Traffic detection system
The traffic detection system uses low-cost LiDAR devices to scan and process point cloud data for accurate vehicle type differentiation and counting, addressing high-cost and inefficiency issues in conventional systems, particularly for motorcycles, and enhancing accuracy in crowded traffic scenarios.
Patent Information
- Application Number
- JP2023079742
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-14
- Publication Date
- 2024-11-26
AI Technical Summary
Conventional traffic detection systems using laser sensors and ultrasonic sensors face issues with high costs, reliance on gantry-type installations, and increased expenses due to the use of multiple detection devices, particularly when distinguishing between different vehicle types, especially motorcycles, and are not efficient in crowded traffic conditions.
A traffic detection system utilizing a low-cost laser irradiation device installed beside the road, employing LiDAR technology to scan and analyze reflected light, processing point cloud data to distinguish vehicle types and count vehicles accurately, with noise filtering and multiple laser setups for enhanced accuracy.
Enables precise detection of vehicle types and counts, even in crowded conditions, reducing costs and improving accuracy by using low-cost LiDAR devices and advanced data processing techniques, effectively distinguishing between motorcycles and other vehicles.
Smart Images

Figure 2024163971000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a traffic detection system that collects various information about vehicles traveling on roads such as expressways. For example, the present invention relates to a traffic detection system that captures moving objects by scanning the road surface, analyzes the outlines and movements of the moving objects, and detects the number of vehicles traveling on the road, the type of vehicle (large vehicle, ordinary passenger car, or motorcycle), the vehicle speed, the vehicle length, congestion, the presence or absence of traffic trouble, etc. [Background technology]
[0002] 2. Description of the Related Art In the prior art, a passing vehicle detection system is known that measures various data such as the number of passing vehicles traveling on a road, the passing speed, and the length of the vehicle. For example, a loop coil type passing vehicle detection system using a device called a "loop coil" is known. The loop coil type traffic detection system is a system in which a conductor called a loop coil is buried underground on a road. When a vehicle with a metal body passes, the inductance of the loop coil buried underground changes. The loop coil type traffic detection system is a system in which two loop coils are buried on the road at a predetermined interval (for example, 6.9 m) and detects the change in inductance of these two loop coils at each of the two points to detect the passing of vehicles at two points and measure the traffic volume. The inductance change of the loop coil can be detected with high accuracy, and the passing time of the vehicle at the two points where the loop coils are buried can be detected. Therefore, by using the loop coil type traffic detection system, not only the number of passing vehicles but also the passing speed and length of each vehicle can be measured with high accuracy. With current technology, it is said that such data can be detected with an accuracy of 99% or more.
[0003] There are also passing vehicle detection systems that use laser sensors or ultrasonic sensors. The laser sensor and ultrasonic sensor shown in Patent Document 1, which is prior art, emit a laser beam or ultrasonic wave across a road at a specified location, as shown in Figures 9(a) and 9(b), measure the time it takes for the emitted laser beam or ultrasonic beam to be reflected and received from an object, such as the road surface or the surface of a vehicle traveling on the road, and measure the distance to the object based on the measured time (Patent Document 1). According to Patent Document 1, if the measurement data is analyzed by a control device, it can be determined that there are no passing vehicles on the road if the object to which the laser beam or ultrasonic beam is irradiated is flat, that is, if there are no vehicles passing on the road and the pattern received is a reflection from the road surface, and if the object to which the laser beam or ultrasonic beam is irradiated has unevenness, it can be determined that there are passing vehicles on the road as a pattern received by reflecting a part of it from a vehicle passing on the road. This passing vehicle detection system using a laser sensor or ultrasonic sensor can be installed in space without damaging the road, so it is said that there are no problems such as the need for construction involving burying in the road, costs, and short durability, which are seen in the above-mentioned system using the passing vehicle detection system using a loop coil.
[0004] A technology disclosed in Patent Document 2 is known as a vehicle detection system using an excellent laser sensor in the prior art. In the technology disclosed in Patent Document 2, two laser sensors are used to accurately detect the number of vehicles passing on a road, the speed of each vehicle, and the length of each vehicle. As shown in FIG. 10, for example, a laser sensor is installed at point A on a road, and a laser is irradiated to a vehicle passing on the road directly below and a reflection is received to detect the intrusion of the vehicle, and a laser sensor is also installed at point B on a road a little away from the road, and a laser is irradiated to a vehicle passing on the road directly below and a reflection is received to detect the intrusion of the vehicle, and by comparing the data from both, the number of vehicles passing on the road, the speed of each vehicle, and the length of each vehicle can be accurately detected.
[0005] As described above, by using the conventional vehicle detection system disclosed in Patent Document 2, it is expected that various data such as the number of passing vehicles, passing speed, and vehicle length of vehicles traveling on a road can be measured with relatively high accuracy.
[0006] In the technology disclosed in Patent Document 2, in addition to vehicle detection using a laser, a surveillance camera that captures visible images is also used as a means for directly monitoring the traffic conditions on the road itself. In fact, road traffic management that monitors the traffic conditions on the road based on images captured by surveillance cameras is being introduced.
[0007] Here, in a traffic management system based on images captured by surveillance cameras, human visual inspection is important, but management that relies solely on human visual inspection is not sufficient. It is necessary to perform image analysis through computer processing to automatically monitor the traffic conditions on the roads and detect traffic troubles that occur on the roads.
[0008] [Patent Document 1] JP 2003-308591 A [Patent Document 2] Patent Publication No. 2022-118950 DISCLOSURE OF THEINVENTION [Problem to be solved by the invention]
[0009] As described above, the vehicle detection systems using lasers in the prior art disclosed in Patent Documents 1 and 2 detect changes in the time it takes for reflected light to be received from the road, and detect the entry and exit of passing vehicles into and out of a laser irradiation surface set on the road. In other words, they detect the ON / OFF of the passage of a vehicle through the laser irradiation surface. Originally, traffic counters evolved from the loop coil type, and were designed to detect whether a vehicle had passed over a loop coil buried in the road by observing changes in the coil current, thus sensing the ON / OFF of the vehicle's passage through the so-called buried loop area. Therefore, it was thought that it would be sufficient to detect whether a vehicle had passed even if the detection means was replaced by changes in the reception time of the reflected light from a laser, which can be installed without contact, rather than changes in the coil current of the loop coil.
[0010] The traffic detection system in Patent Document 2 is a technology that makes it possible to identify not only passing vehicles but also vehicle types by using images captured by a surveillance camera that captures visible images. It is also worthy of praise for its ability to accurately detect passing vehicles even with images captured during nighttime hours.
[0011] However, the traffic detection system in Patent Document 1 is based on the premise that a gantry-type bridge is used for the installation of the laser irradiation device, which results in a problem of high costs. In addition, the traffic detection system in Patent Document 1 basically detects the passage of a vehicle by turning it on and off. However, the traffic detection system in Patent Document 1 can still grasp the vehicle length if two laser irradiation devices are installed.
[0012] The traffic detection system of Patent Document 2 also has the drawback of being costly. The traffic detection system of Patent Document 2 uses a camera device in addition to a laser irradiation device to detect the number of passing vehicles by category, such as large vehicles, regular passenger cars, small light vehicles, and motorcycles, and is a technology that meets the need to know the number of passing vehicles by category, such as large vehicles, regular passenger cars, small light vehicles, and motorcycles, but the use of a camera device in addition to a laser irradiation device increases the cost. In particular, there is a risk of high costs in a system configuration that uses a visible light camera for daytime use and an infrared camera for nighttime use as the camera device.
[0013] In recent years, the measurement resolution of low-cost laser projection devices has improved, making it possible to use low-cost laser projection devices with scanning speeds that are sufficiently fast compared to the speed of passing vehicles. In order to solve the problems in the prior art described above, the traffic detection system of the present invention aims to provide a traffic detection system that uses a low-cost laser irradiation device, not limited to a gantry type, installed near the side road of a road, measures the distance to a moving object on the road from the reception time of the reflected light of the laser irradiation light from the road, and uses the data obtained according to the scanning speed to estimate the type of vehicle passing through, that is, even the type of vehicle by category, such as large vehicle, ordinary passenger car, compact light vehicle, motorcycle, etc., and detects the number of passing vehicles by each category.
[0014] In particular, in cases where vehicle lanes on roads are divided into categories according to vehicle type, for example, when a motorcycle-only lane is set up, the motorcycle-only lane is often set up on the low-speed lane side, and it is expected that by installing a low-cost laser irradiation device on the side of the side road close to the motorcycle-only lane, it will be possible to accurately detect motorcycles passing through the motorcycle-only lane, as well as large vehicles, ordinary passenger cars, and small and light vehicles passing through the higher-speed lane side. [Means for solving the problem]
[0015] In order to achieve the above-mentioned object, the traffic detection system of the present invention includes a LiDAR device that is arranged beside or above a road, irradiates a laser light that scans so as to intersect with the road surface and receives reflected light, and obtains at least irradiation angle data of the scan, distance data to a reflecting object calculated from the measurement time difference, and measurement time data of the scan, a point cloud plot processing unit that plots the irradiation angle data, the distance data, and the measurement time data obtained by the LiDAR device as point cloud data in space-time along a time axis, and a point cloud separation processing unit that estimates and separates a group of point clouds caused by a passing vehicle and its driver passing through the road from a point cloud trajectory formed by the point cloud data plotted by the point cloud plot processing unit, and is characterized in that it is equipped with a vehicle detection unit that detects the group of point clouds estimated and separated by the point cloud separation processing unit as the passing vehicle passing through the road.
[0016] Here, there are two possible types of space-time: The first type of time space is a time space in which a coordinate system formed by the Cartesian coordinates of the position in the road width direction and the height from the road surface obtained at each scanning measurement time is expanded along the time axis. When using this first type of space-time, in plotting point cloud data in space-time by the point cloud plot processing unit 130, if multiple data are acquired in one same scan, the data are sorted and plotted in the road width direction with respect to the Cartesian coordinates corresponding to the measurement time of the scan. In other words, for each scan, the point cloud obtained at the scan time in space-time can be sorted and plotted by the position in the road width direction. The second type of time space is a time space in which a coordinate system formed by polar coordinates of the irradiation angle and distance obtained at each measurement time of the scan is expanded along the time axis. When using this second type of space-time, in plotting point cloud data in space-time by the point cloud plot processing unit 130, if multiple data are acquired in one same scan, they are sorted and plotted by irradiation angle for the polar coordinates corresponding to the measurement time of the scan. In other words, if the point cloud obtained at the scan time in space-time for each scan is plotted by irradiation angle, it is sufficient to sort and plot by position in the road width direction.
[0017] Furthermore, with the above configuration, if the scanning speed of the laser light irradiation is sufficiently fast compared to the speed of the passing vehicle, the irradiation angle data, distance data, and measurement time data obtained by the measurement can be plotted in space-time along the time axis to obtain point cloud data. Although some width error is mixed in the time axis direction depending on the speed of the passing vehicle, this point cloud data will be point cloud data that follows the contour of the passing vehicle. By analyzing this point cloud data in the vehicle detection unit, it is possible to detect passing vehicles passing on the road. In the above configuration, by focusing on the presence or absence of changes on the time axis of the point cloud trajectory formed by the group of point cloud data, the point cloud separation processing unit is able to determine whether the object is a moving vehicle moving on the road or a stationary object on the road surface.
[0018] Here, it is preferable that the point cloud separation processing unit includes a vehicle detection unit, and the vehicle type of the passing vehicle is estimated from a point cloud trajectory formed by the group of point cloud data. If the laser light irradiation device is installed on the side of the road, only a part of the contour on the side opposite to the laser light irradiation device is obtained, but it is expected that the information is sufficient to identify the vehicle type. For example, if the vehicle is a passenger car and the laser light irradiation device is installed on the side of the road on the left side of the road in the traveling direction, the point cloud data only contains a part of the contour of the left side of the passenger car, but the contours of the front end, front, left door area, rear, and rear end of the vehicle are obtained, and it can be said that the information contains enough information to estimate that it is a passenger car even if there is a shift in the time axis direction as described above. The vehicle detection unit estimates the vehicle type by analyzing the point cloud data that contains only a part of the contour. For example, the vehicle detection unit can determine whether the type of a passing vehicle is a four-wheeled vehicle or a two-wheeled vehicle based on the height and length estimated from the point cloud trajectory formed by a group of point cloud data. Also, for example, it is preferable to have a configuration including a vehicle speed estimation unit that estimates the speed of a passing vehicle from the height, length, and change on the time axis estimated from a point cloud trajectory formed by a group of point cloud data. In other words, since a group of point cloud data is a trajectory plotted in space-time with a time axis, the speed of a passing vehicle can be estimated by analyzing the estimated height, length, and changes on the time axis.
[0019] Next, we will discuss the removal of noise expected in the traffic detection system of the present invention. Although foreign objects such as floating debris are very temporary and may have little impact, if the weather is rainy or snowy, the rain and snow may frequently be mixed in as noise. In particular, snow takes longer to pass through the irradiation area of the laser irradiation device than rain, so it may be more likely to be mixed in as noise. When analyzing rain and snow noise, there are two types of noise: noise caused by falling rain and snow, and noise caused by vehicles splashing up water that has fallen to the ground. For the rain and snow filter process, the point cloud separation processor is effective in filtering out objects that are not continuous in time from the objects detected by the LiDAR device, as the former. For the latter, it is effective to filter out objects that are low above the ground and objects with blurred and unclear edges that are detected by the LiDAR device. It is expected that this rain and snow filter process can also effectively remove floating debris. It is expected that this rain / snow filter process will effectively determine whether a group of point cloud data represents a vehicle moving on a road or weather-related noise, i.e., rain or snow passing in front of the laser light projection device.
[0020] Next, in the traffic detection system of the present invention, the accuracy can be improved by using multiple irradiation surfaces from the laser light emitting device with different positions and angles for each irradiation surface, or by using multiple laser light emitting devices with different irradiation directions for the road surface. In other words, by performing point cloud plot processing using the measurement results from multiple laser irradiation surfaces, not only does the vehicle detection accuracy improve, but if the interval between the irradiation surfaces of the laser light emitting devices is known, it becomes easier to estimate the speed of passing vehicles, etc.
[0021] Next, in the traffic detection system of the present invention, it is also preferable to have a count processing unit that counts the number of passing vehicles based on the detection results of the vehicle detection unit. Road managers have a need to know the number of passing vehicles by type, and the count results can be notified to them. Effect of the Invention
[0022] With the above-mentioned configuration, the traffic detection system of the present invention obtains point cloud data by plotting the illumination angle data, distance data, and measurement time data obtained in the measurement in space-time along the time axis. Although some error will be introduced along the time axis depending on the speed of the passing vehicle, by using this point cloud data, the point cloud data will follow the contour of the passing vehicle, and the passing vehicle can be detected. Furthermore, according to the traffic detection system of the present invention, if the shape of the point cloud trajectory is not recognizable as a vehicle, it can be determined to be a non-vehicle object, such as rain / snow or floating debris. In addition, the method using point cloud data related to the traffic detection system of the present invention makes it easier to separate clusters of point cloud data trajectories than other methods, and makes it possible to separate clusters of point cloud data even in congested traffic with small vehicle intervals. Furthermore, according to the traffic detection system of the present invention, detailed vehicle type discrimination is possible by analyzing the point cloud data trajectory. [Brief description of the drawings]
[0023] [Figure 1] FIG. 1 is a simplified diagram illustrating roads and vehicles traveling on the roads from which a traffic detection system 100 collects data. [Diagram 2] FIG. 1 is a diagram illustrating a simple configuration example of a traffic detection system 100 according to a first embodiment of the present invention. [Diagram 3] 1 is a diagram showing a scanning state in which the laser light irradiation device 111 of the LiDAR device 110 repeatedly irradiates and receives laser light. [Figure 4] FIG. 13 is a diagram simply illustrating an example of the result of plotting, by the point cloud plot processing unit 130, point cloud data in space-time along the time axis. [Diagram 5] FIG. 11 is a diagram showing an example display of point cloud data obtained by plotting measurement data obtained by a LiDAR device 110 using a point cloud plot processing unit 130 in a demonstration experiment. [Figure 6] This is a diagram of the same demonstration experiment, where data was visualized from a high vantage point, providing a bird's-eye view of the driving situation. [Figure 7] This figure shows the statistical results of a demonstration experiment of traffic detection on roads in Ho Chi Minh City, Vietnam. [Figure 8] FIG. 13 is a diagram showing the statistical results of counting the traffic volume of each type of vehicle 300 per hour in a demonstration experiment. [Figure 9] FIG. 1 is a diagram showing an example of a conventional traffic detection device using a laser sensor. [Figure 10] FIG. 1 is a diagram showing an example of a conventional traffic detection device using a laser sensor. BEST MODE FOR CARRYING OUT THEINVENTION
[0024] Hereinafter, an embodiment of the traffic detection system of the present invention will be described with reference to the drawings. However, it goes without saying that the scope of the present invention is not limited to the specific applications, shapes, numbers, etc. shown in the following embodiments. Example 1
[0025] As a first embodiment, a basic configuration example of a traffic detection system according to the present invention will be described. 1 is a simplified diagram showing roads and vehicles traveling on the road from which a traffic detection system 100 collects data. As shown in FIG. 1, a structure 210 installed near a road 200, a laser light irradiation area 220, and a vehicle 300 are shown. As long as the traffic detection system 100 is in a communicable state, it may be installed in a management center of a road administrator, rather than in the vicinity of the road 200 shown in FIG. 1. For convenience, in this example, the presence of the traffic detection system 100 is also shown in FIG. 1, and a LiDAR device 110, which is a component of the traffic detection system 100, is further illustrated as being disposed on the structure 210. There is no particular limitation on the type of vehicle 300. In the embodiments described below, a motorcycle will be mainly described. In addition, there may be various types of road 200, but Fig. 1 shows a road with two lanes in each direction as an example. It goes without saying that the traffic detection system 100 of the present invention can also be applied to expressways, general two-way traffic roads, and one-way roads. In the embodiments described below, an example in which a motorcycle-only lane is provided on the low-speed lane side will be mainly described.
[0026] The structure 210 is a structure for mounting at least a part of the configuration of the traffic detection system 100 of the present invention, and is installed near the road 200. In the configuration example of Fig. 1, the structure 210 is a so-called columnar structure erected beside the road, but it may also be an overpass structure installed so as to straddle the road. The system may be used in conjunction with other systems such as a traffic speed control system, and lighting equipment for illuminating the road at night may also be installed.
[0027] The laser light irradiation area 220 is an area on the road 200 where laser light is irradiated from the LiDAR device 110 to detect vehicles passing through the road 200. In the example of Fig. 1, the LiDAR device 110 is installed at a predetermined height and a predetermined angle on the structure 210 toward the laser light irradiation area, and irradiates passing vehicles with laser light.
[0028] Next, the system configuration of the traffic detection system 100 will be described. Fig. 2 is a diagram simply illustrating a configuration example of the traffic detection system 100 of the present invention according to Example 1. The basic configuration of the traffic detection system 100 illustrated in Fig. 2 includes a LiDAR device 110 (a laser light irradiation device 111 and a measurement data analysis unit 112), a point cloud plot processing unit 130, a vehicle detection unit 150, and a statistical processing unit 160. Although a data communication means is not illustrated, the system is configured to be able to transmit and receive data to and from a traffic control center system.
[0029] Each component will be described below. The LiDAR device 110 is disposed beside or above a road, and can measure distance by emitting a scanning laser beam to intersect with a road surface 200 and receiving the reflected light. In this example, a laser beam emitting device 111 is installed at a predetermined height and angle with respect to a structure 210 toward a laser beam emitting area 220, and performs a predetermined scan while repeatedly emitting and receiving a laser beam toward the laser beam emitting area 220. In the configuration example of FIG. 1, the laser beam emitting angle from the laser beam emitting device 111 is set almost directly below.
[0030] The LiDAR device 110 is basically configured with two devices. In this example, as shown in Fig. 1, two LiDAR devices 110 are installed close to each other, and the laser light irradiation area 220 is a surface with a constant width. Here, one is positioned as a main LiDAR device and the other is positioned as a sub LiDAR device. Here, the LiDAR device 110-1 is the main LiDAR device, and the LiDAR device 110-2 is the sub LiDAR device. It is also possible to configure the system with one LiDAR device 110. However, if there is only one LiDAR device 110, the vehicle speed and vehicle length cannot be separated in the plot result by the point cloud plot processing unit 130 described later, and therefore the point cloud separation processing unit 151 of the vehicle detection unit 150 cannot separate point clouds for each vehicle, and the vehicle speed estimation processing unit 154 cannot estimate the vehicle speed. However, it is possible to calculate the average vehicle speed in a pseudo manner. In this embodiment, the LiDAR device 110 is configured with two devices, and a so-called two-dimensional LiDAR device is constructed as an example.
[0031] The LiDAR device 110-1 is arranged at an angle where the scanning plane is perpendicular to the height overlooking the vehicle 300 traveling on the road 200, in order to obtain more accurate data. The sub LiDAR device 110-2 may be arranged at an angle. This is a device for mounting the main LiDAR device 110-1 and the sub LiDAR device 110-2 on one structure 210 in order to share the structure 210, which is expensive to install. However, in terms of accuracy, it is desirable to install the sub LiDAR device 110 so as to look down vertically on the road 200, just like the main LiDAR device 110. In the example shown in FIG. 1, both the main LiDAR device 110-1 and the sub LiDAR device 110-2 are arranged so as to look down vertically on the road 200.
[0032] 3 is a diagram showing a scanning state in which the laser light irradiation device 111 of the LiDAR device 110 repeatedly emits and receives laser light. There are various possible patterns for scanning the laser light. Note that, although FIG. 3 shows an example in which the scanning trajectory forms a plane, it may be a curved surface. 3(a) shows a scanning pattern in which the scanning laser irradiation from the laser light irradiation device 111 rotates. Note that it is also preferable to provide a cover or the like for areas irradiated other than the laser light irradiation area 220 that needs to be scanned so that the laser light is not irradiated to the outside. 3(b) shows a scanning pattern in which the laser irradiation from the laser light irradiation device 111 is repeated in a fan shape. In other words, it rotates and scans until it reaches an end, then turns around and scans in the opposite direction, and continues scanning until it reaches the other end. This movement is repeated. Regardless of whether the scanning method shown in FIG. 3(a) or FIG. 3(b) is used, scanning is performed so as to cover the laser light irradiation area 220 that needs to be scanned.
[0033] 3(a) or 3(b), the laser light emitting device 111 irradiates a laser light to a laser light emitting area 220 and receives reflected light from the laser light emitting area 220 (laser light reflected by the road surface 200 or a passing vehicle 300). It is possible to measure the distance to an object existing on the rotation plane of the scan with a certain angular resolution.
[0034] Here, it is assumed that the laser scanning speed of the laser light irradiation device 111 is sufficiently faster than the speed of the vehicle 300 . For example, the number of rotations of the laser scan is set to 50 to 100 per second. With this number of rotations, information can be acquired every 10 msec to 20 msec. By using the laser light irradiation device 111, it is possible to measure the traveling vehicle 300 as if cutting it into slices for each scan (each rotation).
[0035] The measurement data analysis unit 112 obtains at least the irradiation angle data of the scan, the distance data to the reflecting object calculated from the measurement time difference, and the measurement time data for the scan from the measurement results by the laser irradiation device 111. In other words, in the measurement results of the LiDAR device 110 of the traffic detection system 100 of the present invention, at least a set of three data items, namely, the irradiation angle data, the distance data, and the measurement time data, is obtained for each irradiation and reflection during scanning by the laser irradiation device 111. Here, since the laser scanning speed of the laser light irradiation device 111 is sufficiently faster than the speed of the vehicle 300, the laser light irradiation and the reflected light reception are repeated multiple times in a time series until the tip of the front part of the vehicle 300 enters the laser light irradiation area 220 and the rear end of the rear part of the vehicle 300 leaves the laser light irradiation area 220, and the measurement data analysis unit 112 obtains multiple data items (sets of irradiation angle data, distance data, and measurement time data) in a time series.
[0036] The point cloud plot processing unit 130 can manage the time-space along the time axis for plotting, and plots the analysis data obtained by the measurement data analysis unit 112, i.e., irradiation angle data, distance data, and measurement time data related to scanning, as point cloud data in the time-space along the time axis. The time space along the time axis for plotting managed by the point cloud plot processing unit 130 can be of the following two types. The first type of time space is a time space in which a coordinate system formed by the Cartesian coordinates of the position in the road width direction and the height from the road surface obtained at each scanning measurement time is expanded along the time axis. When using this first type of space-time, in plotting point cloud data in space-time by the point cloud plot processing unit 130, if multiple data are acquired in one same scan, the data are sorted and plotted in the road width direction with respect to the Cartesian coordinates corresponding to the measurement time of the scan. In other words, for each scan, the point cloud obtained at the scan time in space-time is sorted and plotted by the position in the road width direction. The second type of time-space is a time-space in which a coordinate system formed by polar coordinates of the irradiation angle and distance obtained at each measurement time of the scan is expanded along the time axis. When using this second type of space-time, in plotting point cloud data in space-time by the point cloud plot processing unit 130, if multiple data are acquired in one same scan, the data are sorted and plotted by irradiation angle for the polar coordinates corresponding to the measurement time of the scan. In other words, if the point cloud obtained at the scan time in space-time for each scan is plotted by irradiation angle, the data are sorted and plotted by position in the road width direction. In this embodiment, an example of plotting using the first type of space-time (i.e., an example of plotting using a space-time in which a coordinate system formed by the orthogonal coordinates of the position in the road width direction and the height from the road surface obtained at each scanning measurement time is expanded along the time axis) will be described.
[0037] 4 is a simple diagram showing an example of the result of plotting point cloud data in time and space along the time axis by the point cloud plot processing unit 130. The data used for the plot is data obtained by the main LiDAR device 110-1. FIG. 4(a) is a schematic diagram showing the plot of point cloud data in time and space when multiple four-wheeled vehicles and multiple two-wheeled vehicles pass through the laser light irradiation area 220. Figure 4(b) is an actual image taken by the camera. It may appear as if the display angle of the plot of the point cloud data based on the data obtained by LiDAR 110 in Figure 4(a) is different from the shooting angle of the actual camera footage in Figure 4(b), and the images do not exactly correspond to each other; however, since the viewpoint can be freely set in the plot processing of the point cloud data by the point cloud plot processing unit 130, the discrepancy between the two is simply because the display angle in Figure 4(a) does not match the shooting angle in Figure 4(b); the two capture the same traffic situation, just in different forms of expression.
[0038] In Fig. 4(a), the horizontal axis is the time axis. In other words, the time space is a space in which the horizontal axis is the time axis and is expressed by the irradiation angle and the measurement distance. In the example shown in Fig. 4(a), it is assumed that the laser light irradiation area 220 is passed at a constant speed without acceleration or deceleration. It can be seen that the plot trajectory of the point cloud data shown in Fig. 4(a) is a three-dimensional representation of the outer surface of the passing motorcycle and the left side of the rider. In addition, since the laser irradiation device 111 of the LiDAR device 110 is irradiated from a relatively high position on the structure 210, each vehicle 300 is displayed as if it is shrunk in the height direction.
[0039] It should be noted that in the plot of point cloud data in Figure 4(a), the plot trajectory in the direction of travel of the motorcycle does not directly display the longitudinal outline of the outer surface of the left side of the rider, but is the result of plotting along the time axis. Therefore, it should be noted that if the vehicle speed is slow, the plot trajectory along the time axis will be longer, and if the vehicle speed is fast, the plot trajectory along the time axis will be shorter.
[0040] The vehicle detection unit 150 detects whether or not an object is a vehicle 300 based on a group of point cloud data separated from the plot of the point cloud data. Furthermore, the vehicle detection unit 150 can also perform processing for estimating the vehicle type and the vehicle speed. In this example, the vehicle detection unit 150 is configured to include a point cloud separation processing unit 151, a matching processing unit 152, a vehicle type estimation processing unit 153, and a vehicle speed estimation processing unit 154. Note that it is also possible to include other functions, such as a vehicle length estimation processing unit, for estimating information required by the road administrator based on the point cloud data.
[0041] The point cloud separation processing unit 151 executes a process of estimating and separating a group of point clouds caused by a single passing vehicle 300 and its driver passing through the road 200 from the point cloud trajectory formed by the point cloud data plotted by the point cloud plot processing unit 130.
[0042] On roads with low traffic density, each vehicle is plotted intermittently and dispersedly, and a block of point cloud data is easily isolated and separated, but on roads with high traffic density, multiple vehicles may be plotted overlapping each other. For large vehicles and ordinary passenger cars, it is assumed that a certain distance is maintained between each vehicle, but for motorcycles, it is assumed that multiple motorcycles pass in a single lane overlapping each other. A specific measurement example described later will show such a severe condition, that is, an example in which multiple motorcycles pass through a motorcycle-only lane, and the point cloud separation processing unit 151 estimates and separates a group of point clouds caused by one passing vehicle 300 and its driver from the point cloud trajectory formed by the point cloud data plotted by the point cloud plot processing unit 130. The point cloud separation processing unit 151 may, for example, execute a program having the following processing contents.
[0043] "Generation of vehicle candidates and point cloud allocation processing" Vehicle candidates are generated, and measurement points supplied for each scan are assigned to the vehicle candidates. The main criterion for assignment is the distance in space-time to point clouds already assigned to the same vehicle candidate. In other words, points that are close in distance are considered to be points obtained from the same vehicle, and are assigned to the same vehicle candidate. If there is no vehicle candidate to assign to a measurement point, a new vehicle candidate is generated and the measurement point is assigned to it. Here, the point clouds are sorted in chronological order, and further sorted in the road width direction, which helps to prevent a proliferation of vehicle candidates. In addition, even in the plot processing using the second type of space-time, the proliferation of vehicle candidates can be suppressed by sorting them by the illumination angle.
[0044] "Separating a group of point cloud data and determining vehicle candidates" If a vehicle candidate has not been assigned a new point for a certain period of time, it is determined that the vehicle has passed by and is confirmed as a single vehicle. The vehicle height, width, and vehicle position are calculated from the assigned point cloud and output as a vehicle. If the duration is too short, it is deemed to be a reaction caused by rain, snow, or debris, rather than a vehicle, and is deleted. In addition, if there is another vehicle candidate nearby in space-time and it is determined to be the same vehicle, the vehicle candidates that were generated separately are merged into a single vehicle candidate.
[0045] The matching processing module 152 is an optional configuration, and in order to improve the estimation accuracy of a group of point clouds by the point cloud separation processing unit 151, it performs a matching process between the estimation result of vehicle candidates using point cloud data plotted from data acquired from the main LiDAR device 110-1 and the estimation result of vehicle candidates using point cloud data plotted from data acquired from the sub LiDAR device 110-2. The accuracy of estimation is improved by using the matching processing module 152 to check whether the same vehicle 300 in the results of independent estimation correspond to each other based on the data of the LiDAR device 110 acquired independently. Note that, for a vehicle 300 whose estimation results do not match, it is possible that the vehicle 300 is hidden in the shadow due to the passing state of multiple vehicles 300, or is affected by some obstacle on the road. This judgment and correction may be performed manually by the system operator, and it is also preferable to have the point cloud separation processing unit 151 learn the judgment and correction contents.
[0046] The vehicle type estimation processing module 153 estimates the vehicle type based on the characteristics of the vehicle candidate. The vehicle type here refers to a category such as a large vehicle, a passenger car, a compact car, a motorcycle, etc. It is not necessary to estimate the vehicle name by manufacturer. Based on the characteristics and contour silhouettes indicated by a group of point cloud data, it is expected that it will be possible to a certain extent to determine the type of vehicle (large vehicle, regular passenger car, small car such as a kei car, or motorcycle).
[0047] There are several possible processes as a means for the vehicle type estimation processing module 153 to determine the vehicle type of the point cloud data. For example, there is a pattern matching process. A database (not shown) is constructed, and many patterns of plot results of point cloud data in time and space formed when a vehicle passes at an expected speed are stored in the database, and the shapes of the point cloud data in time and space input from the point cloud separation processing unit 151 are identified by pattern matching. Note that, since expansion and contraction occur in the time axis direction depending on the vehicle speed, it is considered that the length in the time axis direction may be adjusted during matching.
[0048] It is also possible to use image recognition processing using machine learning, for example. Even if the recognition accuracy rate using pattern matching processing is low, it is expected that the recognition accuracy rate using pattern matching processing will improve by accumulating correct and incorrect data and subjecting it to machine learning.
[0049] It is expected that the vehicle type estimation processing module 153 can achieve a fairly high recognition accuracy rate if the data plotted in space-time by the point cloud plot processing unit 130 is of such high quality that the outline of a vehicle can be recognized by the human eye. For example, it is expected that the vehicle type estimation processing module 153 can distinguish whether the vehicle type of a passing vehicle is a four-wheeled vehicle or a two-wheeled vehicle from the height and length estimated from the point cloud trajectory formed by a group of point cloud data. The results of plotting in space-time by the point cloud plot processing unit 130 contain distortions in the time axis direction depending on the passing speed of the vehicle, but the vehicle type estimation processing module 153 is able to grasp and determine the shape of the vehicle based on the shape in space-time formed by the point cloud data.
[0050] Here, the vehicle type estimation processing module 153 has a condition that the object to be recognized is a vehicle that is a moving body passing through a road, so anything that is obviously not a vehicle can be removed as noise. For example, the vehicle type estimation processing module 153 can determine whether an object is a vehicle 300 moving on a road or a stationary object on the road 200 based on the presence or absence of change on the time axis of the point cloud trajectory formed by a group of point cloud data. An object that does not change on the time axis can be determined to be an object that appears as a stationary object near the road 200 in the laser light irradiation area 220, and can be removed as noise that is not a vehicle 300.
[0051] Also, for example, the vehicle type estimation processing module 153 can determine whether the vehicle is a moving vehicle moving on the road or rain or snow passing near the front of the laser light projection device 110 based on the presence or absence of a change in the height component of the point cloud trajectory formed by a group of point cloud data, and can remove them as noise. Although dust floating in the air is transient, snow and rain fall continuously for a certain period of time, so there is a possibility that they may be mixed into the plot result in space-time by the point cloud plot processing unit 130, and may be cut out as a group of point cloud data in space-time by the point cloud separation processing unit 151. In particular, it is assumed that snow reflects laser light compared to rain and is received by the laser light projection device 110. Therefore, the vehicle type estimation processing module 153 checks the change in the height component of the point cloud trajectory, and if there is no change in the height component of the point cloud trajectory, it can be determined that it is a vehicle moving horizontally on the road, that is, a passing vehicle, and if there is a change in the height component of the point cloud trajectory, it can be determined that it is falling on the road, that is, rain and snow, and can be removed as noise.
[0052] The vehicle speed estimation processing module 154 estimates the vehicle speed of each detected vehicle 300 . Here, the vehicle detection by the above-mentioned vehicle type estimation processing module 153 is possible by analyzing the measurement data of only one main LiDAR device 110-1. In addition, the length on the time axis that appears in the time space is often perceived as a visualized result in the vehicle length direction, but this is because the vehicle speed and the vehicle length are implicitly recognized in an appropriate association. In principle, the speed and the vehicle length of the vehicle 300 cannot be separated from the measurement data of one main LiDAR device 110-1. For example, if a vehicle with twice the length of a certain vehicle 300 passes by at twice the speed, the measurement data obtained by scanning the LiDAR device 110-1 will be plotted in the time space by the point cloud plot processing unit 130 in the same way, and they cannot be distinguished. Therefore, to detect the speed and vehicle length, another LiDAR device 110 (sub LiDAR device 110-2) is placed at a short distance away and scans and measures independently, and the vehicle length and speed are separated using the position difference and time difference with the main LiDAR device 110-1.
[0053] Here, the length in the traveling direction of the laser light irradiation area 220 formed by the main LiDAR device 110-1 and the sub LiDAR device 110-2 is assumed to be about 3 m to 10 m, and it can be assumed that the vehicle 300 hardly moves in the road width direction at that interval. In the results of actual demonstration experiments, the deviation in the road width direction within the laser light irradiation area 220 was within 0.5 m for many vehicles. Using this knowledge, vehicles with a time difference and a position difference in the road width direction within a certain range or less are matched as the same vehicle.
[0054] In this way, the data provided to the vehicle speed estimation processing module 154 is based on the premise that the vehicle 300 has already been identified through the selection of vehicle candidates by the point cloud separation processing unit 151 and the matching processing of the vehicle candidates by the matching processing unit 152. The vehicle speed estimation processing module 154 uses two point cloud data identified as the same vehicle to obtain the vehicle speed v [m / s] of the vehicle 300 using the following equation 1 from the distance D [m] between the main LiDAR device 110-1 and the sub LiDAR device 110-2 on the road surface, the time T11 when the front end of the vehicle 300 is detected by the main LiDAR device 110-1, and the time T21 when the front end of the vehicle 300 is detected by the sub LiDAR device 110-2. [Formula 1] v=D / (T11-T21) In addition, the vehicle length L is obtained from the time T11 when the front end of the vehicle 300 is detected by the main LiDAR device 110-1, the time T12 when the rear end of the vehicle 300 is detected, and the vehicle speed v using the following equation 2. [Formula 2] L = v(T12-T21)
[0055] For reference, the vehicle length L of the vehicle 300 is about 3m 40cm for a light vehicle, about 4m for a normal passenger car, and about 5m to 25m for a large vehicle. Therefore, for example, if large and small vehicles are to be distinguished, the vehicle 300 can be classified as a large vehicle if its length L is 5m or more, and as a small vehicle if its length is less than 5m. In addition, if the vehicle length L is significantly outside the expected range of the vehicle length L, such as 0.5m or less or 30m or more, it can be assumed that some error has been mixed in the results calculated by [Formula 1] or [Formula 2].
[0056] Next, the statistical processing unit 160 performs various statistical processing based on the detection results of the vehicles 300 by the vehicle detection unit 150, such as counting the total number of passing vehicles 300, counting the passing vehicles 300 by time period, detecting whether or not congestion has occurred, and grasping time periods when congestion is likely to occur. The above is a description of each component of the traffic detection system 100.
[0057] An example of a vehicle detection process using traffic detection system 100 is provided below. FIG. 5 shows an example of point cloud data obtained by installing a laser light irradiation device 110 in a laser light irradiation area 220 in a motorcycle-only lane in a tunnel in Ho Chi Minh City, Vietnam, acquiring measurement data obtained by the measurement data analysis unit 112, and plotting the measurement data by the point cloud plot processing unit 130. In this example, the front side of the figure is a lane exclusively for motorcycles, and the back side is a lane exclusively for automobiles. Ho Chi Minh City in Vietnam is said to have a lot of motorcycle traffic, with about three motorcycles passing by per second in a chaotic and overcrowded state at peak times, and the traffic volume is said to exceed 10,000 vehicles per hour. With a traffic detection system using conventional technology based on Japan's traffic conditions, it is expected that it will be impossible to detect motorcycles in the first place or that the accuracy will be significantly reduced in this type of traffic flow.
[0058] As shown in FIG. 5, motorcycles are traveling in an overcrowded state in the motorcycle-only lane in the foreground of the figure, but it can be seen that the processing by the vehicle detection unit 150, that is, the processing results of the point cloud separation processing unit 151 and the matching processing unit 152, have made it possible to separate and visualize both four-wheeled vehicles and motorcycles by vehicle 300 with high accuracy in time and space. Figure 6 shows similar data visualized from a high vantage point, showing that it is also possible to visualize an overview of the driving situation. 5 and 6 show different lateral time in space-time, and the vehicles 300 are shown as if they are shifted from each other in the lateral direction in both figures.
[0059] Next, the results of statistical processing performed by the statistical processing unit 160 based on the estimation results by the vehicle type estimation processing unit 153 will be shown. Figure 7 shows the statistical results of a demonstration experiment of traffic detection on roads in Ho Chi Minh City, Vietnam. Figure 7 shows the statistical results of verifying the number of passing vehicles per lane per hour. FIG. 7 shows a comparison between the number of vehicles detected using the traffic detection system 100 of the present invention and the number of vehicles detected using a video camera image recognition type vehicle detector installed on-site for comparison. The number of detections from both methods was roughly the same, and it was verified that the traffic detection system 100 of the present invention was able to detect four-wheeled vehicles with a certain degree of reliability.
[0060] When comparing the number of four-wheeled vehicles and two-wheeled vehicles by vehicle type in detail for a 10-minute period during the evening on a weekday when traffic volume is high, the number of vehicles detected by the traffic detection system 100 of the present invention and the number of vehicles detected by visual counts from video camera footage were both 233 four-wheeled vehicles and 1,505 two-wheeled vehicles, which were consistent. Because the numbers are so large, there is a possibility that the number of vehicles detected by the video camera image recognition vehicle detector also contains counting errors, and furthermore, there is a possibility that the traffic detection by the traffic detection system 100 of the present invention may include missed detections or double counts due to false detections, but it was verified that the traffic detection process by the traffic detection system 100 of the present invention is considered to have a reliable accuracy.
[0061] Next, the results of the statistical processing performed by the statistical processing unit 160, which are the counts of the number of passing vehicles 300 by type, are shown. 8 is a diagram showing the results of counting the traffic volume per hour by type of vehicle 300. Four-wheeled vehicles were counted by classifying them into large vehicles and small vehicles based on the difference in vehicle height. As shown in Figure 8, motorcycle traffic volume is particularly high during peak hours in the morning and evening on weekdays, exceeding 10,000 vehicles per hour. Statistical processing by the statistical processing unit 160 revealed that approximately three motorcycles per second were traveling in the single motorcycle-only lane, and it was shown that the traffic detection system 100 of the present invention was able to accurately detect motorcycles even in such a congested situation. [Industrial Applicability]
[0062] The traffic detection system of the present invention can be widely applied as a traffic detection system capable of detecting vehicles passing through general roads and expressways. [Explanation of symbols]
[0063] 100 Traffic Detection System 110 LiDAR device 111 Laser irradiation device 112 Measurement Data Analysis Equipment 130 Point cloud plot processing unit 150 Vehicle detection unit 151 Point cloud separation processing unit 152 Matching processing unit 153 Vehicle type estimation processing unit 154 Vehicle speed estimation processing unit 160 Statistical Processing Unit 200 road 210 Structure 220 Vehicle Detection Area 300 vehicles
Claims
1. A LiDAR device that is disposed beside or above a road, irradiates a laser beam that scans so as to intersect with the road surface, receives reflected light, and obtains at least irradiation angle data of the scanning, distance data to a reflecting object calculated from a measurement time difference, and measurement time data of the scanning; A point cloud plot processing unit that plots the irradiation angle data, the distance data, and the measurement time data obtained by the LiDAR device as point cloud data in space-time along a time axis; A traffic detection system comprising: a point cloud separation processing unit that estimates and separates a group of point clouds caused by a passing vehicle passing through the road from the point cloud trajectory formed by the point cloud data on the plot of the point cloud plot processing unit; and a vehicle detection unit that detects the group of point clouds estimated and separated by the point cloud separation processing unit as the passing vehicle passing through the road.
2. the time space is a time space obtained by expanding, along the time axis, an orthogonal coordinate system formed by a road width direction and a height of the measurement data obtained at each measurement time of the scanning, 2. The traffic detection system according to claim 1, wherein, in plotting the point cloud data in the space-time by the point cloud plot processing unit, if a plurality of data exist at the measurement time of the same scan, the data are sorted and plotted in the direction of the road width in the orthogonal coordinates corresponding to the measurement time of the scan.
3. the time space is a time space obtained by expanding, along the time axis, polar coordinates formed by the irradiation angle and the distance of the measurement data obtained at each measurement time of the scanning, 2. The traffic detection system according to claim 1, wherein, in plotting the point cloud data in the space-time by the point cloud plot processing unit, if a plurality of data exist at the measurement time of the same scan, the data are sorted by the irradiation angle and plotted on the plane corresponding to the measurement time of the scan.
4. The traffic detection system according to claim 1, wherein the vehicle detection unit includes a vehicle type estimation processing unit that determines whether the vehicle type of the passing vehicle is a four-wheeled vehicle or a two-wheeled vehicle based on a height and length estimated from a point cloud trajectory formed by the group of point cloud data.
5. 5. The traffic detection system according to claim 4, wherein the vehicle detection unit includes a vehicle speed estimation unit that estimates a speed of the passing vehicle from a height and a length estimated from a point cloud trajectory formed by the group of point cloud data and a change on the time axis.
6. The traffic detection system according to claim 1, characterized in that the vehicle detection unit determines whether the object is a moving vehicle moving on the road or a stationary object on the road surface based on the presence or absence of a change on the time axis of the point cloud trajectory formed by the group of point cloud data.
7. The traffic detection system according to claim 1, characterized in that the vehicle detection unit determines whether the detected object is a moving vehicle moving on the road or rain or snow passing near the front of the laser light irradiation device based on the presence or absence of a change in the height component of the point cloud trajectory formed by the group of point cloud data.
8. The traffic detection system according to claim 1, further comprising a statistical processing unit that performs statistical processing based on the detection results of the vehicle detection unit, including a counting process of the number of passing vehicles and a counting process of the number of passing vehicles by time period.
9. The traffic detection system described in claim 1, characterized in that a plurality of irradiation surfaces by the LiDAR device are set in the direction of travel of the road, and each of the irradiation surfaces has a different position and angle, or a plurality of the LiDAR devices are arranged in the direction of travel of the road, each of which is arranged in a different position and has a different position and angle of its irradiation surface.
10. The traffic detection system described in claim 9, characterized in that it is further characterized by a matching processing unit that compares the processing results of the point cloud separation processing unit using the detection results of the passing vehicles by the LiDAR device having one of the irradiation surfaces with the processing results of the point cloud separation processing unit using the detection results of the passing vehicles by the LiDAR device having the other irradiation surface.
11. (Method claims) A LiDAR device is disposed beside or on a road, and a laser beam is irradiated to scan the road surface so as to intersect with the road surface, and the reflected light is received, and the distance to the reflecting object is measured from the time difference. A point cloud plot process is performed to obtain point cloud data in space-time along a time axis from the irradiation angle data, the distance data, and the measurement time data related to the measurement of the LiDAR device; executing a point cloud separation process for estimating and separating a group of point clouds caused by one passing vehicle passing through the road from a point cloud trajectory formed by the point cloud data related to the plot in the point cloud plot process; A traffic detection method, comprising: detecting the group of points separated by the point cloud separation processing as the passing vehicles passing through the road.
12. (Program Claims) A LiDAR measurement processing step in which a LiDAR device is disposed beside or on a road, a laser beam is irradiated to scan the road surface so as to intersect with the road surface, the reflected light is received, and the distance to the reflecting object is measured from the time difference; a point cloud plotting process step of plotting the irradiation angle data, the distance data, and the measurement time data in the measurement process step as point cloud data in time space along a time axis; a point cloud separation processing step of estimating and separating a group of point clouds caused by one passing vehicle and its driver passing through the road from a point cloud trajectory formed by the point cloud data of the plot result of the point cloud plot processing step; a vehicle detection processing step of detecting the group of points related to the separation by the point cloud separation processing step as the passing vehicle passing through the road; A traffic detection processing program that causes computer resources to execute the measurement processing step, the point cloud plot processing step, the point cloud separation processing step, and the vehicle detection processing step.
Citation Information
Patent Citations
Vehicle type discriminating device
JP1999203588A
Laser area sensor
JP2009110069A
Vehicle determination system and passing object management system
JP2011186526A
Vehicle type discrimination device and vehicle type discrimination method
JP2017220076A
Vehicle recognition device, vehicle recognition system, and vehicle recognition program
JP2019003513A
Cited By
Magnetic disk device, control method, and learning method
CN121708964A
