection intensity value or the measurement distance corresponding to the target point. In this aspect, the information processing device can calculate an evaluation value that represents an accurate evaluation of the noise point while reducing the amount of calculation required to calculate the evaluation value that evaluates whether the target point is an object point or a noise point.

[0011] In one aspect of the information processing device, the search range includes a spatial range related to the measurement direction and a temporal range related to the measurement time measured by the measurement device. In a preferred example, the evaluation value calculation means selects, based on the search range, the reference point that is temporally or spatially close to the target point set in the current frame from measured points included in a current frame, which is the point cloud information acquired by the acquisition means at the current processing time, and a past frame, which is the point cloud information acquired by the acquisition means at a time prior to the current processing time. In this aspect, the information processing device can accurately calculate the evaluation value of the target point based on neighboring points in space-time determined by the search range.

[0012] In another aspect of the information processing device, when the reflection intensity value is equal to or greater than a first threshold, the evaluation value calculation means sets the search range to be smaller than when the reflection intensity value is less than the first threshold. This aspect makes it possible to suitably reduce the amount of calculation for calculating the evaluation value for a target point that is estimated to have a high reliability as an object point.

[0013] In another aspect of the information processing device, when the measured distance is less than a second threshold, the evaluation value calculation means sets the search range to be smaller than when the measured distance is equal to or greater than the second threshold. This aspect also makes it possible to preferably reduce the amount of calculation for calculating the evaluation value for a target point that is estimated to have a high reliability as an object point.

[0014] In another aspect of the information processing device, the information processing device further includes a noise determination unit that determines the noise points in the point cloud data based on the evaluation value. With this aspect, the information processing device can accurately determine the noise points.

[0015] In another preferred embodiment of the present invention, a control method is executed by an information processing device, which acquires point cloud data, which is a collection of data representing a reflection intensity value and a measurement distance for each point measured by a measurement device, calculates an evaluation value for the target point based on an evaluation function that evaluates whether the target point corresponding to each of the data is an object point that is a measured point of an object or a noise point generated by noise, based on data of reference points present within a search range set with the target point as a reference, and determines the search range based on at least one of the reflection intensity value or the measurement distance corresponding to the target point. By executing this control method, the information processing device can calculate an evaluation value that represents an accurate evaluation of the noise point while reducing the amount of calculation required to calculate an evaluation value that evaluates whether the point is an object point or a noise point.

[0016] In another preferred embodiment of the present invention, a program causes a computer to acquire point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by a measurement device, calculate an evaluation value for the target point based on an evaluation function that evaluates whether the target point corresponding to each of the data is an object point that is a measured point of an object or a noise point generated by noise, based on data of reference points present within a search range set with the target point as a reference, and determine the search range based on at least one of the reflection intensity value or the measurement distance corresponding to the target point. By executing this program, the computer can calculate an evaluation value that represents an accurate evaluation of the noise point while reducing the amount of calculation required to calculate the evaluation value that evaluates whether the point is an object point or a noise point. Preferably, the program is stored in a storage medium. [Example]

[0017] Preferred embodiments of the present invention will now be described with reference to the drawings.

[0018] <First Example> (1) Device configuration 1 shows a schematic configuration of a LIDAR 100 according to a first embodiment. The LIDAR 100 is mounted on a vehicle that provides driving assistance such as autonomous driving, for example. The LIDAR 100 emits laser light over a predetermined angular range in the horizontal and vertical directions and receives light that is reflected by an object and returns (also referred to as "reflected light"), thereby discretely measuring the distance from the LIDAR 100 to the object and generating point cloud information that indicates the three-dimensional position of the object.

[0019] As shown in FIG. 1, the lidar 100 mainly includes a transmitter 1, a receiver 2, a beam splitter 3, a scanner 5, a piezoelectric sensor 6, a controller 7, and a memory 8.

[0020] The transmitter 1 is a light source that emits pulsed laser light toward the beam splitter 3. The transmitter 1 includes, for example, an infrared laser light emitting element. The transmitter 1 is driven based on a drive signal “Sg1” supplied from the controller 7.

[0021] The receiver 2 is, for example, an avalanche photodiode, generates a detection signal “Sg2” corresponding to the amount of received light, and supplies the generated detection signal Sg2 to the controller 7.

[0022] The beam splitter 3 transmits the pulsed laser light emitted from the transmitter 1. The beam splitter 3 also reflects the light reflected by the scanner 5 towards the receiver 2.

[0023] The scanner 5 is, for example, an electrostatically driven mirror (MEMS mirror), and its tilt (i.e., the angle of optical scanning) changes within a predetermined range based on the drive signal "Sg3" supplied from the control unit 7. The scanner 5 reflects the laser light that has passed through the beam splitter 3 toward the outside of the LIDAR 100, and also reflects reflected light that enters from the outside of the LIDAR 100 toward the beam splitter 3. In addition, a point measured by irradiating the laser light within the measurement range of the LIDAR 100, or the measurement data thereof, is also referred to as a "measured point."

[0024] The scanner 5 is also provided with a piezoelectric sensor 6. The piezoelectric sensor 6 detects distortion caused by stress of a torsion bar that supports the mirror portion of the scanner 5. The piezoelectric sensor 6 supplies the generated detection signal "Sg4" to the control unit 7. The detection signal Sg4 is used to detect the orientation of the scanner 5.

[0025] The memory 8 is composed of various types of volatile and non-volatile memories such as RAM (Random Access Memory), ROM (Read Only Memory), and flash memory. The memory 8 stores programs necessary for the control unit 7 to execute predetermined processes. The memory 8 also stores various parameters referenced by the control unit 7. For example, the memory 8 stores information on the probability density function of the evaluation function used in this embodiment (also referred to as "probability density function information"). The memory 8 also stores point cloud information for the latest predetermined number of frames generated by the control unit 7.

[0026] The control unit 7 includes various processors, such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The control unit 7 executes a program stored in the memory 8 to perform predetermined processing. The control unit 7 is an example of a computer that executes a program. The control unit 7 is not limited to being realized by software according to a program, but may be realized by any combination of hardware, firmware, and software. The control unit 7 may also be a user-programmable integrated circuit, such as an FPGA (Field-Programmable Gate Array) or a microcontroller, or may be an ASSP (Application Specific Standard Produce), ASIC (Application Specific Integrated Circuit), or the like.

[0027] The control unit 7 functionally includes a transmission driving block 70, a scanner driving block 71, a point cloud information generation block 72, and a point cloud information processing block 73.

[0028] The transmission drive block 70 outputs a drive signal Sg1 that drives the transmission unit 1. The drive signal Sg1 includes information for controlling the emission time of a laser light-emitting element included in the transmission unit 1 and the emission intensity of the laser light-emitting element. The transmission drive block 70 controls the emission intensity of the laser light-emitting element included in the transmission unit 1 based on the drive signal Sg1.

[0029] The scanner driving block 71 outputs a driving signal Sg3 for driving the scanner 5. This driving signal Sg3 includes a horizontal driving signal corresponding to the resonance frequency of the scanner 5 and a vertical driving signal for vertical scanning. The scanner driving block 71 also monitors a detection signal Sg4 output from the piezo sensor 6 to detect the scanning angle of the scanner 5 (i.e., the emission direction of the laser light).

[0030] Based on the detection signal Sg2 supplied from the receiving unit 2, the point cloud information generation block 72 generates point cloud information indicating the distance (measurement distance) to the object irradiated with the laser light and the received light intensity (reflection intensity value) of the reflected light for each measurement direction (i.e., the emission direction of the laser light), with the lidar 100 as the reference point. In this case, the point cloud information generation block 72 calculates the time from when the laser light is emitted until the receiving unit 2 detects the reflected light as the time of flight of the light. Then, the point cloud information generation block 72 generates point cloud information indicating a set of points corresponding to a combination of the measurement distance according to the calculated time of flight, the emission direction of the laser light corresponding to the reflected light received by the receiving unit 2 (i.e., the measurement direction), and the reflection intensity value based on the reflected light, and supplies the generated point cloud information to the point cloud information processing block 73. Hereinafter, the point cloud information obtained by one scan of all measurement points will be referred to as point cloud information for one frame. Here, the point cloud information can be considered as an image in which each measurement direction is a pixel and the measured distance and reflection intensity value in each measurement direction are pixel values. In this case, the emission direction of the laser light varies depending on the elevation and depression angles of the vertically arranged pixels, and the emission direction of the laser light varies depending on the horizontally arranged pixels. Then, for each pixel, a coordinate value in a three-dimensional coordinate system based on the LIDAR 100 is calculated based on the corresponding pair of emission direction and measurement distance.

[0031] The point cloud information processing block 73 determines noise data generated by erroneously detecting an object in the point cloud information. Hereinafter, measurement points corresponding to data generated by detecting an actual object will be referred to as "object points," and measurement points other than object points (i.e., measurement points corresponding to noise data) will be referred to as "noise points." The point cloud information processing block 73 determines noise points in the current frame using a frame of point cloud information obtained at the current processing time (also referred to as the "current frame") and a frame of point cloud information obtained at a processing time prior to the current processing time (also referred to as the "past frame").

[0032] The point cloud information processing block 73 may delete noise data corresponding to the determined noise points from the point cloud information, or may add flag information indicating whether each measured point is an object point or a noise point to the point cloud information. The point cloud information processing block 73 may also supply the processed point cloud information to an external device outside the LIDAR 100, or to another processing block within the LIDAR 100 that performs obstacle detection, etc. In the former case, the point cloud information may be output to, for example, a device (also referred to as a "driving assistance device") that controls driving assistance such as automatic driving of the vehicle. In this case, for example, the vehicle may be controlled to at least avoid obstacle points based on the point cloud information. The driving assistance device may be, for example, an ECU (Electronic Control Unit) of the vehicle, or an on-board device such as a car navigation device electrically connected to the vehicle. The point cloud information processing block 73 stores the point cloud information for each frame in the memory 8 in association with time information indicating the processing time for each frame. The point cloud information processing block 73 according to the first embodiment is an example of an “evaluation value calculation means,” a “statistics calculation means,” a “threshold setting means,” and a “noise determination means.” Furthermore, the LIDAR 100 excluding the point cloud information processing block 73 is an example of a “measurement device.”

[0033] The lidar 100 is not limited to a scanning type lidar that scans a field of view with laser light, but may also be a flash type lidar that generates three-dimensional data by irradiating a diffused laser light within the field of view of a two-dimensional array sensor.

[0034] (2) Noise detection processing Next, we will explain the noise determination process, which is a process for determining noise points executed by the point cloud information processing block 73. In summary, the point cloud information processing block 73 calculates an evaluation value for each measurement point in the current frame using an evaluation function based on the differences in measurement direction, measurement distance, and measurement time between adjacent measurement points in space and time, and determines measurement points whose evaluation value is less than a threshold as noise points. In this case, the point cloud information processing block 73 determines the threshold using statistics of the probability density function of the evaluation function, thereby accurately determining noise points taking into account the noise detection rate. Then, by accurately determining whether a point is a noise point or an object point, point cloud information is generated that allows accurate detection of distant objects.

[0035] Hereinafter, the above evaluation function will be referred to as the "noise evaluation function," and the evaluation value for each measurement point calculated based on the noise evaluation function will be referred to as the "noise evaluation value." Furthermore, the measurement points that are the targets for calculating the noise evaluation value (i.e., the targets for determining whether they are noise points) will be referred to as "target points," and measurement points in the current frame and past frames that are used to calculate the noise evaluation value other than the target points will be referred to as "reference points." Furthermore, the above-mentioned threshold for the noise evaluation value will also be referred to as the "threshold value Th."

[0036] (2-1) Noise evaluation function First, we will explain the noise evaluation function. The noise evaluation function is a function that takes as input the difference in the measured distance (Euclidean distance) between the target point and the reference point, the frame interval between the target point and the reference point (i.e., the difference in measurement time), and the distance on the frame (image) between the target point and the reference point (i.e., the difference in measurement direction). Hereafter, the difference in the measured distances will be called the "distance index," the frame interval will be called the "time index," and the distance on the frame will be called the "spatial index."

[0037] In this embodiment, as an example of a noise evaluation function, the following formula (1) is used, which treats the distance index, time index, and space index as equals and determines the weight of each index using coefficients (coefficients a to d). The point cloud information processing block 73 calculates a noise evaluation value for each target point using the following formula (1).

[0038]

number

[0039] In addition, in equation (1), by using an exponential function with a base of 2 as the noise evaluation function, it is possible to increase minute value differences and express them as noise evaluation values. Also, by setting "-1" in the exponent part of the noise evaluation function, the noise evaluation value decreases as the distance index, time index, and space index increase.

[0040] Furthermore, the range of the horizontal index h and vertical index v used in equation (1) may be the entire frame, or may be a range within a predetermined range on the frame centered on the target point (i.e., a range of a predetermined measurement direction based on the measurement direction of the target point). Similarly, the range of t used in equation (1) may be set to a range including the processing times of all obtained past frames and the current frame, or may be set to a range including the processing times of a predetermined number of recent past frames and the current frame (i.e., a predetermined range of measurement times based on the current processing time). By appropriately setting the search range, it is possible to reduce the calculation load while maintaining the accuracy of noise point determination. Methods for setting such search ranges will be described in detail in the second and third embodiments.

[0041] By using the noise evaluation function of equation (1), it is possible to treat the distance index, time index, and space index on the same level, increase the difference between these indexes, and determine the weight of each index using coefficients a to d.

[0042] (2-2) Threshold setting according to probability density function Next, the setting of the threshold value Th according to the probability density function of the noise evaluation function will be specifically described.

[0043] Generally, noise points are measured with a certain probability at each measurement distance. Therefore, in this embodiment, when point cloud information is generated in a state where no object is present within the measurement range (field of view) of the LIDAR 100, it is assumed that the output value (evaluation function value) of the noise evaluation function follows a normal distribution at each measurement distance according to the central limit theorem. In this case, by calculating the mean "μ" and variance "σ" of the probability density function of the noise evaluation function, it is possible to set a threshold value Th according to the noise detection rate.

[0044] FIG. 2 is a graph showing the probability density function of the noise evaluation function. In this case, the probability density function of the noise evaluation function is modeled as following a normal distribution. In this case, for example, if "μ + 3σ" is set as the threshold Th, 99.7% of noise points can be correctly determined as noise points. On the other hand, in this case, 0.3% of noise points will be erroneously determined as object points, resulting in a noise detection rate of 0.3%. Therefore, for example, if a noise detection rate of 0.3% is desired, "μ + 3σ" can be set as the threshold Th. Furthermore, even when an arbitrary target noise detection rate is set, the threshold Th for achieving that noise detection rate can be set using the mean μ and the variance (strictly speaking, the standard deviation) σ according to the properties of the normal distribution.

[0045] Next, a supplementary explanation will be given of the method for calculating the mean μ and the variance (standard deviation) σ.

[0046] Here, among the parameters of the noise evaluation function shown in Equation (1), the random variable is the inter-point distance "rdist" (i.e., the difference in the measured distance between the target point and the reference point). Therefore, when a noise point appears at the measured distance dist[m] of the target point, the probability that a point will appear within the inter-point distance rdist[m] is calculated to calculate the mean and variance of the probability density function for each measured distance.

[0047] Here, the mean "E(X)" (= μ) and variance "V(X)" (= σ) of the probability density function "f(x)" of Equation (1) at the measurement distance dist of the target point are 2 ) is calculated. The measured distance of the reference point is defined as "nd". The difference between the measured distance dist of the target point and the measured distance nd of the reference point is defined as "y" (=rdist=dist-nd), and the gate length is defined as "gate". The frequency (number of occurrences) of noise points at distance y is defined as "hist(dist-nd)".

[0048] In this case, the following equations (2) to (4) hold.

[0049]

number

[0050]

number

[0051]

number

[0052] As described above, the point cloud information processing block 73 calculates the mean E(X) (=μ) and variance V(X) (=σ) of the probability density function f(x) based on the point cloud information acquired when no object is present within the field of view of the lidar 100, using the above-mentioned equations (2) to (8). 2 ) and stores them as probability density function information in the memory 8 or the like. Then, in the noise determination process, the point cloud information processing block 73 can appropriately set the threshold value Th that achieves a desired noise detection rate by referring to the probability density function information.

[0053] The point cloud information for calculating the probability density function information may be point cloud information acquired by the LIDAR 100 before the noise determination process is performed, or may be point cloud information acquired in advance by the LIDAR 100 or another similar LIDAR before product shipment. In the former case, the LIDAR 100 generates point cloud information that does not include object points by, for example, facing a direction in which no objects, such as the sky, are present and performing the point cloud information generation process. In yet another example, point cloud information acquired in advance may be provided as an initial value, and scanning points in the frame that do not have object points may be identified in real time using the determination results of the presence or absence of object points in past frames, and the point cloud information of those scanning points may be used to calculate the probability density function information.

[0054] (3) Processing Flow FIG. 3 is an example of a flowchart showing the procedure of processing related to point cloud information (point cloud information processing) executed by the LIDAR 100.

[0055] First, the point cloud information processing block 73 acquires probability density function information (step S01). In this case, the point cloud information processing block 73 acquires a probability density function representing the mean μ and variance (standard deviation) σ of the probability density function of the noise evaluation function from the memory 8. Note that the point cloud information processing block 73 may acquire point cloud information when no object is present within the field of view of the lidar 100, and may perform processing to generate probability density function information based on the point cloud information in step S01.

[0056] Next, the point cloud information processing block 73 sets a threshold value Th for the noise evaluation value based on the probability density function information (step S02). In this case, the point cloud information processing block 73 sets the threshold value Th for achieving a desired noise detection rate using the mean μ and variance (standard deviation) σ included in the probability density function information. In this case, for example, information indicating the relationship between the threshold value Th and the mean μ and variance (standard deviation) σ is stored in advance in the memory 8.

[0057] Then, the point cloud information processing block 73 compares the noise evaluation value calculated for each target point with a threshold value Th to determine whether it is a noise point in the current frame (step S04). In this case, the point cloud information processing block 73 sets each measured point in the current frame as a target point in turn, and calculates the noise evaluation value for the target point using the current frame and the previous frame with reference to equation (1). Then, the point cloud information processing block 73 determines that a target point whose noise evaluation value is equal to or greater than the threshold value Th is an object point, and determines that a target point whose noise evaluation value is less than the threshold value Th is a noise point.

[0058] Then, the point cloud information processing block 73 determines whether or not the point cloud information processing should be terminated (step S05). For example, the point cloud information processing block 73 determines that the point cloud information processing should be terminated when there is a request to stop generating point cloud information or a request to stop the LIDAR 100. Then, if the point cloud information processing block 73 determines that the point cloud information processing should be terminated (step S05; Yes), it terminates the processing of the flowchart. On the other hand, if the point cloud information processing block 73 determines that the point cloud information processing should be continued (step S05; No), it returns the processing to step S03, updates the current processing time to the next processing time, and executes steps S03 and S04.

[0059] As described above, the information processing device 1 according to the first embodiment functions as an information processing device having an acquisition means, an evaluation value calculation means, and a noise determination means. The acquisition means acquires point cloud data, which is a collection of data for each point measured by a measurement device. The evaluation value calculation means calculates a noise evaluation value for each data point in the point cloud data based on a noise evaluation function that evaluates whether the point is an object point that is a measured point of an object or a noise point generated by noise. The threshold setting means sets a threshold for the noise evaluation value based on statistics of the probability density function of the noise evaluation function. The noise determination means determines noise points in the point cloud data based on the noise evaluation value and the threshold. With this aspect, the information processing device 1 can set a threshold based on the ranging probability of noise and accurately identify noise points.

[0060] <Second Example> In the second embodiment, the point cloud information processing block 73 sets the spatiotemporal search range (i.e., the range of each index t, h, v centered on the target point) when calculating the noise evaluation function based on the reflection intensity value (i.e., brightness) measured at each measurement point. Specifically, if the reflection intensity value of the target point is equal to or greater than a predetermined threshold, the point cloud information processing block 73 assumes that the target point is highly reliable as an object point and limits the search range. This allows the point cloud information processing block 73 to perform noise determination processing with high accuracy while reducing the amount of calculation required to calculate the noise evaluation function. Hereinafter, the same components as those in the first embodiment will be appropriately designated by the same reference numerals, and their description will be omitted. The point cloud information processing block 73 in the second embodiment functions as an "acquisition means," an "evaluation value calculation means," a "noise determination means," and a computer that executes a program.

[0061] FIG. 4(A) is a diagram showing a target point P1 of the current frame whose reflection intensity value is less than a threshold and its surrounding measurement points on a virtual plane facing the lidar 100. This virtual plane shows each measurement point at a position corresponding to the corresponding vertical and horizontal measurement direction. FIG. 4(B) is a diagram showing a target point P2 of the current frame whose reflection intensity value is greater than or equal to the threshold and its surrounding measurement points on the virtual plane. A dashed frame 91 indicates a spatial search range set for the target point P1, and a dashed frame 92 indicates a spatial search range set for the target point P2. The above-mentioned threshold is stored in advance in, for example, the memory 8, and is set to an adaptive value that takes into account the relationship between the reflection intensity value and the reliability as an object point.

[0062] Because the reflection intensity value of the target point P1 shown in FIG. 4A is less than the threshold, a normal-sized search range is set with the target point P1 at its center when calculating the noise evaluation value for the target point P1. In FIG. 4A, the horizontal index h has a value range of two points to the left and right of the target point P1, and the vertical index v has a value range of two points above and below the target point P1. The frame index t (not shown) also has a value range such that a predetermined number of recent frames (two frames in this example) are used to calculate the noise evaluation value. The point cloud information processing block 73 then considers points (74 points in this example) within the normal-sized search range set in space-time as reference points and calculates the noise evaluation value. In this way, the search range includes both a spatial range related to the measurement direction, determined by the ranges of the horizontal index h and vertical index v, and a temporal range related to the measurement time, determined by the range of the frame index t.

[0063] On the other hand, since the reflection intensity value of the target point P2 shown in FIG. 4(B) is equal to or greater than the threshold, a reduced-size search range is set with the target point P2 as its center in calculating the noise evaluation value of the target point P2. In FIG. 4(B), the horizontal index h is a value range equivalent to one point to the left and right of the target point P1, and the vertical index v is a value range equivalent to one point above and below the target point P1. The frame index t (not shown) also has a value range such that a predetermined number of recent frames (a number of frames less than the predetermined number of frames in the example of FIG. 4(A), assumed to be one here) are used in calculating the noise evaluation value. The point cloud information processing block 73 then regards points (26 points in this case) within the normal-size search range set in space-time as reference points and calculates the noise evaluation value.

[0064] In this way, the point cloud information processing block 73 reduces the spatiotemporal search range (value range of each index t, h, v) set using the target point of the current frame as a reference in calculating the noise evaluation value of the target point, compared to the normal search range, when the reflection intensity value of the target point is equal to or greater than the threshold. This reduces the number of reference points used for target points that are highly reliable as object points, making it possible to suitably reduce the amount of calculation required to calculate the noise evaluation value.

[0065] 5 is an example of a flowchart showing the procedure of noise point determination processing according to Example 2. The point cloud information processing block 73 executes the processing of this flowchart, for example, in step S04 of the flowchart in FIG.

[0066] First, the point cloud information processing block 73 sets a target point in the current frame (step S11). In this case, the point cloud information processing block 73 sets a measurement point in the current frame for which a noise evaluation value has not yet been calculated as the target point.

[0067] Next, the point cloud information processing block 73 determines a search range for searching for a reference point in space-time based on the reflection intensity value of the target point (step S12). In this case, for example, the point cloud information processing block 73 sets a normal-sized search range when the reflection intensity value is less than a threshold, and sets a reduced-sized search range when the reflection intensity value is equal to or greater than the threshold.

[0068] Then, the point cloud information processing block 73 calculates a noise evaluation value of the target point based on the search range set in step S12 (step S13). In this case, the point cloud information processing block 73 calculates a noise evaluation value based on formula (1) using the values ​​of each index h, v, and t within the set search range.

[0069] Then, the point cloud information processing block 73 determines whether the target point is a noise point (object point) based on the noise evaluation value calculated in step S13 (step S14). In this case, the point cloud information processing block 73 determines whether the target point is a noise point based on the result of comparing the noise evaluation value with the threshold value Th set according to, for example, the first embodiment.

[0070] Then, the point cloud information processing block 73 determines whether all points in the current frame have been set as target points (step S15). If all points in the current frame have been set as target points (step S15; Yes), the point cloud information processing block 73 determines that the noise determination process for the current frame has been completed and ends the processing of the flowchart. On the other hand, if the point cloud information processing block 73 determines that there are points in the current frame that have not been set as target points (step S15; No), it returns the processing to step S11 and sets the points in the current frame that have not been set as target points as target points.

[0071] Note that instead of or in addition to determining the search range based on the reflection intensity value of the target point, the point cloud information processing block 73 may determine the search range based on the measured distance of the target point. Generally, the probability of measuring the distance to an object decreases as the object becomes more distant, while the reliability of measuring the distance to an object that is relatively close is high. Therefore, when the measured distance of the target point is less than the threshold, the point cloud information processing block 73 sets a reduced-size search range. This also enables the point cloud information processing block 73 to perform noise determination processing with high accuracy while suppressing the amount of calculation required to calculate the noise evaluation function.

[0072] FIG. 6 is an example of a flowchart showing the procedure of noise point determination processing according to the second embodiment, taking into account the measurement distance of the target point.

[0073] First, the point cloud information processing block 73 sets a target point in the current frame (step S21). Next, the point cloud information processing block 73 determines a search range in which to search for a reference point in space-time based on at least one of the reflection intensity value or the measured distance of the target point (step S22). In this case, for example, the point cloud information processing block 73 may set a reduced-size search range when the reflection intensity value is equal to or greater than a threshold (first threshold), or may set a reduced-size search range when the measured distance is less than a threshold (second threshold). In another example, the point cloud information processing block 73 may set a reduced-size search range when the reflection intensity value of the target point is equal to or greater than the first threshold and the measured distance is less than a second threshold, or may set a reduced-size search range when either the reflection intensity value of the target point is equal to or greater than the first threshold or the measured distance is less than the second threshold is satisfied. Then, the point cloud information processing block 73 performs the processes of steps S23 to S25 in the same manner as steps S13 to S15.

[0074] As described above, the information processing device 1 according to the second embodiment functions as an information processing device having an acquisition unit and an evaluation value calculation unit. The acquisition unit acquires point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by a measurement device. The evaluation value calculation unit calculates a noise evaluation value for each target point based on a noise evaluation function that evaluates whether the target point corresponding to each data item is an object point that is a measured point on an object or a noise point generated by noise, based on data from reference points present within a search range set using the target point as a reference. In this case, the evaluation value calculation unit determines the search range based on at least one of the reflection intensity value or measurement distance corresponding to the target point. This enables the information processing device 1 to perform noise determination processing with high accuracy while reducing the amount of calculation required to calculate the noise evaluation function.

[0075] <Third Example> In the third embodiment, when the lidar 100 is fixedly installed outdoors or indoors, the point cloud information processing block 73 sets the spatiotemporal search range for calculating the noise evaluation function based on the density of object points in the previous frame (also referred to as the “previous frame”) one processing time before. Specifically, the point cloud information processing block 73 samples measurement points in the previous frame whose measurement direction is the same as or similar to the measurement direction of the target point according to a predetermined rule. Then, if the proportion (also referred to as the “object point proportion”) of the sampled measurement points (also referred to as “sample points”) that are object points is equal to or greater than a predetermined threshold, the point cloud information processing block 73 sets a reduced-size search range for the target point. In this way, the point cloud information processing block 73 suitably reduces the search range for target points that are likely to be object points, thereby reducing the amount of calculation of the noise evaluation value. The point cloud information processing block 73 in the third embodiment functions as an “acquisition means,” an “evaluation value calculation means,” a “noise determination means,” and a computer that executes a program. Furthermore, as described below, the processing of the third embodiment can be executed in combination with the second embodiment. Furthermore, if the LIDAR 100 can acquire information from other sensors, it does not need to be installed in a fixed location. In this case, the LIDAR 100 corrects the position indicated by the point cloud information by the amount of movement of the LIDAR 100.

[0076] Generally, when the lidar 100 is installed in a fixed position, stationary structural objects present within the field of view of the lidar 100 are continuously measured in each frame as a cluster of object points in the same measurement direction. Therefore, in a measurement direction in which object points are densely located in past frames, there is a high possibility that the same object will also be measured in the current frame. Taking the above into consideration, the point cloud information processing block 73 determines the reliability of the target point as an object point based on the density of object points in the previous frame in the measurement direction of the target point, and determines the size of the search range for the target point according to the reliability. This enables the point cloud information processing block 73 to reduce the amount of processing while maintaining the accuracy of determining object points and noise points.

[0077] 7 is a diagram showing a measured point (past target point) P3 and its surrounding measured points on a virtual plane, which have the same measurement direction as the target point in the immediately preceding frame. In addition, in FIG. 7, sample points used to calculate the object point ratio are indicated by hatching.

[0078] In this example, the point cloud information processing block 73 sets sample points in a cross shape with the past target point P3, which has the same measurement direction (i.e., the same scanning point) as the target point, at its center. In other words, the point cloud information processing block 73 sets, as sample points, measured points in the immediately preceding frame whose measurement direction is the same in the horizontal direction and similar in the vertical direction to the measurement direction of the target point, or measured points in the immediately preceding frame whose measurement direction is the same in the vertical direction and similar in the horizontal direction to the measurement direction of the target point. Specifically, the point cloud information processing block 73 sets, as sample points, two measured points each above and below the past target point P3 aligned in the vertical direction, and three measured points each on the left and right sides of the past target point P3 aligned in the horizontal direction.

[0079] The point cloud information processing block 73 then calculates the proportion of sample points determined to be object points out of a total of 10 sample points as the object point proportion. If the object point proportion is equal to or greater than a predetermined proportion, the point cloud information processing block 73 determines that the object points at past target point P3 are dense, and sets a reduced-size search range (see, for example, FIG. 4(B)) for the target point. On the other hand, if the object point proportion is less than the predetermined proportion, the point cloud information processing block 73 determines that the object points at past target point P3 are sparse, and sets a normal-size search range (see, for example, FIG. 4(A)) for the target point. As in the second embodiment, these search ranges include both a spatial range related to the measurement direction and a temporal range related to the measurement time.

[0080] In this way, the point cloud information processing block 73 determines sample points in the past frame by cross-sectional search centered on the past target point, and can calculate the object point ratio that accurately reflects the degree of density in the vicinity of the same measurement direction as the target point in the past frame.

[0081] It should be noted that the point cloud information processing block 73 can determine sample points that are effective for determining the density of object points based on other search rules, instead of determining sample points by cross search.

[0082] 8(A) is a diagram showing the arrangement of sample points of past target point P3 determined based on the second method. Here, the sample points are indicated by hatching. In the second method, the point cloud information processing block 73 determines eight measurement points that are adjacent to the left, right, top, bottom, and diagonally of past target point P3 as sample points. The point cloud information processing block 73 then calculates the proportion of sample points that are determined to be object points out of the total eight sample points as the object point proportion, and sets a search range of a size corresponding to the object point proportion.

[0083] 8(B) is a diagram showing the arrangement of sample points for past target point P3 determined based on the third method. Here, the sample points are indicated by hatching. In the third method, the point cloud information processing block 73 sets a frame 95 of a predetermined size centered on the past target point P3, and determines a total of 16 sample points by sampling from the measurement points of past frames that exist within the frame 95 according to a predetermined rule. Here, the frame 95 indicates a rectangular range that includes two measurement points above and below the past target point P3 and three measurement points to the left and right of the past target point P3. The point cloud information processing block 73 then calculates the proportion of sample points determined to be object points out of the total of 16 sample points as the object point proportion, and sets a search range of a size corresponding to the object point proportion.

[0084] In this way, the point cloud information processing block 73 can calculate the object point ratio that reflects the density of object points in past frames based on sample points determined by a method other than cross search, and accurately determine the search range for target points.

[0085] 9 is an example of a flowchart showing the procedure of noise point determination processing according to Example 3. The point cloud information processing block 73 executes the processing of this flowchart, for example, in step S04 of the flowchart in FIG.

[0086] First, the point cloud information processing block 73 sets a target point in the current frame (step S31). Next, the point cloud information processing block 73 selects a sample point that exists near a past target point in the immediately preceding frame and that has the same measurement direction as the target point in the immediately preceding frame (step S32). In this case, for example, the point cloud information processing block 73 selects the sample point in the immediately preceding frame according to the rules exemplified in FIG. 7, FIG. 8(A), or FIG. 8(B).

[0087] Next, the point cloud information processing block 73 calculates the object point ratio based on the sample points selected in step S32 (step S33). Then, the point cloud information processing block 73 determines a search range for searching for reference points in space-time based on the object point ratio (step S34). In this case, for example, the point cloud information processing block 73 sets a normal-sized search range when the object point ratio is less than a threshold, and sets a reduced-sized search range when the object point ratio is equal to or greater than the threshold.

[0088] Then, the point cloud information processing block 73 calculates a noise evaluation value of the target point based on the search range set in step S34 (step S35). In this case, the point cloud information processing block 73 calculates a noise evaluation value based on formula (1) using the values ​​of each index h, v, and t within the set search range.

[0089] Then, the point cloud information processing block 73 determines whether the target point is a noise point (object point) based on the noise evaluation value calculated in step S35 (step S36). In this case, the point cloud information processing block 73 determines whether the target point is a noise point based on the result of comparing the noise evaluation value with the threshold value Th set according to, for example, the first embodiment.

[0090] Then, the point cloud information processing block 73 determines whether all points in the current frame have been set as target points (step S37). If all points in the current frame have been set as target points (step S37; Yes), the point cloud information processing block 73 determines that the noise determination process for the current frame has been completed and ends the processing of the flowchart. On the other hand, if the point cloud information processing block 73 determines that there are points in the current frame that have not been set as target points (step S37; No), it returns the processing to step S31 and sets the points in the current frame that have not been set as target points as target points.

[0091] In a preferred example, the point cloud information processing block 73 may execute the third embodiment in combination with the second embodiment.

[0092] In this case, in a first example, the point cloud information processing block 73 limits the search range based on the reflection intensity value or the result of the determination of density at past target points. In this case, the point cloud information processing block 73 may switch between limiting the search range based on the reflection intensity value of the target points and limiting the search range based on the result of the determination of density at past target points, depending on the scene. For example, the point cloud information processing block 73 limits the search range based on the reflection intensity value of the target points when a predetermined condition is met under which it is determined that it is appropriate to limit the search range based on the reflection intensity value of the target points. On the other hand, the point cloud information processing block 73 limits the search range based on the result of the determination of density at past target points when a predetermined condition is met under which it is determined that it is appropriate to limit the search range based on the result of the determination of density at past target points.

[0093] In a second example, the point cloud information processing block 73 limits the search range based on the reflection intensity value of the target point, and also limits the search range based on the density determination result of past target points. In this case, the point cloud information processing block 73 sets a reduced-size search range when it is determined that the reflection intensity value of the target point is equal to or greater than a threshold, or when it is determined that past target points are dense. According to these examples, the point cloud information processing block 73 can appropriately limit the search range and reduce the amount of calculation processing.

[0094] As described above, the information processing device 1 according to the third embodiment functions as an information processing device having an acquisition unit and an evaluation value calculation unit. The acquisition unit acquires point cloud data, which is a collection of data representing points measured by a measurement device for each measurement direction. The evaluation value calculation unit calculates a noise evaluation value for a target point based on a noise evaluation function that evaluates whether the target point represented by each piece of data acquired at the current processing time is an object point that is a measured point on the object or a noise point generated by noise, based on data of reference points present within a search range set using the target point as a reference. In this case, the evaluation value calculation unit determines the search range based on the density of object points in the measurement direction of the target point at the processing time before the current processing time. This enables the information processing device 1 to perform noise determination processing with high accuracy while reducing the amount of calculation required to calculate the noise evaluation function.

[0095] <Fourth Example> 10 is a configuration diagram of a LIDAR system according to the fourth embodiment. In the fourth embodiment, the point cloud information processing block 73 of the control unit 7 and functions corresponding to the point cloud information processing block 73 are provided in a device separate from the LIDAR 100X. Hereinafter, elements of the fourth embodiment that are the same as those in the first to third embodiments will be appropriately designated by the same reference numerals, and their description will be omitted.

[0096] The LIDAR system according to the fourth embodiment includes a LIDAR 100X and an information processing device 200. In this case, the LIDAR 100X supplies the point cloud information generated by the point cloud information generation block 72 to the information processing device 200.

[0097] The information processing device 200 has a control unit 7A and a memory 8. The memory 8 stores information necessary for the control unit 7A to execute processing. The control unit 7A functionally has a point cloud information acquisition block 72A and a point cloud information processing block 73. The point cloud information acquisition block 72A receives point cloud information generated by the point cloud information generation block 72 of the LIDAR 100X and supplies the received point cloud information to the point cloud information processing block 73. The point cloud information processing block 73 performs the same processing on the point cloud information supplied from the point cloud information acquisition block 72A as the point cloud information processing block 73 according to each of the above-described embodiments.

[0098] The information processing device 200 may be realized by a driving assistance device. Furthermore, information on parameters necessary for processing may be stored in another device having a memory that can be referenced by the information processing device 200. Even with the configuration of this modified example, the information processing device 200 can accurately process the point cloud information generated by the lidar 100X.

[0099] In the above-described embodiments, the program can be stored using various types of non-transitory computer-readable media and supplied to a controller or the like that is a computer. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic storage media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical storage media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / Ws, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)).

[0100] Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above embodiments. Various modifications within the scope of the present invention that would be understood by those skilled in the art can be made to the configuration and details of the present invention. In other words, the present invention naturally includes various modifications and alterations that would be possible for those skilled in the art based on the entire disclosure, including the claims, and the technical ideas. Furthermore, the disclosures of the above-cited patent documents and other documents are incorporated herein by reference. [Explanation of symbols]

[0101] 1. Transmitter 2. Receiving section 3 Beam Splitter 5. Scanner 6 Piezo Sensors 7, 7A control section 8. Memory 100, 100X Lidar 200 Information processing device

Claims

1. an acquisition means for acquiring point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by the measurement device; and an evaluation value calculation means for calculating an evaluation value for the target point based on an evaluation function that evaluates whether the target point corresponding to each of the data is an object point that is a measured point of the object or a noise point generated by noise based on a distance index, a time index, and a space index that respectively represent the difference in measurement distance, the time difference, and the spatial difference between the target point and data of a reference point that exists within a search range that is set based on the target point and includes a spatial range related to the measurement direction and a temporal range related to the measurement time measured by the measurement device, the evaluation value calculation means determines the search range based on at least one of the reflection intensity value or the measurement distance corresponding to the target point; Information processing device.

2. 2. The information processing device according to claim 1, wherein the evaluation value calculation means selects the reference point that is temporally or spatially close to the target point from measured points included in a current frame, which is the point cloud data acquired by the acquisition means at the current processing time, and a past frame, which is the point cloud data acquired by the acquisition means at a time prior to the current processing time, based on the search range.

3. The information processing apparatus according to claim 1 , wherein the evaluation value calculation means, when the reflection intensity value is equal to or greater than a first threshold, sets the search range to be smaller than when the reflection intensity value is less than the first threshold.

4. The information processing device according to any one of claims 1 to 3, wherein the evaluation value calculation means sets the search range to be smaller when the measured distance is less than a second threshold value than when the measured distance is greater than or equal to the second threshold value.

5. 5. The information processing apparatus according to claim 1, further comprising a noise determination unit that determines the noise points in the point cloud data based on the evaluation value.

6. A control method executed by an information processing device, Acquire point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by the measurement device; calculating an evaluation value for each of the target points based on an evaluation function that evaluates whether the target points corresponding to each of the data are object points that are measured points on the object or noise points generated by noise based on distance indexes, time indexes, and space indexes that respectively represent the difference in measurement distance, time difference, and spatial difference between the data of a reference point that exists within a search range that is set based on the target point and includes a spatial range related to the measurement direction and a temporal range related to the measurement time measured by the measurement device; determining the search range based on at least one of the reflection intensity value or the measurement distance corresponding to the target point; Control method.

7. Acquire point cloud data, which is a collection of data representing the reflection intensity value and measurement distance for each point measured by the measurement device; calculating an evaluation value for each of the target points based on an evaluation function that evaluates whether the target points corresponding to each of the data are object points that are measured points on the object or noise points generated by noise based on distance indexes, time indexes, and space indexes that respectively represent the difference in measurement distance, time difference, and spatial difference between the data of a reference point that exists within a search range that is set based on the target point and includes a spatial range related to the measurement direction and a temporal range related to the measurement time measured by the measurement device; A program that causes a computer to execute a process of determining the search range based on at least one of the reflection intensity value or the measurement distance corresponding to the target point.

8. A storage medium storing the program according to claim 7.

Images