JP7226387B2 - Object detection system - Google Patents

Description

The present invention relates to an object detection system, in particular to the detection of objects on a road surface constructed of precast concrete slabs.

Conventionally, there are technologies for detecting objects on the road, such as vehicles and people, using cameras and lasers. For example, Patent Literature 1 proposes a technique of detecting a vehicle in a surveillance area using a laser radar.

JP 2019-207654 A

However, since the camera uses visible light, it becomes difficult to see the surroundings in bad weather such as heavy rain, snow, and fog, and the ability to detect objects on the road deteriorates. In laser radar, the laser is reflected and scattered by raindrops, snow, and fog. In other words, even when laser radar is used, the ability to detect objects on the road may deteriorate in bad weather.

SUMMARY OF THE INVENTION An object of the present invention is to make it possible to more reliably detect an object on the road even in bad weather.

An object detection system according to the present invention detects an object on a road having a structure formed by using piles erected in a road construction area and precast concrete slabs placed on the piles. an object detection system comprising: a plurality of sensor means arranged on the road; sensor information acquisition means for acquiring sensor information including at least the sensor values of the sensor means; and detection means for detecting an object on the road by referring to the load.

Further, when the load applied to the road surface is gradually increasing or decreasing, the detection means detects the amount of increase in the load per time in the gradually increasing state with no objects on the road surface, based on the load obtained from the sensor value of the sensor means. is assumed to be constant , or the load obtained by subtracting the load obtained by subtracting the load when it is assumed that the amount of decrease in the load per time in the gradually decreasing state with no objects on the road surface is constant. object is detected.

Further, the detection means is a learning model for outputting the presence or absence of an object when the sensor means is attached to the pile, and the distribution of the load applied to the pile when the object exists on the road is input. (2) the object on the road is detected by inputting the distribution of the load applied to the pile obtained from the sensor value of the sensor means;

detecting means for detecting the position, moving direction, and moving speed of the object detected by the detecting means by referring to the distribution of the load applied to the road surface obtained from the sensor values of the sensor means and its transition; and reporting means for reporting the detection of the object to other objects located around the object detected by the means.

Further, there is discriminating means for discriminating the type of the object detected by the detecting means from the load applied to the road surface obtained from the sensor value of the sensor means, and the reporting means discriminates the object as a person by the discriminating means. When the object is determined to be a vehicle by the determination means, the approach of the vehicle is notified to a person positioned in the traveling direction of the vehicle. characterized by

Further, the sensor means is a load sensor or a sensor that measures sensor data that can be converted into a load.

According to the present invention, objects on the road can be detected more reliably even in bad weather.

It is a perspective view which shows the structure of the road in this Embodiment. 1 is an overall configuration diagram of an object detection system according to an embodiment; FIG. 1 is a block configuration diagram of a communication device and an object detection device according to this embodiment; FIG. FIG. 4 is a diagram showing an example of teacher data used for learning a learning model in the present embodiment; FIG. 10 is a diagram showing another example of teacher data used for learning a learning model in the present embodiment; 4 is a flow chart showing object detection processing in the present embodiment. FIG. 5 is a graph showing an example of load transition measured by one load sensor in fine weather; FIG. 5 is a graph showing an example of load transition measured by one load sensor when there is snow cover;

Preferred embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a perspective view showing the structure of a road according to this embodiment. A precast concrete slab (hereinafter referred to as “road slab”) 1 is a plate-shaped concrete that is manufactured in advance in a shape and size that can be transported at a factory or the like for assembly and installation on site. The size of the roadblock 1 is assumed to be 1.2m square, but it is not necessary to be limited to this size. The roadbed 1 in this embodiment is formed in a rectangular shape.

As shown in FIG. 1, the road 2 in this embodiment is constructed by arranging the roadbed 1 on the piles 3 erected on the road body 4, and connecting the corners of the roadbed 1 to the piles 3 by bolts. It has a structure formed by being fixed to The "road body" 4 refers to the lowest layer of the road structure formed by digging down the area where the road is to be constructed. Since the roadbed 1 is laid on the piles 3 erected on the road body 4, it is necessary to dig down to at least the thickness of the roadbed 1 or more. In the present embodiment, as shown in FIG. 1, the road is dug deeply so that a space 5 is formed between the roadbed 1 and the road body 4 .

A load sensor 6 is attached to each pile 3 . The load sensor 6 constantly measures the load applied to the roadbed 1 placed on the pile 3 . In the present embodiment, an example in which the pile 3 is attached to the upper portion is shown, but the mounting position is not limited to this example as long as the load can be measured. Alternatively, it may be installed on the front surface, back surface, or inside of the roadblock 1, for example.

FIG. 2 is an overall configuration diagram of an object detection system according to this embodiment. In addition, in FIG. 2, the situation around the road 2 is shown. A vehicle 7 and a person 8 corresponding to objects to be detected are shown on the road. In addition, a sign board 9 for providing information to a person 8 crossing is installed at the installation position of the pedestrian crossing. Furthermore, FIG. 2 shows a load sensor 6 attached to a stake (not shown), a communication device 10 installed near the road 2, and an object detection device installed at a remote location on the road 2. 20 are shown. The communication device 10 is connected by wire or wirelessly to the load sensors 6 and the signboard 9 installed in the vicinity, collects the sensor data measured by each of the connected load sensors 6, and transmits the sensor data to the object detection device 20. Along with the transmission, it notifies that an object has been detected in the vicinity in accordance with an instruction from the object detection device 20 . The object detection device 20 detects an object on the road 2 by analyzing sensor information transmitted by wire or wirelessly from each of a plurality of communication devices 10 installed around the road 2 . In the present embodiment, it is assumed that the function provided by the object detection device 20 described later is provided on the cloud, but it is not necessary to be limited to this. It may be realized by installing it in a road management sensor or the like.

A plurality of the load sensors 6 and the communication devices 10 are installed on the road 2, but since they perform the same operation, they are illustrated one by one in FIG. 2 for the sake of convenience.

FIG. 3 is a block configuration diagram of a communication device and an object detection device according to this embodiment. Components that are not used for explanation in this embodiment are omitted from the drawings.

The communication device 10 can be realized with the hardware configuration of a conventional general-purpose computer having a communication function. That is, the communication device 10 includes a processor, ROM, RAM, storage and network interface.

The communication device 10 has a sensor data receiving section 11, a sensor information transmitting section 12, a detection information receiving section 13 and a reporting section . The sensor data receiver 11 collects sensor data from the connected load sensor 6 . The sensor information transmission unit 12 transmits sensor information including at least collected sensor data to the object detection device 20 . The detection information receiving unit 13 receives detection information sent when the object detection device 20 detects an object. In accordance with the sent detection information, the notification unit 14 notifies other objects located around the detected object that the object has been detected.

Each component 11 to 14 in the communication device 10 is realized by cooperative operation of a computer that forms the communication device 10 and a program that runs on a processor installed in the computer.

The object detection device 20 can be realized by a hardware configuration such as a conventional general-purpose server computer having a communication function. That is, the object detection device 20 includes a processor, a ROM, a RAM, storage means such as a hardware disk drive (HDD), and a network interface.

The object detection device 20 has a sensor information reception section 21 , an object detection section 22 , a notification control section 23 , a sensor information storage section 24 and a management information storage section 25 . The sensor information receiving unit 21 receives sensor information transmitted from the communication device 10 and stores it in the sensor information storage unit 24 . The object detection unit 22 detects an object on the road and the position of the object on the road by referring to the sensor information received by the sensor information reception unit 21 . The object detection unit 22 also detects the moving direction and moving speed of the object from the history of sensor information stored in the sensor information storage unit 24 . Furthermore, when detecting an object, the object detection unit 22 determines the type of the detected object, that is, whether the object is a person or a vehicle. The report control unit 23 transmits detection information to the communication device 10 when an object is detected.

The management information storage unit 25 stores management information for managing the communication device 10 , the sign board 9 and the load sensor 6 installed on the road 2 . The management information includes, as management information relating to the communication device 10, identification information of the communication device 10 and position information specifying the position where the communication device 10 is installed. In addition, as management information about the load sensor 6, identification information of the load sensor 6, position information specifying the position where the load sensor 6 is installed (when the position of the pile 3 etc. is managed, the attached pile 3). In addition, the management information regarding the sign board 9 includes identification information of the sign board 9 and position information specifying the position where the sign board 9 is installed. Furthermore, the management information includes information that associates each communication device 10 with the sign plate 9 and load sensor 6 connected to the communication device 10 .

Each component 21 to 23 in the object detection device 20 is realized by cooperative operation of a computer that forms the object detection device 20 and a program that runs on a CPU installed in the computer. Moreover, each of the storage units 24 and 25 is implemented by an HDD mounted on the object detection device 20 . Alternatively, RAM or external storage means may be used via a network.

The program used in the present embodiment can be provided not only by communication means but also by being stored in a computer-readable recording medium such as a CD-ROM or USB memory. A program provided from a communication means or a recording medium is installed in a computer, and various processes are realized by the CPU of the computer sequentially executing the program.

If a camera or laser radar is used to detect objects on the road, there is a risk that the ability to detect objects will deteriorate in bad weather. In contrast, in the present embodiment, objects on the road are detected based on the load applied to the road surface, so objects on the road can be detected more reliably even in bad weather. The operation in this embodiment will be described below.

In the present embodiment, the object detection unit 22 detects objects on the road by referring to sensor information, but the object detection unit 22 detects objects using a learning model formed in advance by machine learning. . Learning of this learning model will be described with reference to FIGS. 4A and 4B.

FIGS. 4A and 4B (hereinafter collectively referred to as “FIG. 4”) correspond to conceptual diagrams showing teacher data used for learning a learning model. The learning model used in this embodiment receives the load distribution applied to the piles 3 as input and outputs whether or not an object is detected.

First, how to view FIG. 4 will be described. In FIG. 4, a part of the road 2 is cut out and shown. As shown in (a) of FIG. 4A, the road 2 is provided with a sidewalk 2a and a roadway 2b. A rectangle included in the road 2 is a roadbed 1, and circular stakes 3 are shown at each corner of the roadbed 1. FIG. In FIG. 4, the road 2 on the left side of the drawing shows the size and position of the vehicle 7 and the person 8, and the corresponding road 2 on the right side of the drawing shows the vehicle 7 and the person 8 at the position shown on the left side. It shows the magnitude of the load measured by the load sensor when it is positioned. As shown in (a) of FIG. 4A, in this embodiment, the load is divided into four stages for convenience. The load sensor 6 always measures the weight of at least the roadbed 1 placed on the piles 3 regardless of the presence or absence of the object. will be called the "reference load". In FIG. 4, the piles 3 whose reference loads are being measured are indicated by white circles. Then, the magnitude of the measured load is classified into small, medium, and large by comparison with each predetermined threshold value. In FIG. 4, the heavier the stake 3, the darker the color.

In the present embodiment, only the reference load may be subtracted from the load applied to the pile 3, and learning may be performed using only the load of the object. In the present embodiment, the latter, that is, teacher data based only on the load of the object that does not include the reference load is used.

Here, (b) to (h) shown in FIG. 4 are used as teacher data to make the learning model learn. For example, by inputting the training data of (b) into the learning model, the load distribution and the magnitude of the load when the person 8 is in the position shown in (b) when the object is the person 8 are learned. Similarly, by inputting the teacher data (c) to (e) into the learning model, when the object is a vehicle 7 and the vehicle 7 is a general vehicle 7a, a motorcycle 7b, or a large vehicle 7c, (c ) to (e) of the vehicle 7a to 7c are learned. Furthermore, since a plurality of vehicles continue to travel on the road 2, the teacher data (f) to (h) are used assuming a case where the plurality of vehicles 7 travel on the road 2 so as to be able to deal with this case as well. Train a learning model.

The types of vehicles need not be limited to the general vehicle 7a, the motorcycle 7b, and the large vehicle 7c shown in FIG. The accuracy of object detection is improved by making the learning model learn based on teacher data based on the model and the alignment and running positions of a plurality of vehicles (for example, near the sidewalk on road 2).

The load sensor 6 constantly measures the load applied to the pile 3 . Sensor data indicating the measured load (also referred to as “sensor value” or “load value”) is sent to the communication device 10 . When the sensor data is sent from the load sensor 6, the communication device 10 receives the sensor data, the identification information of the load sensor 6 that measured the sensor data, the measurement date and time, and the position information for specifying the measured position. Information is generated and transmitted to the object detection device 20 . The positional information is also used by the object detection device 20 to specify the destination of the report. The position information may be identification information of the communication device 10 or identification information of the load sensor 6, for example.

When the sensor information is transmitted from the communication device 10, the object detection device 20 executes a process of detecting an object on the road based on the sensor information. Object detection processing according to the present embodiment will be described below with reference to the flowchart shown in FIG. A plurality of objects may exist on the road 2, but here, for convenience of explanation, attention is focused on a predetermined range on the road, and objects are detected within that range. The predetermined range may be, for example, a range in which the load is measured by the load sensor 6 connected to one communication device 10 . Also, when an object exists in that range, the description will focus on detection of one object. Note that, as described with reference to FIG. 4, the object detection unit 22 can detect all objects within a predetermined range by using a learning model.

In the object detection device 20 , the sensor information receiving section 21 receives the sensor information corresponding to each load sensor 6 transmitted from the communication device 10 and stores it in the sensor information storage section 24 . The object detection unit 22 acquires the sensor information of the same measurement date and time within a predetermined range by retrieving it from the sensor information storage unit 24 (step 110). Alternatively, it may be obtained directly from the sensor information receiving unit 21 . The object detection unit 22 refers to the sensor values included in the sensor information and creates a load distribution as shown on the right side of the drawing shown in FIG. 4 (step 120). The arrangement of the load sensors 6 can be specified by comparing the sensor identification information included in the sensor information and the management information stored in the management information storage unit 25 .

Subsequently, the object detection unit 22 performs a correction process for correcting the load represented by the created load distribution as necessary so that the object can be detected even in bad weather (step 130), but the correction process will be described later. Here, the subsequent processing will be described assuming that the correction processing has been completed.

The object detection unit 22 inputs the created load distribution to the learning model and determines whether or not there is an object on the road (step 140). When only the standard load is measured for all the piles 3 as shown in FIG. 4(a) in the created load distribution, in other words, when the load of the object is not detected, the object exists within a predetermined range. I will not. That is, the object detection unit 22 does not detect objects. On the other hand, when the created load distribution matches or resembles one of the load distributions shown on the right side of the drawing in FIG. 4, the object detection unit 22 detects an object within a predetermined range. For example, if the load distribution created in step 120 is the same as or similar to the load distribution shown on the right side of FIG. can be done. The detected object can be distinguished from the general vehicle 7a.

In the present embodiment, instead of simply referring to the load value measured by the load sensor 6, that is, not simply whether or not the load is detected within a predetermined range, but the degree of matching of the load distribution, the object is detected. Therefore, it is desirable to make the learning model learn using more teaching data for the vehicle 7 and the person 8 to be detected. As a result, it is possible to avoid erroneously detecting an object simply placed on the road as a vehicle 7 or a person 8 .

If no object is detected within the predetermined range (N at step 150), the process proceeds to step 110 to acquire the next sensor information. If an object is detected within the predetermined range (Y at step 150), then the object detection unit 22 determines the type of the object detected on the road (step 160). In the present embodiment, the object is determined based on the load measured by the load sensor 6, that is, the weight. More specifically, the load point (center of gravity) applied to the roadbed 1 is obtained from the loads applied to the four piles 3 supporting the roadbed 1 for which the load of the object is detected. Similarly, load points are obtained for the other roadbeds 1 where the load of the object is detected, and the weight of the object is calculated from the load at the load point of each roadbed 1 . Since the weight of the vehicle 7 and the person 8 is clearly different, a predetermined threshold value is set, and the vehicle 7 is determined when the weight is equal to or greater than the threshold, and the person 8 is determined when the weight is less than the threshold. Note that the learning model may be made to output the type of the object.

When the type of object is determined, the object detection section 22 calculates the position, moving direction, and running speed of the object (step 170). As for the position of the object, the center point of the object is obtained from the position of the load point (center of gravity) applied to each roadblock 1 where the load of the object is detected, and the center point is detected as the position of the object. Also, once an object is detected within a predetermined range, it basically moves within the predetermined range after being detected. The object detection unit 22 detects an object as described above based on the received sensor information, but the sensor information storage unit 24 contains sensor information received in the past, so based on the past sensor information The direction of movement of the object can be identified from the detected change in the position of the object, and the movement speed can be calculated from the amount of movement of the object per unit time.

The notification control unit 23 instructs notification by transmitting detection information including the fact that an object has been detected, the type of the detected object, and the moving direction and moving speed to the communication device 10 (step 180). The destination of the detection information is the communication device 10 corresponding to the predetermined range. Alternatively, communication devices 10 corresponding to ranges adjacent to the predetermined range may also be included in the transmission destinations. Alternatively, the information may be stored in the object detection device 20 for the purpose of managing the object detection status.

Note that when another object different from the object detected within the predetermined range is further detected, for example, when the vehicle 7 is detected, the notification control unit 23 detects the person 8 in the traveling direction of the vehicle 7. When it is determined that the road is in a state to be reported, such as when the person 8 is detected and the vehicle 7 traveling toward the walking direction of the person 8 is detected. You may control so that detection information may be transmitted only. Also, the communication device 10 may be provided with this reporting control function.

In the communication device 10, when the detection information receiving unit 13 receives the detection information, the reporting unit 14, when the detected object is the person 8, is traveling toward the detected vehicle 7 or at least the person 8. Effective information is reported to the vehicle 7. Valid information is, for example, that a person 8 has been detected around the driving position, that there is a person 8 crossing the pedestrian crossing ahead, and the like. Further, when the detected object is the vehicle 7 , the reporting unit 14 displays information useful for the person 8 on the sign board 9 installed in the traveling direction of the vehicle 7 . Valid information is, for example, that a vehicle 7 is approaching. Alternatively, when the person 8 carries an information device having a display means, the notification unit 14 may inform the person 8 that the vehicle 7 is approaching by reporting to the information device.

In the present embodiment, the communication device 10 receives instructions from the object detection device 20 and makes a report. or may be marked.

Here, the correction processing in step 130 will be described.

FIG. 6 is a graph showing an example of load transitions measured by one load sensor 6. In FIG. FIG. 6 shows the transition of the load measured when the weather is fine weather, which is not bad weather. In the graph shown in FIG. 6, the horizontal axis is time and the vertical axis is the measured load.

FIG. 6(a) is a graph showing the measurement when there is no object on one roadbed 1 placed on the pile 3 corresponding to the load sensor 6. FIG. Since there is no object present, only a constant load, ie the reference load, is being measured and the load does not change. FIG. 6(b) shows a graph in which the load of the object is detected during a certain period of time t1.

6(c) shows the load transition when only the load of the object is extracted by subtracting the load shown in FIG. 6(a), that is, the reference load, from the load transition shown in FIG. 6(b). It is a graph showing. In the correction process, a load distribution is created by detecting only the load of the object, similar to the teacher data.

Bad weather, such as fog or haze that does not apply a load to the roadblock 1, can be corrected in the same manner as fine weather. In the case of rain, the load of rainwater is applied to the road, but when compared with the weight of the roadbed 1, it is considered to be within the range of error, that is, the noise level of the measured value. Therefore, rainfall does not affect the result of determining the presence or absence of an object on the roadbed 1 and the result of determining the type of object. In other words, fog, haze, and rain can be corrected in the same way as fair weather. Heavy rain with a lot of precipitation does not affect the judgment result.

FIG. 7 is a graph showing an example of load transitions measured by one load sensor 6. In FIG. FIG. 7 shows the transition of the load measured when the bad weather is snow, especially when snow is observed. In the graph shown in FIG. 7, the horizontal axis is time and the vertical axis is the measured load.

FIG. 7(a) is a graph showing a case where measurement is performed with no object on one roadbed 1 placed on the load sensor 6. FIG. Since there is no object, the load gradually increases as snow is observed on the roadbed 1 due to snowfall. Here, for convenience of explanation, it is assumed that the amount of accumulated snow increases constantly, so that the amount of load increase is constant. FIG. 7(b) shows a graph in which the load of the object is detected during a certain period of time t1 when snow cover is observed on the roadbed 1 as in FIG. 7(a).

7(c) is a graph showing the load transition when only the load of the object is extracted by subtracting the load shown in FIG. 7(a) from the load transition shown in FIG. 7(b). . In this embodiment, it is possible to create a load distribution by extracting only the load of an object even when there is snow cover.

It should be noted that whether the load applied to the roadbed 1 is due to accumulated snow or due to the load of an object can be determined by the amount of load increase per time. In the case of snow, the load increases slowly, that is, gradually, whereas in the case of an object, depending on the speed of movement, it increases relatively abruptly compared to the case of snow. Therefore, it can be determined by the amount of load increase per time. More specifically, a threshold may be set for the amount of increase, and the determination may be made by comparing the amount of increase per hour with the threshold.

On the other hand, in the case of thawing snow, correction may be performed in the same manner as in the case of accumulated snow. In the case of thaw, the load gradually decreases when no object exists, but as described with reference to FIG. By subtracting, it is possible to create a load distribution in which only the load of the object is extracted.

As described above, according to the present embodiment, by measuring the load on the road, it is possible to reliably detect an object even in bad weather such as when snow is observed.

In this embodiment, the load sensor 6 of the road 2 is installed to detect the load applied to the pile 3 and the roadbed 1 . This is because the load sensor 6 can obtain the load itself as a sensor value. However, as sensor means other than the load sensor 6, for example, a vibration sensor may be provided on the pile 3 or the roadbed 1. FIG. Alternatively, a strain sensor may be provided on the roadbed 1 . The vibration value measured by the vibration sensor and the strain value measured by the strain sensor can be converted into a load. Therefore, the object detection device 20 converts the sensor data (vibration value or strain value) acquired from the communication device 10 into a load value, and then performs the above-described object detection processing. Note that the conversion of sensor data into load values may be performed in the communication device 10 .

Vibration sensors are not affected by fog or haze. In addition, vibrations due to rainfall and snowfall are only noise and are hardly affected. Strain sensors are immune to fog, haze and rain. Also, the roadbed 1 may be slightly distorted due to snow cover, but it is only noise. In other words, an object can be detected using a vibration sensor or a distorted sensor. Other sensor means may be used as long as the sensor means measures sensor data that can be converted into a load.

1 precast concrete slab (roadbed), 2 road, 2a sidewalk, 2b roadway, 3 pile, 4 road body, 5 space, 6 load sensor, 7 vehicle, 7a general vehicle, 7b motorcycle, 7c large vehicle, 8 people, 9 sign board, 10 communication device, 11 sensor data reception unit, 12 sensor information transmission unit, 13 detection information reception unit, 14 notification unit, 20 object detection device, 21 sensor information reception unit, 22 object detection unit, 23 notification control unit , 24 sensor information storage unit, 25 management information storage unit.

Claims (6)

In an object detection system for detecting an object on a road having a structure formed by using piles erected in a road construction area and precast concrete slabs placed on the piles,
a plurality of sensor means arranged on the road;
sensor information acquisition means for acquiring sensor information including at least the sensor value of the sensor means;
detection means for detecting an object on the road by referring to the load applied to the road surface obtained from the sensor value of the sensor means;
An object detection system comprising: When the load applied to the road surface is gradually increasing or decreasing, the detecting means keeps constant the amount of increase in the load per time in the gradually increasing state with no object on the road surface, based on the load obtained from the sensor value of the sensor means. or the load obtained by subtracting the load obtained by subtracting the load when it is assumed that the amount of decrease in the load per time in the gradually decreasing state is constant when there is no object on the road. 2. The object detection system according to claim 1, wherein the object detection system detects In the case where the sensor means is attached to the pile, the detection means receives as input the distribution of the load applied to the pile when there is an object on the road, and a learning model that outputs the presence or absence of the object, 2. The object detection system according to claim 1, wherein the object on the road is detected by inputting the load distribution applied to the piles obtained from the sensor values of the sensor means. detection means for detecting the position, moving direction, and moving speed of the object detected by the detecting means by referring to the distribution of the load applied to the road surface obtained from the sensor values of the sensor means and its transition;
reporting means for reporting the detection of the object to other objects located around the object detected by the detection means;
The object detection system of claim 1, comprising: a discrimination means for discriminating the type of the object detected by the detection means from the load applied to the road surface obtained from the sensor value of the sensor means;
The notification means notifies a vehicle approaching the person of the existence of the person when the object is determined to be a person by the determination means, and the vehicle when the object is determined to be a vehicle by the determination means. Notify a person located in the direction of travel of the approach of the vehicle,
5. The object detection system according to claim 4, characterized in that: 2. The object detection system according to claim 1, wherein said sensor means is a load sensor or a sensor that measures sensor data that can be converted into load.

Images