JP2013205410A - Outside recognition device for vehicle, and vehicle system using the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To stably identify a vehicle regardless of a state of a captured image.SOLUTION: An HOG (Histogram of Oriented Gradients) (first) feature amount calculation unit 310 calculates a first feature amount from an image captured by a camera 100 (an imaging unit) on the basis of a concentration gradient. An HOF (Histogram of Optical Flow) (second) feature amount calculation unit 320 calculates a second feature amount on the basis of an optical flow. A reliability calculation unit 410 calculates first reliability Rthat is higher as the concentration gradient is greater, and calculates second reliability Rthat is higher as the optical flow is greater. A first vehicle identification unit 420 identifies the vehicle on the basis of the first feature amount, and a second vehicle identification unit 430 identifies the vehicle on the basis of the second feature amount. A vehicle determination unit 400 determines presence/absence of the vehicle on the basis of the identification result of the first vehicle identification unit 420, the identification result of the second vehicle identification unit 430, the first reliability R, and the second reliability R.

Description

  The present invention relates to a vehicle external recognition device for identifying a vehicle existing around a host vehicle and a vehicle system using the same.

  In recent years, aiming to reduce the number of casualties due to traffic accidents, a collision warning system that monitors the front of the vehicle and notifies when there is a danger of a collision, and driving that maintains a distance between vehicles within a set speed range An ACC (Adaptive Cruise Control) is being developed.

  Such a system requires a function of identifying a vehicle existing around the host vehicle and specifying its position. As an example, for example, a vertical edge extending along the vertical direction of the image, which is highly likely to represent the left and right ends of the vehicle, is detected from an image in front of the own vehicle taken by a monocular camera mounted on the own vehicle. Thereafter, a technique for extracting a candidate area of a vehicle by voting an interval w between vertical edges and a center position x between the vertical edges in an xw space has been proposed (Patent Document 1).

JP 2005-156199 A

  However, according to the vehicle detection device described in Patent Document 1, it is difficult to stably identify the vehicle under various conditions. This is because the image captured by the camera has a combination of various conditions such as the weather, time zone, solar radiation state, vehicle color, and the like, so that the vertical edge may not be detected depending on the conditions. Note that the identification target is not limited to the vehicle, and pedestrians and obstacles on the road that may affect the traveling of the host vehicle may also be the identification target. There are challenges.

  The present invention has been made in view of the above circumstances, and provides an external environment recognition device for a vehicle that can stably identify a vehicle regardless of the state of a captured image, and a vehicle system using the same. With the goal.

  The vehicle external environment recognition apparatus and the vehicle system using the same according to the present invention identify a vehicle by combining a plurality of feature amounts and a discriminator from captured images around the vehicle.

That is, the external environment recognition device for a vehicle according to claim 1 of the present invention is mounted on the own vehicle and has an imaging unit that images the periphery of the own vehicle and an image obtained by the imaging unit with a concentration gradient. A first feature amount calculation unit that calculates a first feature amount based on the second feature amount calculation unit that calculates a second feature amount based on an optical flow from a plurality of images captured at different times by the imaging unit. And a first reliability that is higher as the concentration gradient is larger, and a second reliability that is higher as the optical flow is larger, and a vehicle is identified based on the first feature amount. A first vehicle identification unit; a second vehicle identification unit that identifies a vehicle based on the second feature value; an identification result in the first vehicle identification unit; an identification result in the second vehicle identification unit; 1st reliability and said 2nd reliability A vehicle determination unit determines the presence of the vehicle on the basis of,
It is characterized by having.

  According to the vehicle external environment recognizing device according to claim 1 of the present invention configured as described above, the first feature amount calculation unit is based on the density gradient from the surrounding images of the host vehicle captured by the imaging unit. The first feature amount is calculated, the second feature amount calculation unit calculates a second feature amount based on the optical flow from a plurality of images captured at different times by the imaging unit, and the reliability calculation unit The higher the concentration gradient, the higher the first reliability, and the higher the optical flow, the higher the second reliability, and the first vehicle identification unit identifies the vehicle based on the first feature amount. The second vehicle identifying unit identifies the vehicle based on the second feature amount, and the vehicle determining unit identifies the identification result in the first vehicle identifying unit, the identification result in the second vehicle identifying unit, and the first reliability. And the presence / absence of the vehicle is determined based on the second reliability. For an image having a large density gradient in the captured image, that is, an image with a clear edge, the first vehicle identification unit identifies the vehicle based on the first feature amount based on the density gradient, and is also captured. Since the second vehicle discriminating unit identifies the vehicle based on the second feature quantity based on the optical flow for an image having a large optical flow, that is, a clear movement in the image, a single classifier Therefore, it is possible to provide a vehicle external environment recognition device that can identify a vehicle in a wide range of scenes rather than identifying the vehicle.

  Furthermore, the vehicle external environment recognition device according to claim 3 of the present invention is mounted on the host vehicle, and an image capturing unit that captures an image of the periphery of the host vehicle, and a density gradient from among images captured by the image capturing unit. A first feature amount calculation unit that calculates a first feature amount based on the second feature amount calculation unit that calculates a second feature amount based on an optical flow from a plurality of images captured at different times by the imaging unit. And a first vehicle identification unit that identifies a vehicle based on the first feature amount, and a second vehicle identification unit that identifies a vehicle based on the second feature amount, the first feature When the amount is larger than a predetermined value, the presence / absence of a vehicle is determined based on the identification result in the first vehicle identification unit, and when the first feature amount is smaller than the predetermined value, the second vehicle identification unit It is characterized by determining the presence or absence of a vehicle based on the identification result That.

  According to the vehicle external environment recognizing device according to claim 3 of the present invention configured as described above, the first feature amount calculation unit is based on the density gradient from the surrounding images of the host vehicle captured by the imaging unit. When the first feature amount is calculated and the calculated first feature amount is larger than a predetermined value, the first vehicle identification unit identifies the vehicle based on the first feature amount, and based on the identification result When the presence or absence of the vehicle is determined and the calculated first feature value is smaller than a predetermined value, the second feature value calculation unit calculates the second feature value based on the optical flow and calculates the calculated second feature value. When the amount is larger than the predetermined value, the second vehicle identification unit identifies the vehicle based on the second feature amount, and the presence or absence of the vehicle is determined based on the identification result. At least one feature amount of the second feature amount By calculating only, since it is possible to determine the presence or absence of a vehicle, it is possible to provide the environment recognizing apparatus for a vehicle processing load is small.

  According to a fourth aspect of the present invention, there is provided an external environment recognition device for a vehicle, which is mounted on the host vehicle and has an imaging unit that captures an image of the periphery of the host vehicle and a density gradient among images captured by the imaging unit. A first feature amount calculation unit that calculates a first feature amount based on the second feature amount calculation unit that calculates a second feature amount based on an optical flow from a plurality of images captured at different times by the imaging unit. A first vehicle identification unit that identifies a vehicle based on the first feature amount, and a second vehicle identification unit that identifies a vehicle based on the second feature amount, and the second feature When the amount is larger than a predetermined value, the presence / absence of a vehicle is determined based on the identification result in the second vehicle identification unit, and when the second feature amount is smaller than the predetermined value, the first vehicle identification unit It is characterized by determining the presence or absence of a vehicle based on the identification result .

  According to the vehicle external environment recognizing device according to the fourth aspect of the present invention configured as described above, the second feature amount calculation unit is based on the optical flow from the surrounding images of the host vehicle captured by the imaging unit. When the second feature value is calculated and the calculated second feature value is larger than a predetermined value, the second vehicle identification unit identifies the vehicle based on the second feature value, and based on the identification result When the presence or absence of the vehicle is determined and the calculated second feature value is smaller than a predetermined value, the first feature value calculation unit calculates the first feature value based on the density gradient, and the calculated first feature value When the amount is larger than the predetermined value, the first vehicle identification unit identifies the vehicle based on the first feature amount, and the presence or absence of the vehicle is determined based on the identification result. At least one feature amount of the second feature amount By calculating only, since it is possible to judge the presence or absence of a vehicle, it is possible to provide the environment recognizing apparatus for a vehicle processing load is small.

  Furthermore, the vehicle external environment recognition apparatus according to claim 5 of the present invention is mounted on the host vehicle, and has an imaging unit that captures an image of the periphery of the host vehicle, and a density gradient among images captured by the imaging unit. A first feature amount calculation unit that calculates a first feature amount based on the second feature amount calculation unit that calculates a second feature amount based on an optical flow from a plurality of images captured at different times by the imaging unit. And a first reliability that is higher as the concentration gradient is larger, and a second reliability that is higher as the optical flow is larger, and a vehicle is identified based on the first feature amount. A first vehicle identification unit; a second vehicle identification unit that identifies a vehicle based on the second feature value; an identification result in the first vehicle identification unit; an identification result in the second vehicle identification unit; The first reliability and the second reliability In the vehicle external environment recognition device having a vehicle determination unit that determines the presence or absence of a vehicle based on the vehicle and a vehicle behavior detection unit that detects the behavior of the host vehicle, the behavior of the host vehicle detected by the vehicle behavior detection unit Accordingly, when the first feature value is calculated by the first feature value calculation unit and the first feature value is larger than a predetermined value, the vehicle is identified by the first vehicle identification unit. When the first feature amount is smaller than a predetermined value, the second feature amount calculation unit calculates the second feature amount, and the second vehicle identification unit performs vehicle identification. When the second feature value is calculated by the second feature value calculation unit and the second feature value is greater than a predetermined value, the vehicle is determined by the second vehicle identification unit. To identify the presence of a vehicle, When the second feature value is smaller than a predetermined value, the first feature value calculation unit calculates the first feature value, and the first vehicle identification unit identifies the vehicle to determine the presence or absence of the vehicle. It is characterized in that the processing to be performed is switched.

  According to the vehicle external environment recognizing device according to claim 5 of the present invention configured as described above, according to the calculation result of the vehicle behavior calculating unit, from the surrounding images of the host vehicle imaged by the imaging unit, When the first feature value calculation unit calculates the first feature value based on the density gradient and the calculated first feature value is larger than a predetermined value, the first vehicle identification unit determines the vehicle based on the first feature value. And the presence / absence of a vehicle is determined based on the identification result. On the other hand, when the calculated first feature value is smaller than a predetermined value, the second feature value calculation unit calculates the second feature value based on the optical flow. When the calculated second feature value is larger than the predetermined value, the second vehicle identification unit identifies the vehicle based on the second feature value, and determines whether the vehicle is present based on the identification result. The process of determining and the second feature quantity calculation unit When the second feature value based on the optical flow is calculated and the calculated second feature value is larger than the predetermined value, the second vehicle identification unit identifies the vehicle based on the second feature value, and the identification result On the other hand, when the calculated second feature value is smaller than a predetermined value, the first feature value calculation unit calculates the first feature value based on the density gradient and is calculated When the first feature amount is larger than a predetermined value, the first vehicle identification unit switches between the process of identifying the vehicle based on the first feature amount and determining the presence or absence of the vehicle based on the identification result. In addition, since it is possible to determine whether or not there is a vehicle by calculating at least one of the first feature amount and the second feature amount, it is possible to provide a vehicle external environment recognition device with a small processing load. . In particular, the processing load can be further reduced by switching the feature amount to be calculated according to the calculation result of the vehicle behavior calculation unit.

  Furthermore, the vehicle system using the vehicle external environment recognition device according to claim 7 of the present invention is a vehicle system using the vehicle external environment recognition device according to any one of claims 1 to 6, wherein Based on the calculation result of the vehicle relative position calculation unit and the vehicle relative position calculation unit that calculates the positional relationship between the vehicle identified by the vehicle external environment recognition device and the host vehicle, the necessity of outputting an alarm is determined. An alarm output determination unit and an alarm output unit that outputs an alarm when it is determined that the alarm output determination unit needs to output an alarm.

  According to the vehicle system using the vehicle external environment recognition device according to the seventh aspect of the present invention configured as described above, the vehicle relative position calculation is performed based on the position of the vehicle identified with high accuracy by the vehicle external environment recognition device. The unit calculates the relative positional relationship between the host vehicle and the identified vehicle, and based on the relative positional relationship thus calculated, the warning output determination unit determines the necessity of the warning output, and the warning output determination unit However, since the alarm output unit outputs an alarm when it is determined that the alarm output is necessary, it is possible to provide a vehicle system using the vehicle external environment recognition device capable of outputting an alarm with high accuracy.

  A vehicle system using the vehicle external environment recognition device according to claim 8 of the present invention is the vehicle system using the vehicle external environment recognition device according to any one of claims 1 to 6, wherein A vehicle relative position calculation unit that calculates a positional relationship between the vehicle identified by the vehicle external environment recognition device and the host vehicle, and a vehicle that controls the behavior of the host vehicle based on the calculation result of the vehicle relative position calculation unit And a behavior control unit.

  According to the vehicle system using the vehicle external environment recognition device according to the eighth aspect of the present invention configured as described above, the vehicle relative position calculation is performed based on the position of the vehicle identified with high accuracy by the vehicle external environment recognition device. Since the vehicle calculates the relative positional relationship between the host vehicle and the identified vehicle, and the vehicle behavior control unit controls the behavior of the host vehicle based on the relative positional relationship thus calculated, a highly accurate vehicle It is possible to provide a vehicle system using a vehicle external environment recognition device capable of controlling the behavior of the vehicle.

  According to the vehicle external recognition apparatus according to the present invention, since the vehicle is identified using two types of classifiers that perform identification based on two different types of features, the identification is performed using a single classifier. In addition, it is possible to provide a vehicle external environment recognition device that can identify a vehicle in a wide range of scenes and can stably identify the vehicle regardless of the state of an image captured.

  Furthermore, according to the vehicle system using the vehicle external environment recognition device according to the present invention, it is possible to provide a vehicle system using the vehicle external environment recognition device capable of highly accurate alarm output and behavior control of the host vehicle. it can.

It is a block diagram which shows schematic structure of the collision warning apparatus using the external field recognition apparatus for vehicles which concerns on Example 1 of this invention. It is a block diagram which shows the internal structure of the feature extraction part of FIG. It is a schematic flowchart which shows the flow of the process in Example 1 of this invention. It is a schematic flowchart which shows the flow of the vehicle candidate area | region extraction process in Example 1 of this invention. It is a schematic flowchart which shows the flow of the feature calculation process in Example 1 of this invention. It is a schematic flowchart which shows the flow of the vehicle identification process in Example 1 of this invention. It is a figure explaining the operator which detects a vertical edge. (A) shows the 1st example of the image acquired in Example 1 of this invention. (B) shows the 2nd example of the image acquired in Example 1 of this invention. FIG. 8C shows an example of edge information detected from the image of FIG. FIG. 8D shows an example of edge information detected from the image of FIG. It is a block diagram which shows schematic structure of the collision warning apparatus using the external field recognition apparatus for vehicles which concerns on Example 2 of this invention. It is a block diagram which shows the internal structure of the feature extraction part of FIG. It is a schematic flowchart which shows the flow of the process in Example 2 of this invention. It is a block diagram which shows schematic structure of the collision warning apparatus using the external field recognition apparatus for vehicles which concerns on Example 3 of this invention. It is a block diagram which shows the internal structure of the feature extraction part of FIG. It is a schematic flowchart which shows the flow of the process in Example 3 of this invention.

  DESCRIPTION OF EMBODIMENTS Hereinafter, an embodiment of a vehicle system using a vehicle external environment recognition device according to the present invention will be described with reference to the drawings.

  The first embodiment relates to a collision warning device using a vehicle external environment recognition device that monitors the front of a vehicle with a vehicle-mounted camera and outputs a warning when a vehicle that may come into contact with the host vehicle is identified. Is.

  First, the configuration of the apparatus will be described with reference to FIGS. The collision warning device 50 according to the first embodiment is installed in a host vehicle 10 (not shown), and a camera 100 (imaging unit) that captures an image ahead of the host vehicle 10 and a vehicle external environment that identifies vehicles other than the host vehicle 10. When the vehicle is identified, the recognition device 60, a vehicle relative position calculation unit 70 that calculates the direction and distance from the host vehicle 10 to the identified vehicle, and an alarm output determination unit 80 that determines the possibility of a collision. And an alarm output unit 90 that outputs an alarm when the possibility of collision is a predetermined value or more.

  Here, the vehicle external environment recognition device 60 is configured by a computer having a CPU, a memory, an I / O, and the like. Specifically, the vehicle external environment recognition device 60 captures an image captured by the camera 100 (imaging unit) at a constant period. An image acquisition unit 150; a vehicle candidate region extraction unit 200 that detects a candidate region of the vehicle from the captured image; a feature amount calculation unit 300 that calculates a feature amount that represents a vehicle from the vehicle candidate region; A vehicle determination unit 400 that determines whether the candidate area represents a vehicle is provided. In addition, a predetermined process is programmed in the vehicle external environment recognition device 60, and the process is repeatedly executed at a predetermined cycle.

  The vehicle candidate region extraction unit 200 further includes a vertical edge detection unit 210 that detects the positions of the pixels that constitute the vertical edge, and a horizontal interval between the pixels that constitute the vertical edge detected by the vertical edge detection unit 210. The vertical edge interval calculation unit 220 for calculating the vertical edge interval, the vertical edge interval (w) calculated by the vertical edge interval calculation unit 220, and the center position (x) of the left and right vertical edges are voted in the xw space. As a result of the voting, an xw space generation unit 230 that detects a point (x, w) having a peak and sets it as a vehicle candidate region, and a vehicle candidate region setting unit 240 that sets the vehicle candidate region.

  In addition, the feature amount calculation unit 300 further includes a HOG feature amount calculation unit 310 (first feature amount calculation unit) that calculates a HOG (Histogram of Oriented Gradients) feature amount (first feature amount) related to the density gradient based on the histogram. And a HOF feature quantity calculation unit 320 (second feature quantity calculation unit) that calculates a HOF (Histogram of Optical Flow) feature quantity (second feature quantity) related to the optical flow based on the histogram.

  In addition, as shown in FIG. 2, the HOG feature value calculation unit 310 further includes a plurality of partial areas that are laid out so as not to overlap each other in the vehicle candidate areas set by the vehicle candidate area setting unit 240. A gradient information calculation unit 312 that calculates the magnitude and direction of the concentration gradient in the partial region thus set, and the concentration gradient using the concentration gradient magnitude and direction calculated by the gradient information calculation unit 312. A gradient histogram creation unit 314 that creates a direction histogram, and a gradient histogram created by the gradient histogram creation unit 314 calculates a HOG feature amount (first feature amount) based on a density gradient (edge feature) in the image. A density gradient feature amount calculation unit 316 is included.

  Further, as shown in FIG. 2, the HOF feature value calculation unit 320 further includes an optical flow calculation unit 322 that searches for a corresponding region between images taken at different times and calculates an optical flow, and the calculated optical flow. An optical flow histogram creating unit 324 that creates a flow histogram and a HOF feature amount (second feature amount) based on an optical flow representing a feature of motion in an image are calculated based on the magnitude and direction of the calculated optical flow. An optical flow feature quantity calculation unit 326 is included.

  The vehicle determination unit 400 further includes a reliability calculation unit 410 that calculates the reliability based on the HOG feature value and the HOF feature value, and a first vehicle identification unit 420 that identifies the vehicle based on the HOG feature value. Based on the second vehicle identification unit 430 that identifies the vehicle based on the HOF feature amount, the calculated reliability, the identification result in the first vehicle identification unit 420, and the identification result in the second vehicle identification unit 430 It consists of the determination part 440 which determines the presence or absence of a vehicle. Here, the first vehicle identification unit 420 selects an HOG based on a density gradient in the image from among the images of the identification target (vehicle) and other images (background, etc.) captured under various conditions collected in advance. A feature amount (first feature amount) is calculated, the calculated HOG feature amount is learned by a learning algorithm, and the discriminator is designed based on this learning. In addition, the second vehicle identification unit 430 selects an HOF feature based on the optical flow in the image from among the images of the identification target (vehicle) and other images (background, etc.) captured under various conditions collected in advance. An amount (second feature amount) is calculated, the calculated HOF feature amount is learned by a learning algorithm, and the discriminator is designed based on this learning.

  Hereinafter, the operation of the collision warning device 50 according to the present embodiment will be described with reference to the flowcharts of FIGS. 3 to 6 and FIGS. 7 and 8.

  First, the operation of the collision warning device 50 will be described based on the flowchart of FIG. In step S100, an image in front of the host vehicle 10 is captured by the camera 100 (imaging unit). The captured image is I (x, y), and 0 ≦ x ≦ m−1 and 0 ≦ y ≦ n−1. The captured image I (x, y) is captured by the vehicle external environment recognition device 60 by the image acquisition unit 150.

  By step S100, for example, the images shown in FIG. 8A and FIG. 8B are acquired and taken into the vehicle external environment recognition device 60. Here, FIG. 8A is an example of an image captured under an appropriate light beam state. The outermost part of the vehicle and the interior of the vehicle are imaged with high contrast, and good edge information is obtained. Has been obtained. On the other hand, FIG. 8B is an example of an image captured in a backlight environment or a dim environment, and the contrast of the vehicle is low and the outermost contour of the vehicle can be confirmed. It is missing. As described above, in step S100, images of various qualities are acquired according to the external environment.

  Next, in step S120, a vehicle candidate region is extracted from the image I (x, y) by the action of the vehicle candidate region extraction unit 200. Details of the processing performed in step S120 are shown in FIG. 4, and the details will be described later.

  Further, in step S140, a feature amount representing the likelihood of a vehicle is calculated within the candidate vehicle region extracted in step S120 by the operation of the feature amount calculation unit 300. Details of the processing performed in step S140 are shown in FIG.

  Thereafter, in step S150, it is determined by the action of the vehicle determination unit 400 whether or not the vehicle candidate region represents a vehicle using the feature amount calculated in step S140. Details of the processing performed in step S150 are shown in FIG. 6, and the contents will be described later.

  Next, in step S160, the relative position relationship between the host vehicle 10 and the vehicle identified in step S150 is calculated by the operation of the vehicle relative position calculation unit 70. Specifically, first, identification is made from the own vehicle 10 based on the position y of the lower end of the identified vehicle in the image I (x, y), the attachment position and the attachment direction of the camera 100 (imaging unit). Find the approximate distance to the vehicle that was released. Next, the approximate direction of the identified vehicle is determined based on the position x in the left-right direction of the identified vehicle in the image I (x, y), the mounting position and the mounting direction of the camera 100 (imaging unit). Ask. Based on the distance and direction to the identified vehicle and the speed and traveling direction of the host vehicle 10, it is determined whether or not alarm output is necessary by the operation of the alarm output determination unit 80.

  If it is determined that alarm output is required, an alarm is output from the alarm output unit 90 in step S170.

  Next, details of the process of extracting the vehicle candidate area performed in step S120 of FIG. 3 will be described with reference to FIG.

  First, in step S121, an edge (vertical edge) extending in the vertical direction is detected from the image I (x, y). This process is performed in the vertical edge detection unit 210. In general, vertical edges are detected by superimposing a predetermined operator on the image, performing a product-sum operation on the corresponding pixels, and setting the result as the edge strength at the position of the pixel corresponding to the center of the operator. Executed by. In this embodiment, it is assumed that a 3 × 3 operator shown in FIG. The image in which the vertical edge intensity obtained in this way is stored is referred to as an edge image J (x, y). Note that the operator for detecting the vertical edge is not limited to that shown in FIG. 7, and an operator having another coefficient may be used, or an operator having a different size from that in FIG.

  Next, in step S122, in order to search for a point constituting a vertical edge from the edge image J (x, y), the value of the address indicating the position in the image is set to x = 0 and y = 0. .

  In step S123, a process of searching for a pair of vertical edges arranged in the horizontal direction from the edge image J (x, y) is performed. This process is performed in the vertical edge interval calculation unit 220. This is a process performed to extract a candidate area of the vehicle, and is performed by utilizing the fact that a pair of vertical edges arranged in the horizontal direction is likely to constitute the left and right ends of the vehicle.

Specifically, the edge image J (x, y) is scanned from left to right and from top to bottom, and first, a pixel having a vertical edge intensity equal to or greater than a predetermined value is searched. Then, when found pixel _{E 1} corresponding to the condition, the pixel _{E 1 (x E1, y E1} ) as a reference, in the right direction of the pixel _{E 1 (x E1, y E1} ) , preset predetermined width _Within w _th , search is made for a pixel E _{2 in} which the vertical edge intensity that appears next is greater than or equal to a predetermined value. At this time, the paired vertical edges are searched under such a condition that the edge directions are shifted by about 180 °. This is because a vertical edge having a direction shifted by about 180 ° is detected at the left and right ends of the vehicle. Therefore, the vertical edge pair that does not correspond to the left and right ends of the vehicle is not searched as much as possible. It is for improving.

In step S123, when the pixel E ₂ (x _E2 , y _E1 ) is found, the horizontal interval w between the pixel E ₁ and the pixel E ₂ is calculated. In the case of the present embodiment, the interval w in the left-right direction is w = x _E2 −x _E1 .

Next, in step S124, it is determined whether the horizontal interval w of the extracted vertical edge pair is within a predetermined width w _th set in advance. Then, after the pixel E ₂ (x _E2 , y _E1 ) as the first vertical edge pair is found for the pixel E ₁ (x _E1 , y _E1 ), further within the predetermined width w _th The search for vertical edge pairs continues in the x-axis direction.

In step S 125, the calculated horizontal edge w of the vertical edges and the center position x = (x _E1 + x _E2 ) / 2 of the left and right vertical edge composing points are created in the xw space generation unit 230. The corresponding point in the xw space is voted (the gray value of the corresponding pixel is incremented by 1).

  The processing from step S123 to step S125 is performed while incrementing the horizontal position x of the edge image J (x, y) (step S126) and incrementing the vertical position y of the edge image J (x, y). (Step S127) and repeat.

  Next, in step S128, a pixel having a peak, which is generated when there are many vertical edge composing points constituting the vertical edge pair in the generated xw space, is extracted. The pixel (x, w) in the xw space extracted in this way represents a vehicle candidate area.

  In step S129, the left and right ends of the vehicle candidate area are set by the action of the vehicle candidate area setting unit 240. Specifically, based on the position of the pixel (x, w) having a peak extracted from the xw space, the positions of the left and right ends of the region corresponding to the pixel (x, w) are calculated backward, thus The positions of the left and right ends of the reversely calculated area are set on the image I (x, y) as the left and right ends of the vehicle candidate area.

  Next, in step S130, the upper and lower ends of the vehicle candidate area are set by the action of the vehicle candidate area setting unit 240. Specifically, since a lot of edge components extending in the horizontal direction are included in the lower part of the vehicle, the horizontal edge extending in the horizontal direction is set in the region set as the left and right ends of the vehicle candidate region in step S129. A detection process is performed, the detected horizontal edge point is projected in the horizontal direction, and a position where the projected frequency exceeds a predetermined value is set as the lower end position of the vehicle candidate. And the upper end position of a vehicle candidate area | region is estimated from the set lower end position of a vehicle candidate area | region, and the space | interval of a right-and-left end, and the vehicle candidate area | region R enclosed by the rectangle is set.

  Next, details of the process of calculating the feature amount performed in step S140 of FIG. 3 will be described with reference to FIGS.

  First, in step S141, a plurality of partial areas laid out so as not to overlap each other are set for the vehicle candidate areas in the image I (x, y) by the action of the gradient information calculation unit 312. The magnitude and the direction of the density gradient of the image inside the set partial area are calculated.

  Here, the density gradient magnitude G (x, y) and density gradient direction d (x, y) in the pixel (x, y) of the image I (x, y) are calculated as follows. .

G (x, y) = (f _x (x, y) ² + f _y (x, y) ² ) ^1/2 (Formula 1)

d (x, y) = tan ⁻¹ (f _y (x, y) / f _x (x, y)) (Formula 2)
_{However, f x (x, y)} = I (x + 1, y) -I (x-1, y) ( Equation 3)

f _y (x, y) = I (x, y + 1) −I (x, y−1) (Formula 4)

  Next, in step S142, the density gradient direction d (x, y) calculated by (Equation 2) for each partial region set by the gradient information calculation unit 312 by the action of the gradient histogram creation unit 314. Create a histogram for. As a result, as many histograms as the number of partial areas are created. Specifically, the concentration gradient direction d (x, y) is calculated in a range of −180 to 180 degrees, and is adjusted to a range of 0 to 180 degrees, and further divided into 20 degrees. Then, a histogram in nine directions is created. At the time of voting the histogram, the value of the density gradient magnitude G (x, y) of the pixel is voted at the position on the horizontal axis of the histogram corresponding to the density gradient direction d (x, y) of each pixel. As a result, a gradient histogram H (d) in which a value corresponding to the density gradient magnitude G (x, y) is stored on the vertical axis of the histogram is created.

Next, in step S143, a norm _NG representing the magnitude of the density gradient is calculated from the gradient histogram H (d) created in step S142 by the action of the density gradient feature quantity calculation unit 316, and further created. relative gradients histogram H (d), to calculate the entropy E _G representing the direction of the randomness of the gradient stored in the gradient histogram H (d).

The norm (Euclidean norm) _NG is calculated by (Equation 5).

N _G = Σ _d H (d) (Formula 5)

The norm _NG takes a larger value as the density gradient is larger, that is, as the amount of edges is larger.

Further, the entropy _{E G} is calculated by (Equation 6).

E _G = Σ _d (−p (d) log ₂ (p (d))) (Formula 6)

Here, p (d) is the probability calculated from the gradient histogram H (d) that the concentration gradient direction is d (the concentration gradient size occupies the area of the histogram and the concentration gradient direction is d). Ratio). Then, as the direction of the density gradient stored in the gradient histogram H (d) is in various directions, that is, edges in various directions are included in the partial region, and the randomness is high. Tokihodo, entropy _{E G} takes a large value.

Thus, the norm N _G calculated by the equation (5), by the entropy E _G calculated by the equation (6), the appearance of the edges in the partial area is quantified.

For example, when the image shown in FIG. 8A is acquired, the edge information shown in FIG. 8C is obtained. On the other hand, when the image shown in FIG. 8B is acquired, the edge information shown in FIG. 8D is obtained. As shown in FIG. 8 (c), the when the number of edges constituting point is detected, the norm N _G entropy E _G described above takes a large value. On the other hand, as shown in FIG. 8 (d), when the edge constituting point is small, the value of the norm N _G entropy E _G becomes small.

Next, in step S144, by the action of the optical flow calculation unit 322, two images I _t (adjacent to each other in time) among a plurality of images I (x, y) photographed at consecutive different times. x, y), I _{t + Δt} (x, y) is selected, and each pixel in the image I _t (x, y) is considered to have moved from the image I _{t + Δt} (x, y) ( The position of the corresponding pixel) is searched.

A corresponding pixel search method will be described below. First, a point with a large luminance gradient is detected as a feature point from the image I _t (x, y). Specifically, for the image I _t (x, y), an operator who sets a small area in the vicinity of the pixel of interest and obtains the density gradient inside the set small area as an amount representing the density gradient. As a result of the action, a pixel whose magnitude of the obtained density gradient is larger than a predetermined value is set as a feature point. At this time, the direction of the density gradient in the same pixel is also calculated.

Next, the image I _{t (x,} y) of the pixel having the direction size and concentration gradient of the same gradient as the feature points detected from the searches from the image _{I t + Δt (x, y} ). In this process, a search range having a predetermined size is set in the image I _{t + Δt} (x, y), and the feature point detected from the image I _t (x, y) in the set search range. Is performed by searching for pixels having the same density gradient (size and direction).

Threshold values are set for the density gradient magnitude and direction approximation, respectively, and the difference in density gradient magnitude and the difference in density gradient direction are both within the set threshold values. It is determined that a corresponding point has been found. On the other hand, when a corresponding point is not searched, another feature point is detected on the image I _t (x, y).

In this way, the optical flow having the feature point detected from the image I _t (x, y) as the start point and the corresponding point found from the image I _{t + Δt} (x, y) as the end point is determined. The optical flow is a vector quantity and has a size and a direction. Hereinafter, the optical flow detected in the pixel (x, y) of the image I _t (x, _y ) is represented by F _t (x, y) = (g _x , g _y ). At this time, the magnitude F _G (x, y) of the optical flow is expressed by (Expression 7), and the direction F _d (x, y) of the optical flow is expressed by (Expression 8).
F _G (x, y) = g _x ² + g _y ² (Formula 7)
F _d (x, y) = tan ⁻¹ (g _y / g _x ) (Formula 8)

Next, in step S145, the direction of the optical flow is applied to each of the plurality of partial regions set in the image I _t (x, y) by the gradient information calculation unit 312 by the action of the optical flow histogram generation unit 324. A histogram of F _d (x, y) is created. As a result, as many histograms as the number of partial areas are created. Specifically, the optical flow direction F _d (x, y) is calculated in a range of −180 to 180 degrees, and is adjusted to a range of 0 to 180 degrees, and further, this is adjusted by 20 degrees. To create a histogram in 9 directions. At the time of voting the histogram, the value of the optical flow magnitude F _G (x, y) of the pixel is set at the position of the horizontal axis of the histogram corresponding to the optical flow direction F _d (x, y) of each pixel. By voting, an optical flow histogram H (F _d ) in which a value corresponding to the magnitude of the optical flow is stored on the vertical axis of the histogram is created.

Next, in step S146, a norm N _d representing the magnitude of the optical flow is calculated from the created optical flow histogram H (F _d ) by the action of the optical flow feature quantity calculation unit 326, and further created. calculated for optical flow histogram H (F _d), the direction F _{d (x,} y) of the stored optical flows into the optical flow of the histogram H (F _d) entropy E _d representing the clutter of To do.

The norm (Euclidean norm) N _d is calculated by (Equation 9).

N _d = Σ _d H (F _d ) (Formula 9)

The norm N _d takes a larger value as the optical flow is larger, that is, as the motion of the image is larger.

The entropy _Ed is calculated by (Equation 10).

E _d = Σ _d (−p (F _d ) log ₂ (p (F _d ))) (Formula 10)

Here, p (F _d ) is the probability that the optical flow direction is F _d calculated from the optical flow histogram H (F _d ) (the optical flow direction occupying the area of the histogram is F _d ). The ratio of the magnitude of the optical flow). Then, as the direction of the optical flow stored in the optical flow histogram H (F _d ) is in various directions, that is, the optical flows in various directions are included in the partial region, higher at higher randomness, entropy E _d takes a large value.

As described above, the appearance state of the optical flow in the partial region is quantified by the norm N _d calculated by (Equation 9) and the entropy E _d calculated by (Equation 10).

For example, when the image is acquired as shown in FIG. 8 (b), obtained edge information shown in FIG. 8 (d), since there is less in this case edge constituting point, the norm N _G entropy E _G as described above Take a small value. However, since the outermost contour of the vehicle can be confirmed, many optical flows generated by the movement of the vehicle can be detected. As a result, the norm N _d and the entropy E _d take large values.

  Note that the processing from step S141 to S146 is sequentially performed on all the candidate vehicle regions extracted from the image I (x, y).

  Next, details of the vehicle determination process performed in step S150 of FIG. 3 will be described with reference to FIGS.

First, in step S151, the by the action of the reliability calculation unit 410, based on the norm _{N G} entropy _{E G} calculated by the gradient feature amount calculation unit 316 takes a higher value as the norm _{N G} is large, the entropy E _{Based on} the first reliability R ₁ that takes a higher value as _G increases and the norm N _d and entropy E _d calculated by the optical flow feature quantity calculation unit 326, the larger the norm N _{d, the} higher the value, and the entropy A second reliability R _{2 that} is higher as E _d is larger is calculated. That is, the higher the concentration gradient is, and the higher the concentration gradient is in various directions, the higher the first reliability R ₁ is. Then, as the motion of the image is large and the higher contains various directions of the optical flow, the second reliability R ₂ has a high value.

For example, when the image shown in FIG. 8 (a) is acquired, because the norm N _G entropy E _G becomes a large value, first reliability R ₁ is a high value. Further, since the norm N _d and the entropy E _d are large values, the second reliability R ₂ is a high value.

On the other hand, when the image shown in FIG. 8 (b) are acquired, since the norm N _G entropy E _G becomes smaller, the first reliability R ₁ is a low value. Further, since the norm N _d and the entropy E _d are large values, the second reliability R ₂ is a high value.

The density gradient feature quantity calculation unit 316, for one vehicle candidate region from each of a plurality of partial areas, although the norm N _G entropy E _G are respectively calculated, first reliability R ₁ is thus calculated the maximum value of the plurality of the norm N _G that is, may be calculated based on the maximum value of a plurality of entropy E _G, or an average value of the plurality of norms N _G, the average of a plurality of entropy E _G You may calculate based on a value. That is, the first reliability R ₁ is calculated as a numerical value representing the characteristic of the concentration gradient of the vehicle candidate region.

The optical flow feature quantity calculation unit 326 calculates the norm N _d and the entropy E _d from each of the plurality of partial areas for one vehicle candidate area, but the second reliability R ₂ is the maximum value of the plurality of norms N _d, may be calculated based on the maximum value of a plurality of entropy E _d, or an average value of the plurality of norms N _d, the average value of a plurality of entropy E _d You may calculate based on. That is, the second reliability R ₂ is calculated as a numerical value representing the state of the optical flow of the vehicle candidate region.

  Next, in step S152, the first vehicle identification unit 420 identifies the vehicle. This identification process is created in this way by creating a gradient histogram H (d) each time the position of the partial area is shifted by one pixel in the HOG feature quantity calculator 310 (first feature quantity calculator). A plurality of gradient histograms H (d) are normalized, and a HOG feature amount having a number of dimensions corresponding to the size of the vehicle candidate region and the size of the partial region based on the plurality of gradient histograms H (d) thus normalized This is performed by inputting the HOG feature amount thus calculated to the first vehicle identification unit 420.

Further, in step S153, the second vehicle identification unit 430 identifies the vehicle. In this identification processing, the HOF feature value calculation unit 320 (second feature value calculation unit) creates the optical flow histogram H (F _d ) each time while shifting the position of the partial region by one pixel, and thus created. The histograms H (F _d ) of the plurality of optical flows thus obtained are normalized, and based on the histograms H (F _d ) of the plurality of optical flows thus normalized, depending on the size of the vehicle candidate region and the size of the partial region This is done by calculating the HOF feature value having the number of dimensions and inputting the HOF feature value thus calculated to the second vehicle identification unit 430.

Then, in step S154, by the action of the determination unit 440, whether or not high is determined than the first predetermined value _{R th1} first reliability _{R 1} is preset.

When the first reliability R ₁ is determined to be higher than the first predetermined value R _th1 set in advance, in step S155, the discrimination result in the first vehicle identification unit 420, is employed as the final identification result, Based on the identification result in first vehicle identification section 420, the presence or absence of a vehicle is determined.

For example, when the image shown in FIG. 8 (a) is acquired, because the norm N _G entropy E _G becomes a large value, first reliability R ₁ becomes a high value, setting the first reliability R ₁ in advance When it is determined that the value is higher than the first predetermined value _Rth1 , the identification result in the first vehicle identification unit 420 is adopted as the final identification result.

Then, in step S156, by the action of the determination unit 440, whether or not high is determined than the second predetermined value _{R th2} second reliability _{R 2} is preset.

If it is determined that the first reliability R ₁ is lower than the preset first predetermined value R _th1 and the second reliability R ₂ is higher than the preset second predetermined value R _th2 , step S157 is performed. Then, the identification result in the second vehicle identification unit 430 is adopted as the final identification result, and the presence or absence of the vehicle is determined based on the identification result in the second vehicle identification unit 430.

For example, when the image shown in FIG. 8B is acquired, the norm N _d and the entropy E _d are large values. Therefore, the second reliability R ₂ is a high value, and the second reliability R ₂ is set in advance. When it is determined that the value is higher than the first predetermined value R _th2 , the identification result in the second vehicle identification unit 430 is adopted as the final identification result.

Furthermore, step S154, when it is not satisfied any of the conditions in step S156, i.e., the first predetermined value lower than _{R th1} of first reliability _{R 1} is preset, and the second reliability _{R 2} is preset and when the second lower than the predetermined value R _th2 in step S158, the vehicle candidate region of interest is determined not to be a vehicle.

  Then, in step S159, it is determined whether or not the identification has been completed for all the extracted vehicle candidate areas, and if it is determined that the identification has been completed for all the vehicle candidate areas, the step of FIG. The process proceeds to S160. And when identification is not complete | finished with respect to all the vehicle candidate area | regions, an identification process is continued.

In FIG. 6, after both the first vehicle identification unit 420 and the second vehicle identification unit 430 perform the identification process, the vehicle is determined based on the magnitudes of the first reliability R ₁ and the second reliability R _2. However, the present invention is not limited to this. The first reliability R ₁ and the second reliability R ₂ are first evaluated, and the first vehicle identification is performed based on the evaluation results. It is good also as a structure which performs an identification process in the part 420 or the 2nd vehicle identification part 430. FIG. By adopting such a processing configuration, it is not necessary to perform identification processing in both the first vehicle identification unit 420 and the second vehicle identification unit 430, and the first reliability R ₁ calculated by the reliability calculation unit 410 and based on the magnitude of the second reliability R _2, for performing a first vehicle identification unit 420 either discriminator identification process is selection of the second vehicle identification unit 430, executes the identification process efficiently can do.

  In the present embodiment, in order to increase the efficiency of software processing, the feature amount calculation unit 300 calculates the HOG feature amount and the HOF feature amount only for the vehicle candidate region extracted by the vehicle candidate region extraction unit 200. The vehicle determination unit 400 identifies and determines the vehicle, but it is not particularly necessary to extract the vehicle candidate area. The feature amount calculation unit 300 performs the HOG feature amount and the HOF feature amount over the entire captured image. The vehicle determination unit 400 may be configured to identify and determine the vehicle. In particular, if the vehicle external environment recognition device 60 is equipped with a plurality of CPUs and can execute a plurality of image processings in parallel, it may be more efficient to execute the same processing all at once on the entire image. .

As described above, according to the collision alarm device 50 using the vehicle external environment recognition device 60 of the present invention configured as described above, the HOG is selected from the images in front of the host vehicle 10 captured by the imaging unit 100. The feature amount calculator 310 (first feature amount calculator) calculates the HOG feature amount (first feature amount) based on the density gradient, and the HOF feature amount calculator 320 (second feature amount calculator) is the imaging unit 100. The HOF feature quantity (second feature quantity) based on the optical flow is calculated from a plurality of images taken at different times in the above, and the reliability calculation unit 410 increases the first reliability R as the density gradient increases. ₁ and further, the higher the optical flow, the higher the second reliability R ₂ is calculated, and the first vehicle identification unit 420 identifies the vehicle based on the HOG feature (first feature), Second vehicle identification unit 430 is H The vehicle is identified based on the F feature quantity (second feature quantity), and the vehicle determination unit 400 identifies the identification result in the first vehicle identification unit 420, the identification result in the second vehicle identification unit 430, and the first reliability. Since the presence or absence of the vehicle is determined based on R ₁ and the second reliability R ₂ , the first vehicle identification unit 420 is used for an image having a large density gradient in the captured image, that is, an image with a clear edge. Identifies the vehicle based on the HOG feature value (first feature value) based on the density gradient, and the second is applied to an image having a large optical flow, that is, a clear movement in the captured image. Since the vehicle identification unit 430 identifies the vehicle based on the HOF feature value (second feature value) based on the optical flow, the vehicle identification is performed in a wider range of scenes than when the vehicle is identified by a single classifier. This It is possible to provide the environment recognizing apparatus 60 for a vehicle that can.

Further, according to the collision warning device 50 using the environment recognizing apparatus 60 for a vehicle of the present invention configured in this manner, the vehicle determining unit 400, when the first reliability R ₁ is higher than the first predetermined value, perform the determination of the presence or absence of the vehicle based on the identification result of the first vehicle identification unit 420, first reliability R ₁ is lower than the first predetermined value R _th1, and the second reliability R ₂ is a second predetermined value R when higher than _th2, based on the identification result of the second vehicle identification unit 430 performs the determination of the presence or absence of the vehicle, first reliability R ₁ is lower than the first predetermined value R _th1, and the second reliability R ₂ Is lower than the second predetermined value R _th2 , it is determined that there is no vehicle, so that one classifier that matches the characteristics of the image captured at that time is used without using all of the plurality of prepared classifiers. The vehicle can be identified using only the identification It is possible to carry out the management efficiently.

  Furthermore, according to the collision alarm device 50 using the vehicle external environment recognition device 60 according to the present invention configured as described above, the vehicle relative position is determined based on the position of the vehicle identified with high accuracy by the vehicle external environment recognition device 60. The position calculation unit 70 calculates the relative positional relationship between the host vehicle 10 and the identified vehicle, and the warning output determination unit 80 determines the necessity of the warning output based on the relative positional relationship thus calculated. Since the alarm output unit 90 outputs an alarm when the alarm output determination unit 80 determines that the alarm output is necessary, a vehicle system using a vehicle external recognition device capable of outputting an alarm with high accuracy is provided. Can be provided.

  In the present embodiment, the collision alarm device that outputs an alarm when there is a possibility of collision with the identified vehicle based on the position of the identified vehicle has been described as an example. The present invention can be applied not only to the alarm device but also to a system that more actively controls the behavior of the host vehicle 10.

  That is, an inter-vehicle distance measuring unit (vehicle behavior control unit) that measures the inter-vehicle distance between the host vehicle 10 and the preceding vehicle is added to the configuration of FIG. 1, and the own vehicle 10 measured by the inter-vehicle distance measuring unit It is also possible to realize an ACC system that controls the vehicle speed of the host vehicle 10 so as to maintain the inter-vehicle distance of the preceding vehicle. In addition, an automatic brake system that decelerates the host vehicle 10 by operating a braking actuator that brakes the host vehicle 10 when there is a possibility of a collision based on the identified vehicle position can be realized. .

  In this embodiment, the camera 100 (imaging unit) captures an image in front of the host vehicle 10 and identifies the vehicle, but this is not limited to the image in front of the host vehicle 10. That is, an image behind the host vehicle 10 is captured by the camera 100 (imaging unit), and when the host vehicle 10 moves backward, the vehicle behind the host vehicle 10 is identified and collides with the identified vehicle. It may be configured to output an alarm when there is a possibility. Furthermore, the rear image of the host vehicle 10 is captured by the camera 100 (imaging unit), the vehicle is identified from the captured image, and the vehicle 10 is identified when the lane change is performed. When there is a possibility of collision with the vehicle, it is also possible to output a warning.

  Further, in the present embodiment, it has been described that the vehicle is identified by the first vehicle identification unit 420 and the second vehicle identification unit 430 configured by a classifier designed in advance based on a learning algorithm. Such a configuration example is not particularly limited, and may be realized by using, for example, a neural network, a support vector machine (SVM) classifier, an NN (nearest neighbor) classifier, a Bayes classifier, or the like. May be realized.

  In this embodiment, the HOG feature value is used as the first feature value. However, this is not limited to the HOG feature value. That is, any feature quantity based on the density gradient in the captured image I (x, y) can be used instead of the HOG feature quantity. For example, SIFT can obtain a feature that is invariant in size and rotation by obtaining a brightness difference between adjacent rectangular areas and using a Haar like feature quantity that uses the difference as a feature quantity or a brightness distribution in a neighboring area as a density gradient. A feature amount or the like may be used.

  Next, regarding the second embodiment of the vehicle system using the vehicle external environment recognition apparatus according to the present invention, the vehicle front is monitored by a camera mounted on the vehicle, and a vehicle that may contact the host vehicle is identified. A collision warning device using a vehicle external environment recognition device that outputs a warning when an accident occurs will be described as an example.

  First, the configuration of the apparatus will be described with reference to FIG. The collision warning device 52 according to the second embodiment is installed in a host vehicle 10 (not shown), and a camera 100 (imaging unit) that captures an image ahead of the host vehicle 10 and a vehicle external environment that identifies vehicles other than the host vehicle 10. When the vehicle is identified, the recognition device 62, a vehicle relative position calculation unit 70 that calculates the direction and distance from the host vehicle 10 to the identified vehicle, and an alarm output determination unit 80 that determines the possibility of a collision. And an alarm output unit 90 that outputs an alarm when the possibility of collision is a predetermined value or more.

  Here, the vehicle external environment recognition device 62 is configured by a computer having a CPU, a memory, an I / O, and the like, and specifically captures an image captured by the camera 100 (imaging unit) at a constant period. An image acquisition unit 150; a vehicle candidate region extraction unit 200 that detects a candidate region of a vehicle from the captured image; a feature amount calculation unit 350 that calculates a feature amount that represents the vehicle likeness from the vehicle candidate region; A vehicle determination unit 450 that determines whether the candidate area represents a vehicle is provided. The vehicle external environment recognition device 62 is programmed with a predetermined process and repeatedly executes the process at a predetermined cycle.

  The vehicle candidate area extraction unit 200 further calculates a vertical edge detection unit 210 that detects the position of the pixel that forms the vertical edge, and a horizontal interval between the pixels that form the vertical edge detected by the vertical edge detection unit 210. The vertical edge interval calculator 220, the horizontal edge interval (w) of the vertical edge calculated by the vertical edge interval calculator 220, and the center position (x) of the left and right vertical edges are voted in the xw space, As a result of this voting, an xw space generation unit 230 that detects a point (x, w) having a peak and sets it as a vehicle candidate region, and a vehicle candidate region setting unit 240 that sets the vehicle candidate region.

  The feature amount calculation unit 350 further includes a HOG feature amount calculation unit 310 (first feature amount calculation unit) that calculates a HOG (Histogram of Oriented Gradients) feature amount (first feature amount) related to gradient information based on the histogram, Calculated by a HOF feature quantity calculation unit 320 (second feature quantity calculation unit) that calculates a HOF (Histogram of Optical Flow) feature quantity (second feature quantity) relating to the optical flow based on the histogram, and a HOG feature quantity calculation unit 310. The HOG feature value is obtained and determined, and when the HOG feature value is less than a predetermined value, the HOF feature value calculation unit 320 is instructed to calculate the HOF feature value. The feature amount determination unit 330 is configured to acquire the obtained HOF feature amount and determine the size thereof.

  The detailed configuration of the feature amount calculation unit 350 is shown in FIG. Among these, the detailed configurations of the HOG feature value calculation unit 310 and the HOF feature value calculation unit 320 are as shown in FIG. The feature amount determination unit 330 further performs the HOF feature on the HOG feature amount determination unit 332 that determines the size of the HOG feature amount calculated by the HOG feature amount calculation unit 310 and the HOF feature amount calculation unit 320. An HOF feature value calculation instruction unit 334 for instructing the amount calculation, and a HOF feature value determination unit 336 for determining the magnitude of the HOF feature value calculated by the HOF feature value calculation unit 320.

  The vehicle determination unit 450 further includes a first vehicle identification unit 420 that identifies the vehicle based on the HOG feature value, and a second vehicle identification unit 430 that identifies the vehicle based on the HOF feature value. Here, the 1st vehicle identification part 420 and the 2nd vehicle identification part 430 are comprised by the discriminator designed based on the learning algorithm as demonstrated in Example 1. FIG.

  Hereinafter, the operation of the collision warning device 52 according to the present embodiment will be described based on the flowchart of FIG. In addition, about the location which performs the process similar to the process demonstrated in Example 1, description is abbreviate | omitted. Therefore, the portions where the description of the operation of the second embodiment is omitted follow the contents described in the first embodiment.

  Based on the flowchart of FIG. 11, the operation of the collision warning device 52 will be described. In step S200, an image in front of the host vehicle 10 is captured by the camera 100 (imaging unit). The captured image is I (x, y), and 0 ≦ x ≦ m−1 and 0 ≦ y ≦ n−1. The captured image I (x, y) is taken into the vehicle external environment recognition device 62 by the image acquisition unit 150.

  Next, in step S210, a vehicle candidate region is extracted from the image I (x, y) by the action of the vehicle candidate region extraction unit 200. Note that the details of the process performed in step S210 are the same as the contents described in the first embodiment, and thus the description thereof is omitted.

Further, in step S220, the HOG feature amount (first feature amount) based on the density gradient of the image is calculated inside the vehicle candidate region extracted in step S210 by the action of the feature amount calculation unit 350. The details of the processing performed in step S220 are the same as those in steps S141 to S143 described in the first embodiment based on FIG. Then, this through the process of step S220, as HOG features and gradient histogram H (d) is created, the norm _{N G} entropy _{E G} is calculated.

Thereafter, in step S230, the by the action of the characteristic amount determination unit 330, an HOG feature amount calculated in the step S220, the size of the norm _{N G} entropy _{E G} is determined. Specifically, in the HOG features determining unit 332, the size of the norm _{N G} is larger than a predetermined value _{N Gth,} and the size of the entropy _{E G} is a predetermined value _{E Gth} If greater than, the process proceeds to step S240.

On the other hand, the size and magnitude of the entropy _{E G} norm _{N G} is, when does not satisfy the condition, HOF feature value calculation is instructed by the HOF feature quantity calculation instruction unit 334 with respect HOF feature calculation section 320, The HOF feature value calculation unit 320 calculates a HOF feature value (second feature value) based on the motion of the image inside the vehicle candidate area extracted in step S210. Note that the HOF feature amount calculation process is the same as steps S144 to S146 described with reference to FIG. By this processing, an optical flow histogram H (F _d ) is created as the HOF feature value, and the norm N _d and the entropy E _d are calculated.

Next, in step S234, the magnitude of the norm N _d and the entropy E _d that are HOF feature amounts are determined by the action of the feature amount determination unit 330. Specifically, in the HOF feature amount determination unit 336, the magnitude of the norm N _d is greater than a preset predetermined value N _dth and the magnitude of the entropy E _d is a preset predetermined value E _dth. If greater than, the process proceeds to step S250.

On the other hand, when the norm N _d and the entropy E _d do not satisfy the above condition, it is determined that there is no vehicle in the vehicle candidate region, and the process of FIG. 10 is terminated.

In step S230, the an HOG features, the size of the norm _{N G} entropy _{E G} is, if it is determined that both greater than the predetermined value, in step S240, the vehicle by the action of the first vehicle identification unit 420 Identification is performed. This identification processing is performed according to the size of the vehicle candidate area and the size of the partial area calculated by normalizing the plurality of gradient histograms H (d) created by the HOG feature quantity calculation section 310 (first feature quantity calculation section). This is done by inputting a HOG feature amount having a different number of dimensions to the first vehicle identification unit 420. And it is determined whether a vehicle candidate area | region represents a vehicle, and it progresses to step S260.

Furthermore, when it is determined in step S234 that both the norm N _d and the entropy E _d , which are HOF feature quantities, are larger than a predetermined value, in step S250, the second vehicle identification unit 430 operates. Vehicle identification is performed. In this identification process, the size of the vehicle candidate region and the size of the partial region calculated by normalizing the histograms H (F _d ) of the plurality of optical flows created by the HOF feature amount calculation unit 320 (second feature amount calculation unit) are used. This is performed by inputting a HOF feature amount having a dimension number corresponding to the size to the second vehicle identification unit 430. And it is determined whether a vehicle candidate area | region represents a vehicle, and it progresses to step S260.

  Next, in step S260, the relative position relationship between the host vehicle 10 and the vehicle identified in step S240 or step S250 is calculated by the operation of the vehicle relative position calculation unit 70. Specifically, first, identification is made from the own vehicle 10 based on the position y of the lower end of the identified vehicle in the image I (x, y), the attachment position and the attachment direction of the camera 100 (imaging unit). Find the approximate distance to the vehicle that was released. Next, the approximate direction of the identified vehicle is determined based on the position x in the left-right direction of the identified vehicle in the image I (x, y), the mounting position and the mounting direction of the camera 100 (imaging unit). Ask. Based on the distance and direction to the identified vehicle and the speed and traveling direction of the host vehicle 10, it is determined whether or not alarm output is necessary by the operation of the alarm output determination unit 80.

  If it is determined that alarm output is necessary, an alarm is output from the alarm output unit 90 in step S262.

As described above, according to the collision alarm device 52 using the vehicle external environment recognition device 62 of the present invention configured as described above, the image in front of the host vehicle 10 captured by the camera 100 (imaging unit). from the first feature quantity calculating unit 310 calculates the gradient histogram H (d) a first feature based on the concentration gradient norm N _G, entropy E _G, a first feature quantity calculated N _G when E _G is greater than the predetermined value, the first vehicle identification unit 420 performs identification of the vehicle using the HOG features calculated based on H (d) a first feature quantity, the result of the identification On the other hand, if the presence or absence of the vehicle is determined and the calculated first feature values N _G and E _G are smaller than a predetermined value, the second feature value calculation unit 320 uses the second feature value based on the optical flow. Optica being Flow histograms H _{(F d)} and norm _{N d,} to calculate the entropy _{E d,} which is the second feature amount calculated _N d, when _{E d} is greater than the predetermined value, the second vehicle identification unit 430 Since the vehicle is identified using the HOF feature value calculated based on the second feature value H (F _d ), and the presence or absence of the vehicle is determined based on the identification result, the first feature value is obtained. Since H (d), N _G , E _G and the second feature quantity H (F _d ), N _d , E _d can be determined without necessarily calculating all, It is possible to provide a vehicle external environment recognition device with a small load.

In FIG. 6, after both the first vehicle identification unit 420 and the second vehicle identification unit 430 perform the identification process, the vehicle is determined based on the magnitudes of the first reliability R ₁ and the second reliability R _2. However, the present invention is not limited to this. The first reliability R ₁ and the second reliability R ₂ are first evaluated, and the first vehicle identification is performed based on the evaluation results. It is good also as a structure which performs an identification process in the part 420 or the 2nd vehicle identification part 430. FIG. By adopting such a processing configuration, it is not necessary to perform identification processing in both the first vehicle identification unit 420 and the second vehicle identification unit 430, and the first reliability R ₁ calculated by the reliability calculation unit 410 and based on the magnitude of the second reliability R _2, for performing a first vehicle identification unit 420 either discriminator identification process is selection of the second vehicle identification unit 430, executes the identification process efficiently can do.

  Next, regarding a third embodiment of the vehicle system using the vehicle external environment recognizing device according to the present invention, the vehicle front is monitored by a camera mounted on the vehicle, and a vehicle that may contact the host vehicle is identified. A collision warning device using a vehicle external environment recognition device that outputs a warning when an accident occurs will be described as an example.

  First, the configuration of the apparatus will be described with reference to FIG. The collision warning device 54 according to the second embodiment is installed in the host vehicle 10 (not shown), and a camera 100 (imaging unit) that captures an image ahead of the host vehicle 10 and a vehicle external environment that identifies vehicles other than the host vehicle 10. A recognition device 64; a vehicle relative position calculation unit 70 that calculates a direction and a distance from the host vehicle 10 to the identified vehicle when the vehicle is identified; and an alarm output determination unit 80 that determines the possibility of a collision. An alarm output unit 90 is provided that outputs an alarm when the possibility of collision is equal to or greater than a predetermined value.

  Here, the vehicle external environment recognition device 64 is configured by a computer having a CPU, a memory, an I / O, and the like, and specifically captures an image captured by the camera 100 (imaging unit) at a constant period. An image acquisition unit 150; a vehicle candidate region extraction unit 200 that detects a candidate region of a vehicle from the captured image; a feature amount calculation unit 370 that calculates a feature amount that represents the vehicle likeness from the vehicle candidate region; A vehicle determination unit 450 that determines whether the candidate area represents a vehicle is provided. The vehicle external environment recognition device 64 is programmed with a predetermined process and repeatedly executes the process at a predetermined cycle.

  The vehicle candidate area extraction unit 200 further calculates a vertical edge detection unit 210 that detects the position of the pixel that forms the vertical edge, and a horizontal interval between the pixels that form the vertical edge detected by the vertical edge detection unit 210. The vertical edge interval calculator 220, the horizontal edge interval (w) of the vertical edge calculated by the vertical edge interval calculator 220, and the center position (x) of the left and right vertical edges are voted in the xw space, As a result of this voting, an xw space generation unit 230 that detects a point (x, w) having a peak and sets it as a vehicle candidate region, and a vehicle candidate region setting unit 240 that sets the vehicle candidate region.

  The feature amount calculation unit 370 further measures the vehicle speed of the host vehicle 10 and determines whether or not a vehicle speed equal to or higher than a predetermined value is output, and a gradient based on the histogram. HOG feature quantity calculation unit 310 (first feature quantity calculation unit) that calculates HOG (Histogram of Oriented Gradients) feature quantity (first feature quantity) related to information, and HOF (Histogram of Optical Flow) related to optical flow based on a histogram Calculated by the HOF feature value calculation unit 320 (second feature value calculation unit) that calculates the feature value (second feature value), the HOG feature value calculated by the HOG feature value calculation unit 310, or the HOF feature value calculation unit 320 The feature amount determination unit 340 determines the size of the HOF feature amount.

  The detailed configuration of the feature amount calculation unit 370 is shown in FIG. Among these, the detailed configurations of the HOG feature value calculation unit 310 and the HOF feature value calculation unit 320 are as shown in FIG. Then, the feature amount determination unit 340 further includes a HOG feature amount determination unit 342 that determines the size of the HOG feature amount calculated by the HOG feature amount calculation unit 310 and an HOF feature calculated by the HOF feature amount calculation unit 320. A HOF feature amount determination unit 344 that determines the magnitude of the amount is included.

  The vehicle determination unit 450 further includes a first vehicle identification unit 420 that identifies the vehicle based on the HOG feature value, and a second vehicle identification unit 430 that identifies the vehicle based on the HOF feature value. Here, as described in the first embodiment, the first vehicle identification unit 420 and the second vehicle identification unit 430 are configured by classifiers designed in advance based on a learning algorithm.

  Hereinafter, the operation of the collision warning device 54 according to the present embodiment will be described based on the flowchart of FIG. In addition, about the location which performs the process similar to the process demonstrated in Example 1, description is abbreviate | omitted. Therefore, the portions where the description of the operation of the third embodiment is omitted shall follow the contents described in the first embodiment.

  The operation of the collision warning device 54 will be described based on the flowchart of FIG. In step S300, an image in front of the host vehicle 10 is captured by the camera 100 (imaging unit). The captured image is I (x, y), and 0 ≦ x ≦ m−1 and 0 ≦ y ≦ n−1. The captured image I (x, y) is captured by the vehicle external environment recognition device 64 by the image acquisition unit 150.

  Next, in step S310, a vehicle candidate region is extracted from the image I (x, y) by the action of the vehicle candidate region extraction unit 200. Note that the details of the processing performed in step S310 are the same as the contents described in the first embodiment, and thus the description thereof is omitted.

In step S320, the constant k set to control the flow of a series of processes is set to 0. Then, in step S330, the vehicle speed of the host vehicle 10 is determined by the action of the vehicle speed determination unit 305 (vehicle behavior detection unit). v is measured, and it is determined whether or not the measured vehicle speed v is equal to or greater than a predetermined value _vth . When the vehicle speed v is greater than or equal to the predetermined value v _th , the process proceeds to step S340, and when the vehicle speed v is smaller than the predetermined value v _th , the process proceeds to step S390.

In step S340, the HOF feature quantity calculation unit 320 applies the HOF feature quantity (second feature quantity) based on the optical flow of the image to the inside of the vehicle candidate area extracted in step S310 by the action of the feature quantity calculation unit 370. ) Is calculated. Here, the reason for calculating the HOF feature value is that when the vehicle speed v of the host vehicle 10 is larger than the predetermined value v _th , it is expected that the feature of motion in the captured image will increase. . The HOF feature amount calculation process is the same as that in steps S144 to S146 described with reference to FIG. In step S340, an optical flow histogram H (F _d ) is created, a norm N _d and entropy E _d are calculated, and the process proceeds to step S350.

On the other hand, in step S330, when the measured vehicle speed v is less than the predetermined value _{v th,} the process proceeds to step S390, by the action of the characteristic amount calculation unit 370, with respect to the interior of the vehicle candidate region extracted in step S310 The HOG feature value calculation unit 310 calculates a HOG feature value (first feature value) based on the density gradient of the image. The reason for calculating the HOG feature amount is that, when the vehicle speed v of the host vehicle 10 is smaller than the predetermined value _vth , it is expected that the feature of motion in the captured image is reduced. . The HOG feature amount calculation process is the same as that in steps S141 to S143 described with reference to FIG. Then, by this process, gradient histogram H (d) is created, it is calculated norm _{N G} entropy _{E G} is, the process proceeds to step S400.

After the process of step S340, in step S350, the magnitude of the norm N _d and the entropy E _d that are the HOF feature amounts calculated in step S340 are determined by the action of the feature amount determination unit 340. Specifically, in the HOF feature amount determination unit 344, the magnitude of the norm N _d is larger than a preset predetermined value N _dth and the magnitude of the entropy E _d is a preset predetermined value E _dth. If greater than, the process proceeds to step S360.

On the other hand, when the norm N _d and the entropy E _d do not satisfy the above condition in step S350, it is confirmed in step S370 whether the value of the constant k is 1. If because when it is k = 1, the size of the norm N _G entropy E _G is HOG features, and the size of the norm N _d and the entropy E _d is HOF feature amount are both small values, the vehicle It is determined that there is no vehicle in the candidate area, and the process proceeds to A of FIG.

If the value of the constant k is not 1 in step S370, the constant k is set to 1 in step S380, and the process proceeds to B (step S390) in FIG. This is because even when the host vehicle 10 is traveling at a vehicle speed equal to or higher than the predetermined value _vth , the vehicle shown in the captured image is traveling at a vehicle speed substantially equal to the host vehicle 10. This is because the movement of the vehicle does not necessarily increase. In such a case, since the norm N _d and the entropy E _d do not increase, the process proceeds to step S390 to calculate the HOG feature (first feature) based on the density gradient of the acquired image.

When the process proceeds from step S350 to step S360, the vehicle is identified by the action of the second vehicle identification unit 430. In this identification process, the size of the vehicle candidate region and the size of the partial region calculated by normalizing the histograms H (F _d ) of the plurality of optical flows created by the HOF feature amount calculation unit 320 (second feature amount calculation unit) are used. This is performed by inputting a HOF feature amount having a dimension number corresponding to the size to the second vehicle identification unit 430. And it is determined whether a vehicle candidate area | region represents a vehicle, and it progresses to step S440.

After step S390, in step S400, by the action of the characteristic amount determination unit 340, an HOG feature amount calculated in the step S390, the size of the norm _{N G} entropy _{E G} is determined. Specifically, in the HOG features determining unit 342, the size of the norm _{N G} is larger than a predetermined value _{N Gth,} and the size of the entropy _{E G} is a predetermined value _{E Gth} If greater than, the process proceeds to step S410.

On the other hand, in step S400, the size and magnitude of the entropy E _G norm N _G is, when does not satisfy the condition in step S420, the value of the constant k is verified whether or not 1. If because when it is k = 1, the size of the norm N _G entropy E _G is HOG features, and the size of the norm N _d and the entropy E _d is HOF feature amount are both small values, the vehicle It is determined that there is no vehicle in the candidate area, and the process proceeds to A of FIG.

If the value of the constant k is not 1 in step S420, the constant k is set to 1 in step S430, and the process proceeds to C (step S340) in FIG. This is because, even when the host vehicle 10 is traveling at a vehicle speed lower than the predetermined value v _th , the contrast of the captured image is low and the norm N, which is the HOG feature amount, is in a backlight or dusk state. _G and entropy _{E G} is because decreases. In such a case, the process proceeds to step S340, and the HOF feature value (second feature value) based on the optical flow of the captured image is calculated.

  When the process proceeds from step S400 to step S410, the vehicle is identified by the action of the first vehicle identification unit 420. This identification processing is performed according to the size of the vehicle candidate area and the size of the partial area calculated by normalizing the plurality of gradient histograms H (d) created by the HOG feature quantity calculation section 310 (first feature quantity calculation section). This is done by inputting a HOG feature amount having a different number of dimensions to the first vehicle identification unit 420. And it is determined whether a vehicle candidate area | region represents a vehicle, and it progresses to step S440.

  Next, in step S440, the relative positional relationship between the host vehicle 10 and the vehicle identified in step S360 or step S410 is calculated by the operation of the vehicle relative position calculation unit 70. Specifically, first, based on the position y of the lower end of the identified vehicle in the image I (x, y), the attachment position and the attachment direction of the camera 100 (imaging unit) attached to the host vehicle 10. Then, an approximate distance from the own vehicle 10 to the identified vehicle is obtained. Next, the approximate direction of the identified vehicle is determined based on the position x in the left-right direction of the identified vehicle in the image I (x, y), the mounting position and the mounting direction of the camera 100 (imaging unit). Ask. Based on the distance and direction to the identified vehicle and the speed and traveling direction of the host vehicle 10, it is determined whether or not alarm output is necessary by the operation of the alarm output determination unit 80.

  If it is determined that alarm output is required, an alarm is output from the alarm output unit 90 in step S450.

As described above, according to the collision alarm device 54 using the vehicle external environment recognition device 64 of the present invention configured as described above, an image of the front of the host vehicle 10 captured by the camera 100 (imaging unit) is displayed. Among them, the HOG feature quantity calculation unit 310 (first feature quantity calculation unit) has a gradient histogram H (d) that is a first feature quantity based on the density gradient, according to the calculation result of the vehicle speed determination unit 305 (vehicle behavior calculation unit). ) and calculates the norm N _G, entropy E _G, N _G is the first feature amount _calculated, when E _G is greater than the predetermined value, the first vehicle identification unit 420, is the first feature amount The vehicle is identified using the HOG feature amount calculated based on H (d), and the presence or absence of the vehicle is determined based on the identification result. On the other hand, the calculated first feature amounts N _G and E _G When H is smaller than a predetermined value, H F characteristic amount calculating section 320 (second feature quantity calculating section) is calculated histogram H of the optical flow is a second feature based on optical flow and (F _d) norm N _d, entropy E _d, it was calculated When the second feature amounts N _d and E _d are larger than the predetermined values, the second vehicle identification unit 430 uses the HOF feature amount calculated based on the second feature amount H (F _d ). The process of determining the presence or absence of a vehicle based on the identification result, and the HOF feature quantity calculation unit 320 (second feature quantity calculation unit) is H (F _d ), which is the second feature quantity based on the optical flow. And N _d and E _d are calculated, and when the calculated second feature amounts N _d and E _d are larger than a predetermined value, the second vehicle identification unit 430 uses the second feature amount H (F _d ) Based on The vehicle is identified using the HOF feature value that has been issued, and the presence or absence of the vehicle is determined based on the identification result. On the other hand, when the calculated second feature values N _d and E _d are smaller than a predetermined value , HOG feature amount calculation unit 310 (first feature quantity calculating section) is calculated as H (d) a first feature based on the concentration gradient N _G, the E _G, is the first feature quantity calculated N _G, when E _G is greater than the predetermined value, the first vehicle identification unit 420 performs identification of the vehicle using the HOG features calculated based on H (d) a first feature amount, the identification In order to switch between the process of determining the presence / absence of a vehicle based on the result, H (d), N _G , E _G as the first feature quantity, and H (F _d ), N _d , as the second feature quantity, the E _d, can also determine the presence or absence of a vehicle without calculating necessarily all Therefore, it is possible to provide a environment recognizing apparatus for a vehicle processing load is small. In particular, the processing load can be further reduced by switching the feature amount to be calculated according to the calculation result of the vehicle behavior calculation unit.

  As mentioned above, although Example 1, Example 2, and Example 3 of this invention were explained in full detail with drawing, since an Example is only an illustration of this invention, this invention is limited only to the structure of an Example. Of course, design changes and the like within a range not departing from the gist of the present invention are included in the present invention.

50 Collision warning device 60 External recognition device for vehicle 70 Vehicle relative position calculation unit 80 Alarm output determination unit 90 Alarm output unit 100 Camera (imaging unit)
150 image acquisition unit 200 vehicle candidate region extraction unit 210 vertical edge detection unit 220 vertical edge interval calculation unit 230 xw space generation unit 240 vehicle candidate region setting unit 300 feature amount calculation unit 310 HOG feature amount calculation unit (first feature amount calculation unit) )
320 HOF feature value calculation unit (second feature value calculation unit)
400 vehicle determination unit 410 reliability calculation unit 420 first vehicle identification unit 430 second vehicle identification unit 440 determination unit

Claims (8)

An imaging unit that is mounted on the host vehicle and images the periphery of the host vehicle;
A first feature amount calculation unit that calculates a first feature amount based on a density gradient from images captured by the imaging unit;
A second feature amount calculation unit that calculates a second feature amount based on an optical flow from a plurality of images captured at different times by the imaging unit;
A reliability calculation unit that calculates a higher first reliability as the concentration gradient is larger, and calculates a higher second reliability as the optical flow is larger;
A first vehicle identification unit for identifying a vehicle based on the first feature amount;
A second vehicle identification unit for identifying a vehicle based on the second feature amount;
A vehicle determination unit that determines the presence or absence of a vehicle based on the identification result in the first vehicle identification unit, the identification result in the second vehicle identification unit, the first reliability, and the second reliability;
A vehicle external environment recognition device characterized by comprising: The vehicle determination unit
When the first reliability is higher than a first predetermined value, the presence or absence of a vehicle is determined based on the identification result of the first vehicle identification unit,
When the first reliability is lower than a first predetermined value and the second reliability is higher than a second predetermined value, the presence / absence of a vehicle is determined based on the identification result of the second vehicle identification unit. ,
2. The vehicle external environment according to claim 1, wherein when the first reliability is lower than a first predetermined value and the second reliability is lower than a second predetermined value, it is determined that there is no vehicle. Recognition device. An imaging unit that is mounted on the host vehicle and images the periphery of the host vehicle;
A first feature amount calculation unit that calculates a first feature amount based on a density gradient from images captured by the imaging unit;
A second feature amount calculation unit that calculates a second feature amount based on an optical flow from a plurality of images captured at different times by the imaging unit;
A first vehicle identification unit for identifying a vehicle based on the first feature amount;
A second vehicle identification unit that identifies a vehicle based on the second feature amount, and when the first feature amount is greater than a predetermined value, based on the identification result in the first vehicle identification unit An external environment recognition device for vehicles, wherein the presence / absence of a vehicle is determined, and when the first feature value is smaller than a predetermined value, the presence / absence of the vehicle is determined based on an identification result in the second vehicle identification unit. An imaging unit that is mounted on the host vehicle and images the periphery of the host vehicle;
A first feature amount calculation unit that calculates a first feature amount based on a density gradient from images captured by the imaging unit;
A second feature amount calculation unit that calculates a second feature amount based on an optical flow from a plurality of images captured at different times by the imaging unit;
A first vehicle identification unit for identifying a vehicle based on the first feature amount;
A second vehicle identification unit that identifies a vehicle based on the second feature amount, and when the second feature amount is greater than a predetermined value, based on the identification result in the second vehicle identification unit An external environment recognition device for vehicles, wherein the presence / absence of a vehicle is determined, and when the second feature amount is smaller than a predetermined value, the presence / absence of the vehicle is determined based on an identification result in the first vehicle identification unit. An imaging unit that is mounted on the host vehicle and images the periphery of the host vehicle;
A first feature amount calculation unit that calculates a first feature amount based on a density gradient from images captured by the imaging unit;
A second feature amount calculation unit that calculates a second feature amount based on an optical flow from a plurality of images captured at different times by the imaging unit;
A reliability calculation unit that calculates a higher first reliability as the concentration gradient is larger, and calculates a higher second reliability as the optical flow is larger;
A first vehicle identification unit for identifying a vehicle based on the first feature amount;
A second vehicle identification unit for identifying a vehicle based on the second feature amount;
A vehicle determination unit that determines the presence or absence of a vehicle based on the identification result in the first vehicle identification unit, the identification result in the second vehicle identification unit, the first reliability, and the second reliability;
In a vehicle external environment recognition device having a vehicle behavior detection unit that detects the behavior of the host vehicle,
According to the behavior of the host vehicle detected by the vehicle behavior detection unit,
The first feature quantity calculation unit calculates the first feature quantity, and when the first feature quantity is larger than a predetermined value, the first vehicle identification unit identifies the vehicle and determines the presence or absence of the vehicle. When the first feature value is smaller than a predetermined value, the second feature value calculation unit calculates the second feature value, and the second vehicle identification unit performs vehicle identification to determine whether or not there is a vehicle. The process of determining
The second feature quantity calculation unit calculates the second feature quantity, and when the second feature quantity is larger than a predetermined value, the second vehicle identification unit identifies the vehicle and determines the presence or absence of the vehicle. When the second feature value is smaller than a predetermined value, the first feature value is calculated by the first feature value calculation unit, and the vehicle is identified by the first vehicle identification unit. And an outside recognition device for a vehicle characterized by switching the processing to determine. The vehicle behavior detection unit detects a vehicle speed of the host vehicle,
When the vehicle speed is greater than a predetermined value, the second feature amount calculation unit calculates the second feature amount. When the second feature amount is greater than a predetermined value, the second vehicle identification unit The vehicle is identified to determine the presence or absence of the vehicle. When the second feature amount is smaller than a predetermined value, the first feature amount calculation unit calculates the first feature amount, and the first vehicle identification is performed. The vehicle is identified at the part to determine the presence or absence of the vehicle,
When the vehicle speed is smaller than a predetermined value, the first feature amount calculating unit calculates the first feature amount. When the first feature amount is larger than a predetermined value, the first vehicle identifying unit The vehicle is identified to determine the presence or absence of the vehicle, and when the first feature amount is smaller than a predetermined value, the second feature amount calculation unit calculates the second feature amount, and the second vehicle identification unit The vehicle external environment recognition device according to claim 5, wherein the presence of the vehicle is determined by identifying the vehicle. A vehicle relative position calculation unit that calculates a positional relationship between the vehicle identified by the vehicle external environment recognition device and the host vehicle;
Based on the calculation result of the vehicle relative position calculation unit, an alarm output determination unit that determines the necessity of outputting an alarm;
7. The vehicle according to claim 1, further comprising: an alarm output unit that outputs an alarm when it is determined that an alarm output is necessary in the alarm output determination unit. A vehicle system using an external recognition device. A vehicle relative position calculation unit that calculates a positional relationship between the vehicle identified by the vehicle external environment recognition device and the host vehicle;
The vehicle external environment according to claim 1, further comprising: a vehicle behavior control unit that controls the behavior of the host vehicle based on a calculation result of the vehicle relative position calculation unit. A vehicle system using a recognition device.