JP2016016751A - Curve detection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a curve detection device for detecting a curve in a front direction of a vehicle stably, before approaching to the curve.SOLUTION: A fuzzy inference determines whether or not, a candidate of a curve start point Cc in a front direction of a vehicle, detected by images 60, 61 obtained by image processing to images including a lane mark imaged by a camera 16, is the curve start point Cc. More particularly, approach distance Di from distance Dp of the curve start point Cc in a past frame to distance Dn of the curve start point Cc in a present frame, is determined, and the approach distance Di and movement distance Dr of a vehicle 12 determined based on time ΔT between both frames and vehicle speed v, are compared. When distance difference |Di-Dr| is large, it is estimated that, probability that detection is correctly performed is low, and when the distance difference |Di-Dr| is small, it is estimated that, probability that detection is correctly performed is high, and then determined.SELECTED DRAWING: Figure 6

Description

  The present invention relates to a curve detection device that detects a curve ahead of a vehicle based on an image including a lane boundary line (also referred to as a lane marking) imaged by an imaging device mounted on the vehicle, for example, an electric power steering device. The present invention relates to a curve detection device suitable for being mounted on an automatic steering vehicle or an automatic steering assist vehicle including

  In Patent Document 1, the running state of the vehicle is estimated by performing fuzzy inference according to the fuzzy rule based on the rotational angular velocity calculated from the actual motor driving current and the actual motor driving voltage of the steering assist motor and the detected vehicle speed. An electric power steering device is disclosed that corrects the drive current of the steering assist motor based on the estimated value and controls the steering assist motor so that the steering force is adapted to the actual running state of the vehicle (Patent Document 1). Summary).

  According to the technique disclosed in Patent Document 1, for example, in a traveling state where a long distance is traveled for a long time, the steering force is lightened, and on a curved road such as a mountain road, the steering force is slightly increased to increase the response. It is said that the demand can be smoothly met in a certain running state.

JP-A-5-213217

  However, in the electric power steering apparatus using fuzzy inference according to the above-described prior art, the actual motor drive current and the actual motor drive voltage of the steering assist motor are detected in order to control the steering force in accordance with the actual running state of the vehicle. ([0011], [0012] of Patent Document 1), it cannot be applied to a vehicle before entering a curve.

  In other words, in the above-described conventional technology, the curve ahead of the vehicle cannot be detected stably before entering the curve, so there is room for improvement in particular regarding steering assist control when entering the curve.

  The present invention has been made in consideration of such problems, and provides a curve detection device that can stably detect a curve ahead of a vehicle (including a curve start point) before entering the curve. For the purpose.

  The curve detection device according to the present invention is a curve detection device that detects a curve ahead of a vehicle based on an image including a lane boundary line imaged by an imaging device mounted on the vehicle. A curve candidate detection unit that detects a curve candidate that may be a curve ahead of the vehicle by performing image processing, and whether or not the curve candidate detected by the curve candidate detection unit is a curve. And a curve determination unit that determines using fuzzy inference.

  According to the present invention, whether the curve candidate in front of the vehicle detected by performing image processing on the image including the lane boundary line is a curve is determined using fuzzy inference. The curve can be detected stably.

  In this case, the fuzzy inference is calculated based on the approach distance of the curve candidate calculated from the images of different frames captured by the imaging device, the time between the different frames, and the vehicle speed. By including a new membership function consisting of a rule having an antecedent part in the case where the moving distance is substantially equal, a reasonable result for detecting the curve stably can be obtained.

  More preferably, the curve determination unit selects at least two rules among the four rules used for the fuzzy inference, and determines the correctness of the curve start point candidate in the curve. The four rules Is the first rule that is erroneous if the distance from the vehicle to the curve start point candidate calculated from the image of the current frame is too far or too close A second rule that is positive if the distance from the vehicle to the curve start point candidate calculated from the image is an intermediate distance between the distance too far and the distance too close, Calculated from the approach distance of the curve start point candidate calculated from the image of the frame and the past frame to the vehicle, the time between the current frame and the past frame, and the vehicle speed. If there is a large gap between the moving distance of the vehicle and the third rule to be erroneous, and the approach distance of the curve start point candidate calculated from the images of the current frame and the past frame to the vehicle When the moving distance of the vehicle calculated from the time between the current frame and the past frame and the vehicle speed is substantially the same, the fourth rule is set to be positive.

  In this way, the curve determination unit selects at least two of the four rules used for fuzzy inference and determines whether the curve start point candidate is correct or not. Therefore, the probability of being the curve start point is increased. can do. By selecting all four rules, a more robust curve start point can be detected.

  The curve determining unit compares the threshold of the combined output centroid of each membership function of the two selected rules with a threshold to determine the threshold, and the threshold is used as the lane boundary of the captured image. If the image signal representing the line increases or decreases according to the size of the image signal representing the lane boundary line, the size of the image signal representing the lane boundary line changes depending on the surrounding environment of the road such as lane marks and weather. The curve can be determined with certain certainty from the curve candidates.

  Moreover, when the said curve determination part determines the said curve or a curve start point, it is preferable to further generate | occur | produce a vehicle travel assistance start signal from before the said curve or the said curve start point. By generating a vehicle travel support start signal from before the curve start point, it is possible to execute suitable vehicle travel support, for example, reliable automatic steering, steering assistance without a sense of incongruity, timely application of accelerator pedal reaction force, and the like.

  According to the present invention, the effect of being able to stably detect the curve ahead of the vehicle (including the curve start point) before entering the curve is achieved.

1 is a block diagram showing a schematic configuration of a vehicle on which a lane mark recognition device including a curve detection device according to an embodiment of the present invention is mounted. It is a schematic diagram explaining the outline | summary of a structure of the curve determination part which consists of a fuzzy inference part and a defuzzification part. It is a flowchart with which operation | movement description of the curve detection apparatus which concerns on one Embodiment is provided. FIG. 4A is an explanatory diagram of a planar edge image in the past two frames before, FIG. 4B is an explanatory diagram of a planar edge image in the past frame one frame before, and FIG. 4C is an explanatory diagram of a planar edge image in the current frame. FIG. It is explanatory drawing of the example of a process which detects a curve start point candidate from a planar view edge image. It is explanatory drawing which shows the example of a planar view edge image. FIG. 7A is an explanatory diagram of the membership function according to the first rule, and FIG. 7B is an explanatory diagram of the membership function according to the second rule. FIG. 8A is an explanatory diagram of the membership function according to the third rule, and FIG. 8B is an explanatory diagram of the membership function according to the fourth rule. It is explanatory drawing of the rule which concerns on a modification.

  Hereinafter, preferred embodiments of a curve detection device according to the present invention will be described in detail with reference to the accompanying drawings.

  FIG. 1 shows a schematic configuration of a vehicle 12 on which a lane mark recognition device 11 including a curve detection device 10 according to the embodiment is mounted.

  The lane mark recognizing device 11 supports a steering operation by a driver of a vehicle 12 traveling along a lane boundary line (hereinafter referred to as a lane mark) provided on a traveling road, a braking operation by a brake pedal, and the like. Therefore, it is configured as a part of the function of the driving support device 14 mounted on the vehicle 12.

  In this embodiment, an example in which the lane mark recognition device 11 including the curve detection device 10 is applied to the driving support device 14 is shown, but the embodiment may be applied to an automatic driving device.

  The lane mark is a mark indicating a lane boundary (lane division), and the lane mark is a continuous line (also referred to as a deemed continuous line) composed of broken white lines (segments) provided at intervals. In addition to continuous lines such as white lines, continuous marks made up of bots dots, cat's eyes, etc. (also considered as continuous lines) are also included.

  A lane mark recognition device 11 including the curve detection device 10 according to this embodiment basically recognizes a lane mark from a camera 16 (imaging device) that images the front of the vehicle 12 and an image captured by the camera 16. And an electronic control unit 20 (Electronic Control Unit: hereinafter referred to as ECU 20) that detects a curve (including a curve start point) from the lane mark.

  The camera 16 is mounted on the upper part of the front windshield in the vehicle 12 and images the front of the vehicle 12 through the front windshield, and images an image including a traveling path ahead of the vehicle. The camera 16 is, for example, a digital video camera and captures, for example, several to several tens of images (for example, an image of 15 frames) per second and outputs an image signal (video signal) Si. In this case, the camera 16 has its mounting portion as the origin, the vehicle width direction (horizontal direction) of the vehicle 12 is the X axis, the axle direction (traveling direction, forward direction) is the Y axis, and the vehicle height direction (vertical direction, vertical direction). A real space coordinate system with the (direction) as the Z axis is defined.

  The lane mark recognition device 11 includes a speed sensor 22 in addition to the camera 16 and the ECU 20 described above. The speed sensor 22 outputs a vehicle speed v [m / s] that is a speed signal of the vehicle 12.

  The ECU 20 is a computer including a microcomputer, and includes a CPU (Central Processing Unit), a ROM (including EEPROM) as a memory, a RAM (Random Access Memory), an A / D converter, a D / A converter, and the like. Input / output device, a timer as a timer, and the like, and the CPU functions as various function realization units (function realization means) by reading and executing a program recorded in the ROM.

  In this embodiment, the ECU 20 functions as a lane mark recognition unit 36, a curve candidate detection unit 42, a curve determination unit 44, and the like.

  The ECU 20 also includes an image memory 30, a lane mark memory 34, and a curve candidate memory 40 as storage devices.

  The driving support device 14 includes the steering angle sensor 26 in the curve detection device 10 and the lane mark recognition device 11 described above.

  The steering angle sensor 26 outputs a steering angle [deg] that is a steering angle signal of the steering device (steering) of the vehicle 12.

  The ECU 20 determines the traveling direction of the vehicle 12 from the output of the steering angle sensor 26 {the vehicle width direction of the vehicle 12 is the X axis, and the Y axis direction in front of the vehicle 12 is 0 [deg] (reference) X on the XY plane. The inclination in the axial direction, that is, the steering angle} is calculated. The steering angle sensor 26 may be replaced with a yaw rate sensor (not shown).

  Based on the lane mark recognized by the lane mark recognition device 11 and the curve detected by the curve detection device 10 (including the curve start point, the radius of curvature, and the curve end point), the driving support device 14 performs steering ( The vehicle 12 is sandwiched between lane marks on both sides in the width direction of the vehicle 12 by controlling the steering device 50, the braking device 52, and the acceleration device 54 under predetermined conditions such as gripping (not shown). In order not to deviate from the road, that is, the lane, in other words, the vehicle 12 can run substantially in the center of the lane, and the braking device 52 or an accelerator pedal reaction force applying mechanism (not shown) is operated before the curve (curve start point). Carry out driving assistance for the driver.

  As schematically shown in FIG. 2, the curve determination unit 44 according to the main part of this embodiment includes the first, second, third, and fourth rules used for fuzzy inference (each rule is an antecedent part and a rear part). A fuzzy inference unit 70 having four rules, and a defuzzification unit (non-fuzzification unit) 72.

  Here, an outline of the first, second, third, and fourth rules of the fuzzy inference unit 70 will be described.

  The first rule is that the distance Dn from the vehicle 12 to the curve start point candidate calculated from the image of the current frame imaged by the camera 16 is too far or too close. This is a rule for erroneous (false detection: grade “0” to “−1”).

  The second rule is that the distance Dn from the vehicle 12 to the curve start point candidate calculated from the image of the current frame imaged by the camera 16 is an intermediate distance between the distance too far and the distance too close. In some cases, the rule is positive (positive detection: grade “0” to “1”).

  The third rule is that the vehicle speed v is determined by the approach distance Di of the candidate curve start point calculated from the images of the current frame and the past frame captured by the camera 16 and the time ΔT between the current frame and the past frame. When there is a large gap between the travel distance Dr (Dr = ΔT × v) of the vehicle 12 calculated by multiplication, it is determined as an error (false detection: grade “0” to [−1]).

  In this embodiment, the past frame is a frame one frame before, and the time ΔT between the current frame and the past frame (hereinafter referred to as one frame time, reference frame time, elapsed time, or simply time) is ΔT = 0.066 [sec]. Therefore, in the third rule, the approach distance Di of the curve start point candidate calculated from the image to the vehicle 12, the time ΔT between the two most recent frames and the vehicle speed v (the average vehicle speed between the two frames or the current frame) If there is a large gap between the travel distance Dr (Dr = ΔT × v) of the vehicle 12 calculated from the vehicle speed v of the vehicle or the vehicle speed v of the past frame, the error (false detection: grade “0”). To “−1”).

  If the vehicle speed v is 36 [km / h], the vehicle 12 advances 10 [m] in 1 second [sec], and advances 6.6 [m] in 1 frame. An appropriate timing can be set to prompt deceleration before the point. Moreover, since 6.6 [m] is a close distance from the vehicle 12, the detection accuracy of the front distance by the camera 16 is also high. The time ΔT may be changed by changing the number of frames up to the past frame according to the vehicle speed v, the reference frame time ΔT (ΔT = 0.066 [sec]) can be made variable, and the time ΔT is set to the vehicle speed v. However, it may be changed so as to vary stepwise or continuously.

  The fourth rule is that the vehicle 12 calculated from the approach distance Di to the vehicle 12 of the curve start point candidate calculated from the images of the current frame and the past frame, the time ΔT between the current frame and the past frame, and the vehicle speed v. When the movement distance Dr is substantially the same, it is positive (positive detection: grade “0” to “1”).

  The curve determination unit 44 selects at least two rules among the first, second, third, and fourth rules, and then obtains the center of gravity by the center of gravity method in the defuzzification unit 72, and the vehicle curve start point It is determined whether or not the distance Dn is correct.

  Next, basically, with respect to the operation of the vehicle 12 configured as described above, the operation of the fuzzy inference unit 70 of the curve detection device 10 including the curve determination unit 44 is mainly described with reference to the flowchart of FIG. This will be described in more detail.

  In step S <b> 1, the ECU 20 captures an image (multi-valued image) every reference frame time ΔT (ΔT = 0.066 [sec]) that is a predetermined time, that is, every time an image of the vehicle front is captured by the camera 16. A gray scale image, which is an image, is captured (stored) in the image memory 30.

  Next, in step S2, image processing involving lane mark recognition processing by the lane mark recognition unit 36 is executed. In this case, the edge extraction unit 32 takes in an image having a predetermined luminance or higher (brightness on the road surface is higher than the predetermined lightness) from the image captured by the camera 16 and performs binarization processing, and then adds the entire image to 1 Differentiating is performed while scanning line by line in the horizontal direction to extract edges (edge images) of the entire image.

  In step S2, the lane mark recognizing unit 36 extracts an image having characteristics as a lane mark from the entire edge extracted by the edge extracting unit 32, and takes it into the lane mark memory 34 as a lane mark. (Remember.). Further, as will be described later, the lane mark determined to have the characteristics of the start point (inflection point) of the curve is captured (stored) in the curve candidate memory 40 as a curve start point candidate. The lane mark memory 34 and the curve candidate memory 40 store lane marks extracted in a predetermined number of past frames in addition to lane marks extracted in the current frame.

  When an image having a feature as a lane mark is captured (stored) in the lane mark memory 34 and the curve candidate memory 40, the image is converted into a planar view edge (plan view edge image) as seen from above. Then, after the capture, the processing is performed using the planar view edge image.

  4A, FIG. 4B, and FIG. 4C show planar view edge images 59, 60, and 61 captured in the lane mark memory 34, respectively. In FIG. 4A, FIG. 4B, and FIG. 4C, the two thick broken lines that extend in parallel in the vertical direction indicate lane marks that exist on the left and right sides of the lane in which the vehicle 12 travels, and FIG. 4C is a plane edge image 61 of the current frame at the current time point t1, and FIG. 4A is a plane at the past time point (−t1) two frames before. Visual edge images 59 are shown respectively.

  4A, 4B, and 4C, the thin solid line on the horizontal axis represents the mounting position of the camera 16 on the vertical axis (distance axis), that is, the position of the vehicle 12 (front distance = zero value).

  The curve start point Cc is not captured in the planar view edge image 59 at the past time point (−t1) two frames before the current frame, and the curve start point Cc is included in the plan view edge image 60 of the past frame. In the plan view edge image 61 of the current frame, it can be seen that the curve start point Cc is captured where the camera 16 is attached, that is, the vehicle 12 on which the camera 16 is mounted.

  Next, every time each of the planar view edge images 59, 60, 61 is taken into the lane mark memory 34 in step S 3, the curve start point is included in the planar view edge images 59, 60, 61 by the curve candidate detection unit 42. It is determined whether there is a Cc candidate (curve start point Cc candidate), in other words, whether there is a curve start point Cc candidate.

  It is determined by the curve candidate (curve start point candidate) detection unit 42 whether or not a candidate for the curve start point Cc (curve start point Cc candidate) exists (the presence or absence of the curve start point Cc candidate). , 60, 61, based on a predetermined lane mark on either the left side or the right side of the lane (the distance / distance between the left and right lane marks). In this embodiment, a lane mark LM on the right side surrounded by a two-dot chain line is specified in FIGS. 4A, 4B, and 4C, and a curve start point Cc candidate is detected. Note that the curve start point Cc candidate may be detected using lane marks on both the left and right sides of the lane.

  FIG. 5 is an explanatory diagram of a processing example in which the curve candidate detection unit 42 detects a curve start point Cc candidate from the planar view edge image Ie.

  In the detection process of the curve start point Cc candidate, an extension line Lp obtained by extending the lane mark Lm1 detected on the front side of the vehicle 12 to the back side, and lane marks Lm2, Lm3, and Lm4 detected on the back side (necessary) If it detects a position where the difference (distance Ds) in the lateral direction (vehicle width direction) from the zero value increases, the line that connects the end points of the front and rear lane marks is also included. It is detected (recognized) as a starting point Cc candidate (step S3: YES). The planar view edge image Ie in which the lane mark LM where the curve start point Cc candidate is detected is captured (stored) in the curve candidate memory 40.

  FIG. 6 shows planar view edge images 60 and 61 of the lane mark LM at the time point t0 (past frame) and the time point t1 (current frame) taken into the curve candidate memory 40 (hereinafter also simply referred to as images 60 and 61). ).

  Next, in step S4, the curve determination unit 44 first calculates the distance Dn from the vehicle 12 to the curve start point Cc candidate based on the image 61 of the current frame in FIG. Based on this, a distance Dp from the vehicle 12 to the curve start point Cc candidate is calculated. Next, approach to the vehicle 12 during one frame (ΔT = 0.066 [sec]) of the curve start point Cc candidate obtained from the difference between the distances Dn and Dp between the images 61 and 60 of the current frame and the past frame. The distance Di (Di = Dp−Dn) is calculated.

  Next, in step S5, the curve determination unit 44 calculates the moving distance Dr of the vehicle 12 as Dr = ΔT × v = 0.066v from the time ΔT between the current frame and the past frame and the vehicle speed v.

  Next, in step S6, the curve determination unit 44 executes fuzzy inference of the first, second, third, and fourth rules. In the fuzzy inference, a reasonable result can be obtained by executing at least two rules among the first, second, third, and fourth rules.

  FIG. 7A graphically shows the membership function fm1 (Dn) of the first rule generated by empirical rules including experiments (actual running tests) or simulations (machine learning, etc.).

  The horizontal axis is the current frame distance Dn [m] (see FIG. 6), and the vertical axis is the grade G1 (goodness of fit, probability). In practice, the grade G1 takes a value of 0 to −1 representing “false (false detection)”. For this reason, the grade G1 returned from the membership function fm1 (Dn) shown in FIG. 7A drawn with the positive and negative signs reversed for the convenience of comparison and understanding with other rules with respect to the input distance Dn. , Put a negative sign. For example, a grade G1 = −1 with a negative sign is interpreted as being correctly detected (actually a false detection) with a grade of −1.

  In FIG. 7A, when the distance Dn [m] is in the range of 0 to D1 [m], G1 = 1 (false detection because the distance is too close), and in the range of D1 to D2 [m], y = −a11x + b11 (y is a function, x is a variable) and grade G1 decreases from 1 to 0, G1 = 0 is maintained in the range of D2 to D3 [m], and y = a12x−b12 in the range of D3 to D5 [m]. The grade G1 increases from 0 to 1, and in the range exceeding D5 [m], G1 = 1 is set (false detection because it is too far). The distance D5 is set to several tens of meters, for example.

  As described above, the grade G1 returned (output) from the membership function fm1 (Dn) to the defuzzification unit 72 is too close in the range of 0 to D1 [m], so that the misdetected G1 = In the range of D1 to D2 [m], G1 increases from -1 to 0 without false detection, and in the range of D2 to D3 [m], G1 = 0 and D3 to D5 [m ], G1 decreases from 0 to −1, and when it exceeds D5 [m], G1 = −1 because of a false detection because it is too far away.

  FIG. 7B graphically shows the membership function fm2 (Dn) of the second rule generated by an empirical rule including experiment or simulation.

  As in the first rule, the horizontal axis is the current frame distance Dn [m], and the vertical axis is the grade G2 (compliance, probability). The grade G2 takes a value of 0 to 1 representing “positive (positive detection)”. For example, the grade G2 = 1 is interpreted as being correctly detected in the first grade.

  In FIG. 7B, the grade G2 returned (output) from the membership function fm2 (Dn) to the defuzzification unit 72 is set to G2 = 0 when Dn [m] is in the range of 0 to D1 [m]. In the range of D2 [m], G2 increases from 0 to 1 by a function of y = a21x-b21, and in the range of D2 to D4 [m], G2 = 1 (positive detection because of intermediate distance), and D4 to In the range of D6 [m], G2 decreases from 1 to 0 by a function of y = −a22x + b22, and in the range exceeding D6 [m], G2 = 0 is set (0 <D1 <D2 <D3 <D4 <D5 <D6).

  Next, FIG. 8A shows a figure of the membership function fm3 (Dr) of the third rule.

  The horizontal axis of the membership function fm3 (Dr) is the time ΔT between the current frame and the past frame (in this embodiment, the time between the current frame and the previous frame (past frame), ΔT = 0.066). With the movement distance Dr of the vehicle 12 calculated from [sec] and the vehicle speed v (Dr = ΔT × v = 0.066 v [m]), the function value of the membership function fm3 (Dr) is zero (grade G3 = 0).

  The vertical axis is grade G3 (compliance, probability). Note that the shape of the membership function fm3 (Dr) depends on one frame time ΔT and the vehicle speed v.

  As described above, the membership function fm3 (Dr) has the movement distance Dr of Dr = 0.066v and the function value fm3 (0.066v) = 0. The coefficient α generated by an empirical rule including experiment or simulation is α = several times (exemplary), and more specifically, the membership function fm3 (Dr) is 0.066v on the positive side of the moving distance Dr. From grade .alpha..times.v.times.0.066 = 0.066.alpha.v, the grade G3 increases from 0 to 1 with the function y = a32x-b32, and when it exceeds 0.066.alpha.v, G3 = 1 is set. Further, toward the negative side of the moving distance Dr, the grade G3 increases from 0 to 1 with a function of y = −a31x + b31 from 0.066v to −α × v × 0.066 = −0.066αv, and −0. If it falls below 066αv, G3 = 1 is set.

  The horizontal axis input to the membership function fm3 (Dr) is the current frame image 61 (see FIG. 6) captured by the camera 16 and the past frame image 60 (0.066 [sec]) one frame ahead (0.066 [sec]). The approach distance Di (Di = Dp−Dn) (see FIG. 6) of the curve start point Cc candidate calculated from FIG.

  In practice, the grade G3 of the membership function fm3 (Dr) takes a value of 0 to −1 representing a false (false detection) with a large distance difference, so that the moving distance Dr is compared with other rules. For the convenience of understanding, a negative sign is attached to the grade G3 that is returned with the approach distance Di as an input to the membership function fm3 (Dr) shown in FIG. For example, a grade G3 = −1 with a negative sign is interpreted as being correctly detected (actually a false detection) with a grade of −1.

  In this case, the grade G3 returned (output) from the membership function fm3 (Dr) to the defuzzification unit 72 has an approach distance Di [m] calculated from the images 61 and 60 as the membership function. When fm3 (Dr) has the same value (Di = Dr) as the travel distance Dr = 0.066v calculated from one frame time ΔT and the vehicle speed v, G3 = 0 (the probability of erroneous detection being the lowest) ).

  The distance until the approach distance Di [m] calculated from the images 61 and 60 becomes Dr = 0.066αv which is α times longer than the moving distance Dr = 0.066v calculated from the one frame time ΔT and the vehicle speed v. As the difference becomes larger, grade G3 decreases from 0 to −1, and if Dr = 0.066αv is exceeded, grade G3 is interpreted as G3 = −1 (-1 grade was detected correctly. Is false detection).

  The distance difference is until the approach distance Di [m] calculated from the images 61 and 60 becomes Dr = −0.066αv, which is shorter than the moving distance Dr = 0.066v calculated from one frame time ΔT and the vehicle speed v. As grade G3 decreases from 0 to −1 and falls below Dr = −0.066αv, it is interpreted that grade G3 was correctly detected with a grade of G3 = −1 (−1. , False detection).

  FIG. 8B shows the membership function fm4 (Dr) of the fourth rule.

  The horizontal axis of the membership function fm4 (Dr) indicates that the moving distance Dr of the vehicle 12 calculated from the time ΔT and the vehicle speed v between the current frame and the past frame is Dr = ΔT × v = 0.066v [m]. The function value of the membership function fm4 (Dr) is 1 (grade G4 = 1).

  The vertical axis is grade G4 (compliance, probability). Also in this case, it should be noted that the shape of the membership function fm4 (Dr) depends on the one frame time ΔT and the vehicle speed v.

  More specifically, the membership function fm4 (Dr) has a coefficient β of (0 <β <1), for example, between Dr = (1−β) 0.066v and Dr = (1 + β) 0.066v. The function value fm4 ((1-β) 0.066v) = 1 is set, and the coefficient α generated by an empirical rule including experiment or simulation is set to α = several times, and α × v × 0. The grade G4 decreases from 1 to 0 by a function of y = −a42x + b42 until 066 = 0.066αv, and when it exceeds 0.066αv, G4 = 0 is set. Further, toward the negative side of the moving distance Dr, the grade G4 decreases from 1 to 0 with a function of y = a41x + b41 until −α × v × 0.066 = −0.066αv, and when it falls below −0.066αv, G4 = 0 is set.

  Similar to the third rule, the horizontal axis for this membership function fm4 (Dr) is the current frame image 61 (see FIG. 6) captured by the camera 16 and one frame (0.066 [sec]) before. The approach distance Di (Di = Dp−Dn) (see FIG. 6) of the candidate curve start point Cc calculated from the past frame image 60 (see FIG. 6).

  Since the grade G4 of the membership function fm4 (Dr) takes a value of 0 to 1 indicating a positive (positive detection) with a small distance difference, it is assumed that the grade G4 = 1 is correctly detected (positive detection) in the first grade. To do.

  That is, the grade G4 returned (output) from the membership function fm4 (Dr) to the defuzzification unit 72 has an approach distance Di [m] calculated from the images 61 and 60 as the membership function fm4. When (Dr) is between Di = (1-β) 0.066v and Di = (1 + β) 0.066v with respect to the travel distance Dr = 0.066v calculated from one frame time ΔT and the vehicle speed v. , It is assumed that G4 = 1, that is, the probability that the positive detection is the highest, that is, the detection is correct.

  The approach distance Di [m] calculated from the images 61 and 60 becomes Dr = 0.066αv which is α times longer than the moving distance Dr = (1 + β) 0.066v calculated from one frame time ΔT and the vehicle speed v. Thus, the grade G4 decreases from 1 to 0 and the grade G4 becomes G4 = 0 (it cannot be said that the distance difference is small) in the range exceeding Dr = 0.068αv. .

  The approach distance Di [m] calculated from the images 61 and 60 is shorter than the travel distance Dr = (1-β) 0.066v calculated from the one frame time ΔT and the vehicle speed v = 0.068αv. Since the distance difference becomes smaller until grade G4 decreases from 1 to 0, and the travel distance Dr falls below Dr = −0.066αv, grade G4 becomes G4 = 0 (the distance difference is small) I can't say that.)

  When the fuzzy inference is thus performed, in step S7, the curve determination unit 44 performs a defuzzification process, here, a curve start point determination process in the defuzzification unit 72.

  In this defuzzification process, the average (synthetic output center of gravity) Gmean of the grades G1 to G4 obtained by the first, second, third, and fourth rules is compared with the threshold Th, and the average Gmean is compared. Is equal to or greater than the threshold Th (Gmean ≧ Th), the distance Dn from the vehicle 12 from the current frame image 61 to the curve start point Cc candidate is identified as a correct value, and the curve start point Cc candidate is set to the curve start. It is identified as point Cc.

  Note that the fuzzy inference in step S6 may use at least two rules in order to obtain the average Gmean in step S7, and a threshold Th that matches each rule used is set in advance.

  If the average Gmean is compared with the threshold Th and the average Gmean is equal to or greater than the threshold Th (Gmean ≧ Th), the distance Dn from the vehicle 12 from the current frame image 61 to the curve start point Cc candidate is correct. When identifying the curve start point Cc candidate as the curve start point C1, the amplitude (edge strength) of the image signal Si representing the lane mark is not rubbed in fine weather or when the lane mark is not rubbed. If the boundary is clearly visible, the brightness difference (reflection brightness difference) between the road surface image and the lane mark image (lane boundary line image) increases, so the distance Dn and the approach distance Di are The threshold Th may be decreased in consideration of the fact that it can be calculated more reliably. Thus, by increasing / decreasing the threshold Th, the curve start point Cc can be determined from the curve start point Cc candidates with a certain certainty regardless of the road environment.

  When the curve start point C1 is determined in step S8 (step S8: YES), in step S9, the curve determination unit 44 determines the distance Dn between the vehicle travel support start signal and the curve start point C1 as the driving support device 14. Output to.

  Accordingly, the driving support device 14 prevents the vehicle 12 from departing from the lane between the lane marks along the lane mark stored in the lane mark memory 34, in other words, the vehicle 12 moves to the center of the curve lane. The steering device 50, the braking device 52, the acceleration device 54, and the like are controlled so that the vehicle can run, and the driver's driving assistance is executed. When it is determined that the vehicle 12 has not sufficiently decelerated before the curve start point C1, the braking device 52 is operated to assist. Alternatively, a reaction force is applied to an accelerator pedal (not shown) in order to urge the braking device 52 to operate.

[Summary of Embodiment and Modifications]
(1) Outline Description As described above, the curve detection device 10 according to the above-described embodiment performs image processing on the image signal Si of the vehicle-mounted camera 16 and detects the curve start point Cc. At this time, the following first and second methods using fuzzy inference are used to determine whether or not the curve start point Cc has been correctly made.

  First, from the image 61 of the current frame at the time point t1, the position of the curve start point Cc (position where the interpolation line indicated by the broken line in FIG. 6 intersects) is obtained from the vehicle 12, and then the vehicle 12 (camera The distance Dn from the 16 attachment positions) to the curve start point Cc is calculated. In this case, if the calculated distance Dn is too far, it is considered that there is a low possibility that the detected image is correctly detected in relation to the resolution of the captured image (low in the distance and high in the vicinity) (first rule). Further, when the calculated distance Dn is too close, it is considered that there is a low possibility of being correctly detected due to noise or the like (first rule). Further, when the calculated distance Dn is an intermediate distance, it is considered that there is a high possibility that the calculated distance Dn is correctly detected (second rule). This is referred to as the first method.

  Next, the approach from the distance Dp of the curve start point Cc of the past frame (time t0) calculated on the image 60 to the distance Dn of the curve start point Cc of the current frame (time t1) calculated on the image 61 is performed. The distance Di (Di = Dp−Dn) is obtained, and the approach distance Di and the moving distance Dr (Dr = v × ΔT) of the vehicle 12 obtained from the vehicle speed v and the time (one frame time) ΔT between both frames are obtained. In comparison, when the distance difference | Di−Dr | is large, it is considered that the possibility of being correctly detected is low (third rule), and when the distance difference | Di−Dr | I think it is expensive (4th rule). This is referred to as the second method.

  With respect to the first and second methods, the curve start point Cc is stably detected using fuzzy inference.

(2) Detailed Description The curve detection device 10 according to the above-described embodiment detects a curve start point Cc in front of the vehicle based on an image including a lane mark imaged by the camera 16 mounted on the vehicle 12.

  In the curve detection device 10, the curve candidate detection unit 42 starts a curve that may be a curve start point Cc ahead of the vehicle 12 from the images 59, 60, 61 obtained by performing image processing on the image including the lane mark. The point Cc candidate is detected, and the curve determination unit 44 determines whether the curve start point Cc candidate detected by the curve candidate detection unit 42 is the correct curve start point Cc by using fuzzy inference. ing.

  Thus, whether or not the curve start point Cc candidate ahead of the vehicle 12 detected by performing image processing on the image including the lane mark is the correct curve start point Cc is determined using fuzzy inference. Since it is determined, the curve start point Cc can be stably detected.

  In this case, the fuzzy inference is based on the curve start point Cc calculated from the images 61 and 60 of different frames (current frame at time t1 and, for example, past frame at time t0 one frame before) captured by the camera 16. The antecedent part has a case where the approach distance Di to the candidate vehicle 12 is approximately equal to the time ΔT between the different frames (current frame and past frame) and the moving distance Dr of the vehicle 12 calculated from the vehicle speed v. By including the rule (the third rule and / or the fourth rule), it is possible to stably detect the curve start point Cc by including this new membership function {fm3 (Dr) and / or fm4 (Dr)}. A reasonable result can be obtained.

  The curve determination unit 44 selects at least two of the first, second, third, and fourth rules used for fuzzy inference to determine the correctness of the curve start point Cc candidate. Thus, it is possible to increase the probability of being the curve start point Cc.

  In the four rules, when the distance Dn from the vehicle 12 to the curve start point Cc candidate calculated from the image 61 of the current frame is too far or too close, the first rule is the distance that is too close. In this case, the rule is expressed as a membership function fm1 (Dn) or the like.

  The second rule is that the distance Dn from the vehicle 12 to the curve start point Cc candidate calculated from the image 61 of the current frame is an intermediate distance between the distance too far and the distance too close. The rule is expressed as a membership function fm2 (Dn) or the like.

  The third rule is calculated from the approach distance Di of the curve start point Cc candidate calculated from the images 61 and 60 of the current frame and the past frame to the vehicle 12, the time ΔT between the current frame and the past frame, and the vehicle speed v. If there is a large gap between the vehicle 12 and the moving distance Dr of the vehicle 12, the rule expressed by the membership function fm3 (Dr) is assumed to be erroneous.

  The fourth rule is calculated from the approach distance Di of the curve start point Cc candidate calculated from the images 61 and 60 of the current frame and the past frame to the vehicle 12, the time ΔT between the current frame and the past frame, and the vehicle speed v. If the moving distance Dr of the vehicle 12 is substantially the same, the rule is expressed as a positive membership function fm4 (Dr) or the like.

  Note that the curve determination unit 44 determines the combined output centroid Gmean of each membership function {two of fm1 (Dn), fm2 (Dn), fm3 (Dr), fm4 (Dr)} of the selected two rules. When the curve start point Cc is determined by comparing the threshold Th with the threshold Th, the threshold Th may be increased or decreased according to the size of the image signal Si representing the lane mark of the captured image. Since the magnitude of the amplitude of the image signal Si representing the lane mark changes depending on the surrounding environment of the road, such as rubbing of the lane mark, weather, etc., if the threshold value Th is increased or decreased according to the magnitude of the image signal Si, Regardless, the curve start point Cc can be determined with certain certainty from the curve start point Cc candidates.

  When the curve determination unit 44 determines the curve start point Cc, the curve determination unit 44 further generates a vehicle travel support start signal from the front of the curve start point Cc, so that a suitable vehicle 12 travel support, for example, reliable automatic steering, It is possible to execute steering assist without a sense of incongruity, timely application of accelerator pedal reaction force, and the like.

(3) Modification As shown in FIG. 9, as a new rule for fuzzy inference, the curvature radius R of the curve 104 obtained from the map information, the curve 104 of the curve 104 on the image, and the curve start point Cc of the curve 104 A rule that the curve 104 is correctly detected may be introduced if the horizontal divergence distances 106, 108, and 110 with respect to the extension line 102 of the straight path 100 on the near side correspond correctly. By introducing a new rule, the curve start point Cc and the robustness of curve detection can be further improved.

  Further, the present invention is not limited to the above-described embodiment, and it is needless to say that various configurations can be adopted based on the contents described in this specification.

DESCRIPTION OF SYMBOLS 10 ... Curve detection apparatus 11 ... Lane mark recognition apparatus 12 ... Vehicle 14 ... Driving assistance apparatus 16 ... Camera 20 ... ECU
DESCRIPTION OF SYMBOLS 22 ... Speed sensor 26 ... Steering angle sensor 36 ... Lane mark recognition part 40 ... Curve candidate memory 42 ... Curve candidate detection part 44 ... Curve determination part 50 ... Steering device 52 ... Braking device 54 ... Acceleration device 59 ... Past two frames ago Plane view edge image 60 ... Plane view edge image 61 of the previous frame one frame before ... Plane view edge image 70 of the current frame 70 ... Fuzzy inference unit 72 ... Dephasing unit

Claims (5)

In a curve detection device for detecting a curve ahead of a vehicle based on an image including a lane boundary line imaged by an imaging device mounted on the vehicle,
A curve candidate detection unit for detecting a curve candidate that may be a curve ahead of the vehicle by performing image processing on an image including the lane boundary;
A curve determination unit that determines whether or not the curve candidate detected by the curve candidate detection unit is a curve, using fuzzy inference;
A curve detection device comprising: The curve detection device according to claim 1,
The fuzzy inference includes an approach distance to the vehicle of the curve candidate calculated from images of different frames imaged by the imaging device, a moving distance of the vehicle calculated from a time between the different frames and a vehicle speed. A curve detection device comprising a rule having an antecedent part when the two are substantially equal. The curve detection device according to claim 1,
The curve determination unit
Among the four rules used for the fuzzy inference, at least two rules are selected to determine whether the curve start point candidate is correct or incorrect in the curve.
The four rules are:
A first rule that is false if the distance from the vehicle to the curve start point candidate calculated from the image of the current frame is too far or too close;
A second rule that is positive if the distance from the vehicle to the curve start point candidate calculated from the image of the current frame is an intermediate distance between the distance too far and the distance too close ,
There is a large difference between the approach distance of the curve start point candidate calculated from the images of the current frame and the past frame to the vehicle, and the moving distance of the vehicle calculated from the time and the vehicle speed between the current frame and the past frame. If there is a gap, the third rule is false,
Further, the approach distance of the curve start point candidate calculated from the images of the current frame and the past frame to the vehicle, and the moving distance of the vehicle calculated from the time between the current frame and the past frame and the vehicle speed are approximately. A curve detecting device comprising a fourth rule that is positive if they are the same. In the curve detection device according to claim 3,
The curve determination unit
When the curve is determined by comparing the composite output centroid of each membership function of the two selected rules and the threshold value, the threshold value is set to the magnitude of the image signal representing the lane boundary line of the captured image. A curve detection device characterized by increasing or decreasing in response. In the curve detection apparatus according to any one of claims 1 to 4,
The curve determination unit
When the curve or the curve start point is determined, a vehicle travel support start signal is further generated from before the curve or the curve start point.

Images