JP5821687B2 - Vehicle detection device and vehicle detection method - Google Patents

Description

  The present invention relates to a vehicle detection device and a vehicle detection method for detecting a vehicle based on a captured image.

  Conventionally, as a system for detecting a vehicle based on a captured image, for example, there is an axle detection system that detects an axle from three-dimensional data acquired using a three-dimensional image pickup device, and Patent Literature 1 is disclosed.

JP 2011-108223 A

  However, in the conventional axle detection system described above, since the vehicle is detected from the ratio of tires in the entire vehicle histogram, the ratio varies depending on the type of vehicle such as a truck or a regular vehicle. There was a problem that it was necessary to deal with it.

  Therefore, the present invention has been proposed in view of the above-described situation, and a vehicle detection device and a vehicle detection method that can detect any type of vehicle regardless of the vehicle type by the same process. The purpose is to provide.

  The present invention relates to an imaging unit capable of imaging a side surface of a traveling vehicle, and a processing region setting for setting, as a processing region, a region including a feature point indicating characteristics as a vehicle with respect to an image captured by the imaging unit. A motion detection unit that detects a motion of a feature point based on images of processing regions set in a plurality of images having different imaging times, and an average that calculates an average value of the motion of the feature point detected by the motion detection unit A motion estimation unit, a relative motion detection unit that obtains a relative motion of a feature point by subtracting the average value of the motion from the motion of the feature point, and a relative motion obtained for each feature point in the processing region. And a rotation determination unit that detects a vehicle by determining whether the image of the processing region is a rotating body based on the distribution of the histogram.

  According to the vehicle detection device and the vehicle detection method of the present invention, the rotational motion of the tire is detected by determining whether or not the image of the processing region is a rotating body. Can be detected by the same process.

It is a block diagram which shows the structure of the vehicle detection apparatus which concerns on one Embodiment of this invention. It is a figure which shows the structure at the time of mounting the vehicle detection apparatus which concerns on one Embodiment of this invention in a vehicle. It is a figure which shows an example of the motion of the feature point detected by the vehicle detection apparatus which concerns on one Embodiment of this invention. It is a figure which shows an example of the relative motion detected by the vehicle detection apparatus which concerns on one Embodiment of this invention. It is a figure for demonstrating the relationship between the horizontal motion detected by the vehicle detection apparatus which concerns on one Embodiment of this invention, and a vertical motion. It is a figure for demonstrating the calculation method of the angular velocity by the vehicle detection apparatus which concerns on one Embodiment of this invention. It is a figure which shows an example of the histogram detected by the vehicle detection apparatus which concerns on one Embodiment of this invention. It is a flowchart which shows the procedure of the vehicle detection process by the vehicle detection apparatus which concerns on one Embodiment of this invention. It is a figure for demonstrating the effect by the vehicle detection apparatus which concerns on one Embodiment of this invention.

  Hereinafter, an embodiment to which the present invention is applied will be described with reference to the drawings.

[Configuration of driving support device]
FIG. 1 is a block diagram illustrating a configuration of a vehicle detection device according to the present embodiment. As shown in FIG. 1, the vehicle detection device 1 according to the present embodiment sets an image processing unit 2 that can image a side surface of a traveling vehicle, and a processing region for an image captured by the image capturing unit 2. A processing region setting unit 3, a motion detection unit 4 that detects motion of feature points based on images of processing regions set in a plurality of images having different imaging times, and motion of feature points detected by the motion detection unit 4 For each feature point in the processing area, an average motion estimator 5 for obtaining the average value, a relative motion detector 6 for obtaining the relative motion of the feature points by subtracting the average value of the motion from the motion of the feature points, and A histogram creation unit 7 that creates a histogram based on the obtained relative movement, and a rotation determination unit 8 that determines whether the image of the processing region is a rotator based on the distribution of the histogram are provided. .

  The vehicle detection device 1 according to the present embodiment is mounted on the host vehicle 20 as shown in FIG. 2, for example, and images of the host vehicle 20 are captured by the imaging unit 2 installed toward the side of the host vehicle 20. Detect vehicles running side by side.

  The imaging unit 2 is a camera attached to the side of the host vehicle 20 so that the optical axis is horizontal and perpendicular to the traveling direction of the host vehicle, and high-speed imaging of about 500 fps so that the rotation of the tire can be detected. Is possible.

  The processing region setting unit 3 sets a plurality of regions including feature points that are candidates for rotating bodies indicating the characteristics of the vehicle as processing regions for the image captured by the imaging unit 2. Usually, the processing area is defined by four parameters: position (horizontal position and vertical position) and size (horizontal size and vertical size). Therefore, by setting the four parameters to various values, the processing region setting unit 3 sets a processing region according to the parameter value.

  The motion detection unit 4 detects the horizontal motion and the vertical motion of the feature points based on the images in the processing region set in a plurality of images having different imaging times. In general, a motion can be expressed by an optical flow. For example, Non-Patent Document 1 discloses a method for detecting an optical flow in detail.

[Non-Patent Document 1]
Bruce D. Lucas and Takeo Kanade, "An Iterative Image Registration Technique with an Application to Stereo Vision", IJCAI, pp.674-679, 1981.
Although the method disclosed in Non-Patent Document 1 is widely used, it is difficult to continuously perform processing at the frame rate of a high-speed camera due to high processing cost. Therefore, in this embodiment, a motion detection method using a high-speed camera disclosed in Patent Document 2 is used.

[Patent Document 2]
In the method disclosed in Japanese Patent Laid-Open No. 2005-209155, the horizontal movement shown in FIG. 3B and the horizontal movement shown in FIG. Since the vertical motion shown can be detected, the motion detection unit 4 outputs the horizontal motion and the vertical motion, respectively.

  The average motion estimator 5 sets a predetermined center region 30 at the center of the processing region as shown in FIG. 3B, and calculates the average value of the horizontal motion detected in the center region 30 to obtain feature points. The average value of movement is calculated.

  The relative motion detection unit 6 subtracts the average value of the motion detected by the average motion estimation unit 5 from the motion of the feature point detected by the motion detection unit 4, thereby calculating the relative motion of the feature point with respect to the central region 30. Looking for. For example, when the relative movement of the feature points is obtained based on the horizontal movement shown in FIG. 3B, the result shown in FIG. 4 is obtained.

  The histogram creation unit 7 creates a histogram based on the relative movement obtained for each feature point in the processing region. First, the angular velocity is calculated from the relative motion detected by the relative motion detector 6. Here, as shown in FIG. 5, when the angular velocity is ω, the distance from the center is r, and the angle is θ, the velocity is rω. When this is decomposed in the horizontal direction and the vertical direction, rωcos θ and rωsin θ are obtained. These are the horizontal motion and the vertical motion actually observed by the motion detector 4. Then, the angular velocity ω can be obtained by dividing each of these rω cos θ and rω sin θ by r cos θ and r sin θ. In addition, when the method of Non-Patent Document 1 is adopted in the motion detection unit 4, the detected motion (optical flow) is rω, so that the angular velocity ω can be obtained simply by dividing by r. However, if any of r, cos θ, and sin θ takes a value as close to zero as possible, it becomes unstable when division is performed. Therefore, it is necessary to prevent further processing from being performed at such a feature point. is there. Therefore, in the present embodiment, as shown in FIG. 6, a region obtained by removing the central region 65 (which may be the same as the central region 30 used in the average motion detecting unit 5) from the processing region 60 is divided into four equal parts, thereby obtaining a region 61. In the region 63, the angular velocity ω is obtained only from the horizontal motion, and in the regions 62 and 64, the angular velocity ω is obtained only from the vertical motion. The angular velocity ω thus obtained creates a histogram as shown in FIG. 7 by performing a voting process that divides the sections with a predetermined value width and accumulates the values of the angular velocities ω for each section. The histogram is preferably smoothed to eliminate the influence of noise.

  The rotation determination unit 8 determines whether or not the image of the processing region is a rotating body based on the distribution of the histogram created by the histogram creation unit 7, and thereby detects the vehicle. When the image of the processing area is a rotating body, the angular velocity ω has the same value at all points, and the histogram has a sharp and sharp shape. Therefore, using this property, the rotation determination unit 8 determines that the image of the processing area is a rotating body when the variance of the histogram is equal to or less than a predetermined value. However, when the histogram includes a plurality of peaks, it is difficult to consider it as a single rotating body, so it is determined that it is a non-rotating body. When it is determined that the image of the processing area is a rotating body, the rotation determination unit 8 determines that the captured image is a vehicle and detects the vehicle.

[Vehicle detection processing]
Hereinafter, the procedure of the vehicle detection process according to the present embodiment will be described with reference to the flowchart shown in FIG.

  As shown in FIG. 8, in the vehicle detection process according to the present embodiment, first, in step S <b> 101, the side of the host vehicle 20 is imaged by the imaging unit 2, and the side surface of the vehicle traveling around the host vehicle 20 is detected. Take an image.

  Next, in step S <b> 102, the processing area setting unit 3 sets a processing area for the image captured by the imaging unit 2. This processing region is a region that includes feature points that indicate the characteristics of the vehicle, and more specifically, is a region that includes feature points that are candidates for rotating bodies. The processing area setting unit 3 defines a processing area with four parameters of position (horizontal position and vertical position) and size (horizontal size and vertical size), and this parameter is set to a predetermined value from an initial value to an end value. The following processing is executed while changing the step size. In step S103, the motion detection unit 4 detects the motion of the feature point based on the images of the processing areas set in the plurality of images having different imaging times. The movement of the feature point detected here detects the movement of the feature point in two directions, a horizontal movement in a direction parallel to the road surface and a vertical movement in the vertical direction.

  Next, in step S104, the average motion estimation unit 5 sets a predetermined center region 30 at the center of the processing region, and calculates the average value of the horizontal motion detected in the center region 30, thereby calculating the motion of the feature point. Find the average value.

  When the average value of motion is thus obtained, in step S105, the processing area setting unit 3 determines whether or not the average value of motion is less than a predetermined threshold value set in advance. For example, a value close to 0 can be set as the threshold value. When the average value of the motion is less than the threshold value, it is determined that the processing region does not include a rotating body, and the process returns to step S102. The process proceeds to S106 and the process is continued.

  In step S <b> 106, the relative motion detection unit 6 subtracts the average value of the motion detected by the average motion estimation unit 5 from the motion of the feature point detected in the processing region, thereby relative to the center region 30. Seeking a special move. In step 107, the histogram creation unit 7 creates a histogram based on the relative motion obtained for each feature point in the processing region. As a specific method for creating a histogram, first, an angular velocity is calculated from a relative motion, and this angular velocity is accumulated for each predetermined value width to create a histogram.

  Next, in step S108, the rotation determination unit 8 determines whether the image of the processing area is a rotating body based on the histogram. Specifically, when the variance of the histogram is equal to or less than a predetermined value, the image in the processing area is determined to be a rotating body. And if it determines with it being a rotary body, the rotation determination part 8 will determine with the imaged image being a vehicle, and will detect a vehicle.

  When the determination as to whether or not the object is a rotating body is made in this way, in step S109, the processing area setting unit 3 sets the values of four parameters (horizontal position, vertical position, horizontal size, vertical size) that define the processing area. It is determined whether the end value has been reached. The vehicle detection process according to the present embodiment is performed while changing four parameters from an initial value to an end value with a predetermined step size. Therefore, when the parameter does not reach the end value, the process returns to step S102 to change the step size and repeat the above-described processing. When the parameter reaches the end value, the vehicle detection processing according to the present embodiment ends. To do.

[Effect of the embodiment]
As described above, according to the vehicle detection device according to the present embodiment, the rotational motion of the tire is detected by determining whether or not the image of the processing area is a rotating body, so that it is possible regardless of the vehicle type. Even a vehicle of any vehicle type can be detected by the same process. For example, as shown in FIG. 9, the shape of a normal passenger car and the shape of a large truck are greatly different, but the characteristics of a tire that is a rotating body are not changed. Therefore, in the conventional method shown in Patent Document 1, since the vehicle is detected from the ratio of tires in the histogram of the entire vehicle, it has to be dealt with according to the difference in the type of vehicle such as a truck or a normal vehicle. According to the vehicle detection device of the present invention, it is possible to detect any vehicle by the same process regardless of the type of vehicle.

  Further, according to the vehicle detection device according to the present embodiment, since the movement of the feature point in two directions, that is, the horizontal movement in the direction parallel to the road surface and the vertical movement in the direction perpendicular to the road surface is detected, The movement can be detected separately. As a result, the horizontal movement component that includes the movement of the vehicle in addition to the rotation of the tire and the vertical movement component that does not include the movement of the vehicle can be detected separately, and the vehicle can be detected more accurately.

  Furthermore, according to the vehicle detection device according to the present embodiment, since the setting of the processing region is canceled when the average value of movement is less than a predetermined threshold, it is possible to easily exclude objects other than the rotating body from the detection target. This can reduce the time and cost required for vehicle detection.

  Further, according to the vehicle detection device according to the present embodiment, the angular velocity is obtained from the relative movement, and the histogram is created by accumulating the angular velocity for each predetermined value width. A histogram can be created based on the angular velocities that are equal even if detected from the vehicle, and thus the vehicle can be accurately detected.

  Furthermore, according to the vehicle detection device according to the present embodiment, when there are a plurality of peaks in the histogram, it is determined that the image of the processing region is not a rotating body. It can be easily determined whether or not it is a rotating body by utilizing the fact that there is one peak.

  As mentioned above, although the vehicle detection apparatus of this invention was demonstrated based on embodiment of illustration, this invention is not limited to this, The structure of each part is replaced with the thing of the arbitrary structures which have the same function. Is possible.

DESCRIPTION OF SYMBOLS 1 Vehicle detection apparatus 2 Imaging part 3 Processing area | region setting part 4 Motion detection part 5 Average motion estimation part 6 Relative motion detection part 7 Histogram creation part 8 Rotation determination part 20 Own vehicle 30 Central area | region

Claims (6)

Imaging means capable of imaging a side surface of the traveling vehicle;
Processing area setting means for setting, as a processing area, an area including a feature point indicating characteristics as a vehicle for the image captured by the imaging means;
Motion detection means for detecting motion of the feature points based on images of the processing region set in a plurality of images having different imaging times;
Average motion estimation means for obtaining an average value of motion of the feature points detected by the motion detection means;
A relative motion detecting means for obtaining a relative motion of the feature point by subtracting an average value of the motion from the motion of the feature point;
A histogram creating means for creating a histogram based on the relative movement obtained for each feature point in the processing region;
A vehicle detection apparatus comprising: a rotation determination unit that detects a vehicle by determining whether or not an image of the processing region is a rotating body based on the distribution of the histogram.   The vehicle detection device according to claim 1, wherein the motion detection unit detects a motion of the feature point in two directions of a horizontal motion in a direction parallel to a road surface and a vertical motion in a vertical direction.   The vehicle detection device according to claim 1, wherein the processing region setting unit cancels the setting of the processing region when the average value of the movement is less than a predetermined threshold.   The said histogram preparation means produces | generates a histogram by calculating | requiring angular velocity from the said relative motion, and accumulating the said angular velocity for every predetermined value range, The Claim 1 characterized by the above-mentioned. Vehicle detection device.   The said rotation determination means determines with the image of the said process area | region not being a rotary body, when the some peak exists in the said histogram, The any one of Claims 1-4 characterized by the above-mentioned. Vehicle detection device. An imaging step of capturing an image of the side of the traveling vehicle with an imaging means;
A processing region setting step for setting a region including a feature point indicating a feature as a vehicle as a processing region for the image captured in the imaging step;
A motion detection step of detecting a motion of the feature point based on images of the processing region set in a plurality of images having different imaging times;
An average motion estimation step for obtaining an average value of the motion of the feature points detected in the motion detection step;
A relative motion detection step of obtaining a relative motion of the feature point by subtracting an average value of the motion from the motion of the feature point;
A histogram creation step for creating a histogram based on the relative movement obtained for each feature point in the processing region;
A vehicle detection method comprising: a rotation determination step of detecting a vehicle by determining whether an image of the processing region is a rotating body based on the distribution of the histogram.

Images