JP2022069052A - Moving body detection system - Google Patents

Abstract

To provide a moving body detection system that can limit a scope of moving bodies to be detected to detect a moving body accurately.SOLUTION: In a body detection system 1 according to the present invention, a recognition device 10 comprises: moving body detection means 11 for detecting the presence of a vehicle C when a light source with constant illuminance penetrates into a closed domain S in an imaging area R1; closed domain passage detection means 12 for detecting passing through of the vehicle C from change in the average illuminance; filtering means 13 for excluding a moving body as a moving body not passing through the closed domain S when the average illuminance does not change in the closed domain S; and means 14 for determining movement along neighboring closed domains for determining whether the vehicle C that has passed through the closed domain S has passed through a neighboring closed domain T neighboring it at its upper side within a prescribed time.SELECTED DRAWING: Figure 2

Description

The present invention relates to a moving body detection system that detects the movement of a moving body from an image or video image captured by an image pickup device.

Conventionally, as a method of detecting the movement of a moving object, a method of calculating the difference in pixel values of frames included in captured moving image data and performing binarization processing (frame difference method), or an image to be detected from within the frame. A method of extracting the feature points of the above and tracking the feature points in time series (optical flow method) and the like are known.

The following motion detection system of Patent Document 1 is a system mounted on, for example, a car or the like to detect moving objects moving around itself. This motion detection system includes a motion detection device, a camera (imaging unit) as an image pickup means, a speaker, and a monitor.

When the motion detection system acquires a plurality of image frames, the motion detection device determines the brightness of the first image frame among the plurality of image frames and the second image frame acquired before the first image frame. The preprocessing frame including the ON pixel obtained from the absolute value of the difference is obtained.

Further, the motion detecting device aggregates the number of ON pixels for each region in the preprocessing frame, and obtains the time-dependent change of the center of gravity obtained from the number of ON pixels in the plurality of preprocessing frames. This makes it possible to detect a moving object and a moving direction (Patent Documents 1 / paragraphs 0029 to 0039, FIGS. 4 and 8).

Japanese Unexamined Patent Publication No. 2017-220020

However, in the case of the moving object detection system of Patent Document 1, all the moving objects existing in the image frame are detected, and the change with time is investigated. This means that tree branches and the like swaying due to the wind in the image frame are also detected targets.

Therefore, for example, when it is desired to check the moving direction of the vehicle as a moving body, there is a problem that the movement of the moving body, which does not need to be checked this time, becomes an obstacle and the detection accuracy is lowered.

In view of the above problems, it is an object of the present invention to provide a moving body detection system capable of detecting a moving body with high accuracy by limiting the moving body to be detected.

In order to solve the above problems, the present invention is a moving body detection system including a recognition device that captures a predetermined area using an imaging device and recognizes the movement of the moving body from the captured image or video.
The recognition device includes a moving body detecting means for detecting the presence of the moving body when a light source having a constant illuminance invades a closed region obtained by dividing the imaging region into a horizontal direction and a vertical direction, and a specific identification within the imaging region. From the change in the average illuminance in the closed region, the closed region passage detecting means for detecting the passage of the moving body invading the closed region and invading the closed region, and the case where the average illuminance in the closed region does not change. As a moving body that does not pass through the closed region, a filtering means that excludes it from the detection target and whether the moving body that has passed through the closed region has passed the adjacent closed region adjacent to the upper side of the closed region within a specified time. A means for determining movement between adjacent closed areas for determination, and
It is characterized by having.

In the moving body detection system of the present invention, the moving body detecting means of the recognition device detects the presence of a moving body (vehicle or the like) when a light source in the captured image or video enters a closed region in the imaging region. Further, the closed region passage detecting means detects a change in the average illuminance in a specific closed region. Specifically, when a moving object having a certain illuminance invades the closed region, the average illuminance in the closed region increases, and when the moving object invades, the average illuminance returns to the original value. Detects moving objects passing through.

For moving objects that move only in the closed region but do not pass through, the average illuminance in the closed region does not change, so the filtering means excludes them from the detection target. Further, the movement determining means between adjacent closed regions determines that the moving body that has passed through the closed region has passed the adjacent closed region adjacent to the upper side of the closed region within a specified time, thereby determining the direction of movement and its movement. Recognize that is occurring continuously. As described above, this moving object detection system can detect a moving object with high accuracy only for the moving object approaching the image pickup apparatus.

In the moving object detection system of the present invention
The recognition device has a position of the closed region and a focal length based on the size and focal length of the image pickup sensor of the image pickup device, the size of the object to be imaged, and the distance from the image pickup device to the object to be imaged. It is preferable to have a closed region determining means for determining the size.

For example, when the closed area is too large for the moving body, even if the moving body invades the closed area, the change in the average illuminance is small and difficult to understand. Therefore, the closed region determining means determines the size of the closed region based on the size and focal length of the image pickup sensor, the size of the object (moving body) to be imaged, and the distance from the image pickup device to the object to be imaged. Calculate and determine. As a result, the mobile object detection system can generate a closed region of an appropriate size that can easily capture changes in the average illuminance.

Further, in the moving body detection system of the present invention,
The recognition device includes a road identification means for distinguishing a road area in the imaging region from an area other than the road area using a machine-learned recognition model, and a road closure in which the closed area is set in the road area. It is preferable to have an area setting means.

The road identification means can accurately discriminate between the road area in the imaging region and the area other than the road area by using the machine-learned recognition model. Further, since the closed area setting means in the road sets the closed area in the road area, it is possible to reliably detect the moving body existing on the road.

Further, in the moving body detection system of the present invention,
The recognition device includes an image pickup control means for instructing control of an angle and a range of the image pickup region of the image pickup device.
The image pickup control means divides the horizontal direction of the image pickup area into at least two or more areas, adjusts the area having the largest area of the road area among the areas so as to be in the center of the image pickup area, and then adjusts the area. It is preferable to zoom in on the imaging area.

The imaging control means divides the horizontal direction of the imaging region and instructs the adjustment of the angle and range of the imaging region so that the area having the largest recognized road area is in the center of the imaging region. After that, the image pickup control means zooms in on the image pickup area. This makes it possible to accurately fit the vehicle on the road within the imaging region.

Further, in the moving body detection system of the present invention,
The recognition device includes a machine learning moving object detecting means for detecting the presence or absence of the moving object using a machine-learned recognition model, and an environmental brightness determining means for determining the brightness of the imaging environment from the captured image or video. Equipped with
It is preferable that the moving object is detected by the machine learning moving object detecting means and / or the moving object detecting means based on the brightness determined by the environmental brightness determining means.

The environmental brightness determination means can detect the brightness of the imaging environment from the captured image or video and determine daytime, nighttime, and evening. Further, the recognition device includes a machine-learned moving object detecting means for detecting the presence or absence of a moving object by using a machine-learned recognition model. The moving body detection system uses the machine learning moving body detecting means in the daytime, uses the above-mentioned moving body detecting means at night, and uses both the moving body detecting means and the machine learning moving body detecting means in the evening. The detection means is switched according to the brightness of the imaging environment. As a result, the moving object detection system can accurately detect the moving object according to the imaging environment.

Further, in the moving body detection system of the present invention,
The recognition device includes a closed region generation means that divides the imaging region into a horizontal direction and a vertical direction to generate a plurality of the closed regions, and an average illuminance detection that detects a change in the average illuminance for each of the closed regions. A downward movement vector that extracts a highlight area consisting of the means and a region having an average illuminance equal to or higher than a predetermined threshold among the closed regions, and the locus of the center coordinates of the highlight area moves below the imaging region. It is preferable to have a vector determination means for determining whether or not.

According to this configuration, the average illuminance detecting means detects a change in the average illuminance for each of the closed regions generated by the closed region generating means. This makes it possible to determine whether or not the moving body exists in the closed region. Further, the vector determination means extracts the highlight area and determines whether or not the locus of the center coordinates thereof is a downward movement vector. Then, when it is a downward movement vector, it can be determined that the moving body is moving in a direction approaching the recognition device.

Further, in the moving body detection system of the present invention,
The vector determination means has a counter, and when the locus of the center coordinates of the highlight area is a vector that moves downward in the imaging region, the value of the counter is added and the vector moves upward. It is preferable to subtract the value of the counter and determine that the downward movement vector is obtained when the value of the counter becomes a predetermined value or more.

According to this configuration, the vector determination means has a counter, and by adding the counter values, it is determined whether or not the vector is a downward movement vector. This makes it possible to improve the determination accuracy of the downward movement vector.

The figure explaining the outline of the moving body detection system which concerns on 1st Embodiment. The figure explaining the image data of an image pickup apparatus. The figure explaining the internal structure of the recognition apparatus and the image pickup apparatus of 1st Embodiment. (A) The figure (1) explaining the closed area passage detection processing of a vehicle. (B) Average illuminance graph in a closed region (1). (A) The figure (2) explaining the closed area passage detection processing of a vehicle. (B) Average illuminance graph in the closed region (2). (A) The figure (3) explaining the closed area passage detection processing of a vehicle. (B) Average illuminance graph in the closed region (3). (A) The figure (1) explaining the process of determining a closed area size. (B) The figure (2) explaining the process of determining a closed area size. The figure explaining the internal structure of the recognition apparatus and the image pickup apparatus of 2nd Embodiment. (A) A diagram illustrating a road area determination process (before viewpoint adjustment, 1x magnification). (B) The figure explaining the determination process of a road area (area division, equal magnification). (C) The figure explaining the determination process of the road area (the same magnification after adjusting the viewpoint). (D) The figure explaining the road area determination process (5x zoom after viewpoint adjustment). (A) A diagram illustrating machine learning mobile object detection (daytime). (B) A diagram illustrating machine learning mobile object detection (evening). (C) The figure explaining the moving body detection (evening). (D) The figure explaining the moving object detection (at night). The figure explaining the cell generation process. The figure explaining the average illuminance detection processing. The figure explaining the movement vector determination process. (A) Highlight area. (B) Highlight area (after ΔT). The figure explaining the moving object detection process between adjacent layers.

[First Embodiment]
First, with reference to FIG. 1, the outline of the mobile body detection system 1 according to the first embodiment of the present invention will be described.

The moving object detection system 1 includes a recognition device 10 and an image pickup device 50. The moving object detection system 1 is mounted on, for example, a special vehicle V that performs construction work or collects fallen objects on a highway at night, and has a function of detecting a vehicle C approaching the special vehicle V.

The recognition device 10 and the image pickup device 50 are connected by wire or wirelessly, and can communicate with each other. The image pickup device 50 captures an image of the rear of the road, acquires image data (real-time video), and transmits the image data to the recognition device 10. Then, the recognition device 10 executes image processing of the image data captured by the image pickup device 50.

The recognition device 10 analyzes the presence / absence of a vehicle, the direction in which the vehicle approaches the recognition device 10, the direction in which the vehicle moves away from the recognition device 10, and the like from the image data.

When the recognition device 10 recognizes the vehicle C in the figure as a moving body approaching the special vehicle V, a dangerous situation may occur for the worker P who works in the vicinity of the special vehicle V. Therefore, the recognition device 10 communicates with the mobile terminal owned by the worker P and outputs a warning sound or vibration to notify the situation. Further, an alarm may be attached to the special vehicle V to notify the worker P.

Next, the image data captured by the image pickup apparatus 50 will be described with reference to FIG. 2.

The outer frame of the figure (photograph) is a region (imaging region R1) of image data captured by the imaging device 50. The imaging region R1 may include roads, surrounding trees, road signs, guardrails, buildings, and vehicles on the road. Further, the frame (solid line) in the road indicates a closed area S (hereinafter, also referred to as “cell”). Although the details will be described later, the moving body detection system 1 recognizes a moving body passing through the closed region S and analyzes whether or not the movement is continuously generated.

The region adjacent to the upper side of the closed region S is the adjacent closed region T. The vehicle C approaching the image pickup device 50 will move below the image pickup region R1. If the vehicle C exists in the adjacent closed region T immediately before entering the closed region S, it can be said that the vehicle moves in the direction approaching the image pickup apparatus 50, and is therefore subject to monitoring.

Here, with reference to FIG. 3, the internal configuration of the recognition device 10 and the image pickup device 50 of the first embodiment will be described.

The recognition device 10 includes a moving body detecting means 11, a closed area passage detecting means 12, a filtering means 13, an adjacent closed area movement determining means 14, and a closed area determining means 15.

Further, the image pickup apparatus 50 includes an image pickup means 51. The image pickup means 51 is a video camera capable of capturing a moving image or a still image, and the generated image data is transmitted to the recognition device 10.

The moving object detecting means 11 of the recognition device 10 analyzes the received image data and detects the moving object. Since the moving body detecting means 11 detects the moving body depending on the presence or absence of a light source having a constant illuminance (headlight in the case of a vehicle), the moving body detecting means 11 functions even at night when the outer shape of the moving body cannot be recognized.

Further, the closed region passage detecting means 12 determines whether or not the moving body has passed through the closed region S from the change in the average illuminance in the closed region S described above. The average illuminance is quantified by converting the image data into a gray scale (black and white image). For example, the numerical value is set to "0 (black)" to "255 (white)", and the average value of the numerical values in the closed region S is calculated. The higher the value, the higher (brighter) the average illuminance in the closed region S.

Here, with reference to FIGS. 4A to 4C, the details of the detection of the moving body (vehicle) and the passage determination of the closed region of the moving body will be described.

The closed regions are arranged in a matrix in the imaging region R. Here, the upper closed region is referred to as a closed region S1, S2, S3, and the lower closed region, which is an adjacent closed region of the closed regions S1, S2, S3, is referred to as a closed region T1, T2, T3, respectively.

FIG. 4A shows a state immediately before the moving vehicle C (circle) enters the closed areas S1 to S3. As shown in FIG. 4A (a), here, the average illuminance of the closed regions S1 to S3 is low.

FIG. 4A (b) shows an average illuminance graph (solid line) in the closed region S1 and an average illuminance graph (broken line) in the closed region T2, respectively. As shown in the figure, the average illuminance of the closed regions S1 and T2 maintains the level 1.

The asterisk in the closed region S3 indicates the moving body M. Although the details will be described later, the moving body M is an object (tree branch or vehicle) moving in the closed region S3. As for the closed regions S2, T1 and T3, since there is neither an invading moving body nor an originally existing moving body, the average illuminance in the closed region does not change.

FIG. 4B shows a state after the vehicle C has entered (gate-in) the closed region S1. As shown in FIG. 4B (a), when the vehicle C enters the closed region S1, the average illuminance of the closed region S1 becomes higher than that of the other closed regions. Therefore, the moving body detecting means 11 detects the presence of the vehicle C in the closed region S1. The vehicle C is traveling in the lower right direction of the closed region S1.

FIG. 4B (b) shows an average illuminance graph (solid line) of the closed region S1. As shown in the figure, the average illuminance of the closed region S1 becomes level 3 during the period in which the vehicle C is invading, but then decreases to level 1 as the vehicle C invades. Further, the average illuminance graph (broken line) of the closed region T2 does not change at level 1 because the vehicle C does not exist in the closed region T2.

FIG. 4C shows a state after the vehicle C has invaded (gated out) from the closed region S1 and entered the closed region T2. As shown in FIG. 4C (a), since the vehicle C exists in the closed region T2, the average illuminance in the closed region T2 is higher than that in the other closed regions.

FIG. 4C (b) shows an average illuminance graph (solid line) of the closed region S1. Partly overlapping the period of FIG. 4B (b), the average illuminance becomes level 1 after the vehicle C has invaded the closed region S1. Further, the average illuminance graph (broken line) of the closed region T2 is level 1 during the period when the vehicle C does not exist in the closed region T2, but rises to level 3 as the vehicle C invades. As described above, the closed region passage detecting means 12 detects that the vehicle C has invaded and invaded the closed region S1 from the change in the average illuminance in the closed region S1, that is, the passage of the closed region S1.

Returning to FIG. 3, in the moving determination means 14 between adjacent closed regions of the recognition device 10, the vehicle C invades from the closed region S (any of S1 to S3) as in the above example, and the adjacent closed region T (T1). It is detected whether the event until the invasion into (any of T3) has occurred within the specified time. This detection is used to detect the continuity of movement of the vehicle C and the direction of movement.

On the other hand, the moving body M only moves in the closed region S3 and does not advance to the outside of the closed region S3 (see the closed region S3 in FIGS. 4A to 4C). Therefore, the average illuminance of the closed region S3 remains a constant value. When the average illuminance in the closed region S3 does not change, the filtering means 13 of the recognition device 10 excludes it from the detection target as a moving object (so-called noise) that does not pass through the closed region S3.

Next, the process of determining the closed area size will be described with reference to FIG.

In the mobile body detection system 1, it is necessary to set the size of the closed region S described above to an appropriate range. For example, when the size of the closed region S is too large with respect to the size (height) of the vehicle C, the change in the average illuminance is extremely small even if the vehicle C enters the closed region S, and it is difficult to detect.

Therefore, in the mobile body detection system 1, the closed area determining means 15 of the recognition device 10 determines the size of the closed area S. Specifically, the closed region determining means 15 includes the vertical size and focal length of the image pickup sensor of the image pickup device 50, the size of the object to be imaged, and the distance from the image pickup device 50 to the object to be imaged. The optimum size of the closed region S is determined based on.

As shown in FIG. 5A, the recognition device 10 generates a closed region S in the imaging region R2 in order to recognize the vehicle D. Then, when the vehicle D is detected in the closed area S, the size of the closed area S is determined. The estimated height of the imaging region R2 is H (m).

As shown in FIG. 5B, the estimated height H (m) of the image pickup region R2 with respect to the vehicle D traveling at a point separated by a distance L (m) from the image pickup device 50 is the image pickup device 50 (imaging means 51). The vertical size of the image sensor 55 of the image sensor 55 is Ih (mm), and the focal length F (mm) of the lens 56 is given by the following equation.
H = L × (Ih / F) ・ ・ ・ (Equation 1)
The focal length F is proportional to the zoom magnification (F = k × [magnification]).

For example, when F = 4.7 × 10 = 47 (mm) and Ih = 3.6 (mm), for vehicle D 200 m (L = 200) before, H≈15.3 from (Equation 1). It becomes (m).

When the height h of the vehicle D is 3.0 (m), in order to obtain a closed region S having a height (4.5 (m)) 1.5 times (N = 1.5) of the height h,
HC = (h × N) / _H ... (Equation 2)
Therefore, the height HC (%) of the closed region _S may be suppressed to about 30% of the imaging region R2. By such processing, the optimum size of the closed region S is determined.

[Second Embodiment]
In the moving object detection system 1, the above-mentioned function of passing detection in a closed region cannot be realized unless a moving object such as a vehicle to be detected is housed in the imaging region of the imaging device 50. Therefore, it is necessary to control the direction and the area of the image pickup apparatus 50 to fit the approaching moving object within the area.

FIG. 6 shows the internal configuration of the recognition device 10 and the image pickup device 50 of the second embodiment. In addition to the configuration described in the first embodiment (see FIG. 3), the recognition device 10 includes a road recognition means 17, a road closed area setting means 18, an image pickup control means 19, and a machine learning moving object detecting means 20. The environment brightness determining means 21, the closed region generating means 22, the average illuminance detecting means 23, and the vector determining means 24 are provided.

The road recognition means 17 uses the first machine-learned recognition model to discriminate between a road area in the imaging region and an area other than the road area (buildings, walls, trees, etc.). Then, the closed area setting means 18 in the road sets the above-mentioned closed areas S and T in the road area. As a result, the moving object detection system 1 can reliably detect a moving object (vehicle) existing on the road.

Next, the road area determination process will be described with reference to FIG. 7.

FIG. 7A shows an image pickup region R3 at the same magnification (without zoom) when the image pickup apparatus 50 images the rear of the road. This is the state at the time when the image pickup is started (before the viewpoint adjustment), and the viewpoint of the image pickup apparatus 50 is at the center (solid line circle) of the image pickup region R3.

After that, as shown in FIG. 7B, the road recognition means 17 divides the horizontal direction of the imaging region R3 into three areas of L (Left) -Zone, C (Center) -Zone, and R (Right) -Zone. do. The road recognition means 17 further divides each area into three in the vertical direction, determines T (Top) -Area, M (Middle) -Area, and B (Bottom) -Area from above, and sets the road in each area. Recognize how many pixels are included. The portion recognized as the road area by the road recognition means 17 is given a predetermined color.

In the example of FIG. 7B, when a road area of 47% in the L-Zone, 26% in the C-Zone, and 11% in the R-Zone is recognized, the ratio of the road area in the imaging region R3 is occupied. Is 28% of the average. In addition, detailed information is obtained on how much area is determined to be a road area in the L-Zone, such as 15% for T-Area, 70% for M-Area, and 15% for B-Area. You may.

After that, the image pickup apparatus 50 is controlled so that the center of the L-Zone or the center of the R-Zone becomes the image pickup center. The standard specifications for remote control of the image pickup device 50 are defined by the Open Network Video Interface (hereinafter referred to as ONVIF (registered trademark)). In ONVIF control, it is possible to specify how many seconds the image pickup means 51 is rotated according to the rotation angles to be adjusted in the left-right direction and the up-down direction. Similarly, with ONVIF control, it is possible to specify how many seconds the zoom up or zoom down is performed. Further, by this ONVIF control, the center of the imaging region R3 is controlled by PTZ (Pan, Tilt, Zoom).

The image pickup apparatus 50 includes, in addition to the image pickup means 51 described in the first embodiment (see FIG. 3), a control means 52 for transmitting a control signal to the image pickup means 51. The control means 52 receives an instruction (control signal) from the image pickup control means 19 of the recognition device 10 and performs the above-mentioned PTZ control.

In this example, the road recognition means 17 of the recognition device 10 uses the first machine-learned recognition model to recognize the road area from the imaging region R3. As a result of calculating the road area, it is recognized that there are many road areas on the left side of the image pickup area R3 (curve in the left direction), so that the image pickup control means 19 instructs PTZ control.

Specifically, the imaging control means 19 acquires the area of the road area recognized by the first machine-learned recognition model, and the center (broken line circle) of the L-Zone having the largest area of the road area is determined by ONVIF control. Adjust the position so that it is in the center of the screen. Specifically, the image pickup control means 19 performs ONVIF control so that the image pickup means 51 rotates to the left at an angle of 1/3 of the viewing angle. As a result, as shown in FIG. 7C, the center of the imaging region R4 moves to the center of the L-Zone in the imaging region R3.

After that, as shown in FIG. 7D, the image pickup control means 19 instructs to increase the magnification (5x zoom), and ONVIF control performs time zooming up according to the increase in the magnification. As a result, the moving object detection system 1 can accurately accommodate the moving object (vehicle) on the road within the imaging region.

Next, with reference to FIG. 8, moving object detection according to the brightness of the imaging environment will be described.

The moving object detecting means 11 described in the first embodiment analyzes the image data to detect the moving object, but when the detection target is a vehicle, the presence or absence of the moving object is detected by the light of the headlight. Therefore, the moving object detection including the moving object detecting means 11, the closed region passage detecting means 12, the filtering means 13, and the moving object detecting means 14 between adjacent closed regions is applied at night when the brightness (brightness) of the imaging environment is low. Is preferable.

On the other hand, in the daytime when the image quality is high, the image data is analyzed using a second machine-learned recognition model (for example, a moving object recognition model provided by Intel as OpenVINO®). , It is easiest to detect the presence or absence of a moving object. The machine learning moving object detecting means 20 of the recognition device 10 is in charge of detecting the vehicle by the second machine-learned recognition model (see FIG. 6).

The environmental brightness determination means 21 of the recognition device 10 analyzes the image data or determines the brightness of the imaging environment from the average illuminance of the image data, and whether the time when the image data is captured corresponds to daytime, evening, or nighttime. To judge. When the recognition device 10 is provided with a real-time clock, it is possible to perform daytime, evening, and nighttime determination from the date and time.

When the environmental brightness determining means 21 determines that the imaging time is "daytime", it is controlled to detect the presence or absence of the vehicle by the above-mentioned machine learning moving object detection. Further, when the environmental brightness determining means 21 determines that the imaging time is "nighttime", it is controlled so as to detect the presence or absence of the vehicle by the moving object detection of the first embodiment.

Further, when the environmental brightness determining means 21 determines that the imaging time is "evening", it is controlled to detect the presence or absence of a vehicle by a hybrid method of machine learning moving object detection and moving object detection of the first embodiment. As a result, the moving object detection system 1 can accurately detect the moving object according to the brightness of the imaging environment.

Next, the approaching moving object detection process will be described with reference to FIGS. 9 to 12. This is a process of detecting a moving object approaching the image pickup apparatus 50 by a method different from that of the first embodiment, and the closed region generation means 22, the average illuminance detecting means 23, and the vector determination means 24 of the recognition device 10 are used. Used (see FIG. 6).

FIG. 9 shows image data (imaging range R5) captured by the imaging device 50. The closed region generation means 22 of the recognition device 10 generates a closed region (cell) in the imaging region R5.

In order to generate cells, the closed region generation means 22 first divides the imaging region R5 into layers. Here, the "layer" is an image area obtained by dividing the imaging region R5 in the vertical direction. In the example of FIG. 9, the closed region generation means 22 vertically divides the imaging region R4 into seven (Layer_1 to Layer7). After that, the closed area generation means 22 divides each layer in the horizontal direction to generate cells. As a result, for example, Cell_A in which Layer_3 is divided and Cell_B in which Layer_7 is divided are generated.

The vertical height of the layers can be set to any value, but it is preferable that the upper layer of the imaging region R5 is lower in height than the lower layer. By reducing the cell size above the imaging region R5, the cell size in the approaching direction (downward) becomes large, so that the accuracy of determining whether a distant approaching vehicle exists in the cell is improved.

Next, the average illuminance detecting means 23 of the recognition device 10 detects a change in the average illuminance for each of the cells and determines the presence of an approaching vehicle. As shown in FIG. 10, the average illuminance detecting means 23 first executes an average illuminance calculation process on a plurality of cells in the imaging region R6 (for example, six cells in the region X). In the average illuminance calculation process, the average illuminance is calculated for each cell and the minimum illuminance is updated.

The calculation of the average illuminance is the same as that of the first embodiment, but first, the image data is quantified by converting it into a gray scale (black and white image). Further, the average value of the numerical values (0 to 255) in the cell is calculated. The higher the number, the higher (brighter) the average illuminance in the cell.

After that, the average illuminance detecting means 23 executes the average illuminance change evaluation process. In this process, two conditions are determined and evaluated: [Condition 1] (average illuminance) ≧ (designated threshold value) and [Condition 2] (average illuminance-minimum illuminance)> designated threshold value. The average illuminance detecting means 23 estimates that an approaching vehicle exists in the cell when [Condition 1] and [Condition 2] are satisfied.

For example, when the designated threshold value is "100", the minimum illuminance value of Cell_C in the region X is "30", and the average illuminance value of Cell_C is "237", the average illuminance (237) ≥ designated threshold value (100). ), And since the average illuminance (237) -minimum illuminance (30)> designated threshold value (100), [Condition 1] and [Condition 2] are satisfied. Therefore, the average illuminance detecting means 23 estimates that an approaching vehicle exists in Cell_C.

Next, the vector determination means 24 of the recognition device 10 uses the function provided by a general-purpose image processing software package such as OpenCV to determine the center coordinates of the highlight area (illuminance is equal to or higher than the specified threshold value) in the cell. It is determined whether or not the locus of the center coordinates is a vector (downward movement vector) that moves downward in the image pickup region R6.

As shown in FIG. 11A, the vector determination means 24 first executes subdivision cell average illuminance processing on Cell_C in the imaging region R6. In this process, cells are divided horizontally and vertically to generate subdivided cells (for example, 16 subdivided cells), and the average illuminance of the subdivided cells is calculated.

Next, the vector determination means 24 executes the highlight area calculation process. In FIG. 11B (a), the region HA in Cell_C shows the outer shape of the highlighted subdivided cell calculated in this process. Further, the vector determination means 24 executes the highlight vector processing. In this process, the center coordinate O ₁ of the highlight area (region HA) is determined, and the locus of the center coordinate O ₁ is investigated.

FIG. 11B (b) shows the region HA after ΔT from the state of FIG. 11B (a). Here, the region HA in Cell_C is deformed as the vehicle moves, and the center coordinates move to the point _O2 . At this time, the vector determination means 24 makes an approach determination from the angle of the movement vector (O ₁ → O ₂ ).

The vector determination means 24 includes a counter, and the movement vector may be obtained from the counter value. When the locus of the center coordinates of the region HA is a movement vector that moves below the imaging region R6, the value of the counter is added. On the other hand, in the case of a movement vector in which the locus of the center coordinates moves upward, the value of the counter is subtracted. Then, when the value of the counter becomes equal to or higher than a predetermined value, the vector determination means 24 determines that the vector is a downward movement vector.

Finally, the adjacent closed region movement determination means 14 of the recognition device 10 determines whether or not the vehicle has passed between the adjacent layers within a designated time. FIG. 12 shows a plurality of cells (nine cells in the region Y) in the imaging region R6. For example, when the cell to be processed is Cell_E existing in Layer_4, the average illuminance detecting means 23 detects the change in average illuminance with respect to Cell_E and detects the presence of an approaching vehicle.

When it is determined that an approaching vehicle exists in Cell_E, in the adjacent closed region movement determination means 14, the layer adjacent to one above Cell_E, that is, the cell (Cell_C or Cell_D) of Layer_3 is within the specified time (for example, 1 to 1 to). 2 seconds), it is determined whether the presence of an approaching vehicle is detected. By this determination, the continuity of movement of the vehicle and the direction of movement are detected, so that it is possible to detect a moving object approaching the image pickup apparatus 50 as a result.

The present invention is not limited to the above-described embodiment, and may take various forms as long as it belongs to the technical scope of the present invention. Although an example of a vehicle is shown as a moving body, a motorcycle, an agricultural vehicle, a bicycle, or the like may be used.

1 Mobile detection system 10 Recognition device 11 Mobile detection means 12 Closed area passage detection means 13 Filtering means 14 Adjacent closed area movement determination means 15 Closed area determination means 17 Road recognition means 18 Road closed area setting means 19 Imaging control means 20 Machine learning moving object detecting means 21 Environmental brightness judging means 22 Closed area generating means 23 Average illuminance detecting means 24 Vector judging means 50 Imaging device 51 Imaging means 52 Control means 55 Imaging sensor 56 Lens C, D Vehicles R1 to R6 Imaging region S , S1 to S3 Closed area T, T1 to T3 Adjacent closed area V Special vehicle

Claims (7)

A mobile body detection system including a recognition device that captures a predetermined area using an image pickup device and recognizes the movement of the moving body from the captured image or video.
The recognition device is
A moving body detecting means for detecting the presence of the moving body when a light source having a constant illuminance invades a closed area obtained by dividing the imaging region into a horizontal direction and a vertical direction.
A closed region passage detecting means for detecting the passage of the closed region in which the moving object invades and invades the closed region from the change in the average illuminance in the specific closed region in the imaging region.
When the average illuminance in the closed region does not change, a filtering means for excluding it from the detection target as a moving object that does not pass through the closed region, and
A means for determining movement between adjacent closed regions, which determines whether or not the moving body that has passed through the closed region has passed through an adjacent closed region adjacent to the upper side of the closed region within a specified time.
A mobile detection system characterized by being equipped with. In the mobile detection system according to claim 1,
The recognition device has a position of the closed region and a focal length based on the size and focal length of the image pickup sensor of the image pickup device, the size of the object to be imaged, and the distance from the image pickup device to the object to be imaged. A moving object detection system characterized by having a closed area determination means for determining the size. In the mobile detection system according to claim 1 or 2.
The recognition device is
A road identification means for discriminating between a road area in the imaging region and an area other than the road area using a machine-learned recognition model.
A road closed area setting means for setting the closed area in the road area,
A mobile detection system characterized by being equipped with. In the mobile detection system according to claim 3,
The recognition device includes an image pickup control means for instructing control of an angle and a range of the image pickup region of the image pickup device.
The image pickup control means divides the horizontal direction of the image pickup area into at least two or more areas, adjusts the area having the largest area of the road area among the areas so as to be in the center of the image pickup area, and then adjusts the area. A moving object detection system characterized by zooming in on the imaging area. In the mobile detection system according to any one of claims 1 to 4,
The recognition device is
A machine learning moving object detection means that detects the presence or absence of the moving object using a machine-learned recognition model,
The environment brightness determination means for determining the brightness of the imaging environment from the captured image or video is provided.
A moving body detection system characterized in that the moving body is detected by either the machine learning moving body detecting means and / or the moving body detecting means based on the brightness determined by the environmental brightness determining means. .. In the mobile detection system according to any one of claims 1 to 5,
The recognition device is
A closed region generation means that divides the imaging region in the horizontal direction and the vertical direction to generate a plurality of the closed regions.
An average illuminance detecting means for detecting a change in the average illuminance for each of the closed regions,
A highlight area consisting of a region having an average illuminance equal to or higher than a predetermined threshold value is extracted from the closed region, and the locus of the center coordinates of the highlight area becomes a downward movement vector that moves below the imaging region. Vector judgment means to determine whether or not
A mobile detection system characterized by being equipped with. In the mobile detection system according to claim 6,
The vector determination means has a counter and has a counter.
When the locus of the center coordinates of the highlight area is a vector moving downward in the imaging region, the value of the counter is added, and when the trajectory is a vector moving upward, the value of the counter is subtracted.
A moving object detection system characterized in that when the value of the counter becomes equal to or higher than a predetermined value, it is determined as the downward movement vector.

Images