JP4764278B2 - Collision risk determination device, collision risk determination method, and pedestrian identification method - Google Patents

Description

  The present invention relates to a collision risk determination device, a collision risk determination method, and a pedestrian identification method for risk determination that determine a risk of a vehicle colliding with a pedestrian within a travel prediction range of the vehicle traveling direction. About.

  2. Description of the Related Art Conventionally, a vehicle that detects a pedestrian in the travel prediction range in the traveling direction on the vehicle side and generates a warning to the driver has been put into practical use. In order to detect a pedestrian, for example, an obstacle detection device using a neural network (NN: Neural Network) (for example, see Patent Document 1 below) is used.

  The obstacle detection device described above detects a pedestrian as an obstacle. The moving direction of the vehicle is photographed at predetermined time intervals, and the obtained image data is binarized to obtain pedestrian shape data. Pedestrian movement pattern is extracted and detected as a pedestrian movement pattern from temporal changes in shape data between multiple visible images, and pedestrians are compared by comparing the pedestrian movement pattern and the movement pattern learned by the neural network. Identifying.

At this time, the neural network is made to learn a series of movement data accumulated in time series in advance as a motion pattern of a detection target pedestrian. For example, for actual humans, shape data and movement data are measured in advance under conditions such as physique, clothing, walking, and direction, and these data are learned as human movement patterns as neural network teacher data. Let me. Then, pattern recognition is performed using the shape data and movement data obtained from the visualization camera, and it is determined whether or not the person is an actual person.
JP 2004-145660 A (paragraphs 0016 to 0022)

  However, since the obstacle detection device described above is based on the premise that the pedestrian is photographed with a visualization camera or the like, the pedestrian is hidden behind the obstacle or is affected by light such as backlight. However, there is a problem in that it is impossible to determine the degree of danger that a vehicle will collide with a pedestrian because each pedestrian cannot be detected.

  In addition, since the obstacle detection device described above is only targeted when a pedestrian is standing, even if the posture of the pedestrian changes due to squatting, falling down or running while crossing the road In order to make it possible to detect the pedestrian, a variety of motion patterns had to be learned by a neural network, or many neural networks had to learn each motion pattern. As described above, when an apparatus for generating an alarm for a driver is configured using an obstacle detection apparatus that can cope with various postures of a pedestrian, there is a problem that the apparatus cost is significantly increased.

  The present invention was made in order to solve the above-mentioned problems, and its purpose is that an upright pedestrian falls, is hidden behind an obstacle, or is affected by light such as backlight, An object of the present invention is to provide a collision risk determination device and a collision risk determination method capable of determining the risk of a vehicle colliding with a pedestrian even when it is not found on the way.

  Another object of the present invention is to provide a collision risk determination device and a collision risk determination method capable of suppressing a significant increase in device cost.

  Another object of the present invention is to provide a pedestrian identification method capable of identifying a pedestrian even when an upright pedestrian is not found on the way.

Further, the collision risk determination device according to the present invention, the pedestrian detection means for detecting the pedestrian by detecting the pedestrian by pattern recognition by imaging the travel prediction range of the traveling direction of the vehicle,
Obstacle detection means for detecting an obstacle including the pedestrian located in a travel prediction range in the traveling direction of the vehicle, and detecting the position of the obstacle;
Pedestrian information storage means for storing the position of the pedestrian in time series for each pedestrian detected by the pedestrian detection means;
Obstacle information storage means for storing the position of the obstacle for each obstacle detected by the obstacle detection means;
When the pedestrian that has been detected by the pedestrian detection unit is no longer detected by the pedestrian detection unit, the position of the pedestrian stored in the pedestrian information storage unit is changed from time-series changes to the pedestrian. Pedestrian tracking means for tracking a person's movement trajectory, estimating the current position of the pedestrian, and storing the pedestrian information;
Pedestrians / obstacles that judge the pedestrians and non-pedestrians in the obstacles by comparing the information stored in the obstacle information storage unit with the information stored in the pedestrian information storage unit Object verification means;
The current position of the pedestrian estimated by the pedestrian tracking means, the position stored in the obstacle information storage means of the obstacle determined to be a non-pedestrian by the pedestrian / obstacle checking means, A risk determination means for determining the risk of the vehicle colliding with the pedestrian that is no longer detected by the pedestrian detection means, taking into account the relative speed with the vehicle position and the speed of the vehicle. And
Prepare.

In addition, the collision risk determination device according to the present invention captures a travel prediction range in the traveling direction of the vehicle, detects a pedestrian by pattern recognition, and detects an pedestrian position from an external pedestrian detection unit. Pedestrian information storage means for storing the position of the pedestrian in time series for each pedestrian detected by the pedestrian detection means in response to information;
In response to information from an external obstacle detection means for detecting an obstacle including the pedestrian located in the predicted travel range of the vehicle in the traveling direction, and detecting the position of the obstacle, the obstacle detection means Obstacle information storage means for storing the position of the obstacle for each detected obstacle ;
When the pedestrian that has been detected by the pedestrian detection unit is no longer detected by the pedestrian detection unit , the position of the pedestrian stored in the pedestrian information storage unit is changed from time-series changes to the pedestrian. Pedestrian tracking means for tracking a person's movement trajectory and estimating the current position of the pedestrian;
Pedestrians / obstacles that judge the pedestrians and non-pedestrians in the obstacles by comparing the information stored in the obstacle information storage unit with the information stored in the pedestrian information storage unit Object verification means;
The current position of the pedestrian estimated by the pedestrian tracking means, the position stored in the obstacle information storage means of the obstacle determined to be a non-pedestrian by the pedestrian / obstacle checking means, A risk determination means for determining the risk of the vehicle colliding with the pedestrian that is no longer detected by the pedestrian detection means , taking into account the relative speed with the vehicle position and the speed of the vehicle. And
Prepare.

Further, the collision risk determination method according to the present invention includes a step of detecting a pedestrian by pattern recognition by imaging a travel prediction range in the traveling direction of the vehicle, and detecting the position of the pedestrian;
Detecting an obstacle including the pedestrian located in a travel prediction range in the traveling direction of the vehicle, and detecting the position of the obstacle;
Storing the position of the pedestrian in time series for each detected pedestrian;
Storing the position of the obstacle for each detected obstacle;
When the pedestrian detected in the step of detecting the position is not detected, the movement position of the pedestrian is tracked from a time-series change in the stored position of the pedestrian, Estimating and storing the current position;
The stored position information of the obstacle and the current position information of the pedestrian whose position has been detected or the pedestrian whose position has been estimated are collated with each other, and the pedestrian in the obstacle And determining a non-pedestrian,
Prepare.

Further, the pedestrian identification method according to the present invention includes a step of detecting a pedestrian by pattern recognition by imaging a travel prediction range in the traveling direction of the vehicle, and detecting the position of the pedestrian;
Detecting an obstacle including the pedestrian located in a travel prediction range in the traveling direction of the vehicle, and detecting the position of the obstacle;
Storing the position of the pedestrian in time series for each detected pedestrian;
Storing the position of the obstacle for each detected obstacle;
When the pedestrian detected in the step of detecting the position is not detected, the movement position of the pedestrian is tracked from a time-series change in the stored position of the pedestrian, Estimating and storing the current position;
The stored position information of the obstacle and the current position information of the pedestrian whose position has been detected or the pedestrian whose position has been estimated are collated with each other, and the pedestrian in the obstacle And determining a non-pedestrian,
Prepare.

According to the collision risk determination device according to the present invention, a pedestrian within the travel prediction range in the traveling direction of the vehicle and an obstacle including the pedestrian are detected, and each position of the pedestrian and the obstacle is detected. Further, the position of the pedestrian is stored for each pedestrian, the position of the obstacle is stored for each obstacle, and the pedestrian and the non-pedestrian in the obstacle are determined and stored by a predetermined determination . For the currently detected pedestrian, taking into account the relative speed between the current position of the pedestrian, the position of the obstacle determined to be a non-pedestrian by comparison, and the position of the vehicle, including the speed of the vehicle. Since the risk of collision of the vehicle is determined, the risk of collision of the vehicle with the currently detected pedestrian can be predicted.

  Moreover, according to the collision risk determination device according to the present invention, when the pedestrian detected by the pedestrian detection unit is no longer detected by the pedestrian detection unit, the walking stored in the pedestrian information storage unit. The pedestrian's position is tracked from time-series changes, the pedestrian's movement trajectory is tracked, the pedestrian's current position is estimated, and the estimated pedestrian's current position is compared with the obstacles that are determined as non-pedestrians Since the relative relationship between the position and the position of the vehicle and the speed of the vehicle is taken into account as necessary, the risk of the vehicle colliding with the pedestrian is judged. Even if the vehicle cannot be found on the way, it is possible to determine the risk of the vehicle colliding with the pedestrian.

  Further, according to the collision risk determination device according to the present invention, the pedestrian information storage means detects the pedestrian by pattern recognition, images the travel prediction range in the traveling direction of the vehicle, and detects the position of the pedestrian. In response to information from the external pedestrian detection means, the position of the pedestrian is stored in time series for each pedestrian detected by the pedestrian detection means, and the obstacle information storage means For each obstacle detected by this obstacle detection means in response to information from an external obstacle detection means that detects obstacles including pedestrians located in the predicted travel range and detects the position of this obstacle Since the position of the obstacle is stored, even if the photographing means, the pedestrian detection means and the obstacle detection means are installed in a predetermined place other than the vehicle, an upright pedestrian cannot be found on the way. Even if It is possible to determine the risk of the vehicle collides with the pedestrian.

  Further, according to the pedestrian identification method according to the present invention, when the detected pedestrian is no longer detected, the movement trajectory of the pedestrian is tracked from the time-series change in which the position of the pedestrian is stored, Since the current position of the pedestrian is estimated and collated with the position of the obstacle, the pedestrian can be identified even when an upright pedestrian is not found on the way.

<First Embodiment>
Hereinafter, the present invention will be described in detail based on preferred embodiments shown in the drawings. FIG. 1 is a block diagram showing the configuration of the first embodiment of the collision risk degree judging device according to the present invention. The photographing means 1, the pedestrian detection means 2, the obstacle detection means 3, and the pedestrian information of the current frame. A storage means 4A, an obstacle information storage means 4B, a pedestrian information storage means 4C, a pedestrian / obstacle checking means 5, a pedestrian tracking means 6 and a risk level determination means 7 are provided. An alarm signal is applied to the installed danger alarm means 8. In addition, the “travel prediction range” used in this specification means that when a traveling vehicle advances in the traveling direction as it is, a predetermined value (for example, 1 m in total allocated to both sides) is added to the width of the vehicle. It is the forward driving range that is predicted to be the driving trajectory of a wide vehicle when it is assumed that a wide vehicle travels, and the `` running lane '' is a traffic rule defined by the white line of the road etc. It is a driving range that should be protected.

  Among the components shown in FIG. 1, the photographing means 1 is a vehicle (not shown in FIG. 1, but is used in the same meaning as the automobile 10 shown in FIG. 2 to be described later. For example, a visualization camera and a travel prediction range in the traveling direction of the vehicle, which can capture a travel prediction range in the traveling direction of the vehicle and recognize a pattern from the image. For example, a radar transmission / reception unit that can measure the number of obstacles including pedestrians and the distance from the vehicle is provided. Note that a stereo camera can be used in place of the radar transmission / reception unit, and in this case, one of the stereo cameras can also be used as a visualization camera instead of the CCD camera.

  The pedestrian detection means 2 binarizes the image data obtained by the visualization camera of the photographing means 1 and extracts it as obstacle shape data, and moves temporal changes in the shape data between a plurality of visible images. A pedestrian is identified by comparing this movement pattern with a movement pattern of a pedestrian previously learned by a neural network. Further, the pedestrian detection means 2 calculates the position of each identified pedestrian for each frame of the photographing means 1 and stores it in the pedestrian information storage means 4A of the current frame as a part of the pedestrian information. As a method for calculating the distance from the vehicle to the pedestrian by the pedestrian detection means 2, for example, the geometric information is converted from the positional information of the pedestrian's foot part in the coordinates on the image to the positional information in the world coordinates with the vehicle as the origin. There is a way to convert.

  The obstacle detection means 3 receives the image data obtained by the radar transmission / reception unit of the photographing means 1 and based on the information about each block image obtained by dividing each image into a plurality of areas, the obstacle (pedestrian) Is calculated for each frame of the photographing means 1 and stored in the obstacle information storage means 4B as obstacle information.

Here, the position of the pedestrian calculated by the pedestrian detection means 2 and the position of the obstacle calculated by the obstacle detection means 3 will be described. FIG. 2A is a schematic plan view showing how pedestrians, obstacles including pedestrians, and automobiles are arranged with respect to the predicted driving range of the automobile. In FIG. 2A, it is assumed that the automobile 10 travels from the lower side to the upper side at a speed V per hour (in the following description, V is also used as a relative speed between the car and the pedestrian). At that time, let W be the width of the above-described travel prediction range (range surrounded by a one-dot chain line) of the automobile 10. The imaging means 1 shown in FIG. 1 captures the predicted travel range, but here it is present in the predicted travel range of the automobile 10 in order to explain in detail the positions of obstacles including pedestrians and pedestrians. Obstacles O ₁ , O ₂ , O ₃ , and O ₄ are shown enlarged. Of these obstacles O ₁ , O ₂ , O ₃ and O ₄ , the obstacle O ₂ is a pedestrian P ₁ and the obstacle O ₃ is a pedestrian P ₂ .

Then, in order to detect the positions of these obstacles O ₁ , O ₂ , O ₃ , O ₄ and pedestrians P ₁ , P ₂ from the image, the center in the left-right direction at the forefront of the automobile 10 is set as the origin 0, An axis Z is defined in the traveling direction of the automobile 10 so as to pass through the origin 0, and an axis X is defined in the lateral width direction. As a result, an XZ coordinate system and an azimuth coordinate system represented by θ are set. In this case, the left side of the axis Z is expressed as a negative value in the XZ coordinate system and the azimuth coordinate system, and the right side of the axis Z is expressed as a positive value in the XZ coordinate system and the azimuth coordinate system. Note that the width W of the travel prediction range in the traveling direction of the automobile 10 is not limited to the width of the automobile 10 plus 1 m, and can be appropriately changed and set according to the structure of the road.

2A shows the details of the position of the pedestrian P ₁ when the pedestrian detection means 2 mounted on the automobile 10 calculates the positions of the pedestrians P ₁ and P ₂ , and the obstacle detection means 3 Details of the position of the obstacle O ₁ in the case of calculating the positions of the obstacles O ₁ , O ₂ , O ₃ , and O ₄ are shown. That is, the pedestrian detection means 2 uses the azimuth coordinate system distance d from the own vehicle, the minimum x coordinate x ₁ in the XZ coordinate system from the own vehicle, and the XZ coordinate system from the own vehicle as the position of the pedestrian P _1. The maximum x coordinate x ₂ , the minimum θ direction θ ₁ in the direction coordinate system from the own vehicle, and the maximum θ direction θ ₂ in the direction coordinate system from the own vehicle are calculated. On the other hand, the obstacle detection means 3 determines the position of the obstacle O ₁ as the shortest distance d _min in the azimuth coordinate system from the own vehicle, the longest distance d _max in the azimuth coordinate system from the own vehicle, and the azimuth coordinates from the own vehicle. Minimum θ azimuth θ _{1 in} the system, maximum θ azimuth θ _{2 in} the azimuth coordinate system from the host vehicle, and minimum x coordinate x _{1 in} the XZ coordinate system from the host vehicle (omitted in FIG. 2 because the drawing becomes complicated) Then, the maximum x coordinate x _{2 in} the XZ coordinate system from the own vehicle is calculated (omitted in FIG. 2 because the drawing becomes complicated).

FIG. 2B shows the pedestrian detection means 2 detecting pedestrians P ₁ and P ₂ based on the image data obtained by the visualization camera of the photographing means 1, and for these pedestrians P ₁ and P _2. As described above, the distance d in the azimuth coordinate system from the own vehicle, the minimum x coordinate x ₁ in the XZ coordinate system from the own vehicle, the maximum x coordinate x _{2 in} the XZ coordinate system from the own vehicle, It shows that the minimum θ azimuth θ ₁ in the azimuth coordinate system and the maximum θ azimuth θ ₂ in the azimuth coordinate system from the own vehicle are calculated and the calculation result is output.

FIG. 2C shows obstacles O ₁ , O ₂ , O ₃ , O ₄ based on the image data obtained by the radar transmission / reception unit of the photographing means 1, and these obstacles O ₁ , O ₂ , As described above for O ₃ and O ₄ , the minimum θ direction θ ₁ in the direction coordinate system from the own vehicle, the maximum θ direction θ ₂ in the direction coordinate system from the own vehicle, and the direction coordinate system from the own vehicle The shortest distance d _{min in} the vehicle, the longest distance d _max in the azimuth coordinate system from the vehicle, the minimum x coordinate x ₁ in the XZ coordinate system from the vehicle, and the maximum x coordinate x _{2 in} the XZ coordinate system from the vehicle. Is calculated, and the calculation result is output.

  The pedestrian information storage means 4A for the current frame updates and stores the calculation result of the pedestrian detection means 2 as pedestrian information each time the travel prediction range in the traveling direction of the vehicle is photographed by the visualization camera of the photographing means 1. In the obstacle information storage unit 4B, the position calculated by the obstacle detection unit 3 is updated and stored as obstacle information each time the radar transmission / reception unit of the imaging unit 1 captures the travel prediction range in the traveling direction of the vehicle. .

  Here, for convenience of explanation, a data storage unit of the pedestrian information storage unit 4C and a data storage unit of the obstacle information storage unit 4B will be described. FIG. 6A is a chart showing the configuration of the storage unit of the pedestrian information storage unit 4C, and is associated with each pedestrian ID (pedestrian identification code) by the visualization camera of the imaging unit 1 at a predetermined time interval. A plurality of pieces of pedestrian information photographed in (1) are stored in the order of the current frame t, the previous frame t-1, the previous frame t-2,. The entire stored data shown in FIG. 6A is hereinafter referred to as a pedestrian list.

FIG. 6B is a chart showing the pedestrian information of one frame, for example, the current frame t of the pedestrian ID 1, and is associated with “numbers 1 to 16” in the “variable data” column on the right side. The information or code described in the “content” column is written. Among these, “Appear”, “d” to “θ ₂ ”, “est_x ₁ ” to “est_v _θ ” in the “variable data” column are written by the pedestrian tracking means 6, which will be described in detail later, and “Danger_type” to “Danger_type” to “ “Hidden_by_Obs” is written by the risk determination means 7 described later in detail. The information written in “d” to “θ ₂ ” in the “variable data” column matches the pedestrian list (not shown) written in the pedestrian information storage unit 4A of the current frame.

FIG. 7 (a) is a chart showing the data storage unit of the obstacle information storage unit 4B. The radar transmission / reception unit of the photographing unit 1 corresponds to the obstacle numbers of obstacles including pedestrians at predetermined time intervals. The obstacle information of the current frame detected in is stored. The entire stored data shown in FIG. 7A is hereinafter referred to as an obstacle list. FIG. 7B is a chart showing the obstacle information of the current frame of one obstacle. Each of the “variable data” fields is associated with “numbers 1 to 7” in the “content” field on the right side. The described information or code is written. Of these, “d _min ” to “Ox ₂ ” in the “variable data” column is written with the value calculated by the obstacle detection means 3, and “Is_Pede” is written with the verification result by the pedestrian / obstacle verification means 5. It is.

  Next, the pedestrian / obstacle checking means 5, the pedestrian tracking means 6, and the risk level judging means 7 will be described in order. The pedestrian / obstacle collation means 5 includes an obstacle whose obstacle information is stored in the obstacle information storage means 4B for the first time and a pedestrian whose pedestrian information is stored in the pedestrian information storage means 4A of the current frame. Judgment is made based on whether the position of the obstacle and the position of the pedestrian are substantially the same for all combinations, and in the following description, the determination and the determination are used in the same meaning) The determination result is written in the “Is_Pede” field of the obstacle list of the obstacle information in the obstacle information storage means 4B, and the obstacle and pedestrian information in which the obstacle information is stored in the obstacle information storage means 4B are used next time. Combining pedestrians whose pedestrian information is stored in the storage means 4C, for each combination, it is determined whether or not the position of the obstacle and the position of the pedestrian substantially coincide with each other. The judgment result is displayed as the obstacle information storage means 4B. This is written in the “Is_Pede” column of the list.

The pedestrian tracking means 6 writes the pedestrian information stored in the pedestrian information storage means 4A for the current frame for the first time as part of the pedestrian information for the current frame t. From the next time, for each pedestrian whose pedestrian information is stored in the pedestrian information storage means 4C, it is determined whether or not the pedestrian is stored in the pedestrian information storage means 4A of the current frame. If it is, the flag in the “Appear” column of the pedestrian list in the pedestrian information storage means 4C is set to “TRUE”. On the other hand, if the pedestrian is not stored in the pedestrian information storage means 4A of the current frame, the new position of the pedestrian is estimated and the estimated position information is stored in “est_x” of the pedestrian list of the pedestrian information storage means 4C. ₁ ”to“ est_v _θ ”are written, and the flag in the“ Appear ”column of the pedestrian list of the pedestrian information storage means 4C is set to“ FALSE ”because it cannot be detected in the current frame.

  The risk determination means 7 determines whether or not the pedestrian not found in the current frame is within the predicted travel range of the vehicle, and if so, It is determined whether or not the vehicle is behind. If the vehicle is behind an obstacle other than a pedestrian, the possibility of jumping out of the obstacle is determined. Furthermore, if it is determined that the pedestrian is not behind an obstacle other than the pedestrian (including the case where there is no obstacle other than the pedestrian in the same direction as the pedestrian), the vehicle travels from this obstacle. For both cases of jumping into the predicted range, the predicted collision time (TTC: hereinafter simply the collision time) to the pedestrian within the predicted travel range of the host vehicle using the relative speed V between the host vehicle and the pedestrian. In addition, the risk state is determined, and each determination result is written in the “Danger_type” to “Hidden_by_Obs” column of the pedestrian list of the pedestrian information storage means 4C, and according to the risk level. A signal for operating the danger warning means 8 is output. The signal for activating the danger warning means 8 is a signal for issuing an alarm to the driver, activating a braking device, or activating a protection device for protecting a pedestrian.

  About operation | movement of the collision risk determination apparatus comprised as mentioned above, the pedestrian and obstacle collation means 5, the pedestrian tracking means 6, and the risk determination means 7 are comprised by one or several arithmetic processing units, for example, CPU. 3 and 4A to 4E and FIGS. 5A to 5F showing specific processing procedures in this case.

FIG. 3 is a flowchart showing a specific processing procedure when the pedestrian / obstacle checking means 5 is constituted by an arithmetic processing unit. In FIG. 3, one obstacle O _i (i = 1, 2,...) Is selected from the obstacle list stored in the obstacle information storage means 4B in step 11, and in step 12, the pedestrian information storage means 4C walks. One pedestrian P _j (j = 1, 2,...) Is selected from the person list. Since the position information initially stored in the pedestrian information storage unit 4C is the same as the position stored in the pedestrian information storage unit 4A for the current frame, the pedestrian list in the pedestrian information storage unit 4A for the current frame. A single pedestrian P _j may be selected.

Then, "Appear" column of the flag of the pedestrian P _j at step 13 pedestrian list of the pedestrian information storing unit 4C whether there the current frame is determined by whether "TRUE" or "FALSE". If it is determined that the pedestrian P _j is in the current frame, in step 14, the shortest distance O _i .d _min of the obstacle O _i and the distance P _j .d to the pedestrian P _j are substantially equal (this specification). is used in the sense that substantially equal includes if they match in writing), and the obstacle O _i Min θ orientation O _i .θ ₁ minimum θ direction P _j .Shita ₁ and is approximately equal to the pedestrian P _j of In addition, it is determined whether or not the condition that the maximum θ direction O _i .θ _{2 of} the obstacle O _i is substantially equal to the minimum θ direction P _j .θ _{2 of the} pedestrian P _j is satisfied. When established, the obstacle O _i is determined to be a pedestrian, and in step 16 the flag in the “Is_Pede” column of the obstacle list of the obstacle information storage means 4B is set to “TRUE” and the process proceeds to step 17. If the condition is not satisfied, the process proceeds to step 17 as it is.

On the other hand, if it is determined in step 13 that the pedestrian P _j is not in the current frame, in step 15, the minimum x coordinate O _i .x ₁ of the obstacle O _{i and} the minimum x coordinate of the pedestrian P _j not in the current frame are determined. P _j .Est_x ₁ and is approximately equal, and the maximum x coordinate P _j .Est_x ₂ and is substantially equal to the pedestrian P _j to the maximum x coordinate O _i .x ₂ obstacle O _i not in the current frame, and, minimum θ direction P _j .Est_shita ₁ and is approximately equal to the minimum θ orientation O _i .θ ₁ and not to the current frame pedestrian P _j obstacle O _i, and the maximum θ azimuth orientation of the obstacle O _i O _i .θ ₂ and the maximum θ direction P _j .est_θ _{2 of the} pedestrian P _j not in the current frame is determined whether or not the condition is satisfied, and when the condition is satisfied, the obstacle O _i walks In step 16, the flag in the “Is_Pede” column of the obstacle list of the obstacle information storage means 4B is set to “TRUE”. The process proceeds to the step 17 Te, the process proceeds to the processing of step 17 when the condition is not satisfied.

In step 17, the process for the pedestrian P _i combined with one obstacle O _i selected from the obstacle information storage unit 4B is executed up to the last pedestrian in the pedestrian list of the pedestrian information storage unit 4C. If it has not been executed to the end, the process returns to the step 12. If it has been executed to the end, the process proceeds to a step 18. In step 18, it is determined whether or not the process for the obstacle O _i selected in step 11 has been executed to the end of the obstacle list of the obstacle information storage means 4B. Return to the end, and if it has been executed to the end, the series of processing is terminated. By the processing shown in the flowchart of FIG. 3, all the obstacles O _i stored in the obstacle list of the obstacle information storage unit 4B are collated to determine whether they are pedestrians. The flag in the “Is_Pede” column of the obstacle list of the information storage unit 4B is set to “TRUE”.

FIG. 4A is a flowchart showing a specific processing procedure when the pedestrian tracking means 6 is constituted by an arithmetic processing device. In this case, all the data of the pedestrian list of the pedestrian information storage unit 4C and all the data of the pedestrian list (not shown) of the pedestrian information storage unit 4A of the current frame are acquired, and the process is started. In Figure 4A, at step 21 to select one of the pedestrian P _i from the pedestrian list of the pedestrian information storing unit 4C, at step 22, as will be described in greater detail below, the pedestrian the pedestrian information storing unit 4A of the current frame against the pedestrian P _j the list stored in the pedestrian P _i and a flag "Appear" column of the pedestrian list of the pedestrian information storing unit 4C when there are large pedestrian P _j correlation "TRUE "in the, the pedestrian list of the pedestrian information storing unit 4C when large pedestrian P _j of the correlation between the pedestrian P _i is not" run G1 process to Appear "column flag" FALSE ".

Then, the pedestrian information storing unit 4C pedestrian list of the pedestrian P _i of "Appear" column of the flag of "TRUE" or performs "FALSE" Kano determined in step 23. In the case of "FALSE" at step 24, as will be described in greater detail below, it executes the G2 processing for estimating a new position of the pedestrian P _i that is not found. Then, the estimated pedestrian information is stored in the pedestrian information storage unit 4C as estimated pedestrian information. Next, in step 25, as will be described in detail later, G4 processing is performed to determine whether or not the pedestrian who has estimated the new position has gone too far from the front and front area of the vehicle, that is, whether or not the pedestrian has gone too far. Execute. Here, G4_flag is set to “TRUE” if it has gone too far, and G4_flag is set to “FALSE” if it has not gone too far. Then, G4_flag it is determined whether "TRUE" or "FALSE" at step 26, to delete the pedestrian P _i and pedestrian information to estimate if "TRUE" the process proceeds to step 30, G4_flag is "FALSE "", The process proceeds to step 30 as it is.

On the other hand, when the flag in the “Appear” column of the pedestrian P _i in the pedestrian list of the pedestrian information storage means 4C is “TRUE” in step 23, in step 28, as will be described in detail later, G3 processing for updating the pedestrian information in the pedestrian information storage means 4C is executed using the information stored in the pedestrian information storage means 4A as the latest pedestrian information, and then the pedestrian information not shown in step 29 storage means 4A of the "Exit_PedeList" column of the flag of the pedestrian P _i in the "TRUE" the process proceeds to step 30.

Next, it is determined whether the execution of the processing for the pedestrian P _i chosen from the pedestrian list of the steps 30 in the pedestrian information storing unit 4C to the end of the pedestrian list of the pedestrian information storing unit 4C, until the end If not, another pedestrian is selected from the pedestrian list of the pedestrian information storage means 4C at step 21, and the processing of steps 22 to 30 is repeated. If it has been executed to the end of the pedestrian list of the pedestrian information storage means 4C in step 30, the process proceeds to step 31.

  In step 31, one pedestrian is selected from the pedestrian list of the pedestrian information storage means 4A of the current frame, and then in step 32 the pedestrian information is flagged in the “Exist_PedeList” column of the pedestrian information storage means 4C. It is determined whether it is “TRUE” or “FALSE”. If it is “TRUE”, the process proceeds to step 34, and if it is “FALSE”, the pedestrian is added to the pedestrian list of the pedestrian information storage means 4C. Proceed to step 34. In step 34, it is determined whether or not the processing for the pedestrian selected from the pedestrian list of the pedestrian information storage means 4A for the current frame in step 31 has been executed to the end of the pedestrian list of the pedestrian information storage means 4A for the current frame. If it has not been executed until the end, in step 31, another pedestrian is selected from the pedestrian list in the pedestrian information storage means 4A for the current frame, and the processing in steps 31 to 34 is repeated. And when it determines with having been performed to the last of the pedestrian list of the pedestrian information storage means 4A of the present frame at step 34, a pedestrian tracking process is complete | finished.

  By the process shown in FIG. 4A, the pedestrian list in the pedestrian information storage unit 4C is updated with the pedestrian list in the pedestrian information storage unit 4A in the current frame. If the pedestrian list is not in the current frame, the position is estimated and estimated. The pedestrian information is stored in the pedestrian information storage means 4C, and the information of the pedestrian who has left outside the front and front side areas of the own vehicle is deleted from the pedestrian list of the pedestrian information storage means 4C, and the current frame If there is a pedestrian newly appearing in the pedestrian information storage means 4A, the pedestrian information is stored in the pedestrian information storage means 4C.

FIG. 4B is a flowchart showing a detailed processing procedure of the G1 processing performed in step 22 shown in FIG. 4A. Here, one pedestrian P _j is selected from the pedestrian list of the pedestrian information storage means 4A for the current frame in step 221, and the image information of the pedestrian P _{j in} the current frame selected in step 221 is selected in step 222. step 21 performs the correlation calculation between the image information of the pedestrian P _i of the pedestrian information storing unit 4C chosen from the pedestrian list (see FIG. 4A). Next, in step 223, it is determined whether or not the obtained correlation value Corr Val exceeds a predetermined threshold Corr Threshold. If the correlation value Corr Val exceeds the threshold Corr Threshold, the pedestrian information storage means 4C is determined in step 224. The pedestrian P _i stored in the pedestrian list of the current frame is recognized as the pedestrian P _j in the pedestrian list of the pedestrian information storage means 4A of the current frame, and the pedestrian P _i of the pedestrian information storage means 4C is recognized. Set the flag in the “Appear” column to “TRUE”. Next, if it is determined in step 223 that the correlation value Corr Val does not exceed the threshold Corr Threshold, and if the flag is set to “TRUE” in step 224, the processing of step 225 is executed. In step 225, whether the flag of each "Appear" column of the pedestrian P _i pedestrian list of the pedestrian information storing unit 4C to all "TRUE", or selected from the pedestrian information storing unit 4A of the current frame It is determined whether or not the processing for the pedestrian P _{j has} been executed to the end of the pedestrian list of the pedestrian information storage means 4A of the current frame. If the processing has not been executed to the end, the processing of steps 221 to 225 is repeated until the end. If it is being executed, the series of processing is terminated. By the processing of FIG. 4B, it is determined whether or not the pedestrian P _{i in} the pedestrian list of the pedestrian information storage unit 4C is in the current frame.

FIG. 4C is a flowchart showing a detailed processing procedure of the G2 process performed in step 24 shown in FIG. 4A. The pedestrian information stored in the pedestrian information storage unit 4C in step 241 is stored in a plurality of past pedestrian information. Based on this, a new position is estimated. This estimation is performed by averaging the time changes of a plurality of position information, or by using linear interpolation or the like. Accordingly, the estimated minimum x coordinate est_x ₁ in the XZ coordinate system from the own vehicle, the estimated maximum x coordinate est_x ₂ in the XZ coordinate system from the own vehicle, and the estimated minimum θ direction est_θ _{1 in} the azimuth coordinate system from the own vehicle. The estimated maximum θ azimuth est_θ _{2 in} the azimuth coordinate system from the own vehicle and the estimated x coordinate speed est_v _{x in} the XZ coordinate system from the own vehicle are calculated and stored as the current position information of the pedestrian in the prone state Is done. These pieces of position information become the pedestrian information in the prone state in the process of step 15 in FIG.

FIG. 4D is a flowchart showing a detailed processing procedure of G4 processing performed in step 25 shown in FIG. 4A. Here, whether or not the magnitude (absolute value) of the estimated minimum θ direction θ ₁ and the estimated maximum θ direction θ ₂ of the pedestrian in the pedestrian list of the pedestrian information storage unit 4C exceeds 120 ° in step 251. It is determined whether or not the pedestrian has gone too far. That is, the range of ± 120 ° with respect to the Z axis of the coordinate system shown in FIG. 2 means a wide angle range including not only the front of the automobile 10 but also the area on the front side of the vehicle, If it is determined in step 251 that the size of the estimated minimum θ azimuth θ ₁ and the estimated maximum θ azimuth θ ₂ exceeds the range of ± 120 °, G4_flag is set to “TRUE” in step 252, and if it does not exceed Sets G4_flag to “FALSE”, recognizes that the pedestrian is lying, and ends the process.

  FIG. 4E is a flowchart showing a detailed processing procedure of the G3 process performed in step 28 shown in FIG. 4A. Here, the position information and the image information of the pedestrian information storage unit 4C are substituted by the pedestrian information of the pedestrian information storage unit 4A of the current frame at step 281. Subsequently, at step 282, the position information and image information are substituted by step 281. According to the new information, the pedestrian tracking means 6 updates the parameter of the filter for estimating the new position in step 24 (see FIG. 4A) and ends the process. Thereby, the position information of the pedestrian information storage unit 4C and the parameter for estimating the new position are updated.

  By the way, in the description so far, both pedestrians and non-pedestrians have been described as obstacles. However, in order to facilitate understanding and to avoid complicated explanations, in the following, other than pedestrians An object will be described as an obstacle.

  FIG. 5A is a flowchart showing a specific processing procedure when the risk level judging means 7 is constituted by an arithmetic processing unit. 5A, in step 41, a pedestrian who has not been found from the pedestrian list of the pedestrian information storage means 4C, that is, a pedestrian in a lying state is selected. Subsequently, in step 42, as will be described in detail later, the pedestrian enters the travel prediction range of the own vehicle, that is, a range of width W obtained by adding 1 meter (m) to the width of the car shown in FIG. Execute S1 processing to determine whether or not

  Next, at step 43, S1_flag indicating the result of the S1 process at step 42 is “TRUE” indicating that a pedestrian in the concealed state is in the travel prediction range of the own vehicle, or the travel prediction of the own vehicle. It is determined whether it is “FALSE” indicating that it is not within the range, and if it is “TRUE”, it is determined in step 44 whether the pedestrian is behind the obstacle as will be described in detail later. , S2 processing is executed. If “FALSE”, the processing returns to step 41. Subsequently, in step 45, S2_flag indicating the result of the S2 process in step 44 is “TRUE” indicating that a pedestrian in the back is behind the obstacle, or behind the obstacle. If it is “FALSE” indicating that the pedestrian is not present, the S3 process is executed in step 46 to determine whether the pedestrian may jump out of the position of the obstacle, and “FALSE” "", The process proceeds to step 48.

  In step 46, as will be described in detail later, based on the estimated position information of the pedestrian and the position information of the obstacle, the determination of the possibility that the pedestrian behind the obstacle will jump out into the travel prediction range of the own vehicle. When there is a possibility of jumping out, S3_flag is set to “TRUE”, and when there is no possibility of jumping out, S3_flag is set to “FALSE”.

  Next, in step 47, S3_flag is “TRUE” indicating that the pedestrian may jump out from behind the obstacle, or there is no possibility that the pedestrian jumps out from behind the obstacle. In the case of “TRUE”, the process of step 48 is executed, and in the case of “FALSE”, the process returns to the process of step 41.

  If it is determined in step 42 that S1_flag is “FALSE” indicating that the vehicle is not in the predicted travel range of the vehicle, the dangerous state is “Type 0” and the pedestrian list of the pedestrian information storage means 4C. Is stored in the “Danger_type” field. If it is determined in step 45 that S2_flag is “TRUE” indicating that there is a pedestrian lying behind the obstacle, the danger state is “Type 1” in the pedestrian information storage means 4C. If it is stored in the “Danger_type” column of the pedestrian list and S2_flag is “FALSE” indicating that there is no pedestrian behind the obstacle in step 45, the danger state is “Type” 2 ”is stored in the“ Danger_type ”column of the pedestrian list of the pedestrian information storage means 4C.

  Next, in step 48, when S2_flag is “FALSE” and there is no possibility that a pedestrian will jump out from behind the obstacle, or S3_flag is “TRUE” and there is a possibility that pedestrian will jump out from behind the obstacle. When there is, the S4 process for calculating the collision time for each pedestrian is executed as will be described in detail later. The collision time TTC measured in this step 48 is stored in the “TTC” column of the pedestrian list of the pedestrian information storage means 4C, and is used for the risk determination. Subsequently, in step 49, as will be described in detail later, S5 processing is executed to determine whether the collision time TTC is close and dangerous because it is smaller than the predetermined value. When it is determined in step 49 that the collision time TTC is smaller than the predetermined value, S5_flag is set to “TRUE”, and when it is determined that the collision time TTC is not smaller than the predetermined value, S5_flag is set to “FALSE”.

  Next, in step 50, it is determined whether S5_flag is "TRUE" indicating that it is close and dangerous, or "FALSE" indicating that it is not dangerous. If "TRUE", an alarm is generated in step 51 Alternatively, the brake is applied by operating the braking device. On the other hand, if “FALSE”, it is determined in step 52 whether or not the processing for the pedestrian selected in step 41 has been executed to the end of the pedestrian list of the pedestrian information storage means 4C. If not, the process returns to step 11, and if it is executed to the end, the series of processes is terminated.

FIG. 5B is a flowchart showing a detailed processing procedure of the S1 process performed in step 42 shown in FIG. 5A. Here, the estimated position information est_x ₁ and est_x ₂ of the pedestrian P _i stored in the pedestrian information storage unit 4C is acquired, and the absolute value | est_x ₁ | of the minimum x coordinate estimated in step 421 is W / 2. It is determined whether or not (W: a predicted travel range of the vehicle, for example, a value obtained by adding 1 m to the vehicle width), and if not less than W / 2, the absolute value of the maximum x coordinate estimated by moving to step 422 It is determined whether or not | est_x ₂ | is W / 2 or less. If it is determined in step 422 that the absolute value | est_x ₂ | of the maximum x coordinate is not less than W / 2, in step 424, S1_flag indicating that the state is not in danger is set to “FALSE”, and the process is terminated. If it is determined in steps 421 and 422 that each is equal to or less than W / 2, S1_flag indicating the danger state is set to “TRUE”, and the process ends. As a result of the processing in steps 421 to 424, it is determined that even a part of the pedestrian in the prone state is in a dangerous state if it is within the predicted travel range of the own vehicle.

  In the flowchart of FIG. 5B, even if some of the pedestrians are in a lying state, it is determined that the vehicle is in a dangerous state if it is within the predicted travel range of the host vehicle. However, instead of the predicted travel range, the road is defined by a white line. It can also be configured to recognize that the vehicle is in a dangerous state if it is in a traveling lane that should be observed in accordance with the traffic rules.

FIG. 5C is a flowchart showing a detailed processing procedure of S2 processing performed in step 44 shown in FIG. 5A. Here, the estimated minimum θ direction est_θ ₁ , estimated maximum θ direction est_θ ₂ , and estimated distance est_d (omitted in FIG. 6 due to the large number of items) of the prone pedestrian P _i have already been acquired, In step 441, the position information of the obstacle O _j , that is, the minimum angle azimuth θ ₁ , the maximum angle azimuth θ ₂ and the shortest distance d _min are acquired. Then, not greater than the estimated minimum angle orientation P _i .Est_shita ₁ minimum angle orientation O _j .θ ₁ obstacle O _j pedestrian P _i at step 442, and the estimated maximum angle orientation of the pedestrian P _i when P _i .Est_shita ₂ is not greater maximum angle orientation O _j .θ ₂ obstacle O _j, i.e., determines whether or not containing pedestrian P _i within the range of angles to view the obstacle O _j To do.

If it is determined that the pedestrian P _i is within the range of the angle at which the obstacle O _j is viewed, the process proceeds to step 443, and the shortest distance O _j .d _{min of the} obstacle O _j is estimated as P _i . It is determined whether or not the distance is less than P _i .est_d. That is, it is determined in step 442 whether or not the pedestrian P _i is in front of or behind the obstacle O _j , and in step 443 it is determined whether or not the pedestrian P _i is behind the obstacle O _j . Thus, when it is determined in step 443 that the pedestrian P _i is behind the obstacle O _j , S2_flag in the “Hidden_by_Obs” column of the pedestrian list of the pedestrian information storage unit 4C is set to “2” in step 444. Set to “TRUE” to finish the process. On the other hand, if it is determined in step 442 that the pedestrian P _i is not in front of or behind the obstacle O _j , or if it is determined in step 443 that the pedestrian P _i is not behind the obstacle O _j , In step 445, S2_flag in the “Hidden_by_Obs” column of the pedestrian list in the pedestrian information storage unit 4C is set to “FALSE”, and the process is terminated. Note that the pedestrian P _i whose S2_flag is “FALSE” includes a pedestrian that is no longer detected in the middle of the obstacle O _j .

FIG. 5D is a flowchart showing a detailed processing procedure of the S3 processing performed in step 46 shown in FIG. 5A. Here, the estimated minimum x-coordinate est_x ₁ , maximum x-coordinate est_x ₂ , estimated x-coordinate speed est_v _x (calculated from time-series stored position information) and obstacles O _j minimum x-coordinate O _j .Ox _1, the maximum x-coordinate O _j .Ox ₂ is acquired. Therefore, the minimum x-coordinate Next_x ₁ of the next frame of the pedestrian estimated to be concealed state to the minimum x coordinate Est_x ₁ of the current frame in step 461, a frame of visualization camera speed Est_v _x estimate x coordinate Obtained by adding the value obtained by multiplying the time Frame Time. Further, the maximum x coordinate Next_x ₂ of the next frame of the pedestrian estimated to be concealed state to the maximum x coordinate Est_x ₂ of the current frame, the frame time visualization camera speed Est_v _x estimated x coordinate Frame Time It is obtained by adding the value multiplied by. As a result, the position while the pedestrian is behind the obstacle is sequentially obtained, and by performing these calculations, the time when the pedestrian in the prone state comes out behind the obstacle and the predicted travel range A position is required.

Next, in step 462, whether the absolute value of the minimum x coordinate next_x ₁ of the next frame is W / 2 or less, or whether the absolute value of the maximum x coordinate next_x ₂ of the next frame is W / 2 or less. Is determined. In other words, it is determined whether the estimated position in the next frame of the prone pedestrian is within or outside the predicted travel range. If it is within the travel prediction range, the process proceeds to step 463.

Next, in step 463, the minimum x coordinate next_x ₁ of the next frame of the prone pedestrian P _i is less than or equal to the minimum x coordinate Ox ₁ of the obstacle, and the maximum x coordinate next_x ₂ is the maximum of the obstacle. It is determined whether or not the x-coordinate Ox ₂ or more, that is, whether or not the prone pedestrian P _i goes outside the obstacle in the next frame. S3_flag indicating the possibility of the person jumping out of the obstacle is set to “TRUE”, and the process is terminated.

On the other hand, when the absolute value of the absolute value of the minimum pedestrian P _i saphenous state at the next frame x-coordinate Next_x ₁ is W / 2 or maximum x-coordinate Next_x ₂ is determined to exceed W / 2 in step 462 If it is determined in step 463 that the next frame is behind the obstacle, in step 465, S3_flag indicating the possibility of the pedestrian jumping out of the obstacle is set to “FALSE”, and the process is terminated. That is, when the estimated position of the next pedestrian in the next frame is outside the predicted travel range in step 462, or in step 463, the next pedestrian P _i in the next state is an obstacle. If it is determined that the not go outside, the vehicle has finished the processing in S3_flag the "FALSE" as there is no possibility of collision to the pedestrian P _i saphenous state. The processing of steps 461 to 465 may cause a pedestrian behind the obstacle to jump out of the area where the obstacle is present in the next frame. the car is in "TRUE" the S3_flag it is determined that there is a possibility of a collision to the pedestrian P _i of the saphenous state.

  FIG. 5E is a flowchart showing a detailed processing procedure of S4 processing performed in step 48 shown in FIG. 5A. Here, the determination result of the S3 process in step 46 in FIG. 5A, the distance d to the pedestrian determined not to be behind the obstacle in step 443 or the estimated distance est_d to the pedestrian in the lying state, and the own vehicle The difference between the traveling speed and the moving speed of the pedestrian is acquired, and the processing of step 481 is executed. In step 481, the distance d or the estimated distance est_d is divided by the relative speed V to calculate the collision time TTC.

  FIG. 5F is a flowchart showing a detailed processing procedure of S5 processing performed in step 49 shown in FIG. 5A. Here, it is assumed that “−1” is set as an initial value in the “C-alarm” column of the pedestrian list of the pedestrian information storage means 4C shown in FIG. The initial value in the “C-alarm” column of this pedestrian information is the value of a counter that counts that the process of FIG. 5A has been repeated on condition that the collision time is less than 3 seconds. In this case, it is determined whether or not the collision time TTC stored in the “TTC” column of the pedestrian list of the pedestrian information storage means 4C in step 491 is less than 1 second. In step 492, a sudden brake operation signal is generated. If it is determined in step 491 that the collision time TTC is not shorter (longer) than 1 second, it is determined in step 493 whether or not the collision time TTC is shorter than 3 seconds. When it is determined in step 493 that the collision time TTC is shorter than 3 seconds, it is determined in step 494 whether or not the value in the “C-alarm” column is “−1”. In step 495, an alarm signal is generated. Subsequently, in step 496, the value in the “C-alarm” column is incremented by one. If it is determined in step 493 that the collision time TTC is shorter than 3 seconds, the alarm signal is generated because the value in the “C-alarm” column is “0” instead of “−1”. In step 496, the value in the “C-alarm” column is incremented by 1 to “1”. Then, each time the process of FIG. 5A is repeated, the value in the “C-alarm” column increases.

  Next, in step 497, it is determined whether or not the value in the “C-alarm” column exceeds “15”. If not, the process ends without generating an alarm signal. On the other hand, if it is determined that the value in the “C-alarm” column exceeds “15”, an alarm signal is generated in step 498 and the value in the “C-alarm” column is set to zero. By repeating such processing, an alarm signal is output every time the value in the “C-alarm” column reaches 16. That is, an alarm signal is output every time the process of the risk determination means shown in FIG. 5A is repeated 16 times.

As described above, the first embodiment of the present invention has been described. However, the travel prediction range in the traveling direction of the vehicle is imaged with a visualization camera, and a pedestrian is detected by pattern recognition. When an obstacle including a pedestrian located is detected by the radar transmission / reception unit, the following four states A to D are assumed. In the following description, the obstacle will be used again to include a pedestrian.
A: The pedestrian was detected by the visualization camera at the time of the previous detection and detected as an obstacle by the radar transmission / reception unit, but the pedestrian fell down, so it was not detected by the visualization camera at the next detection, and the radar transmission / reception unit When detected as an obstacle only by.
B: At the time of the previous detection, the pedestrian was detected by the visualization camera and detected as an obstacle by the radar transmission / reception unit, but because the pedestrian fell, it was not detected as a pedestrian by the visualization camera at the next detection, When it is not detected as an obstacle by the radar transceiver.
C: At the previous detection, the pedestrian was detected by the visualization camera and detected as an obstacle by the radar transceiver, but at the next detection, the pedestrian was hidden behind an obstacle other than the pedestrian. When only obstacles other than pedestrians are detected by the radar transceiver unit without being detected by the visualization camera.
D: In the case of C, a pedestrian hidden behind an obstacle may jump out of the travel prediction range of the own vehicle.

In the first embodiment, the cases A to D are dealt with as follows.
In the case of A: A new position of the pedestrian that is no longer detected by the visualization camera in the G2 process of step 24 in FIG. 4A is estimated. Next, it is determined whether or not the pedestrian whose new position has been estimated by the G4 process in step 25 of FIG. 4A has gone too far from the front and front side areas of the own vehicle. G4_flag is set to “TRUE”. Next, it is determined in step S2 of FIG. 5A whether or not the vehicle is within the predicted travel range of the vehicle. If it is determined that the vehicle is within the predicted range, step 44 in FIG. It is determined that it is not behind, and the risk is determined in step 48 of FIG. 5A.
In the case of B: It is not determined in step 15 in FIG. 3 that the obstacle is a pedestrian. However, if the new position is estimated in step 24 of FIG. 4A and it is not determined that the pedestrian who estimated the new position in step 25 of FIG. 4A has gone too far from the front and front side areas of the vehicle, the step of FIG. In step S2 of 42, it is determined whether or not the vehicle is within the predicted travel range. If so, it is determined in step 44 of FIG. 5A that it is not behind an obstacle other than a pedestrian, and in step 48 of FIG. 5A. A risk assessment is made.
Case C: A new position is estimated in step 24 of FIG. 4A. Then, if it is not determined that the pedestrian whose new position has been estimated in step 25 in FIG. 4A has gone too far from the front and front side areas of the vehicle, step 42 in FIG. In step 44 of FIG. 5A, it is determined that the vehicle is behind an obstacle other than a pedestrian. If the vehicle is behind, the position is sequentially changed in step 24 of FIG. 4A. The estimated position is updated and stored as time series information.
In the case of D: Based on the position estimated in the case of the above C, the possibility of jumping out from the obstacle is determined in step 46 of FIG. 5A, and the walk coming out from behind the obstacle other than the pedestrian The risk of the vehicle colliding with the person is determined.

  In the first embodiment, the photographing means 1, the pedestrian detection means 2, and the obstacle detection means 3 are mounted on a vehicle and are used as components of the collision risk determination device. However, the photographing means 1, The pedestrian detection means 2 and the obstacle detection means 3 may be installed at a predetermined place other than the vehicle. In this case, the pedestrian information storage means 4A, the obstacle information storage means 4B, and the pedestrian information in the current frame. The storage unit 4C, the pedestrian / obstacle checking unit 5, the pedestrian tracking unit 6, and the risk level determination unit 7 can constitute a collision risk level determination device having the same function as described above.

  Further, in the first embodiment, the position of the pedestrian and the obstacle is an azimuth coordinate system based on an axis set in the traveling direction of the vehicle, and a secondary planar coordinate system developed in the traveling direction of the vehicle. Since both are used, there is an advantage that processing such as position collation, position estimation, correlation calculation, and whether or not the vehicle is within the predicted travel range of the vehicle can be executed by a CPU with relatively small processing capability. is doing. However, in the case of using a CPU having a higher calculation speed, whether the position is detected using only the azimuth coordinate system or the position is detected using only the secondary plane coordinate system, the same as described above. Can perform various processes.

  In the first embodiment, the pedestrian detection means 2 binarizes the image data obtained by the visualization camera of the photographing means 1 and extracts it as obstacle shape data. A pedestrian is identified by detecting a temporal change in shape data as a movement pattern. Instead of this pedestrian detection means 2, it is also possible to identify a pedestrian by pattern recognition from a still image.

<Second Embodiment>
In the first embodiment described above, as described with reference to FIG. 5A, pedestrians not found in step 41 are extracted from the pedestrian list of the pedestrian information storage means 4C, and in step 42, the pedestrians are automatically identified. It is determined whether or not the vehicle is within the predicted travel range of the vehicle. When S1_frag indicating that the determination result is within the predicted travel range of the vehicle is “TRUE”, steps 44, 45, and 46 are further performed. , 47 is executed, and then the degree of risk is determined. However, this risk level judgment is not limited to pedestrians that have not been found, and it is also possible to make a risk level judgment for pedestrians that have been found. This will be described with reference to the flowchart of FIG. 5A.

  The process shown in the flowchart of FIG. 5A extracts each pedestrian not found in step 41, that is, a pedestrian whose flag in the “Appear” column of the pedestrian list in the pedestrian information storage means 4C is “FALSE”. In step 42, it is determined whether or not this pedestrian is within the predicted travel range of the vehicle. For the found pedestrian, as a process corresponding to step 41, a pedestrian whose flag in the “Appear” column of the pedestrian list of the pedestrian information storage means 4C is “TRUE” is extracted, and then step 42 In the same manner as described above, it is determined whether or not the found pedestrian is within the predicted travel range of the vehicle. If it is determined that S1_flag is “TRUE” indicating that the vehicle is within the predicted travel range of the host vehicle, the process proceeds to step 48 in FIG. 5A and steps 48, 49, 50, 51, and 52 are performed. And when S1_flag is “FALSE” indicating that the vehicle is not in the predicted travel range of the vehicle, the process returns to step 41.

  Thus, when determining the risk of collision of a vehicle with a pedestrian that has been found, it is possible to appropriately change the threshold for determining the risk of collision, that is, the threshold for the collision time. Thus, by determining the risk of collision of the vehicle with the currently detected pedestrian, the risk of collision of the vehicle with the pedestrian found in the current frame can be predicted.

<Third Embodiment>
The first embodiment described above is intended for pedestrians that have not been detected, and the second embodiment is directed to pedestrians that have been detected. In combination with the embodiment, the risk level can be determined for all the states in which the detected person is no longer detected.

It is a block diagram which shows the structure of 1st Embodiment of the collision risk determination apparatus which concerns on this invention. In order to explain the operation of the first embodiment of the collision risk determination device according to the present invention, how obstacles including pedestrians and pedestrians and automobiles are arranged with respect to the predicted driving range of the automobile. It is a schematic top view which shows whether it is, and explanatory drawing for demonstrating the example of a detection of the obstruction containing a pedestrian and a pedestrian. It is a flowchart which shows the specific process sequence at the time of comprising the pedestrian and obstruction collation means which comprise the 1st Embodiment of this invention with an arithmetic processing unit. It is a flowchart which shows the specific process sequence at the time of comprising the pedestrian tracking means which comprises the 1st Embodiment of this invention with an arithmetic processing unit. It is a flowchart which shows the detailed process sequence of one step in FIG. 4A. It is a flowchart which shows the detailed process sequence of one step in FIG. 4A. It is a flowchart which shows the detailed process sequence of one step in FIG. 4A. It is a flowchart which shows the detailed process sequence of one step in FIG. 4A. It is a flowchart which shows the specific process sequence at the time of comprising the risk degree determination means which comprises the 1st Embodiment of this invention with an arithmetic processing unit. It is a flowchart which shows the detailed process sequence of one step in FIG. 5A. It is a flowchart which shows the detailed process sequence of one step in FIG. 5A. It is a flowchart which shows the detailed process sequence of one step in FIG. 5A. It is a flowchart which shows the detailed process sequence of one step in FIG. 5A. It is a flowchart which shows the detailed process sequence of one step in FIG. 5A. It is the table | surface which showed the content of the pedestrian list of the pedestrian information storage means which comprises the 1st Embodiment of this invention, and the pedestrian information memorize | stored in it. It is the table | surface which showed the content of the obstruction list of the obstruction information storage means which comprises the 1st Embodiment of this invention, and the obstruction information memorize | stored in it.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Imaging | photography means 2 Pedestrian detection means 3 Obstacle detection means 4A Pedestrian information storage means 4B Obstacle information storage means 4C Pedestrian information storage means 5 Pedestrian / obstacle collation means 6 Pedestrian tracking means 7 Risk level Judgment means 8 Danger warning means 10 Automobile (vehicle, own vehicle)

Claims (14)

A pedestrian detection means for detecting a pedestrian by pattern recognition by imaging a travel prediction range in the traveling direction of the vehicle, and detecting the position of the pedestrian;
Obstacle detection means for detecting an obstacle including the pedestrian located in a travel prediction range in the traveling direction of the vehicle, and detecting the position of the obstacle;
Pedestrian information storage means for storing the position of the pedestrian in time series for each pedestrian detected by the pedestrian detection means;
Obstacle information storage means for storing the position of the obstacle for each obstacle detected by the obstacle detection means;
When the pedestrian that has been detected by the pedestrian detection unit is no longer detected by the pedestrian detection unit, the position of the pedestrian stored in the pedestrian information storage unit is changed from time-series changes to the pedestrian. Pedestrian tracking means for tracking a person's movement trajectory, estimating the current position of the pedestrian and storing it in the pedestrian information storage means;
Pedestrians / obstacles that judge the pedestrians and non-pedestrians in the obstacles by comparing the information stored in the obstacle information storage unit with the information stored in the pedestrian information storage unit Object verification means;
The current position of the pedestrian estimated by the pedestrian tracking means, the position stored in the obstacle information storage means of the obstacle determined to be a non-pedestrian by the pedestrian / obstacle checking means, A risk determination means for determining the risk of the vehicle colliding with the pedestrian that is no longer detected by the pedestrian detection means, taking into account the relative speed with the vehicle position and the speed of the vehicle. And
Equipped with a collision risk judgment device. The risk determination means confirms that the pedestrian whose position is estimated to enter the travel lane, and determines the risk of the vehicle colliding with the pedestrian in the travel lane. The collision risk determination device according to claim 1 . The risk determination means confirms that the pedestrian whose position has been estimated is within a predicted travel range of the vehicle, and the vehicle collides with the pedestrian within the predicted travel range. The collision risk determination device according to claim 1 , wherein the risk level is determined. The risk determination means calculates a predicted collision time until the vehicle collides with the pedestrian whose position is estimated, and the estimated walking time when the predicted collision time is shorter than a preset threshold value. The collision risk determination device according to claim 1, wherein it is determined that the risk of the vehicle colliding with a person is high. The risk determination means determines whether there is an obstacle other than the pedestrian in front of the same direction as the pedestrian whose position is estimated, and the pedestrian whose position is estimated when it is determined that there is no obstacle The collision risk determination device according to claim 1 , wherein the risk of collision of the vehicle is determined. The risk determination means determines whether or not the pedestrian whose position is estimated is in the same direction as an obstacle other than the pedestrian, and whether or not it is hidden behind an obstacle other than the pedestrian. The pedestrian is in the same orientation as the obstacles other than the pedestrian, and when hidden behind the obstacles other than the pedestrian, both end positions of the obstacles other than the pedestrian, Based on the movement speed of the pedestrian whose position is estimated, the time and position where the pedestrian whose position is estimated comes out behind the obstacle other than the pedestrian is estimated, and the obstacle other than the pedestrian The collision risk determination device according to claim 1 , wherein the risk of collision of the vehicle against the pedestrian coming out from behind is determined. The pedestrian tracking means captures a travel prediction range in the traveling direction of the vehicle at a predetermined time interval when the pedestrian whose position is estimated is hidden behind an obstacle other than the pedestrian. The collision risk determination device according to claim 6 , wherein the position of the pedestrian that is no longer detected is estimated, and the position stored in the storage unit is updated. The risk determination means generates a signal for selectively operating an alarm device for determining a risk level and generating a warning to a driver, a braking device for suddenly braking the vehicle, and a protection device for protecting a pedestrian. The collision risk determination device according to claim 1 . The risk determination means includes
When the pedestrian that existed in the past does not exist in the latest information from the time-series information of the position of the pedestrian stored in the pedestrian information storage means, it was estimated by the pedestrian tracking means Means for determining whether the current position falls within a predicted travel range of the vehicle;
When it is determined that the current position falls within the predicted travel range of the vehicle, it is estimated that the current position is present at the current position from the current position and the position of the obstacle stored in the obstacle information storage means. Means for determining whether the pedestrian is behind any obstacle stored in the obstacle information storage means;
When it is determined that the pedestrian is behind any one of the obstacles, the pedestrian is detected from the time-series position change of the pedestrian stored in the pedestrian information storage means. A means of judging whether or not there is a possibility of jumping out of the object,
And determining that the pedestrian jumps out of the obstacle when the pedestrian is determined not to be behind any of the obstacles and means for determining whether or not there is a possibility of jumping out of the obstacle by collision risk judging device of claim 1, which is configured to determine the risk when the. In response to information from an external pedestrian detection means that detects a pedestrian by pattern recognition and images a travel prediction range in the traveling direction of the vehicle, and is detected by the pedestrian detection means. Pedestrian information storage means for storing the position of the pedestrian in time series for each pedestrian,
In response to information from an external obstacle detection means for detecting an obstacle including the pedestrian located in the predicted travel range of the vehicle in the traveling direction, and detecting the position of the obstacle, the obstacle detection means Obstacle information storage means for storing the position of the obstacle for each detected obstacle;
When the pedestrian that has been detected by the pedestrian detection unit is no longer detected by the pedestrian detection unit, the position of the pedestrian stored in the pedestrian information storage unit is changed from time-series changes to the pedestrian. Pedestrian tracking means for tracking a person's movement trajectory and estimating the current position of the pedestrian;
Pedestrians / obstacles that judge the pedestrians and non-pedestrians in the obstacles by comparing the information stored in the obstacle information storage unit with the information stored in the pedestrian information storage unit Object verification means;
The current position of the pedestrian estimated by the pedestrian tracking means, the position stored in the obstacle information storage means of the obstacle determined to be a non-pedestrian by the pedestrian / obstacle checking means, A risk determination means for determining the risk of the vehicle colliding with the pedestrian that is no longer detected by the pedestrian detection means, taking into account the relative speed with the vehicle position and the speed of the vehicle. And
Equipped with a collision risk judgment device. Detecting a pedestrian by pattern recognition by imaging a travel prediction range in the traveling direction of the vehicle, and detecting the position of the pedestrian;
Detecting an obstacle including the pedestrian located in a travel prediction range in the traveling direction of the vehicle, and detecting the position of the obstacle;
Storing the position of the pedestrian in time series for each detected pedestrian;
Storing the position of the obstacle for each detected obstacle;
When the pedestrian detected in the step of detecting the position of the pedestrian is no longer detected, the movement trajectory of the pedestrian is tracked from a change in the position of the pedestrian stored in time series, Estimating and storing the current position of the pedestrian;
Comparing each stored position of the obstacle and the estimated current position of the pedestrian to determine a pedestrian and a non-pedestrian in the obstacle;
The relative position between the position of the obstacle determined to be a non-pedestrian by verification, the estimated current position of the pedestrian, and the position of the vehicle, also including the speed of the vehicle, Determining the risk of the vehicle colliding with the pedestrian that is no longer detected in the step of detecting the position of the pedestrian,
Provided collision risk judgment method. Detecting a pedestrian by pattern recognition by imaging a travel prediction range in the traveling direction of the vehicle, and detecting the position of the pedestrian;
Detecting an obstacle including the pedestrian located in a travel prediction range in the traveling direction of the vehicle, and detecting the position of the obstacle;
Storing the position of the pedestrian in time series for each detected pedestrian;
Storing the position of the obstacle for each detected obstacle;
When the pedestrian detected in the step of detecting the position is not detected, the movement position of the pedestrian is tracked from a time-series change in the stored position of the pedestrian, Estimating and storing the current position;
The stored position information of the obstacle and the current position information of the pedestrian whose position has been detected or the pedestrian whose position has been estimated are collated with each other, and the pedestrian in the obstacle And determining a non-pedestrian,
A provided pedestrian identification method. The positions of the pedestrian and the obstacle are the azimuth and distance of the azimuth coordinate system based on the axis set in the traveling direction of the vehicle and / or the secondary of the secondary planar coordinate system developed in the traveling direction of the vehicle. The collision risk determination device according to claim 1 , wherein the collision risk determination device is a coordinate. 14. The collision risk determination device according to claim 13 , wherein the azimuth of the azimuth coordinate system and the secondary coordinates of the planar coordinate system are represented by a minimum value and a maximum value with respect to a contour of the video pattern.

Images