Control device, work machine, and data processing method

Abstract

Provided is a control device capable of detecting a low-reflection moving object. This control device comprises a processor. The processor acquires point cloud data indicating an object present in a detection region, and determines, from the point cloud data, a point at which the reflection intensity is less than a threshold value and which is a moving point, as a point indicating a low-reflection moving object.

Description

Control device, work machine, and data processing method

[0001] The present disclosure relates to a control device, a work machine, and a data processing method.

[0002] Conventionally, work machines such as bulldozers that operate manually or automatically at work sites have been known (see Patent Document 1).

[0003] Japanese Unexamined Patent Publication No. 2020-128155

[0004] Incidentally, at work sites where the above-described work machines operate, objects such as dust, snow, rain, fog, smoke, and steam are easily generated. Such an object is an object with low reflectivity that is in motion (hereinafter referred to as a "low-reflection moving object"). From the viewpoint of improving work safety and efficiency, it is desired to detect such low-reflection moving objects.

[0005] The present disclosure has been made in view of such circumstances, and aims to provide a control device, a work machine, and a data processing method capable of detecting low-reflection moving objects.

[0006] One aspect of the control device according to the present disclosure is a control device including a processor, wherein the processor acquires point cloud data indicating an object existing in a detection area, and determines, from the point cloud data, points whose reflection intensity is less than a threshold and which are moving points as points indicating a low-reflection moving object.

[0007] One aspect of the work machine according to the present disclosure includes the above-described control device.

[0008] One aspect of the data processing method according to the present disclosure includes the steps of: acquiring point cloud data indicating an object existing in a detection area; and extracting, from the point cloud data, points having low reflection intensity and being moving points as points indicating a low-reflection moving object.

[0009] According to the present disclosure, a control device and a work machine capable of detecting low-reflection moving objects can be provided.

[0010] Figure 1 is a side view of a work machine according to an embodiment. Figure 2 is a plan view of the work machine. Figure 3 is a block diagram showing the main components of the work machine. Figure 4 is a schematic diagram showing a state in which rocks are present on the movement path of the work machine. Figure 5 is a schematic diagram showing a state in which a vehicle is present on the movement path of the work machine. Figure 6 is a schematic diagram showing a state in which a puddle of water is present on the movement path of the work machine. Figure 7 is a schematic diagram showing a state in which dust is present on the movement path of the work machine. Figure 8 is a flowchart showing the process of non-obstacle detection processing. Figure 9 is a schematic diagram showing a buffered point cloud data frame.

[0011] The control device, work machine, and data processing method relating to this disclosure will be described below with reference to the drawings. The same reference numerals will be used for the same components. The information described below, along with the attached drawings, is for illustrative purposes only and does not represent the only possible embodiment.

[0012] [Embodiments] A control device, a work machine, and a data processing method according to embodiments of the present disclosure will be described with reference to Figures 1 to 9. Figure 1 is a side view of a work machine 1 on which the control device is mounted. Figure 2 is a top view of the work machine 1. Figure 3 is a block diagram showing the main configuration of the work machine 1.

[0013] The work machine 1 is a bulldozer. However, the work machine 1 is not limited to a bulldozer. The work machine 1 may be a variety of mobile work machines, including, for example, a dump truck, a hydraulic excavator, and a wheel loader.

[0014] Work machine 1 is a construction machine used for excavation, leveling, and transportation work at civil engineering and construction sites, as well as mining sites. First, let's briefly explain the configuration of work machine 1.

[0015] In the following explanation, the Cartesian coordinate system (X, Y, Z) shown in each figure may be used to describe the structure of the work machine 1.

[0016] Specifically, the X direction coincides with the front-to-back direction of the work machine 1. The X-positive side is the front of the work machine 1. The X-negative side is the rear of the work machine 1.

[0017] Furthermore, the Y direction refers to the left-right and width directions of the work machine 1. The Y-direction + side is the left side when viewing the work machine 1 from behind. The Y-direction - side is the right side when viewing the work machine 1 from behind.

[0018] The Z direction is the vertical direction of the work machine 1. The Z-positive side is the upper side of the work machine 1. The Z-negative side is the lower side of the work machine 1.

[0019] The work machine 1 has a vehicle body 11, a traveling device 12, a blade device 13, and a ripper device 14.

[0020] The vehicle body 11 has a driver's cab 111. The driver's cab 111 is equipped with a display unit 112 and an operating unit 113. The driver (not shown) operates the work machine 1 by operating the operating unit 113 to input operating signals.

[0021] The running gear 12 has a pair of left and right crawlers positioned at both ends of the vehicle body 11 in the vehicle width direction. Note that only the right crawler of the pair of left and right crawlers is shown in Figure 1.

[0022] The blade device 13 is a device used for tasks such as excavating and leveling soil. The blade device 13 is located on the front side of the vehicle body 11.

[0023] The ripper device 14 is a device used to dig up the ground or to break up hard ground. The ripper device 14 is located at the rear of the vehicle body 11.

[0024] Furthermore, the work machine 1 has a front object detection sensor 15, a rear object detection sensor 16, a position sensor 17, and a posture sensor 18.

[0025] The front object detection sensors 15 are installed, for example, at two locations on the left and right sides of the upper front of the vehicle body 11, such that the area in front of the work machine 1 is the detection area. The front object detection sensors 15 detect information about objects present in front of the work machine 1. The area detected by the front object detection sensors 15 is referred to as the front detection area. The front detection area is an example of a detection area.

[0026] The rear object detection sensors 16 are installed, for example, at two locations on the left and right sides of the upper rear of the vehicle body 11, such that the area behind the work machine 1 is the detection area. The rear object detection sensors 16 detect information about objects located behind the work machine 1. The area detected by the rear object detection sensors 16 is called the rear detection area. The rear detection area is an example of a detection area.

[0027] The objects detected by the front object detection sensor 15 and the rear object detection sensor 16 are various objects present at the work site. The objects detected by the front object detection sensor 15 and the rear object detection sensor 16 may include terrain, moving vehicles, and infrastructure.

[0028] The front object detection sensor 15 and the rear object detection sensor 16 are three-dimensional sensors capable of detecting, for example, the shape, position, and distance of an object to be detected in three dimensions.

[0029] Specifically, the front object detection sensor 15 and the rear object detection sensor 16 are LiDAR sensors (LiDAR: Light Detection and Ranging) that can measure the distance to an object by irradiating it with laser light and receiving the reflected light.

[0030] The three-dimensional data representing the three-dimensional shape of the detected object (in other words, information about the object) includes point cloud data consisting of multiple points. The point cloud data is a set of points corresponding to the reflection positions of a single laser beam emitted by the front object detection sensor 15 and the rear object detection sensor 16. Each point stores information about the object to which the laser beam was reflected.

[0031] Specifically, the information about the object stored at each point includes information about the relative distance between the front object detection sensor 15 and the rear object detection sensor 16 and the object, information about the relative position, and information about the relative height. In addition, the information about the object includes information about the reflectivity of the object.

[0032] Furthermore, the information regarding the relative position and relative height between the front object detection sensor 15 and the rear object detection sensor 16 and the object, which is stored at each point, may be three-dimensional coordinates based on a coordinate system (in other words, a local coordinate system) that is based on the front object detection sensor 15 and the rear object detection sensor 16.

[0033] The position sensor 17 detects the position of the work machine 1. The position sensor 17 is installed, for example, on the top of the vehicle body 11. The position sensor 17 has, for example, a GNSS receiver and uses the Global Navigation Satellite System (GNSS) to detect the position of the work machine 1.

[0034] GNSS includes the Global Positioning System (GPS). By using GNSS, the position of the work machine 1 in a global coordinate system defined by latitude, longitude, and altitude coordinate data can be detected. A global coordinate system is a coordinate system fixed to the Earth.

[0035] The attitude sensor 18 is an IMU (Inertial Measurement Unit). The attitude sensor 18 detects the inclination angle of the vehicle body 11 with respect to the horizontal plane, and / or the direction in which the vehicle body 11 is facing. The attitude sensor 18 is located on the rear side of the driver's cab 111.

[0036] Next, the functional configuration of the work machine 1 will be described with reference to Figure 3. Figure 3 is a block diagram showing the functional configuration of the work machine 1.

[0037] The work machine 1 has a control unit 2 on its body 11. The control unit 2 is an example of a control device and its main components include a CPU (Central Processing Unit) 21, RAM (Random Access Memory) 22, ROM (Read Only Memory) 23, etc.

[0038] The CPU 21 corresponds to an example of a processor. It reads a program corresponding to processing content from the ROM 23, loads the program into the RAM 22, and performs centralized control of the operation of each element of the work machine 1 in cooperation with the loaded program.

[0039] In addition, the control unit 2 is provided with an auxiliary memory 24, and calculation results and the like obtained by the CPU 21 are stored in the auxiliary memory 24.

[0040] Note that all or part of the control unit 2 may be formed of hard-wired circuits such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field-Programmable Gate Array).

[0041] The control unit 2 acquires detection values from the front object detection sensor 15, the rear object detection sensor 16, the position sensor 17, and the attitude sensor 18.

[0042] The control unit 2 acquires an operation signal input by a driver from an operation unit 113 provided in the driver's cab 111 of the vehicle body 11. The control unit 2 controls the traveling device 12, the blade device 13, and the ripper device 14 based on the acquired operation signal.

[0043] The control unit 2 may acquire an operation signal from a remote control terminal (not shown). In this case, the control unit 2 is communicatively connected to the remote control terminal via the communication unit 114.

[0044] The remote control terminal is arranged outside the work machine 1. In order to remotely control the work machine 1, the driver inputs an operation instruction to the remote control terminal. The remote control terminal transmits an operation signal to the work machine 1 based on the operation instruction input by the driver.

[0045] In addition, the control unit 2 controls a display unit 112 provided in the driver's cab 111 of the vehicle body 11.

[0046] The control unit 2 is communicatively connected to the management device 3 via the communication unit 114. The control unit 2 transmits information to the management device 3 via the communication unit 114. In addition, the control unit 2 receives information from the management device 3 via the communication unit 114.

[0047] The work machine 1 having the above configuration is moved by the traveling device 12. An operator operates the traveling device 12 via direct operation or remote control to move the work machine 1. Note that the work machine 1 may also move by autonomously driving the traveling device 12.

[0048] Furthermore, an operator may operate the blade device 13 and the ripper device 14 via direct operation or remote control to cause the work machine 1 to perform a desired work. Note that the work machine 1 may also autonomously drive the blade device 13 and the ripper device 14 to perform the desired work.

[0049] Furthermore, the control unit 2 controls the traveling operation of the work machine 1 based on point cloud data detected by the front object detection sensor 15 and the rear object detection sensor 16. Such control is referred to as automatic travel control for the work machine.

[0050] Automatic travel control is control autonomously implemented by the control unit 2 based on detection results from the front object detection sensor 15 and the rear object detection sensor 16.

[0051] In automatic travel control, the control unit 2 detects an object present in the traveling direction of the work machine 1 based on information related to objects detected by the front object detection sensor 15 and the rear object detection sensor 16, in other words, the point cloud data.

[0052] Specifically, the control unit 2 detects an object present in front of the work machine 1 based on point cloud data detected by the front object detection sensor 15. Furthermore, the control unit 2 detects an object present behind the work machine 1 based on point cloud data detected by the rear object detection sensor 16.

[0053] Then, the control unit 2 controls the traveling operation of the work machine 1 in accordance with the detected object.

[0054] Specifically, in automatic travel control, when the control unit 2 detects an obstacle in the traveling direction of the work machine 1, it brakes the work machine 1. Such control corresponds to an example of collision avoidance control.

[0055] Furthermore, in automatic driving control, if the control unit 2 detects an obstacle in the direction of travel of the work machine 1, it changes the path of the work machine 1 to avoid the obstacle. This type of control is also an example of collision avoidance control.

[0056] By the way, at the work site, there are both obstacles for the work machine 1 and objects that do not pose an obstacle (in other words, non-obstacle objects).

[0057] Obstacles present at the work site include, for example, rocks 41 (see Figure 4). At the work site, rocks 41 may be present in the path of the work machine 1. If the work machine 1 collides with a rock 41, the work machine 1 may be damaged.

[0058] Such rocks 41 are generally objects that have a high reflectance (hereinafter simply referred to as "reflectance") detected by the front object detection sensor 15 and the rear object detection sensor 16, and that do not move. Rocks 41 are an example of a highly reflectable stationary object.

[0059] Furthermore, highly reflective stationary objects are not limited to rocks. Highly reflective stationary objects may include various objects that have high reflectivity and do not move. Among stationary objects such as rocks, objects with a black surface (hereinafter referred to as "black stationary objects") have low reflectivity. These black stationary objects, like the puddle 43 described later, are examples of objects that have low reflectivity and do not move (in other words, low-reflectivity stationary objects).

[0060] Furthermore, obstacles present at the work site include, for example, other vehicles 42 (see Figure 5). At the work site, other vehicles 42 may be moving along the path of the work machine 1. If the work machine 1 collides with a vehicle 42, the work machine 1 may be damaged.

[0061] Such a vehicle 42 is a moving object with high reflectivity. Vehicle 42 is an example of a highly reflecting moving object.

[0062] Furthermore, highly reflective moving objects are not limited to vehicles. Highly reflective moving objects may include various objects that have high reflectivity and are in motion. A stationary vehicle may be considered a highly reflective stationary object.

[0063] Another example of an obstacle present at the work site is a puddle 43 (see Figure 6). At the work site, a puddle 43 may form in the path of the work machine 1. If the work machine 1 enters a puddle 43, the travel device 12 of the work machine 1 may get stuck.

[0064] The puddle 43 irregularly scatters or absorbs the laser light emitted from the front object detection sensor 15 and the rear object detection sensor 16. As a result, the reflection intensity detected by the front object detection sensor 15 and the rear object detection sensor 16 decreases. Also, the puddle 43 is essentially stationary.

[0065] Therefore, puddle 43 is an object with low reflectivity and is not moving. Puddle 43 is an example of a low-reflectivity stationary object.

[0066] On the other hand, non-obstacles present at the work site include, for example, dust 44 (see Figure 7). At the work site, dust 44 may be generated on the path of the work machine 1 due to strong winds.

[0067] However, even if the work machine 1 collides with the dust 44, no problem occurs to the work machine 1. Therefore, the dust 44 is not fundamentally an obstacle to the work machine 1 (in other words, it is a non-obstacle).

[0068] The particles constituting the dust 44 are small and have an irregular shape, resulting in low reflectivity detected by the front object detection sensor 15 and the rear object detection sensor 16. Furthermore, the particles constituting the dust 44 are moving. Therefore, the dust 44 is an object that is both low reflectivity and moving. Such dust 44 is an example of a low-reflectivity moving object.

[0069] Note that the low-reflectance moving object is not limited to dust 44. The low-reflectance moving object may include, for example, snow, rain, fog, smoke, and vapor. The low-reflectance moving object may include various objects that have low reflectance and are in motion.

[0070] As described above, various obstacles and non-obstacles exist at the work site. If the control unit 2 can accurately identify obstacles and non-obstacles, the control unit 2 can perform appropriate collision avoidance control.

[0071] On the other hand, if the control unit 2 cannot accurately identify obstacles and non-obstacles, the control unit 2 cannot perform appropriate collision avoidance control.

[0072] Specifically, when the control unit 2 detects dust 44 as an obstacle, the control unit 2 performs collision avoidance control. However, since dust 44 is not an obstacle, collision avoidance control is unnecessary. Such unnecessary collision avoidance control is undesirable because it reduces work efficiency.

[0073] Therefore, the control unit 2 according to this embodiment has a function to determine low-reflection moving objects as non-obstacles based on point cloud data detected by the front object detection sensor 15 and the rear object detection sensor 16.

[0074] The following describes the process by which the control unit 2 determines low-reflection moving objects as non-obstacles based on the point cloud data generated by the front object detection sensor 15 and the rear object detection sensor 16 (hereinafter referred to as the "non-obstacle determination process").

[0075] First, let's explain the overview of the non-obstacle detection process. The control unit 2 acquires point cloud information (in other words, point cloud data) indicating objects present in a predetermined detection area (in other words, the front detection area or the rear detection area). This function of the control unit 2 corresponds to its function as an acquisition unit.

[0076] Next, the control unit 2 extracts points with low reflectivity and that are moving from the point cloud information (in other words, point cloud data) as points indicating low-reflectivity moving objects. This function of the control unit 2 corresponds to the function of a control unit.

[0077] Hereafter, points that are moving in the point cloud data will be referred to as "moving points," and points that are not moving in the point cloud data will be referred to as "static points." The method for determining whether a point is moving or not will be described later.

[0078] Next, the control unit 2 determines that the points representing the extracted low-reflection moving objects are non-obstacles. This function of the control unit 2 corresponds to the function of the determination unit.

[0079] Next, with reference to Figure 8, the specific process of non-obstacle detection will be explained. The main component of the non-obstacle detection process shown in Figure 8 is the control unit 2 (specifically, the CPU 21). The non-obstacle detection process performed by the control unit 2 while the work machine 1 is moving forward will be explained below.

[0080] Some of the processes shown in Figure 8 may be omitted to the extent that it does not contradict the technical requirements. Furthermore, the order of the processes shown in Figure 8 may be changed to the extent that it does not contradict the technical requirements.

[0081] First, in step S101 of Figure 8, the control unit 2 acquires GNSS data from the position sensor 17.

[0082] Next, in step S102 of Figure 8, the control unit 2 acquires point cloud data from the front object detection sensor 15. This process is performed by the acquisition unit. The point cloud data acquired by the control unit 2 in step S102 is local coordinate data based on a local coordinate system with the front object detection sensor 15 as the reference point.

[0083] Next, in step S103 of Figure 8, the control unit 2 transforms the point cloud data acquired from the front object detection sensor 15 into global coordinate data based on the global coordinate system. This process is called the coordinate transformation process.

[0084] Specifically, in step S101, the control unit 2 uses the GNSS data acquired from the position sensor 17 to transform the point cloud data into global coordinate data.

[0085] Hereafter, point cloud data that has undergone coordinate transformation processing will be referred to as global point cloud data. Point cloud data before coordinate transformation processing may also be referred to as local point cloud data.

[0086] Next, in step S104 of Figure 8, the control unit 2 buffers the global point cloud data at a first time interval (for example, 0.1 seconds). This process is called the buffering process.

[0087] Note that the first time is not limited to 0.1 seconds. The first time may be determined appropriately according to the performance of the front object detection sensor 15.

[0088] Control unit 2 accumulates global point cloud data every hour through buffering. The global point cloud data accumulated every hour by control unit 2 is called a point cloud data frame.

[0089] Specifically, the control unit 2 buffers the point cloud data frames F1 to F6 shown in Figure 9 at each first time interval. The number of point cloud data frames is not particularly limited. The first time interval may be set according to the movement conditions of the work machine 1 (specifically, the movement speed and movement range).

[0090] This first time is determined experimentally in advance for each movement condition of the work machine 1, and stored in the memory unit. The control unit 2 retrieves the first time corresponding to the movement condition of the work machine 1 from the memory unit and sets it.

[0091] If machine 1 is working in the same location for a long period of time, setting the first time interval to a longer duration and using more older point cloud data frames enables robust decision-making. On the other hand, if machine 1 is moving at high speed, the importance of newer point cloud data frames containing data closer to the current time becomes higher than that of older point cloud data frames. Therefore, if machine 1 is moving at high speed, the first time interval should be set to a shorter duration.

[0092] Next, in step S105 of Figure 8, the control unit 2 divides the point cloud data frame accumulated in step S104 into two-dimensional cells arranged in a grid of predetermined size. This process is called grid division. Note that the cells may be three-dimensional cells.

[0093] Furthermore, the control unit 2 performs a process to associate the positions of cells between the accumulated point cloud data frames. In this example, the global coordinates of the cells are matched between the point cloud data frames.

[0094] Figure 9 is a schematic diagram showing buffered point cloud data frames F1 to F6. Figure 9 shows point cloud data frames F1 to F6. The numbered frames represent newer data.

[0095] Point cloud data frame F1 is the oldest point cloud data frame among point cloud data frames F1 to F6. Point cloud data frame F6 is the newest point cloud data frame among point cloud data frames F1 to F6.

[0096] Point cloud data frames F1 to F6 are all the point cloud data frames buffered in step S104. Therefore, the number of point cloud data frames to be buffered is 6. However, the number of point cloud data frames to be buffered is not particularly limited.

[0097] Each of the point cloud data frames F1 to F6 has a predetermined number of cells (36 in this embodiment). Note that Figure 9 is a schematic diagram showing a buffered point cloud data frame. Therefore, the number of cells is not limited to 36. Figure 9 shows a state in which the global coordinates of the cells of point cloud data frames F1 to F6 coincide.

[0098] Next, in step S106 of Figure 8, the control unit 2 calculates the cell occupancy rate R for the most recent point cloud data frame F6. This process is called the occupancy rate calculation process.

[0099] The cell occupancy rate R is an index that indicates the proportion of points present in a specific cell (hereinafter referred to as the "target cell") during a predetermined time. The cell occupancy rate R is the proportion of point cloud data frames in which one or more points are present in the target cell, out of all point cloud data frames accumulated in step S104. The cell occupancy rate R is calculated by the following formula (1).

[0100] R: Cell occupancy rate (%) N1: Number of point cloud dataframes containing points in the target cell N2: Total number of buffered point cloud dataframes

[0101] Referring to Figure 9, the cell occupancy rate R will be explained in more detail. First, since the number of frames in the buffered point cloud data frames F1 to F6 is 6, N2 in equation (1) above is 6.

[0102] Furthermore, if the target cell is cell C1 of point cloud data frames F1 to F6, then only cell C1 in point cloud data frame F6 contains a point. Therefore, N1 in equation (1) above is 1. Thus, the cell occupancy rate R of cell C1 in the most recent point cloud data frame F6 is approximately 17%.

[0103] Furthermore, if the target cell is cell C2 in point cloud data frames F1 to F6, then all cells C2 in point cloud data frames F1 to F6 contain points. Therefore, N1 in equation (1) above is 6. Thus, the cell occupancy rate R of cell C1 in the most recent point cloud data frame F6 is 100%.

[0104] Furthermore, if the target cell is cell C3 of point cloud data frames F1 to F6, then point P is included in cell C3 of point cloud data frames F3 to F6. Therefore, N1 in equation (1) above is 3. Thus, the cell occupancy rate R of cell C3 in the most recent point cloud data frame F6 is 50%.

[0105] In this manner, the control unit 2 calculates the cell occupancy rate R for all cells in the most recent point cloud data frame F6. The control unit 2 then stores the calculated cell occupancy rate R.

[0106] Next, in step S107 of Figure 8, the control unit 2 extracts moving points from all points P included in the most recent point cloud data frame F6. This process is called the moving point extraction process. Note that in Figure 9, only some of the points P included in the point cloud data frame F6 are shown.

[0107] The control unit 2 determines whether or not each point P included in the point cloud data frame F6 is a moving point.

[0108] Specifically, the control unit 2 determines that points P in the point cloud data frame F6 that are included in cells where the cell occupancy rate R is less than a predetermined first threshold are moving points.

[0109] Meanwhile, the control unit 2 determines that points P in the point cloud data frame F6 that are included in cells where the cell occupancy rate R is equal to or greater than a predetermined first threshold are static points.

[0110] In this example, the first threshold is 50%. However, the first threshold is not limited to 50%. The first threshold may be determined appropriately depending on the performance of the front object detection sensor 15 and the rear object detection sensor 16.

[0111] Here, we will briefly explain why a point P located within a cell whose cell occupancy rate R is less than a predetermined first threshold is determined to be a moving point. A moving point can be understood as a point that represents an object in motion.

[0112] Cell C2 (see Figure 9), which has a high cell occupancy rate R, contains point P in all of the point cloud data frames F1 to F6. Since the global coordinates of cell C2 are the same across point cloud data frames F1 to F6, it is highly likely that point P contained in cell C2 has not moved, at least during the time when point cloud data frames F1 to F6 were acquired. Therefore, the control unit 2 determines that point P contained in cell C2 in point cloud data frame F6 is a static point (a point that is not moving).

[0113] On the other hand, cell C1 (see Figure 9), which has a low cell occupancy rate R, does not contain any points at least during the time when point cloud data frames F1 to F5 were acquired. Under these circumstances, point P that enters cell C1 in point cloud data frame F6 is highly likely to be a moving object. Therefore, the control unit 2 determines that point P included in cell C1 of point cloud data frame F6 is a moving point.

[0114] The control unit 2 extracts the point P, which it has determined to be a moving point, from the point cloud data frame F6. Hereafter, the moving point extracted by the control unit 2 in step S107 will be referred to as the moving point of the point cloud data frame F6 (in other words, the most recent point cloud data frame).

[0115] In the moving point extraction process described above, the control unit 2 extracts moving points using the relationship between the cell occupancy rate and the first threshold as the occupancy condition. However, the control unit 2 may also extract moving points using the relationship between the cell occupancy time and the first time threshold as the occupancy condition.

[0116] In this case, cell occupancy time is an indicator that shows the time a point exists in a specific target cell. Cell occupancy time is the time in all point cloud data frames accumulated in step S104 that one or more points are contained in the target cell.

[0117] When the relationship between cell occupancy time and the first time threshold is used as the occupancy condition, the control unit 2 may determine that points P in the point cloud data frame F6 that are included in cells whose cell occupancy time is less than the first time threshold are moving points.

[0118] Next, in step S108 of Figure 8, the control unit 2 extracts points indicating low-reflectivity objects from the moving points of the point cloud data frame F6 extracted in step S107. This process is called the low-reflectivity point extraction process.

[0119] The control unit 2 determines whether point P represents a low-reflectivity object for all moving points included in the point cloud data frame F6. Note that point P in Figure 9 represents a point included in a cell of the point cloud data frame.

[0120] Specifically, the control unit 2 determines that points in the point cloud data frame F6 whose reflection intensity is less than a predetermined second threshold are points indicating low-reflectivity objects.

[0121] On the other hand, the control unit 2 determines that any moving point in the point cloud data frame F6 whose reflection intensity is above a predetermined second threshold is not a point indicating a low-reflectivity object.

[0122] The second threshold may be determined experimentally depending on the characteristics of the front object detection sensor 15 and the rear object detection sensor 16. Alternatively, the second threshold may be determined depending on the measurement environment and installation location of the front object detection sensor 15 and the rear object detection sensor 16. The second threshold may also be determined appropriately depending on the characteristics of the object detected as a low-reflectivity object.

[0123] The control unit 2 determines whether a moving point in the point cloud data frame F6 is a point indicating a low-reflectivity object, based on the information regarding the reflectivity stored in the moving point of the point cloud data frame F6 and a second threshold. This information regarding reflectivity is detected by the front object detection sensor 15.

[0124] The control unit 2 extracts points determined to be low-reflectivity objects from the moving points of the point cloud data frame F6. The points extracted by the low-reflectivity point extraction process described above are moving points with a reflectivity below the second threshold (in other words, moving points with low reflectivity) among all the points included in the point cloud data frame F6.

[0125] The control unit 2 extracts moving points with low reflectivity as points indicating low-reflectivity moving objects. In other words, the control unit 2 extracts moving points with low reflectivity as points indicating dust 44 (see Figure 7).

[0126] Next, in step S109 of Figure 8, the control unit 2 determines that the object indicated by the moving point with low reflectivity extracted in step S108 (i.e., a low-reflectivity moving object) is a non-obstacle. Then, the control unit 2 terminates the non-obstacle determination process. The control unit 2 repeatedly executes the non-obstacle determination process at appropriate timings.

[0127] Next, we will briefly explain the control that the control unit 2 performs after the non-obstacle detection process described above.

[0128] The control unit 2 may perform the collision avoidance control described above based on the results obtained from the non-obstacle detection process.

[0129] Specifically, the control unit 2 removes moving points with low reflection intensity that were determined to be non-obstacles in step S109 above from the most recent point cloud data frame F6. This type of processing is called noise filtering.

[0130] The point cloud data frame F6, from which moving points with low reflection intensity have been removed by noise filtering, is called the filtered point cloud data frame. The control unit 2 performs collision avoidance control based on the filtered point cloud data frame.

[0131] The filtered point cloud data frame contains few or no points indicating dust particles 44. Therefore, even when dust particles 44 are generated in front of the work machine 1 (see Figure 7), the control unit 2 does not identify the dust particles 44 as an obstacle. As a result, the control unit 2 does not perform unnecessary collision avoidance control for the dust particles 44.

[0132] Furthermore, the control unit 2 may generate an image showing objects present in the detection region (specifically, the front detection region or the rear detection region) based on the filtered point cloud data frame. The control unit 2 may also display the generated image showing the objects on the display unit 112.

[0133] The filtered point cloud data frame contains few or no points indicating dust particles 44. Therefore, even when dust particles 44 are generated in front of the work machine 1 (see Figure 7), the display unit 112 displays an image showing objects other than dust particles 44 present in the detection area. By looking at the display unit 112, the user can identify objects other than dust particles 44 present in the detection area.

[0134] Furthermore, the control unit 2 may generate a terrain map of the work site based on the filtered point cloud data frame.

[0135] The terrain map generated based on the filtered point cloud data frame contains little to no information indicating dust 44. Therefore, the control unit 2 can generate a map that accurately represents the terrain of the work site.

[0136] The control unit 2 may acquire information about dust 44 (hereinafter referred to as "dust information") from moving points with low reflectivity that were determined to be non-obstacles in step S109.

[0137] The dust information may include, for example, information regarding the location where the dust was generated (hereinafter referred to as "dust location information"), and / or information regarding the time when the dust was generated (hereinafter referred to as "dust time information").

[0138] The control unit 2 may also generate a terrain map that associates dust location information and dust time information. Such a terrain map can provide the user with dust location information and dust time information along with terrain information. By looking at the terrain map, the user can understand the locations and times where dust may be generated.

[0139] The control unit 2 may also send filtered point cloud data frames to the management device 3 via the communication unit 114. The management device 3 may generate a topographic map of the work site based on the filtered point cloud data frames received from the control unit 2.

[0140] The terrain map generated based on the filtered point cloud data frame does not contain information indicating dust particles 44. Therefore, the control device 3 can generate a map that accurately represents the terrain of the work site.

[0141] Furthermore, the control unit 2 may send the above-mentioned dust information along with the filtered point cloud data frame to the management device 3.

[0142] The control device 3 may generate a topographic map that associates dust location information and dust time information. Such a topographic map can provide the user with dust location information and dust time information along with the topographic information.

[0143] (Other Embodiments) In the embodiments described above, the control unit 2, which is one embodiment of the control device of the present disclosure, is mounted on the work machine 1. However, the control device may be mounted on a terminal (hereinafter referred to as an "external terminal") that is connected to the work machine 1 via a network (not shown), such as a management device 3. In such a configuration, the control device acquires point cloud data detected by the front object detection sensor 15 and / or rear object detection sensor 16 from the work machine 1 via the network. The control device then extracts points with low reflectivity and that are moving points from the acquired point cloud data as points indicating low-reflectivity moving objects. A control device mounted on an external terminal can also perform the non-obstacle determination process described with reference to Figure 8.

[0144] Furthermore, if the control device is mounted on an external terminal, the control device may generate a topographic map of the detection area based on filtered point cloud data obtained by removing points indicating low-reflection moving objects from the point cloud data acquired from the work machine. In this case, the external terminal may be connected to multiple work machines via a network. The control device may also acquire point cloud data from multiple work machines. The control device may generate a topographic map of the detection area based on filtered point cloud data obtained by removing points indicating low-reflection moving objects from the point cloud data acquired from multiple work machines.

[0145] Furthermore, the control device may be mounted on a detection device (e.g., LiDAR) that is fixedly installed at the work site. In other words, the control device may be mounted on a device that does not move. The configuration of such a control device is substantially the same as the configuration of the control unit 2 according to the above embodiment.

[0146] (Operation and Effects of this Embodiment) According to the control unit 2 of this embodiment as described above, low-reflectivity moving objects such as dust, snow, rain, fog, smoke, and steam can be detected. The work machine 1 can perform appropriate collision avoidance control based on the detection results of the control unit 2. In addition, the management device 3 can generate a terrain map that accurately represents the terrain of the work site based on the detection results of the control unit 2. Other operations and effects obtained from the control unit 2 of this embodiment are as described.

[0147] 1. Working Machinery 11. Body 111. Driver's Cab 112. Display Unit 113. Operation Unit 114. Communication Unit 12. Running Gear 13. Blade Device 14. Ripper Device 15. Front Object Detection Sensor 16. Rear Object Detection Sensor 17. Position Sensor 18. Attitude Sensor 2. Control Unit 21. CPU 22. RAM 23. ROM 24. Auxiliary Memory 3. Management Device 41. Rock 42. Vehicle 43. Puddle 44. Dust F1-F7. Point Cloud Data Frame C. Cell P. Point

Claims

1. A control device comprising a processor, wherein the processor acquires point cloud data indicating objects present in a detection region, and determines from the point cloud data that points whose reflectance intensity is below a threshold and are moving points are points indicating low-reflection moving objects.

2. The control device according to claim 1, wherein the processor removes points representing the low-reflectivity moving object from the point cloud data.

3. The control device according to claim 1, wherein the processor buffers the point cloud data in the first time to obtain a plurality of point cloud data frames, divides the plurality of point cloud data frames into cells of a predetermined size, calculates the cell occupancy rate of all cells in the latest point cloud data frame, and extracts points included in the cells in the latest point cloud data frame whose cell occupancy rate is less than a predetermined value as the moving points.

4. The control device according to claim 1, wherein the processor detects an obstacle in the detection region based on filtered point cloud data obtained by removing points indicating the low-reflectivity moving object from the point cloud data.

5. The control device according to claim 1, wherein the processor generates a terrain map of the detection region based on filtered point cloud data obtained by removing points indicating the low-reflectivity moving object from the point cloud data.

6. The control device according to claim 5, wherein the processor obtains information from the point cloud data regarding the location where the low-reflection moving object originated, and generates the terrain map associated with the location information.

7. A work machine equipped with the control device described in claim 1.

8. A data processing method comprising the steps of: acquiring point cloud data indicating objects present in a detection region; and extracting points from the point cloud data that have low reflectivity and are moving points, as points indicating low-reflectivity moving objects.

9. The data processing method according to claim 8, further comprising the step of removing points indicating the low-reflectance moving object from the point cloud data.

10. The data processing method according to claim 8, further comprising: before the extraction step, the steps of: buffering the point cloud data in a first time to obtain a plurality of point cloud data frames; dividing the plurality of point cloud data frames into cells of a predetermined size; calculating the cell occupancy rate of all cells in the latest point cloud data frame; and extracting points in the latest point cloud data frame that are included in the cells whose cell occupancy rate is less than a predetermined value as the moving points.

11. The data processing method according to claim 8, further comprising the step of detecting an obstacle present in the detection region based on filtered point cloud data obtained by removing points indicating the low-reflectance moving object from the point cloud data.

12. The data processing method according to claim 8, further comprising the step of generating a topographic map of the detection region based on filtered point cloud data obtained by removing points indicating the low-reflection moving object from the point cloud data.

13. The data processing method according to claim 12, further comprising the step of obtaining information from the point cloud data regarding the location where the low-reflectivity moving object originated, prior to the step of generating the terrain map.

Images