JP4361381B2 - Vehicle detection device - Google Patents

Description

  The present invention relates to a vehicle detection device that detects a traveling vehicle in an image.

  As a conventional vehicle detection device that detects a vehicle from an image, for example, there is one disclosed in Patent Document 1. In this vehicle detection device, first, a horizontal edge and a vertical edge are extracted from an image taken by a camera. The horizontal edge is a characteristic sudden change boundary where the luminance changes suddenly in a direction orthogonal to the traveling direction of the vehicle, and the vertical edge is a characteristic sudden change boundary where the luminance changes suddenly in the traveling direction of the vehicle. Next, from the obtained horizontal edges / longitudinal edges, those moving in the traveling direction of the vehicle at a constant speed are extracted as a constant speed moving horizontal edge and a constant speed moving vertical edge. Thereafter, image elements that move uniformly throughout the entire image are removed as a background, and the remaining constant speed horizontal edges and constant speed vertical edges are registered as vehicle candidates. The constant-velocity moving horizontal edge is information that identifies the area in the width direction of the vehicle, while the constant-velocity moving vertical edge specifies the area in the traveling direction of the vehicle, so both edges registered as vehicle candidates The vehicle is detected by using the obtained partial region as a vehicle region.

JP 2003-51088 A (paragraphs 0009 to 0011)

The conventional vehicle detection device is configured as described above, and since the vehicle is detected by extracting the constant velocity moving horizontal edge and the constant velocity moving vertical edge, a large number of edge components with slightly different speeds are generated. There was a problem that they could be combined and processing took time.
Also, if the lighting conditions are extremely bright or dark depending on the lighting conditions, for example, the edge components are split or combined, resulting in an unstable appearance, and it is difficult to stably extract velocity information from the edge components. There was a problem.

  The present invention has been made to solve the above-described problems, and stabilizes the position of the vehicle even when a large number of edge components with slightly different speeds occur or when the appearance state of the edge components is unstable. An object of the present invention is to obtain a vehicle detection device that can detect correctly and efficiently.

The vehicle detection device according to the present invention corresponds to an image input means for inputting an image, an edge component extraction means for extracting an edge component from the input image, an edge component extracted in the past, and a newly extracted edge component In addition, a motion amount extraction unit that extracts a motion amount for each edge component, a motion amount frequency distribution generation unit that generates a frequency distribution of the extracted motion amount, and a high frequency in the input image from the generated motion amount frequency distribution. A high-frequency motion amount extraction unit that extracts a motion amount that occurs frequently as a high-frequency motion amount, and the high-frequency motion amount extraction unit that extracts the motion amount from the edge component from which the motion amount has been extracted by the motion amount extraction unit. Edge component selection means for selecting an edge component having a motion amount included in the range of the high frequency motion amount selected, position coordinates of the selected edge component, and frequency distribution of the edge component at this position coordinate A vehicle position determining means for determining a location of the determined vehicle whether Luo vehicle, a license plate detecting means for detecting the number plate in the input image using the edge components extracted by the edge component extracting unit, detecting License plate recognition means for recognizing characters in the license plate, the edge component extraction means calculates edge strength from the input image, binarizes the edge strength to create an edge binary image, A connected region is extracted as one edge component from the binary projection data created for each partial region obtained by dividing the edge binary image .

  The present invention is very likely to be a part of the vehicle body even when a large number of edge components with slightly different velocities are generated, or even when the edge components are split or combined and appear unstable. Edge components can be extracted, and the vehicle position can be detected stably and correctly.

Hereinafter, an embodiment of the present invention will be described.
Embodiment 1 FIG.
1 is a block diagram showing a configuration of a vehicle detection apparatus according to Embodiment 1 of the present invention. In FIG. 1, an image input unit 1 inputs an image, and an edge component extraction unit 2 extracts an edge component from the input image. The motion amount extraction means 3 extracts the motion amount for each edge component by associating the edge component extracted in the past with the newly extracted edge component, and the motion amount frequency distribution creation means 4 extracts the motion amount of the extracted motion amount. A frequency distribution is created, and the high-frequency motion amount extraction unit 5 extracts a motion amount that occurs frequently in the input image as a high-frequency motion amount from the generated frequency distribution of motion amounts. The edge component selection unit 6 selects an edge component having a motion amount included in the range of the high frequency motion amount extracted by the high frequency motion amount extraction unit 5 from the edge components from which the motion amount has been extracted by the motion amount extraction unit 3. The vehicle position determination means 7 determines the presence or absence of the vehicle from the position coordinates of the selected edge component and the frequency distribution of the edge component at this position coordinate, and determines the position of the vehicle.

Next, the operation will be described.
FIG. 2 is a flowchart showing a process flow of the vehicle detection apparatus according to the first embodiment of the present invention. In step ST11, the image input means 1 inputs an image.
FIG. 3 is a diagram showing an example of an input image. In the input image 101, the head portion (vehicle head) of the vehicle, the lower left of the vehicle, the shadow of the vehicle, and the road surface below can be seen. As shown in FIG. 3, the coordinates in the input image 101 take the X axis in the right direction (horizontal direction) and the Y axis in the downward direction (vertical direction) with the upper left corner of the image as the origin. Here, it is assumed that a black and white grayscale image in which one pixel value is represented by 256 levels from 0 to 255 is input as the input image 101.

In step ST <b> 12 of FIG. 2, the edge component extraction unit 2 extracts edge components from the input image 101.
FIG. 4 is a flowchart showing a flow of detailed processing of edge component extraction. In step ST21 of FIG. 4, the edge component extraction means 2 calculates the edge strength for each pixel. This edge strength is a value indicating the steepness of the light and shade fluctuation in the input image 101. For example, see “Introduction to Computer Image Processing” pp. It can be calculated by various methods described in 119 to 125. As a specific example, it is possible to take a method in which the input image 101 is differentiated and its absolute value is used as the edge strength, or a gradient is used as the edge strength. In the first embodiment, the absolute value obtained by differentiating the input image 101 in the horizontal direction is set as the edge strength, and is obtained by the following equation (1).
E = ABS (I (x-1, y) -I (x, y)) (1)
Here, E is an edge strength, ABS is a function that takes an absolute value, and I (x, y) indicates a pixel value of the input image 101 at coordinates (x, y).

In step ST22, the edge component extraction unit 2 binarizes the edge strength and creates an edge binary image. For example, binarization is described in “Introduction to Computer Image Processing” pp. In the first embodiment, a fixed threshold value is set in advance, and “1” is set when the edge strength is equal to or higher than the threshold value, and “0” is set when the edge strength is lower than the threshold value. It is assumed that binarization is performed.
FIG. 5 is a diagram showing an example of the edge binary image, and the edge binary image 102 shown in FIG. 5 is obtained from the input image 101 shown in FIG. In the edge binary image 102, the pixel value “1” is shown in black and the pixel value “0” is shown in white.

In step ST23 of FIG. 4, the edge component extraction unit 2 divides the edge binary image 102 into partial areas, and obtains a projection value by projecting in the horizontal direction for each partial area.
FIG. 6 is a diagram showing an example of area division of the edge binary image at the time of edge component extraction. In this example, the partial area has a form in which the edge binary image 102 is divided into n equal parts in the vertical direction. Then, the edge component extraction means 2 takes the sum of the pixel values on the same horizontal line for each of the partial areas numbered 1 to n to obtain a projection value.

In step ST24 of FIG. 4, the edge component extraction means 2 binarizes the projection value and creates projection binary data composed of elements of “0” or “1”. The binarization method is the same as the binarization of the edge strength. In step ST <b> 25, the edge component extraction unit 2 registers each continuous portion (connected region) of “1” from the projection binary data as one edge component.
FIG. 7 is a diagram showing an example in which edge components are extracted from one partial area of the edge binary image. The horizontal projection value shown in FIG. 7B is obtained from the partial area shown in FIG. As a result of binarizing this projection value, four edge components shown in FIG. 7C are extracted.

In step ST13 of FIG. 2, the motion amount extraction means 3 associates the edge component extracted in the past with the newly extracted edge component, and extracts the motion amount for each edge component. For example, the motion amount extraction unit 3 compares the edge component of the previous input image with the edge component of the latest input image 101, and when a predetermined condition is satisfied, the edge component is the same. The coordinate difference is extracted as a motion amount. For example, the same condition is assumed to satisfy all of the following three conditions.
(Condition 1) Belonging to partial areas with the same number.
(Condition 2) The length should be the same value.
(Condition 3) The distances to other edge components adjacent to the upper side and the lower side should be the same value.

  FIG. 8 is a diagram illustrating an example of association when extracting the motion amount of the edge component. Here, reference numerals 111 to 115 denote edge components, and 116 denotes the amount of movement. Further, 111 to 113 are edge components extracted from the previous input image, 114 to 115 are edge components newly extracted from the latest input image 101, and the numbers of the partial areas to which they belong are the same. As a result of the association, since the edge component 112 and the edge component 115 satisfy all the above three conditions, they are regarded as the edge components of the same object, and the coordinate difference 116 between these lower ends is extracted as the motion amount.

The above condition 1 is based on the premise that the vehicle moves in a substantially vertical direction, but this is changed based on information on the traveling direction of the vehicle set in advance, such as when there is a possibility of oblique movement. May be. For example, assuming that the partial region number of the edge component extracted in the past is i, and the partial region number of the edge component in the latest input image 101 is j, if there is a shift of partial regions up to one left and right, j is Corresponding to somewhere in the range of i-1 to i + 1, if the following equation (2) is used instead of condition 1, it is possible to cope with skew of one partial region.
i−1 ≦ j and j ≦ i + 1 (2)

Further, when the vehicle always moves from the upper left to the lower right of the image, the following equation (3) is used instead of condition 1.
i + α ≦ j and j ≦ i + β (α and β are constants, 0 ≦ α <β) (3)
Furthermore, the comparison result of the Y coordinate can also be added to the condition 1. For example, if the vehicle moves from the top to the bottom of the image, assuming that the lower end Y coordinate of the edge component extracted in the past is m and the lower end Y coordinate of the edge component in the latest input image 101 is n, the following equation (4) ) Is added to condition 1.
m + γ ≦ n (γ is a constant for adjustment, 0 ≦ γ) (4)
Thus, if condition 1 is appropriately determined, efficient association becomes possible. Further, the contents of condition 2 and condition 3 may be changed so that the association can be appropriately performed.

In step ST14 of FIG. 2, the motion amount frequency distribution creating unit 4 creates a frequency distribution of the motion amount extracted by the motion amount extracting unit 3.
FIG. 9 is a diagram for explaining an example of a motion amount frequency distribution and an example of a method of extracting a high frequency motion amount. FIG. 9A is a diagram showing an example of the frequency distribution of the motion amount created by the motion amount frequency distribution creating means 4, and the number of edge components corresponding to the motion amount on the vertical axis is shown as the frequency on the horizontal axis. Yes. In FIG. 9A, a lump at a position with a small amount of motion corresponds to a non-moving edge component on the road surface or the like, while a lump spreading slightly at a position with a large amount of motion represents an edge binary image 102. The edge component has been divided due to blurring, etc., and it has been extracted as if there were movements that did not actually occur. Since the area occupied by the vehicle in the input image 101 is large, it can be considered that the largest lump corresponds to the actual movement of the vehicle in the frequency distribution of the movement amount shown in FIG.

In step ST15 of FIG. 2, the high-frequency motion amount extraction means 5 calculates the sum of frequencies in a motion amount range with a predetermined width, and sets a range in which this value is maximum as the high-frequency motion amount.
FIG. 9B is a diagram for explaining an example of a method for extracting a high-frequency motion amount. In FIG. 9B, 121 indicates a motion amount tolerance, and 122 and 123 indicate the maximum frequency sum. The amount of movement with a range between is shown. That is, in FIG. 9B, when the sum of the range of the motion amount 122 to the motion amount 123 is maximized as a result of obtaining the partial sum of the frequency distribution with the predetermined motion amount tolerance 121. The movement amounts 122 and 123 are set as the minimum and maximum values of the high-frequency movement amount. However, when the maximum value of the partial sum of the frequency distributions is less than the predetermined threshold, it is regarded as “no vehicle” in the input image 101 and the subsequent processing is not performed.

  In step ST16 of FIG. 2, the edge component selection unit 6 selects an edge component whose motion amount is included in the range of the high-frequency motion amount from the edge components whose motion amount is extracted in step ST13. Since the edge component selected here has a movement amount substantially equal to the movement of the vehicle, the possibility of being part of the vehicle body is extremely high.

  In step ST17, the vehicle position determination means 7 determines the presence or absence of the vehicle from the position coordinates of the edge component selected by the edge component selection means 6 and the frequency distribution of the edge component at this position coordinate, and determines the position of the vehicle. In the first embodiment, the position of the vehicle head is detected, the position where edge components are concentrated in a narrow range of vertical coordinates is detected, and this is set as the vehicle position. In this case, for example, as in steps ST14 to ST15, it is possible to create a vertical coordinate frequency distribution and detect the position of the largest block.

  FIG. 10 is a diagram for explaining a method of determining the vehicle position from the vertical coordinate frequency distribution of the selected edge component. 131 indicates the vertical coordinate tolerance, and 132 and 133 have the maximum frequency sum. It shows the vertical coordinates with a range in between. In FIG. 10, the vehicle position determination means 7 obtains a partial sum of frequency distributions with a predetermined vertical coordinate tolerance 131, for example, the sum of the range of vertical coordinates 132 to vertical coordinates 133 is maximized. In this case, the vertical coordinate 133 closest to the lower end of the image is set as the vehicle position. However, the vehicle position determination means 7 determines that “there is a vehicle” when the maximum value of the partial sum of the frequency distributions is equal to or greater than a predetermined threshold value. It is determined that there is no vehicle inside.

  As described above, according to the first embodiment, the edge component extracting unit 2 extracts an edge component from the input image, and the motion amount extracting unit 3 extracts the edge component extracted in the past and the newly extracted edge component. Is extracted from the frequency distribution for each edge component, the frequency distribution of the motion amount is created by the motion amount frequency distribution creation means 4, and the frequency distribution of the high frequency motion amount extraction means 5 is input from the created frequency distribution The amount of motion that occurs frequently in the image is extracted as the high-frequency motion amount, and the high-frequency motion amount extracting unit 5 extracts the high-level motion amount extracting unit 5 from the edge components from which the motion amount is extracted. An edge component having a motion amount included in the range of the frequency motion amount is selected, and the vehicle position determination means 7 determines the presence / absence of the vehicle from the position coordinate of the selected edge component and the frequency distribution of the edge component at this position coordinate. The position of If there are many edge components with slightly different velocities, or if the edge components are unstable or appear to be unstable due to splitting or combining, the possibility of being part of the vehicle is extremely high. A high edge component can be extracted, and the effect that the position of the vehicle can be detected stably and correctly is obtained.

  In the first embodiment, the absolute value obtained by differentiating the image in the horizontal direction is used as the edge strength. For example, the absolute value obtained by differentiating in the vertical direction is used as the edge strength, or both the horizontal value and the vertical value are used. The edge strength may be obtained from the above, or a calculation formula based on another edge detection method may be applied. Moreover, although the range of the high-frequency motion amount is set to a predetermined width (fixed width), the width may be changed according to the frequency distribution of the motion amount. For example, if the frequency is concentrated in a narrow range, the width may be narrow, and if the frequency is scattered widely, the width may be wide.

  In the first embodiment, a method of binarizing with a fixed threshold is used for binarizing the edge strength and the projection value. However, other binarization methods may be used for this. . Further, a method of setting an intermediate range between two threshold values to ‘1’ instead of a method of setting a large value equal to or greater than the threshold value to ‘1’ may be used.

  Further, in the first embodiment, the edge component is extracted from the projection value in the partial region. However, another method is used in which a connected region of “1” is extracted from the differential binary image and each is used as an edge component. It may be used.

Embodiment 2. FIG.
FIG. 11 is a block diagram showing a configuration of a vehicle detection device according to Embodiment 2 of the present invention. In FIG. 11, the edge component extraction unit 2 combines and deletes the projection binary data when obtaining the projection binary data from the input image and extracting the edge component. Further, the edge component classification means 8 classifies the edge components from the position coordinates of the edge components that are combined or deleted by the edge component extraction means 2, and the motion amount frequency distribution creation means 4 weights according to the classified edge components. The edge component selection means 6 is extracted by the high frequency motion amount extraction means 5 from the edge components from which the motion amount is extracted by the motion amount extraction means 3. An edge component having a motion amount included in the range of the high-frequency motion amount is selected based on the classification result by the edge component classification means 8. Further, the vehicle passing situation estimation means 9 estimates the passing situation of the vehicle by referring to the history information of the high-frequency motion amount extracted by the high-frequency motion quantity extraction means 5, and the estimated vehicle passing situation and the vehicle position determining means. The final vehicle detection result is determined according to the presence / absence of the vehicle determined by 7. Other components are the same as those in FIG. 1 of the first embodiment.

Next, the operation will be described.
FIG. 12 is a flowchart showing a process flow of the vehicle detection apparatus according to the second embodiment of the present invention. In step ST31, the image input means 1 inputs an image, and in step ST32, the edge component extraction means 2 extracts an edge component from the input image 101.
The flowchart showing the detailed processing flow of edge component extraction is the same as that shown in FIG. 4 of the first embodiment, and the processing content is the same as the processing content from steps ST21 to ST24 in FIG. In the second embodiment, the processing content of step ST25 is different. In step ST25, the edge component extraction means 2 combines and deletes the projection binary data, and registers the result as an edge component. The process of combining and erasing the projection binary data is intended to suppress the influence of fragmentation due to blurring of edge components due to brightness fluctuations, noise, and the like.

  FIG. 13 is a diagram for explaining projection binary data combining / deleting processing performed at the time of edge component extraction. The projection binary data 141 of the vehicle body part is converted into the first combination determination threshold value 143 and the second combination. A state is shown in which the projection binary data 142 of the noise portion is deleted by the erasure determination threshold value 144 by combining with the determination threshold value 145. Here, it is assumed that the first combination determination threshold value 143 <the second combination determination threshold value 145.

  The edge component extraction unit 2 applies the first combination determination threshold value 143 to the projection binary data in FIG. 13A, and determines the distance between the upper and lower adjacent connected regions (from the lower end position of the upper connected region). When the length (up to the upper end position of the lower connection area) is equal to or less than the threshold value, the connection areas are combined into one connection area. This result is shown in FIG. 13B, in which fine divisions in the projected binary data 141 of the vehicle body portion are combined.

  Next, the edge component extraction unit 2 applies the erasure determination threshold value 144 to this FIG. 13B, and erases the connected region whose length is equal to or less than the threshold value. This result is shown in FIG. 13C, and the connected region of the projection binary data 142 of the noise portion is deleted. Finally, the edge component extraction unit 2 applies the second combination determination threshold value 145 to FIG. 13C, and when the distance between the connected regions adjacent in the vertical direction is equal to or less than the threshold value, The connected areas are combined into one connected area. This result is shown in FIG. 13 (d). While the division of the projection binary data 141 of the vehicle body portion is eliminated, the connection binary area is completely eliminated in the projection binary data 142 of the noise portion, and the remaining one A connected area is registered as an edge component.

  Here, if the projection binary data in FIG. 13A is erased as it is with the threshold value 144 for erasure determination, the central portion of the projection binary data 141 of the vehicle body part is lost, or the second combination as it is. When combining with the threshold value for determination 145, the projection binary data 142 of the noise portion remains as one large connected area, but as described above, the threshold value is switched (in the above example, the battle value is gradually increased). However, by repeating the coupling and erasing, it becomes possible to eliminate the division caused by the blur and to remove the noise.

In step ST33, the edge component classification unit 8 classifies the edge component from the position coordinates of the edge component extracted by the edge component extraction unit 2. In the second embodiment, the edge components are classified into the following A, B, C, and D.
A: Highly likely to be in front of the vehicle.
B: The movement of the lower end position is highly likely to reflect the movement of the vehicle.
C: The movement of the upper end position is highly likely to reflect the movement of the vehicle.
D: Other than the above.

In this classification, a condition based on the position coordinates of the edge component, that is, the length / position of the edge component and the distance to the adjacent edge component is applied, and the classification is performed as follows, for example.
(Classification 1) If the length of the edge component is within a predetermined range and there is no other edge component within the predetermined range on the upper side or the lower side, it is classified as A.
(Classification 2) If the length of the edge component is not less than a predetermined value and there is no other edge component in the lower predetermined range, it is classified as B.
(Category 3) If the length of the edge component is not less than a predetermined value and there is no other edge component in the upper predetermined range, it is classified as C.
(Category 4) If each condition of classification 1 to classification 3 is not satisfied, it is classified as D.

  FIG. 14 is a diagram showing an example of edge components obtained from an input image. Edge components 151 and 152 corresponding to the vehicle body side surface, edge components 153 and 154 corresponding to the vehicle body front surface, and edge components 155 to 158 corresponding to the road surface are shown. Show. In the example of FIG. 14, the edge components 153 and 154 correspond to A, the edge components 151 and 152 correspond to B, the edge component 155 corresponds to C, and the other edge components are D. It becomes.

  In step ST34 of FIG. 12, the motion amount extraction unit 3 extracts the motion amount for each edge component by associating the previously extracted edge component with the newly extracted edge component. In the second embodiment, only the edge components classified into any one of A, B, and C are associated with each other, and the conditions 1, 2, and 3 used in the first embodiment are satisfied. In addition, the edge components that are classified the same are regarded as the same, and the coordinate difference is set as the amount of movement. Here, as for the reference position for taking the coordinate difference, those classified into A and B are the lower end positions, and those classified into C are the upper end positions. For example, although the edge component 155 of FIG. 14 classified as C is an edge component of the road surface, the change in the upper end position indicates the movement of the vehicle, so the change amount is used.

  In step ST35, the motion amount frequency distribution creating means 4 creates a motion amount frequency distribution. In the second embodiment, the value added as the frequency in order to increase the vehicle detection accuracy according to the classification result in step ST33. Assume that (weight) is changed for each edge component. For example, A which is highly likely to be a vehicle adds 2 for one edge component, and B and C add 1. In step ST36, the high frequency motion amount extraction means 5 extracts the high frequency motion amount from the frequency distribution in the same procedure as in the first embodiment. In step ST37, the edge component selection means 6 selects an edge component for vehicle position estimation. In the second embodiment, the motion amount is included in the high-frequency motion amount range and is classified as A. Select.

  In step ST38, the vehicle position determination means 7 determines the vehicle position by determining the presence or absence of the vehicle in the same procedure as in the first embodiment. In step ST39, the vehicle passage state estimation means 9 extracts the high-frequency motion amount. The passing situation of the vehicle is estimated with reference to the history information of the high-frequency motion amount extracted by the means 5. As a method of estimating the vehicle passing situation, for example, referring to the history information of the high-frequency movement amount (value in the past constant cycle number), when the same high-frequency movement amount continues, “passing the vehicle” If it does not continue, “no vehicle passing” is adopted.

In step ST40, the vehicle passage situation estimation means 9 determines the final vehicle detection result from the presence / absence of the vehicle determined in step ST38 and the vehicle passage situation estimated in step ST39. For example, the vehicle passage state estimation means 9 outputs the vehicle detection result as “vehicle appearance” with the vehicle position when either of the following conditions a or b is satisfied, and when neither of the conditions is satisfied: If the state of “passing through the vehicle” continues, it is determined that “the vehicle is present”, and if not, it is determined that “the vehicle is not present”.
(Condition a) When the vehicle position determination means 7 determines that “there is a vehicle” for the first time in a situation where “passing through the vehicle” continues based on the history information of the high-frequency motion amount.
(Condition b) A case where the vehicle position determination means 7 determines that “there is a vehicle” in the situation of “no vehicle passing” based on the history information of the high-frequency motion amount.

  In this way, “there is a vehicle” can occur a plurality of times during the passage of one vehicle, but “the appearance of the vehicle head” occurs only once. Only the determination result of the vehicle position determination means 7 may detect a truck bed or the like as a separate vehicle, and may double count the number of passing vehicles. Since the final determination is made in consideration of the result, even if a vehicle with a long vehicle length passes, it can be prevented from erroneously detecting multiple times.

  As described above, according to the second embodiment, the edge component extraction unit 2 extracts the edge component from the input image, the edge component classification unit 8 classifies the edge component from the position coordinates of the edge component, and extracts the motion amount. The means 3 extracts the motion amount for each edge component by associating the edge component extracted in the past with the newly extracted edge component, and the motion amount frequency distribution creating means 4 weights according to the classified edge component. The frequency distribution of the extracted motion amount is generated by changing the frequency, and the high-frequency motion amount extraction means 5 extracts the high-frequency motion amount that is frequently generated in the input image from the generated frequency distribution. The edge component having the motion amount included in the range of the high frequency motion amount extracted by the high frequency motion amount extraction unit 5 from the edge components from which the motion amount extraction unit 3 has extracted the motion amount by the edge component selection unit 6 The edge Selection is made based on the classification result by the classification unit 8, and the vehicle position determination unit 7 determines the presence or absence of the vehicle from the position coordinates of the selected edge component and the frequency distribution of the edge component at this position coordinate, and determines the position of the vehicle. As a result, even when a large number of edge components with slightly different speeds occur, or even when the edge components are split or combined and appear unstable, the edge is very likely to be part of the vehicle body. The components can be extracted, and the effect that the position of the vehicle can be detected stably and correctly is obtained.

  Further, according to the second embodiment, the edge component extraction unit 2 combines and deletes the projection binary data, and extracts the result as an edge component, thereby fading the edge component due to brightness fluctuations and the like. The effect of being able to suppress the influence of the division and noise due to is obtained.

  Furthermore, according to the second embodiment, the movement amount of the vehicle can be stably extracted by switching the reference position (upper end / lower end) and extracting the movement amount of the edge component based on the classification result of the edge component. The effect is obtained.

  Further, according to the second embodiment, the vehicle passage state estimation means 9 determines the vehicle detection result in consideration of the passage state of the vehicle, so that it depends on the type of the vehicle that passes, including a vehicle with a long vehicle length. The effect that a vehicle can be detected stably without being acquired.

  In the second embodiment, the procedure for combining and erasing projection binary data is made up of three steps of combining → erasing → combining, but this may be repeated more frequently. In addition, the edge components are classified into A, B, C, and D, but this may be classified into five or more types under more detailed conditions, or A, B, D, A, C, Three types of D may be used.

  Furthermore, although the weight in creating the frequency distribution is determined from the edge component classification result, the weight may be changed using other conditions. For example, the weight at the center of the image may be increased, the weight at the end may be decreased, and the weight may be changed according to the amount of motion.

  Furthermore, as a method of estimating the vehicle passage situation, a method of referring to the history information of the high-frequency motion amount is used, but this may be another method. For example, a method that considers that a vehicle is passing the vehicle within a predetermined time immediately after it is determined that the vehicle has appeared, or a vehicle's estimated passing time is determined from the amount of high-frequency movement at the time when it is determined that the vehicle has appeared. There is also a method to consider. Furthermore, as the vehicle detection result, three types of output “car appearance”, “with vehicle” and “without vehicle” are defined, but this may be further refined and the types may be increased, and the judgment conditions may be changed. good.

Embodiment 3 FIG.
FIG. 15 is a block diagram showing a configuration of a vehicle detection device according to Embodiment 3 of the present invention. In the figure, the license plate detection means 10 detects the license plate using the edge component extracted by the edge component extraction means 2, and the license plate recognition means 11 recognizes the characters in the license plate. Other components are the same as those in FIG. 1 of the first embodiment.

Next, the operation will be described.
FIG. 16 is a flowchart showing a process flow of the vehicle detection apparatus according to the third embodiment of the present invention. In steps ST51 to ST57, vehicle detection is performed in the same procedure as steps ST11 to ST17 in FIG. 2 of the first embodiment.

  In step ST58, the license plate detecting means 10 detects the license plate using the edge component extracted by the edge component extracting means 2 in step ST52. As a method of detecting this license plate, for example, the method described in IEICE Technical Report PRMU 96-46 “Development of License Plate Recognition Device” (hereinafter simply referred to as literature) is used. In this case, as described in section 2.1 “Plate cutout section” of the above document, the projection value of the image differentiated in the horizontal direction, that is, the edge component in the first embodiment is used, and the license plate A pair indicating an upper part and a lower part is detected, and using the input image 101 input in step ST51, the processing of Section 2.2 “Plate binarization unit” in the above document is performed. As described above, since the edge component can be used for both vehicle detection and license plate detection by using the edge component also for cutting out the license plate region, the overall configuration is very efficient. When a plurality of license plate areas are detected in step ST58, the license plate detection means 10 selects the one closer to the vehicle position determined by the vehicle position determination means 7 in step ST57.

  In step ST59, the license plate recognition means 11 recognizes the characters in the detected license plate. As the recognition method, for example, the method described in section 2.3 “Character recognition” in the above-mentioned document is used.

  As described above, according to the third embodiment, the same effect as that of the first embodiment can be obtained, and the license plate detection means 10 can detect the license plate using the edge component for detecting the vehicle. When the license plate recognition means 11 recognizes the characters in the detected license plate, the vehicle detection and the license plate detection can be performed efficiently.

  Further, according to the third embodiment, the license plate detection means 10 uses the information on the vehicle position determined by the vehicle position determination means 7 also for the license plate detection, thereby increasing the detection accuracy of the license plate. The effect that it can be obtained.

  In the third embodiment, the method described in the literature is used as the license plate detection method, but another method may be used as long as it uses edge information. In addition, the license plate recognition method may be different from those described in the literature. In addition, when a plurality of license plate areas are detected, the selection is made according to the vehicle position. For example, the license plate detection is performed from the beginning to limit the target area of the license plate detection to the periphery of the vehicle position. Another form may be used.

Embodiment 4 FIG.
FIG. 17 is a block diagram showing a configuration of a vehicle detection device according to Embodiment 4 of the present invention. In the figure, the license plate detection means 10 detects the license plate using the edge component extracted by the edge component extraction means 2 when the vehicle passage state estimation means 9 determines “car appearance”. Plate recognition means 11 recognizes characters in the license plate. Other components are the same as those in FIG. 11 of the second embodiment.

Next, the operation will be described.
FIG. 18 is a flowchart showing a process flow of the vehicle detection apparatus according to Embodiment 4 of the present invention. In steps ST61 to ST70, vehicle detection is performed in the same procedure as steps ST31 to ST40 in FIG. 12 of the second embodiment.

  In step ST71, the license plate detection means 10 checks the vehicle detection result by the vehicle passage situation estimation means 9 in step ST70, and in the case of “car appearance”, in step ST72, the license plate detection means 10 is the same as that of the third embodiment. Similarly, the license plate is detected. In vehicles such as trucks where the body is long and characters are displayed on the loading platform, a part of the loading platform may be mistakenly detected as a license plate, but only in the case of `` car head appearance '' By detecting the license plate, it is possible to prevent erroneous detection of areas other than the license plate, such as characters displayed on the vehicle body.

  In step ST73, the detection result of the license plate is checked by a control means (not shown). If the license plate is detected, the license plate recognition means 11 is detected in step ST74 as in the third embodiment. If the license plate is recognized, but the license plate is not detected, it is assumed that the detected vehicle does not have a license plate or is in a hidden position, and through a warning device (not shown) A notification that the license plate cannot be detected is issued, or an image storage means (not shown) stores the detected vehicle image.

  As described above, according to the fourth embodiment, the same effects as those of the second embodiment can be obtained, and the edge component for detecting the vehicle when the license plate detecting means 10 is “appearance of the vehicle head” is used. The license plate is detected, and the license plate recognition means 11 recognizes the characters in the detected license plate, so that vehicle detection and license plate detection can be efficiently performed. It is possible to prevent the erroneous detection.

  In addition, according to the fourth embodiment, when a license plate cannot be detected, a warning device is used to notify the vehicle and store an image of the vehicle to which a license plate cannot be detected. The effect that it can be performed is acquired.

  In the fourth embodiment, in step ST75 of FIG. 18, the warning vehicle and the image storage are performed for the vehicle to which the license plate cannot be detected, but this is different from that performed in an unsteady state. Processing may be performed. For example, the phenomenon that the license plate cannot be detected even if the vehicle can be detected can occur even if the imaging parameters such as the aperture value of the camera deviate from the desired range, so the number of times that the vehicle can be detected and the license plate cannot be detected is counted. However, it is conceivable to perform automatic adjustment processing of imaging parameters if a predetermined number or more occur continuously.

It is a block diagram which shows the structure of the vehicle detection apparatus by Embodiment 1 of this invention. It is a flowchart which shows the flow of a process of the vehicle detection apparatus by Embodiment 1 of this invention. It is a figure which shows the example of the input image of the vehicle detection apparatus by Embodiment 1 of this invention. It is a flowchart which shows the flow of the detailed process of edge component extraction of the vehicle detection apparatus by Embodiment 1 of this invention. It is a figure which shows the example of the edge binary image in the vehicle detection apparatus by Embodiment 1 of this invention. It is a figure which shows the area | region division example of the edge binary image at the time of edge component extraction in the vehicle detection apparatus by Embodiment 1 of this invention. It is a figure which shows the example which extracts an edge component from one partial area | region of the edge binary image in the vehicle detection apparatus by Embodiment 1 of this invention. It is a figure which shows the example of matching at the time of extracting the motion amount of the edge component in the vehicle detection apparatus by Embodiment 1 of this invention. It is a figure for demonstrating the example of the frequency distribution example of the motion amount in the vehicle detection apparatus by Embodiment 1 of this invention, and the example of the extraction method of a high frequency motion amount. It is a figure for demonstrating the method to determine a vehicle position from the frequency distribution of the vertical coordinate of the selected edge component in the vehicle detection apparatus by Embodiment 1 of this invention. It is a block diagram which shows the structure of the vehicle detection apparatus by Embodiment 2 of this invention. It is a flowchart which shows the flow of a process of the vehicle detection apparatus by Embodiment 2 of this invention. It is a figure for demonstrating the combination / erasure | elimination process of the projection binary data performed in the case of edge component extraction in the vehicle detection apparatus by Embodiment 2 of this invention. It is a figure which shows the example of the edge component obtained from the input image in the vehicle detection apparatus by Embodiment 2 of this invention. It is a block diagram which shows the structure of the vehicle detection apparatus by Embodiment 3 of this invention. It is a flowchart which shows the flow of a process of the vehicle detection apparatus by Embodiment 3 of this invention. It is a block diagram which shows the structure of the vehicle detection apparatus by Embodiment 4 of this invention. It is a flowchart which shows the flow of a process of the vehicle detection apparatus by Embodiment 4 of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Image input means, 2 Edge component extraction means, 3 Motion amount extraction means, 4 Motion amount frequency distribution creation means, 5 High frequency motion amount extraction means, 6 Edge component selection means, 7 Vehicle position determination means, 8 Edge component classification means , 9 Vehicle passage state estimation means, 10 license plate detection means, 11 license plate recognition means, 101 input image, 102 edge binary image, 111, 112, 113, 114, 115 edge component, 116 motion amount, 121 motion amount Tolerance, 122,123 Movement amount sandwiching a range where the frequency sum is maximum, 131 Vertical coordinate tolerance, 132,133 Vertical coordinate sandwiching a range where the frequency sum is maximum, 141 Projection of body part 2 Value data, 142 Projection binary data of noise part, 143 First combination determination threshold value, 144 Erasing determination threshold value, 145 2 of the coupling determination 闘値, edge components corresponding to 151, 152 vehicle body side, the edge component corresponding to the vehicle body front 153 and 154, the edge component corresponding to 155,156,157,158 road.

Claims (8)

An image input means for inputting an image;
Edge component extraction means for extracting edge components from the input image;
A motion amount extraction means for extracting a motion amount for each edge component by associating the edge component extracted in the past with the newly extracted edge component;
A motion amount frequency distribution creating means for creating a frequency distribution of the extracted motion amount;
A high-frequency motion amount extraction means for extracting, as a high-frequency motion amount, a motion amount that occurs frequently in the input image from the generated motion amount frequency distribution;
Edge component selection means for selecting an edge component having a motion amount included in the range of the high frequency motion amount extracted by the high frequency motion amount extraction means from among the edge components from which the motion amount has been extracted by the motion amount extraction means. When,
Vehicle position determination means for determining the presence or absence of a vehicle from the position coordinates of the selected edge component and the frequency distribution of the edge component at the position coordinates and determining the position of the vehicle ;
License plate detection means for detecting a license plate in the input image using the edge component extracted by the edge component extraction means;
A license plate recognition means for recognizing characters in the detected license plate;
The edge component extraction means calculates edge strength from an input image, binarizes the edge strength to create an edge binary image, and generates projected binary data for each partial region obtained by dividing the edge binary image. A vehicle detection device that extracts a connected region as one edge component from a vehicle. 2. The vehicle detection apparatus according to claim 1, wherein the edge component extracting means extracts the edge component by repeatedly combining and erasing the projected binary data based on a predetermined threshold . Motion amount extraction means vehicle detection apparatus according to claim 1, characterized in that the correspondence of an edge component based on the traveling direction information of the vehicle which is set in advance. 2. The vehicle detection according to claim 1, wherein the motion amount frequency distribution creating means creates the frequency distribution of the motion amount extracted by the motion amount extracting means by changing the weight according to the position of the edge component and the motion amount. apparatus. Edge component classification means for classifying the edge component from the position coordinates of the edge component extracted by the edge component extraction means;
The motion amount extraction means selects an edge component to be subjected to motion amount extraction based on the classification result of the edge component classification means,
The edge component selection means selects an edge component having a motion amount included in the range of the high-frequency motion amount extracted by the high-frequency motion amount extraction means from the edge components from which the motion amount has been extracted by the motion amount extraction means. 2. The vehicle detection device according to claim 1, wherein the selection is made based on a classification result by the component classification means . The vehicle passage state is estimated by referring to the history information of the high-frequency motion amount extracted by the high-frequency motion amount extraction unit, and the vehicle is detected based on the estimated vehicle passage state and the presence / absence of the vehicle determined by the vehicle position determination unit. The vehicle detection apparatus according to claim 1, further comprising vehicle passage state estimation means for determining a result . License plate detection means for detecting the license plate in the input image based on the vehicle detection result by the vehicle passage state estimation means using the edge component extracted by the edge component extraction means;
7. The vehicle detection apparatus according to claim 6, further comprising license plate recognition means for recognizing characters in the detected license plate . 8. The vehicle detection device according to claim 7 , wherein when the license plate cannot be detected by the license plate detection means, the fact that the license plate cannot be detected is reported to the outside, or the detected vehicle image is stored .

Images