JP4905405B2 - Vehicle sensing system - Google Patents

Description

  The present invention relates to a vehicle sensing system.

  A vehicle sensing system that senses the presence or absence of a vehicle on a road is known in order to check traffic volume and vehicle occupancy time. Patent Document 1 discloses such a vehicle sensing system.

The system of Patent Document 1 includes a sensor that detects a vehicle passing through a monitoring area on a road. This sensor is a thermopile element and senses infrared rays emitted from detection objects such as vehicles and roads. This sensing result is given to the system as an input level.
Since the temperature of a vehicle is different from that of a road, the amount of infrared rays emitted from the vehicle is also different from that of the road. For this reason, when the vehicle passes through the monitoring area of the sensor, the input level from the sensor changes according to the temperature. It is possible to detect a vehicle passing through the monitoring area by the change in the input level.

More specifically, in the system of Patent Document 1, a value based on a difference between an input level obtained from a sensor and a background level indicating a road temperature is used as a comparison value, and the comparison value is compared with a threshold value. The presence or absence of a vehicle is determined.
Further, Patent Document 1 discloses a background level calculation means for changing the background level according to the previous vehicle determination result using the obtained input level.
Japanese Patent No. 3719438

  When the background level is changed according to the vehicle determination result, for example, when there is a vehicle, the background level is not learned, and when there is no vehicle, the background level is learned based on the input level.

By making the background level change not by a constant value but by learning, it is possible to cope with changes in the true value of the background level (background temperature) due to a change in temperature or the like.
In addition to such background level learning, it is desirable to take detailed measures so that the background level does not deviate from the true value as much as possible to further improve vehicle detection accuracy.

  Therefore, an object of the present invention is to provide a new technique for dealing with a background level deviating from a true value in a vehicle sensing system.

The present invention uses a value based on a difference between an input level obtained from a detection unit that detects a temperature of a monitoring area on a road and a background level based on a road temperature level as a comparison value, and the comparison value and the threshold value Determining means for determining the presence or absence of a vehicle, learning means for performing a learning process of the background level based on the input level while the determination of the absence of a vehicle is being performed, the comparison value and the background level based on the comparison result of the learning threshold, the vehicle sensing, characterized in that it comprises a, and the background level control Gosuru control means execution or cessation of learning process during which the determination without vehicle System.

As one of the causes that the background level deviates from the true value, the input level when there is a vehicle is determined by erroneously determining that there is no vehicle even though the vehicle exists in the monitoring area. Learning as a background level.
According to the present invention, the execution or pause of the background level learning process during the determination of the absence of the vehicle is controlled based on the comparison result between the comparison value and the threshold for background level learning. Even if the determination is made, the background learning is controlled in consideration of the comparison value.
Therefore, it can suppress that the input level when a vehicle exists is used as a background level, and it can suppress that a background level deviates from a true value.

  It is preferable that the background level learning threshold is set according to the value of the comparison value when it is determined that there is a vehicle. In this case, the background level learning threshold is appropriately set.

  When the comparison value is greater than a background level learning threshold, the control means preferably pauses the background level learning process while it is determined that there is no vehicle. In this case, if the comparison value is larger than the background level learning threshold value, the learning process is suspended. Therefore, it is possible to prevent background level learning in a state where the comparison value is relatively large and a vehicle may exist.

  According to the present invention, it is possible to suppress the background level from deviating from the true value, or to grasp the possibility that the background level deviates from the true value.

Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.
[First Embodiment]
1 and 2 show a vehicle sensing system 1 according to the first embodiment. This vehicle sensing system 1 is for sensing a predetermined area on the road as a monitoring area A, and when a vehicle V exists in the monitoring area A.

  The vehicle sensing system 1 includes a detection unit 2 that detects the temperature of the monitoring area A in order to detect a vehicle. The input level (input voltage value corresponding to the average temperature of the monitoring region A) obtained from the detection unit 2 is given to the sensing processing unit 3. The sensing processing unit 3 can determine the presence or absence of a vehicle in the monitoring area, and can output a signal indicating whether there is a vehicle (sensing ON) or no vehicle (sensing OFF) as a sensing result.

  The detection result is transmitted to a traffic signal controller or a traffic control center (not shown). The traffic signal controller is installed in the vicinity of the vehicle sensing system 1. In a traffic control center or the like, the detection results from a plurality of vehicle detection systems are totaled to calculate traffic volume and vehicle occupancy time, and traffic congestion judgment and vehicle average speed calculation are performed.

The detection unit 2 is composed of a thermopile element that detects the temperature of the monitoring region by sensing infrared rays emitted from a vehicle or a road. As shown in FIG. 1, the detection unit 2 is attached to a roadside column 4 in order to set the monitoring area A on the road R.
The detection unit 2 is connected to the sensing processing unit 3, and the input level obtained from the detection unit 2 is transmitted to the sensing processing unit 3.
The sensing processing unit 3 is disposed in a housing 5 attached to the support column 4. The detection processing unit 3 may be configured to be housed in the same housing as the detection unit 2, and the arrangement of the detection unit 2 and the detection processing unit 3 is not particularly limited to the configuration shown here.

  The detection processing unit 3 is configured by hardware including a computer having a CPU, a storage unit, and the like, and performs processing for vehicle detection by executing a computer program stored in the storage unit.

  Here, as a basic principle for determining the presence or absence of the vehicle V in the monitoring area A (sensing determination), the road temperature when there is no vehicle is provided as the background level, and the difference between the input level and the background level is determined. If there is a vehicle, it is determined that the vehicle is present (sensing ON), and if there is no difference between the input level and the background level, it is determined that there is no vehicle (sensing OFF).

  However, in the present embodiment, the sensing processing unit 3 determines the difference between the input level and the background level in order to perform the sensing judgment more accurately rather than simply comparing the input level and the background level. A value based on this is calculated as a comparison value, and the comparison value is compared with a threshold value (threshold value for sensing determination) to obtain a sensing result (sensing ON / sensing OFF).

Therefore, the sensing processing unit 3 has a function as a comparison value calculation unit 31 that calculates a value based on the difference between the input level and the background level as a comparison value (see FIG. 3A).
Furthermore, the sensing processing unit 3 has a function as a determination unit 32 that compares the comparison value and the threshold value to determine the presence or absence of the vehicle (see FIG. 3B).

  In the present embodiment, the comparison value calculation unit 31 calculates a small sum of products of differences between the input level and the background level as a value (comparison value) indicating a difference between the input level and the background level. This sum of products reduces the influence of noise from the simple difference between the input level and the background level and emphasizes the change of the simple difference to facilitate comparison with the threshold value.

As shown in FIG. 3A, in the comparison value calculation unit 31, first, the difference calculation unit 31a calculates the difference between the input level and the background level in order to calculate the product sum. Further, an integral value of the absolute value of the difference is obtained by the integral calculation unit 31b.
This integral value corresponds to the time average value of the difference, and a value obtained by reducing the influence of noise from the difference can be obtained. Further, a calculation for obtaining a differential value of the absolute value of the difference is performed by the differential calculation unit 31c. The differential value indicates the amount of change in the difference.

The integrated value and the differentiated value are added by the adding unit 31d, and this added value becomes a sum of products (comparison value).
The micro product sum is obtained by adding a differential value indicating a change to an integral value in which noise is reduced, and therefore, the influence of noise is reduced and the change is emphasized.

Since the determination unit performs determination using this sum of products, it is possible to perform detection determination more accurately. However, the comparison value is not limited to the sum of products as long as it is based on the difference between the input level and the background level.
The comparison value calculation unit 31 also outputs the integral value. This integral value is also a value based on the difference between the input level and the background level, and is used for threshold learning described later.

As illustrated in FIG. 3B, the determination unit 32 uses the comparison value (sum of products) calculated by the comparison value calculation unit 31 to perform a detection determination that outputs a detection result (sensing ON / sensing OFF). Do.
If the comparison value is larger than the threshold value (comparison value> threshold value), it is determined that the sensor is ON (there is a vehicle). If the comparison value is smaller than the threshold value (comparison value <threshold value), It is determined that the sensing is OFF (no vehicle).

  However, in this embodiment, in order to prevent chattering of the sensing result immediately after switching from sensing ON to sensing OFF or immediately after switching from sensing OFF to sensing ON, “threshold Lo” (first threshold value) is used as the threshold for sensing determination. ) And “threshold Hi” (second threshold) are introduced.

The threshold value Lo, which is the first threshold value, is for determining sensing OFF from the sensing ON state, and the threshold value Hi, which is the second threshold value, is for determining sensing ON from the sensing OFF state. is there.
In other words, the threshold Lo is used to determine the absence of the vehicle (sensing OFF) while the determination that the vehicle is present (sensing ON), and the threshold Hi is the determination of the absence of the vehicle (sensing OFF). ) Is used to determine whether the vehicle is present (sensing ON).

  The threshold value Lo is normally set to a value that is reduced by a predetermined value from the threshold value Hi (except in the case of a rapid increase mode described later), thereby preventing chattering at the time of sensing ON / OFF switching.

As shown in FIG. 4, the sensing processing unit 3 further functions as a background level learning unit 33 for learning a background level, and a threshold learning unit (threshold adjustment unit for learning threshold values (threshold Lo and threshold Hi)). ) 34 as a function.
The sensing processing unit 3 also has a function as a learning control unit 35 for controlling the learning process of the background level learning unit 33 and / or the threshold learning unit 34.

The background level (road surface temperature) varies greatly depending on the environment, such as when the road becomes sunny or shaded by sunlight. Deviation occurs.
Therefore, it is conceivable to learn the background level as shown in FIG. 5A (learning method according to a comparative example).
In the learning method shown in FIG. 5 (a), the background level learning unit 33 follows the input level because the input level should indicate the temperature of the road R when the vehicle is not in the monitoring area A and the sensing is OFF. To do background level learning.

In the learning method shown in FIG. 5 (a), when the vehicle is present in the monitoring area A, when the sensing is ON, the input level should mainly indicate the temperature of the vehicle V. The background level at the time of switching from sensing OFF to sensing ON is maintained without learning the background level.
As described above, in the learning method according to the comparative example, the presence or absence of learning is switched based only on the sensing result (sensing ON / sensing OFF).
This learning switching is performed by the learning control unit 35. In the learning method according to the comparative example, the learning control unit 35 can perform the switching control of the presence / absence of learning in the background level learning unit 33 by acquiring only the sensing result.

  On the other hand, in the background level learning method of the present embodiment, as shown in FIG. 5B, a case where learning is not performed is set even when sensing is OFF. The background level learning method shown in FIG. 5B will be described later.

Further, in the present embodiment, as described above, learning (adjustment) of not only the background level but also threshold values (threshold value Lo and threshold value Hi) is performed.
Here, when sensing is OFF (when there is no vehicle), in principle, the threshold value compared with the comparison value for determining the presence or absence of the vehicle may be a constant value.
That is, when the steady state with sensing OFF (no vehicle) continues, the difference between the input level and the background level (comparison value) becomes 0 in principle. When the vehicle passes through the monitoring area A in this state, the input level changes, and the difference (comparison value) between the input level and the background level increases.
For this reason, when the sensing is OFF, the threshold value is determined so that the difference between the input level and the background level can be determined to be greater than 0 by the vehicle when the vehicle enters the monitoring area. It ’s enough.
Therefore, as described above, the threshold value may be a constant value in principle when sensing is OFF.

However, actually, even if the sensing is OFF, the difference (comparison value) between the input level and the background level does not always become zero due to factors such as disturbance, and varies depending on time.
For this reason, in this embodiment, in order to determine sensing ON more accurately, as shown in FIG. 6, when sensing is OFF, learning is performed with a target obtained by adding a constant value to a comparison value (sum of product). . That is, in a steady state when sensing is OFF, a value obtained by adding a constant value to the comparison value is the threshold value.

More specifically, the threshold value learning unit 34 of the present embodiment learns the threshold value Hi by adding a constant value to the comparison value (sum of products) for the threshold value Lo when sensing is OFF. Is a value obtained by subtracting a certain value from the threshold value Hi.
As a result, in the present embodiment, when the sensing is OFF, the threshold value Hi and the threshold value Lo are slightly higher than the comparison value (sum of products) while maintaining the interval of the constant value, and the comparison value changes. Will follow.

In addition, when the sensing is ON (when the vehicle is present), the threshold value is a comparison value (difference between the input level and the background level) that is large because the vehicle exists in the monitoring region. It is sufficient if it can be determined that it has become smaller by not existing. Therefore, in principle, the threshold value when sensing is ON may be the same as the threshold value when sensing is OFF.
However, even when the threshold value is changed to follow the comparison value when sensing is OFF as described above, the comparison value when sensing is ON indicates the difference between the input level and the background level. The threshold value cannot follow the comparison value as in the case of OFF. Therefore, it is conceivable to hold the threshold value when switching from sensing OFF to sensing ON during sensing ON.

However, since the following problem cannot be dealt with only by holding the threshold value while sensing is ON, in this embodiment, the threshold learning unit (threshold adjustment unit) 34 learns (adjusts) the threshold value while sensing is ON. I do.
That is, when the vehicle V stops in the monitoring area A for a long time, such as when waiting for a signal, there may be a difference in road surface temperature before and after passing the vehicle.
In other words, since the road surface temperature changes under the influence of the vehicle V being stopped for a long time, the road surface temperature immediately before the detection ON from the detection OFF (just before the vehicle V enters the monitoring area A) and the detection OFF from the detection ON. There is a difference between the road surface temperature immediately after the vehicle V is reached (immediately after the vehicle V passes through the monitoring region A).

  Even in such a case, since the background level learning unit 33 cannot perform background level learning while sensing is ON, the background level learning unit 33 holds the background level when sensing is turned on from sensing OFF. As a result, when the vehicle V exits the monitoring area A and changes from sensing ON to sensing OFF, the value of the background level is a true value (road surface temperature changed by being affected by the vehicle V stopped in the monitoring area A). It will deviate from.

In this way, while the sensing is ON (while the vehicle V is in the monitoring area A), the background level gradually deviates from the true value as time passes. This divergence between the background level and the true value affects the comparison value (sum of products) indicating the difference between the input level and the background level.
In other words, the comparison value increases as the time lapses while sensing is ON.
Therefore, considering the occurrence of a deviation from the true value of the background level, during detection ON, the “sensitivity” for determining detection OFF is increased over time, and detection is performed even if the degree of decrease in the comparison value is small. It is necessary to be able to determine OFF.

Therefore, in the present embodiment, during the sensing ON, 90% of the integral value (a value based on the difference between the input level and the background level) is set to increase the “sensitivity” for determining the sensing OFF with the passage of time. Learning to gradually increase the threshold with the goal of. Here, this learning is referred to as “normal learning”.
This normal learning is not intended to set the threshold to 90% of the integral value itself, but reflects the deviation of the background level included in the integral value in the threshold and responds to the gradual increase of the deviation. This is for gradually increasing the threshold value.
Therefore, in this normal learning, the increase degree of the threshold is relatively gradual.

  More specifically, as shown in FIG. 6, the threshold learning unit 34 of the present embodiment has an integral value (value based on the difference between the input level and the background level) of 90 for the threshold Hi when sensing is ON. Learning is performed with% as a target, and the threshold Lo is set to a value obtained by subtracting a predetermined value from the threshold Hi.

Furthermore, in the present embodiment, the threshold learning unit 34 performs not only normal threshold learning but also rapid threshold increase processing when sensing is ON.
As shown in FIG. 6, this rapid increase process is a process that is executed when a predetermined time (200 msec) has passed since the detection is turned ON, and greatly increases the threshold value.
That is, until the predetermined time (200 msec) elapses, the thresholds (threshold Hi and threshold Lo) gradually increase by the normal learning process, and when the predetermined time (200 msec) is reached, only the threshold Lo is moderated. It increases with an increase greater than the increase.

This rapid increase process is a process for preventing the sensing ON time from becoming longer than the time during which the vehicle is actually present in the monitoring area A even when there is snow on the road surface or when it is raining.
That is, when there is snow on the road surface or in the case of rain, as shown in FIG. 7 (a), even if the vehicle is in the monitoring area A, the detection unit The input level obtained from 2 gradually decreases, and a phenomenon that does not easily return to the background level occurs.

The reason why such a phenomenon occurs is unknown, but when this phenomenon occurs, the comparison value (sum of products) indicating the difference between the input level and the background level is actually relatively large even after the vehicle has passed. Value. In other words, since the input level is increased even if the vehicle is not actually present, the comparison value (sum of products) indicates a value as if the vehicle is present.
As a result, as shown in FIG. 7B, the comparison value (sum of products) falls below the threshold value at the time t2 after the time t1 when the vehicle actually passes. For this reason, the sensing ON time is originally longer than the time when the sensing should be ON.

  In FIG. 7B, the threshold value is always increased by the normal learning process while sensing is ON, and the threshold value is learned while adding a constant value to the sum of products during sensing OFF. The threshold of the case is shown. For ease of understanding, in FIG. 7, the threshold value is shown as a single line without distinguishing between the threshold value Hi and the threshold value Lo, and the same applies to other figures.

Here, in the normal learning process, the threshold value is gradually increased in order to gradually increase the sensitivity of the detection OFF determination in accordance with the influence of the background level deviation. Therefore, the normal learning process gradually increases the threshold value. Even so, the problem that the sensing ON time becomes too long due to an abnormality in the input level due to snow or rain as described above cannot be sufficiently addressed.
Moreover, when the sensing is ON and the normal learning process is not performed and the threshold value when the sensing is ON is held at the threshold value when the sensing is ON, the above input level abnormality occurs. Then, the sensing ON time becomes longer than that in FIG.

  Therefore, in this embodiment, as shown in FIGS. 7C and 8, the threshold learning unit 34 enters the normal learning mode until a predetermined time (200 msec) elapses after the sensing is turned on (step S <b> 1). Although the threshold value is increased by the normal learning process (step S2), when a predetermined time (200 msec) has elapsed (step S1), the threshold value learning unit 34 shifts to a rapid increase mode in which the threshold value Lo is greatly increased. The threshold value is rapidly increased (step S3).

  As a result of this rapid increase processing, as soon as the comparison value (sum of products) slightly decreases from the maximum value, the comparison value falls below the threshold value Lo, and the sensing ON time is shortened compared to FIG. 7B. As a result, the sensing ON time can be prevented from becoming too long.

In the present embodiment, after the threshold Lo is rapidly increased by the rapid increase process, the threshold is held at the value after the rapid increase as long as the sensing ON continues (step S4). After the rapid increase process, the normal learning process may be performed again. However, since the threshold value Lo is increased by the rapid increase process, the necessity of performing the normal learning process after the rapid increase process is low. Therefore, as described above, by maintaining the value after the rapid increase process, the calculation for the normal learning process can be omitted, and the calculation load can be reduced.
Note that the processing in FIG. 8 ends at that point when the sensing result is sensing OFF.

In the present embodiment, the value of the threshold Lo when it is raised by the rapid increase process is the smaller of the values (1) and (2).
(1) {(integral value immediately after sensing ON) + (maximum integral value during sensing ON−integral value immediately after sensing ON) × 0.2} × correction value (2) integral value immediately after sensing ON × correction Value + constant value

  The value after raising the threshold value Lo is not limited to the above (1) and (2), and is a relatively large value (integrated value) within a range not exceeding the maximum integrated value (or sum of products) during sensing ON. Value (or about 80% of the sum of products)). In the case of (1) and (2) above, since (2) is generally smaller, the “constant value” in (2) is the upper limit for the increase.

Here, the timing of executing the rapid increase processing is not limited to 200 msec after the sensing is turned on, but the comparison value (sum of products) is sufficiently large after a predetermined time since the sensing is turned on. It should be executed at the time (near the timing when the comparison value takes the maximum value).
This is because if the timing of the rapid increase process is too early or too late, the threshold value cannot be increased sufficiently because the comparison value is small, and it is difficult to prevent the sensing ON time from being too long.
In this embodiment, since the threshold value Lo for rapid increase is learned for the comparison value, even if the timing of rapid increase processing is too early and the comparison value at that time is small, the comparison value is increased thereafter. As the threshold Lo increases, the influence of the operation timing of the rapid increase process being too early can be limited.

For example, assuming that the size of the monitoring area A is about 75 cm in the vehicle traveling direction and about 100 cm in the vehicle width direction, and the vehicle length is 6 m, for example, after the vehicle V of 60 km / h approaches the monitoring area A The time required to completely exit is (0.75 [m] +6 [m]) / (60 [km / h] × 1000/3600) = 0.405 [sec].
Therefore, at the timing of 200 msec, it is expected that the vehicle V is present at the center position in the vehicle traveling direction in the monitoring area A, and the comparison value (sum of products) is maximized.

  Thus, in view of the size of the monitoring area A and the assumed speed of the vehicle V, the rapid increase process is executed at a timing at which the comparison value is expected to take the maximum value, so that the comparison value is slightly smaller than the maximum value of the comparison value. It is possible to adjust the threshold Lo to a low value (for example, 80% of the maximum value).

  Note that the learning control unit 35 switches the learning method in threshold learning. In the threshold learning, the learning method can be switched as long as the time from sensing ON / OFF and sensing ON is known. Therefore, when the learning control unit 35 controls the threshold learning unit 33, only the sensing result is obtained. If you can get.

  Now, if the above-mentioned rapid increase processing is performed, it is possible to prevent the sensing ON time from becoming long when there is snow or rain. However, when there is no snow or rain, depending on the input level, there is a vehicle in the monitoring area A. Regardless, sensing may be turned off.

  Such a situation occurs, for example, when the input level changes as shown in FIG. 9A while the vehicle is present in the monitoring area A. In FIG. 9A, the input level is the maximum value at the initial stage in the time zone in which the vehicle V actually exists in the monitoring area A (the “vehicle presence” time zone in FIG. 9A). After that, the input level is slightly lower than the maximum value, but is maintained at a certain level in the time zone in which the vehicle V is actually present in the monitoring area A. When the vehicle V no longer exists from the monitoring area A, the input level further decreases.

  The change in the input level as shown in FIG. 9A indicates that the temperature of the hood of the vehicle V, which is high in the initial stage of the time zone in which the vehicle V actually exists in the monitoring area A, is detected. This occurs when the temperature of the ceiling surface of the vehicle V, which is relatively low, is detected.

  When the input level reaches the maximum value in the initial stage, the sum of products (comparison value) also reaches the maximum value at that timing, but when the vehicle V is relatively slow, at the timing of the initial stage, After 200 msec has elapsed, the threshold is rapidly increased, and the threshold is rapidly increased to the vicinity of the maximum value of the sum of products (step S3 in FIG. 8). And a threshold value is hold | maintained by the value after a rapid increase (step S4 of FIG. 8).

  However, as described above, the sum of products (comparison value) falls below the maximum value when the temperature of the ceiling surface of the vehicle V is detected, and the vehicle V actually exists even though the vehicle V exists. First, at the timing of t3 in FIG. 9B, the sum of small products falls below the threshold value (threshold value Lo), and the sensing is turned off from sensing ON.

Here, when the method of the comparative example shown in FIG. 5A is adopted as the background level learning method, when the sensing is turned off, the background level follows the input level as shown in FIG. 9A. Learning is done. That is, although the vehicle V actually exists in the monitoring area A, the input level at that time is erroneously learned as the background level, and the learned background level becomes higher than the true background level.
When the learning value of the background level reaches the input level (timing t4 in FIG. 9B), the sum of small products (comparison value) becomes zero.

  In addition, when the vehicle V does not actually exist from the monitoring area A, the input level is lowered to the true background level and is lower than the learned background level. For this reason, the difference between the input level and the background level increases, and the product sum (comparison value) indicating the difference increases after the timing 4.

When the sum of small products (comparison value) increases and the sum of small products exceeds a threshold value (threshold value Hi), sensing is turned ON. That is, at the timing when the vehicle V no longer exists, the sensing is turned on.
As described above, when the threshold value is rapidly increased, in the case of a change in the input level as shown in FIG. 9, the sensing is turned off even when the vehicle is present, or the sensing is turned on when the vehicle is not present. “Sense reversal” occurs.

This sense inversion is caused by the background level learning being performed unconditionally if the background level learning method is sensing OFF as in the learning method according to the comparative example (FIG. 5A).
That is, the determination of sensing ON / OFF (sensing result) is not limited to the case of FIG. 9, and may not necessarily be complete and accurate. However, in the learning method according to the comparative example, the learning control unit 35 controls the presence / absence of background level learning based only on the sensing result.

Therefore, in the learning method according to the comparative example, if the detection determination is incorrect, the background learning is also incorrect. As a result, sensing inversion is caused.
Therefore, in the background level learning method according to the embodiment of FIG. 5B, other than the sensing result is introduced as a learning execution / pause switching condition.

  Specifically, in the background level learning method of FIG. 5B, a sum of products (comparison value) is introduced in addition to the sensing result as a learning execution / pause switching condition. When the detection is turned off, the background level learning is not performed immediately, but the background level learning is performed after the sum of products (comparison value) becomes sufficiently small, so that the vehicle V actually exists in the monitoring area A. Learning the background level can be suppressed.

  More specifically, in this embodiment, a background level learning threshold is introduced, and if the sum of products (comparison value) is larger than the background level learning threshold as shown in FIG. However, if the sum of small products (comparison value) is smaller than the background level learning threshold when sensing is OFF, the background level is learned.

  As the background level learning threshold (hereinafter referred to as “holding Lo”), the value of the sum of products when sensing is turned on is set every time sensing is turned on. Specifically, as shown in FIG. 10, when the sensing result changes from sensing OFF to sensing ON (step S11), a threshold value (threshold value Lo) when sensing is turned on is set as holding Lo (step S12). ). In other words, the comparison value when the sensing is turned on is set as the holding Lo. The set retention Lo is used for comparison with the sum of products.

  The value of the holding Lo is maintained while the sum of small products exceeds the holding Lo, and is cleared when the sum of small products falls below the holding Lo (step S13).

FIG. 11 shows the sensing determination result when the background level learning method according to the embodiment shown in FIG. 5B is executed when there is an input level similar to FIG.
As shown in FIG. 11B, when the sum of small products becomes smaller than the threshold increased by the rapid increase process at timing t3, the vehicle V actually exists, but the sensing is turned off. However, even when the sensing is turned off, the sum of products exceeds the retention Lo, so that the possibility of vehicle presence is estimated.

  In the learning method according to the embodiment of FIG. 5B, the background level learning is not performed when the product sum exceeds the retention Lo (see FIG. 11A). As shown in FIG. 11 (b), the sum of small products is less than the retention Lo at almost the same time as when the vehicle V disappears from the monitoring area A and the sensing is turned off. Therefore, the time for learning the background level even when the vehicle is present is eliminated or very short.

As described above, since the temperature of the vehicle is hardly learned as the background level, it is possible to suppress the background level from deviating from the true value of the background level.
As a result, as shown in FIG. 11B, when the vehicle V does not actually exist from the monitoring area A, the difference between the input level and the background level (small product sum; comparison value) is suppressed from increasing. Therefore, it is possible to prevent the inversion of sensing when the vehicle V is not actually present but the sensing is ON.
In this way, as long as the rapid increase processing is performed, the sensing inversion in which sensing OFF is performed even if the vehicle V exists cannot be avoided, but the sensing inversion in which sensing ON is actually performed when the vehicle V does not exist can be prevented. Compared to the case of 9, the detection determination accuracy is improved.
If the background level is shifted, the influence may extend to the detection determination of the following vehicle. However, according to the present embodiment in which the background level learning method is brushed up, the background level is shifted from the true value. Can be further reduced. As a result, the influence on the detection determination of the following vehicle can be limited.

[Second Embodiment]
12 to 16 show a second embodiment of the present invention. In the second embodiment, the function of the background follow learning unit 36 and other functions 37 and 38 are added to the first embodiment as functions of the sensing processing unit 3. Note that points not particularly described in the second embodiment are the same as those in the first embodiment.

  In the second embodiment, the background level learning unit (first learning unit) 33 and the background follow learning unit (second learning unit) 36 are provided as functions for learning the background level. The background level learning unit 33 performs learning using the first learning method illustrated in FIG. The first learning method shown in FIG. 13A is the same as the learning method shown in FIG. Further, the background follow learning unit 36 performs learning by the second learning method illustrated in FIG. The second learning method shown in FIG. 13 (b) is basically the same as the learning method shown in FIG. 5 (a). However, background follow learning is faster than the background level (at 5 times faster). ) Learn to follow the input level.

The learning method of the background follow learning unit 36 is to always learn the background level (background follow) during the sensing OFF period. On the other hand, the learning method of the background level learning unit 33 does not execute the learning process depending on the conditions even during the sensing OFF period. For this reason, the learning method of the background follow learning unit 36 has a longer period for learning the background level than the learning method of the background level learning unit 33.
Thus, in the second embodiment, the background level and the background follow can be learned by different learning methods.

  Furthermore, the detection processing unit 3 according to the second embodiment determines the background level based on the function as the matching determination unit 37 that determines whether the background follow and the background level match and the determination result of the matching determination unit 37. Has a function as a displacement determination unit 38 that determines whether the background level is deviated from the true value of the background level.

  The coincidence determination unit 37 obtains a difference between the background level and the background follow when the sum of products (comparison value) is larger than the holding Lo, that is, while the sensing is OFF but the background level is not learned. If this difference is smaller than the predetermined value, the match determination unit 37 determines that the background level and the background follow match. On the other hand, if the difference is equal to or greater than a predetermined value continuously for a predetermined time (for example, 1 second), the coincidence determination unit 37 determines that the background level and the background follow are inconsistent.

If it is determined that the background level does not match the background follow, the background level may be shifted from the true value of the background level.
Furthermore, the deviation determination unit 38 changes the absolute value of the differential value of the input level (a differential value (differential value ()) in a state where the inconsistency determination is made (a state in which the background level is suspected to be deviated from the true value). If the change in the absolute value) is considered to have passed, the background level is determined to be “deviation” from the true value.
Further, the deviation determination unit 38 has a certain number of changes in the differential value of the input level several times (1 or If there are multiple times, it is determined that the background level is “deviation” from the true value.
When the shift determination unit 38 determines that “shift”, the background level learning unit 33 substitutes the background follow value for the background level.

The background level deviation detection and correction method will be described below with reference to FIGS.
As described in the first embodiment, the background level learning unit 33 does not perform background level learning when sensing is ON, and as long as the sum of small products is larger than the retained Lo even when sensing is OFF. , Do not do background level learning.
For this reason, as shown in FIG. 14A, if the true value of the background level changes before and after the vehicle V passes through the monitoring area A, the background level deviates from the true value.

  If the background level recognized by the system (background level learning unit 33) deviates from the true value when the detection is turned off from the detection ON, even if the input level has dropped to the true value of the background level, the input A difference between the level and the background level occurs, and the sum of products (comparison value) indicating the difference becomes relatively large as shown in FIG. 14B. As a result, even if the vehicle V finishes passing through the monitoring region A, the state where the sum of products exceeds the retention Lo continues. For this reason, the background level learning unit 33 cannot learn the background level even when the vehicle is not present, and cannot eliminate the background level shift.

Even in such a state, if the change in the input level at the time of subsequent vehicle passage is sufficiently large, the sensing determination can be performed normally as shown in FIG.
However, as shown in FIG. 15, when a vehicle having a small change in input level passes through the monitoring area A after the background level deviates from the true value, the sum of products cannot exceed the threshold, When the detection should be turned on, a “missing detection” that does not turn on the detection occurs.

  Therefore, in the second embodiment, when the coincidence determination unit 37 calculates the absolute value of the difference between the background level and the background follow in a state where the sum of products is larger than the retention Lo, Furthermore, if the deviation determination unit 38 changes the absolute value of the differential value of the input level more than a predetermined number of times, it is considered that the vehicle V has passed through the monitoring area A, and the background level deviates from the true value. It is determined that

  When it is determined that the background level is deviated from the true value, as shown in FIG. 16A, the background level learning unit 33 performs the background follow-up while learning the sum of products is larger than the retained Lo. The value of is considered to be more reliable, and background follow is substituted for the value of the background level to eliminate background level deviation.

  Thus, the background level shift can be eliminated, and the value of the sum of products can be reduced to be smaller than the holding Lo, and the background level learning unit 33 can learn the background level.

  Thus, according to the second embodiment, even if the background level deviates from the true value, it takes some time, but the deviation can be corrected.

The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the meanings described above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.
For example, in the above embodiment, learning of the threshold value may not be performed. Further, as threshold values, two types of threshold values Hi and threshold value Lo may not be provided, and they may be combined into one. Furthermore, the two learning methods shown in the second embodiment are merely examples, and other learning methods may be used.

It is a perspective view which shows the vehicle detection system installed in the road. It is a block diagram of a vehicle sensing system. It is a block diagram which shows a part of function of a sensing process part, (a) shows a comparison value calculation part, (b) shows a determination part. It is a block diagram which shows the function regarding learning among the functions of a detection process part. It is a table | surface which shows the background level learning method performed by the sensing process part, (a) is a comparative example, (b) is an Example. It is a table | surface which shows the threshold value learning method performed by the sensing process part. It is explanatory drawing at the time of snowfall or rain, (a) shows the relationship between the input level and the background level, (b) shows the relationship between the fine product sum and the threshold value when the rapid increase processing is not performed, and (c) shows the rapid increase The relationship between the sum of microproducts and the threshold when processing is performed is shown. It is a flowchart which shows the learning process of the threshold value at the time of sensing ON. It is a figure which shows the example which a sensing determination occurs when performing a rapid increase process, (a) shows the relationship between an input level and a background level, (b) shows the relationship between a sum of small products, and a threshold value. It is a flowchart which shows the setting process of holding | maintenance Lo. It is a figure which shows the case where background level learning is restrict | limited by holding | maintenance Lo, (a) shows the relationship between an input level and a background level, (b) shows the relationship between a sum of small products, and a threshold value. It is a block diagram which shows the function regarding learning among the functions of the sensing process part which concerns on 2nd Embodiment. It is a table | surface which shows the background level learning method and background follow learning method which concern on 2nd implementation liquid. It is explanatory drawing in case a background level deviates from a true value, and an input level is comparatively large, (a) shows the relationship between an input level and a background level, (b) shows the relationship between a sum of small products and a threshold value. (C) shows the sensing result. It is explanatory drawing when it is a case where a background level deviates from a true value and an input level is comparatively small, and a detection omission occurs, (a) shows the relationship between an input level and a background level, (b) is a micro product. The relationship between the sum and the threshold is shown, and (c) shows the sensing result. It is explanatory drawing at the time of correct | amending the deviation | shift from the true value of a background level, (a) shows the relationship between an input level and a background level, (b) shows the relationship between a sum of small products, and a threshold value, (c) is The sensing result is shown.

Explanation of symbols

1: vehicle sensing system 2: detection unit 3: sensing processing unit 4: support 31: comparison value calculation unit 32: determination unit 33: background level learning unit 34: threshold learning unit 35: learning control unit 36: background follow learning unit 37 : Coincidence determination unit 38: deviation determination unit R: road A: monitoring area V: vehicle

Claims (3)

A value based on the difference between the input level obtained from the detection unit for detecting the temperature of the monitoring area on the road and the background level based on the temperature level of the road is used as a comparison value, and the comparison value is compared with the threshold value. Determination means for determining the presence or absence of a vehicle;
Learning means for performing the learning process of the background level based on the input level while it is determined that there is no vehicle;
Based on the result of comparison between the comparison value and the background level learning threshold, and the background level control Gosuru control means execution or cessation of learning process during which the determination without vehicle,
A vehicle sensing system comprising: The background level learning threshold, the vehicle sensing system according to claim 1, wherein the set according to the value of the comparison value when it is determined there vehicle. 3. The vehicle detection system according to claim 1, wherein, when the comparison value is larger than a background level learning threshold value, the control unit pauses the background level learning process while it is determined that there is no vehicle. .

Images