JP2013125327A - Curvature estimation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a curvature estimation device for estimating a curvature of a road with high precision.SOLUTION: A curvature estimation device estimates a curvature of a road on the basis of an image taken by imaging means. A vehicle travels along a road (traffic lane) according to steering of a driver in consideration of meander by the steering of the vehicle driver. However, the traffic lane has periodic and minute meanders if the traffic lane is observed from a traveling vehicle. It is assumed that a meander of the traffic lane is generated by the steering of the driver, a road model in which the curvature of the road is an input while the steering angle of the steering is an output is constructed, real vehicle traveling data is applied to a Kalman filter where the road model is incorporated, and the curvature is estimated on the basis of a result of driving of the Kalman filter to the curvature of the road obtained from the image. The frequency of meanders by the steering of the vehicle driver is preferably applied to the road model.

Description

  The present invention relates to a curvature estimation apparatus that estimates the curvature of a road based on an image captured by an imaging unit.

  Patent Document 1 describes a lane departure prevention support device that detects lane section lines on both the left and right sides of a host vehicle and assists in preventing lane departure of the host vehicle based on the detection result of the lane section line, and is captured by a camera. It is disclosed that the curvature of the lane is detected from the image ahead of the host vehicle.

JP 2011-118828 A

  As described above, when estimating the curvature of a road (lane), a camera is usually used. When the image captured by the camera is processed and the curvature is estimated, noise (for example, noise of edge processing on the image, camera noise) may be included. Therefore, the accuracy of curvature estimation decreases.

  Then, this invention makes it a subject to provide the curvature estimation apparatus which estimates the curvature of a road with high precision.

  A curvature estimation apparatus according to the present invention is a curvature estimation apparatus that estimates the curvature of a road based on an image captured by an imaging unit, and inputs a steering angle of steering based on meandering by steering of a vehicle driver. The result of driving the Kalman filter against the road curvature obtained from the image captured by the imaging means by constructing a road model that outputs the road curvature and applying the actual vehicle driving data to the Kalman filter incorporating the road model Based on the above, the curvature is estimated.

  Geometric roads cannot be constructed with Kalman filters. Therefore, the curvature estimation device takes into account meandering caused by steering by the driver of the vehicle, and constructs a road model with the steering angle as an input and the road curvature as an output. The vehicle travels along the road (lane) according to the driver's steering, but assuming that the lane is observed from the running vehicle, the lane is periodically meandering, and the lane meander is It is thought that it is generated by driver's steering. Therefore, since this lane change (meandering) is associated with the rudder angle according to the driver's steering, a road model composed of the lane curve and the rudder angle is constructed, and the Kalman filter is applied by using this road model. It can be controlled. In this curvature estimation model, actual vehicle travel data (steering angle, yaw rate (lateral acceleration), yaw angle (lateral velocity), etc.) is applied to the Kalman filter incorporating the road model, and the curvature of the road obtained from the image is calculated. On the other hand, the curvature is estimated based on the result of driving the Kalman filter. Thus, in the curvature estimation apparatus, a Kalman filter that incorporates a road model in which the curvature of the road is associated with meandering by the driver's steering is configured, so that the calculation load is reduced with a simple configuration (calculation) and is obtained from the image. Noise contained in the curvature can be suppressed, and the curvature of the road can be estimated with high accuracy.

  In the curvature estimation apparatus of the present invention, it is preferable to calculate the meander frequency by the steering of the vehicle driver and apply the meander frequency to the road model. Since the steering characteristics are different depending on the driver, the meandering frequency is also different. Therefore, in this curvature estimation device, a meandering frequency by steering for each driver is obtained, and the meandering frequency is applied to a road model to constitute a Kalman filter incorporating a road model according to the driver's steering characteristics. The curvature of the road can be estimated with higher accuracy.

  In the curvature estimation apparatus of the present invention, it is preferable to estimate the curvature using a monotonically increasing function or a monotonically decreasing function when the road is an entrance / exit of a curve. In normal vehicle travel, the vehicle is incut at the entrance of the curve and bulges at the exit of the curve. Therefore, in this curvature estimation device, it is possible to estimate the curvature of the road with higher accuracy by estimating the curvature using a monotonically increasing function or a monotonically decreasing function in order to consider the running characteristics at the curve entrance. it can.

  In the curvature estimation apparatus of the present invention, it is preferable that the Kalman filter is not driven when the road is a straight line. When the road is a straight road, the meandering is larger than the noise, and the error increases when the Kalman filter is used. Therefore, in this curvature estimation apparatus, when the road is a straight line, the curvature error is suppressed without driving the Kalman filter for the curvature of the road obtained from the image.

  According to the present invention, by configuring a Kalman filter that incorporates a road model in which the curvature of the road is associated with meandering by the driver's steering, the calculation load is reduced with a simple configuration (calculation) and is included in the curvature required from the image. Noise can be suppressed, and the curvature of the road can be estimated with high accuracy.

It is a block diagram of the curvature estimation apparatus which concerns on this Embodiment. It is a figure which shows the curvature model of the road used by this Embodiment. It is a flowchart which shows the flow of the curvature estimation process in ECU of FIG.

  Embodiments of a curvature estimation apparatus according to the present invention will be described below with reference to the drawings. In addition, the same code | symbol is attached | subjected about the element which is the same or it corresponds in each figure, and the overlapping description is abbreviate | omitted.

  The curvature estimation apparatus according to the present embodiment is mounted on a vehicle and estimates the curvature of a road (particularly, a white line constituting a lane) based on an image captured by an in-vehicle camera. The estimated curvature information is used in various driving support devices (for example, a lane departure prevention device and a lane keeping device) mounted on the vehicle, a driver model used for a dozing detection device, and the like. In this embodiment, the curvature of the white line constituting the lane drawn on the road is estimated as the curvature of the road. Since the white line is drawn along the road, the curvature of the white line corresponds to the curvature of the road.

  With reference to FIG.1 and FIG.2, the curvature estimation apparatus 1 which concerns on this Embodiment is demonstrated. FIG. 1 is a configuration diagram of a curvature estimation apparatus according to the present embodiment. FIG. 2 is a diagram showing a curvature model used in the present embodiment.

  The curvature estimation apparatus 1 uses vehicle information (steering angle, yaw rate (lateral acceleration), etc.) correlated with the road curvature in addition to the road curvature from the camera image, and the curvature associated with meandering by the driver's steering. Build a model (road model) and incorporate the curvature model into the Kalman filter. Then, the curvature estimation apparatus 1 receives the steering angle by the driver's steering, inputs the curvature as an output, drives the Kalman filter, and estimates the curvature based on the observation result of the Kalman filter.

  Since vehicle information (steering angle, yaw rate, etc.) correlated with the curvature is used in addition to the curvature information based on the camera image in this way, the road curve at the entrance / exit of the curve can be obtained by using the vehicle information correlated with the curvature. It can cope with the slowing of the waveform.

  A Kalman filter is used to suppress curvature noise due to the camera image. A necessary and sufficient condition for the Kalman filter to be established is that the system to be observed is controllable and observable. However, if the geometric road is modeled as it is, the geometric curvature of the road will not change even if the driver steers the vehicle to the right or left, so the curvature at the steering angle and yaw rate (lateral acceleration) is Impossible control. Therefore, a Kalman filter cannot be configured. However, since the actual vehicle travels along the road, there is a significant correlation between the curvature and the steering angle (yaw rate, lateral acceleration). Therefore, a Kalman filter can be configured if it is a controllable state equation.

When the vehicle actually travels on the road, it is not a travel locus that is completely parallel to the white line that constitutes the lane. It meanders with a slight amplitude (for example, an amplitude of the order of 10 ⁻² m) in the width direction of the road. Therefore, when the white line is observed from the vehicle, the white line appears to meander. Therefore, it is assumed that the vehicle travels almost along the road, and when the white line is seen from the running vehicle, it is observed that the white line is meandering substantially periodically, and the white line meandering is generated by the driver's steering. Since the change in the white line is associated with the rudder angle, the system composed of the white line (curvature) and the rudder angle or yaw rate (lateral acceleration) can be controllable.

Therefore, the curvature model of the road (white line) shown in FIG. 2 is considered. Here, the white line WL is observed from the vehicle MV using a vehicle fixed coordinate system. V is the speed of the vehicle MV. a is the lateral acceleration of the vehicle MV (the dimension of a is obtained by multiplying the curvature ρ of the white line WL by V ² ). y is the lateral position of the white line WL with respect to the vehicle MV. v is the lateral speed of the white line WL with respect to the vehicle MV. α is the lateral acceleration of the white line WL with respect to the vehicle MV. When the yaw angle of the vehicle MV with respect to the white line WL is θ, there is a relationship θ = v / V. The curvature estimation apparatus 1 constructs a curvature model on the assumption that a road (white line) meanders in such a vehicle fixed coordinate system, and constructs a Kalman fill using the road curvature model.

  The meandering by the driver's steering is periodic meandering, but the meandering period varies depending on the driver. Therefore, the meander frequency is calculated, and the meander frequency is used in the curvature model. Further, since the meandering is meandering in the vicinity of the LSB, the frequency analysis has a low SN ratio. Therefore, since the meandering frequency is a single vibration, the meandering frequency is calculated from the overall power ratio of each value of the lateral acceleration a and the lateral velocity v (white line yaw angle). Further, when the curvature cannot be detected by the detection by the camera image, the curvature estimation error becomes large. Therefore, when it cannot be detected, the curvature is estimated with an emphasis on the Kalman filter estimation function.

  However, in the case of the entrance / exit of the curve, it does not meander like during normal curve travel, but travels in-cut at the entrance and swells at the exit. Therefore, a monotonically increasing function is used at the entrance of the curve, and a monotonically decreasing function is used at the exit. Further, when the road is a straight line, meandering is larger than noise included in the curvature of the camera image, so that the error becomes larger when the Kalman filter is driven. Therefore, when the road is a straight line, the curvature detected by the camera is used as it is without driving the Kalman filter, or the noise setting value in the Kalman filter is adjusted.

The equation of the white line curvature model shown in FIG. 2 is the differential equation of equation (1). Further, the equation of the curvature model of the white line that is a function of monotonic increase and monotonic decrease at the entrance and exit of the curve is the differential equation of equation (2). The ρ _filter in the expressions (1) and (2) is the difference between the detection position of the front (or rear) of the camera image and the position of the vehicle (detection position of the steering angle, etc.) with respect to the curvature ρ _cam of the detection by the camera image. This is the curvature that has been filtered to remove the time difference, and is the curvature at the vehicle position. ζ is a meandering attenuation ratio. ω is the meander frequency. G is a gain of lateral acceleration with respect to the steering angle. τ is a time constant of lateral acceleration with respect to the steering angle. Note that “·” above each symbol indicates first-order time differentiation.

  Now, the curvature estimation apparatus 1 will be specifically described. As shown in FIG. 1, the curvature estimation apparatus 1 includes a camera sensor 10, a steering angle sensor 11, a lateral acceleration sensor 12, a vehicle speed sensor 13, and an ECU [Electronic Control Unit] 20.

  The camera sensor 10 is a sensor for recognizing a white line constituting a lane, and includes a camera (imaging means) and an image processing device. The camera is a camera using a CCD [Charge Coupled Device] or the like. The camera is attached to the front or rear of the vehicle. At this time, the camera is attached so that the optical axis direction thereof coincides with the traveling direction of the vehicle. The camera captures an area including a road ahead or behind the vehicle at regular time intervals, acquires the captured image, and outputs the image information to the image processing apparatus. The camera has a wide imaging range in the left-right direction, and can sufficiently capture the white lines on the left and right sides (a pair) indicating the traveling lane. The camera may be a color camera or a monochrome camera.

Each time image information is input from the camera, the image processing apparatus recognizes a pair of white lines indicating the lane in which the vehicle is traveling from the image. As this recognition method, for example, there is a method based on edge processing because a difference in luminance between a road surface and a white line drawn thereon is large. Then, the image processing apparatus calculates the curvature ρ _cam of one of the recognized pair of white lines. Further, the image processing apparatus calculates the horizontal position y (the horizontal distance from the vehicle to the white line) of one of the white lines. In the image processing apparatus, the lateral velocity v is calculated using the time series data of the lateral position y. Further, the image processing apparatus calculates the lateral acceleration α using the time series data of the lateral velocity v. Also, in the image processing apparatus, when the white line can be recognized and the curvature can be detected, the curvature lost flag flag_lost is set to 0, and when the white line cannot be recognized and the curvature cannot be detected (when lost), The curvature lost flag flag_lost is set to 1. Then, the camera sensor 10 outputs the curvature lost flag flag_lost and data of the curvature ρ _cam , the lateral position y, the lateral velocity v, and the lateral acceleration α to the ECU 20 every predetermined time. Note that the curvature ρ _cam , the lateral position y, the lateral velocity v, and the lateral acceleration α use the front (or rear) image captured by the camera, so the curvature ρ _cam , the lateral position y, the lateral velocity v, and the lateral acceleration α are used. Is detected at a predetermined position in front (or rear) of the host vehicle, and there is a time difference between the detected position in front (or rear) and the host vehicle position.

  The rudder angle sensor 11 is a sensor that detects the rudder angle of the steered wheels that changes according to steering by the driver. The rudder angle sensor 11 detects the rudder angle δ at regular intervals and outputs data of the rudder angle δ to the ECU 20. The lateral acceleration sensor 12 is a sensor that detects the lateral acceleration of the vehicle. The lateral acceleration sensor 12 detects the lateral acceleration a at regular time intervals and outputs data of the lateral acceleration a to the ECU 20. The vehicle speed sensor 13 is a sensor that detects the speed of the vehicle. The vehicle speed sensor 13 detects the vehicle speed V at regular time intervals and outputs data on the vehicle speed V to the ECU 20.

The ECU 20 is an electronic control unit including a CPU [Central Processing Unit], a ROM [Read Only Memory], a RAM [Random Access Memory], and the like. The ECU 20 inputs data from each of the sensors 10, 11, 12, and 13 at regular intervals, and estimates the curvature ρ of the white line using the input vehicle travel data. As described above, the data to be input at regular intervals include the curvature ρ _cam from the camera sensor 10, the lateral position y, the lateral velocity v, the lateral acceleration α, the curvature lost flag flag_lost, the steering angle δ from the steering angle sensor 11, The lateral acceleration a from the lateral acceleration sensor 12 and the vehicle speed V from the vehicle speed sensor 13 are shown.

The ECU 20 reads a preset setting value at startup. The set values include stability factor A, steering gear ratio N, wheel base 1, distance L between camera and curvature measurement point, curvature threshold value ρo at the vehicle position, and forward position (or rearward position) detected by camera sensor 10. ) Curvature threshold value ρo _cam , meandering attenuation ratio ζ0 (for example, 0.5), crosonoid divergence coefficient ζp (for example, 0.5), crosonoid attenuation coefficient ζn (for example, 0.5), meander frequency observation time constant T , Time constant τ of lateral acceleration with respect to the steering angle, curvature dispersion σ _ρ ² , lateral acceleration dispersion σ _a ² , lateral velocity dispersion σ _v ² , lateral position dispersion σ _y ² , bottom weighting coefficient base W_lost (<< 1) , There is a straight road weighting coefficient W_str (>> 1). These set values are set for vehicle-specific values such as a wheel base, and other values are set in advance by experiments or the like and stored in the ROM in the ECU 20.

Further, the ECU 20 sets an initial value at the time of activation. The initial value, 0 is set as the incremental .DELTA.v2 ^* of the mean square value of the lateral speed, 0 is set as the incremental .DELTA.a2 ^* of the mean square value of the lateral acceleration, the 0 transients flag (curve entrance flag) tr Substitution is performed, the curvature ρ is set to 0, and the straight state flag straight is set to 0. tr is a flag of 0/1, and a case of 1 is a curve entrance. straight is a 0/1 flag, and a case of 1 is a straight road. The increment Δ is an increment from the previous value.

In the ECU 20, integral values for the square value of the lateral velocity v and the square value of the lateral acceleration a are sequentially obtained, and the ratio of each integral value (each value of the lateral acceleration and the lateral velocity (white line yaw angle)). The meandering frequency ω is obtained from the overall power ratio of Specifically, the mean square value v2 ^* (previous value) of the lateral velocity, the lateral velocity v and the meander frequency observation time constant T input this time are used, and the increment Δv2 of the mean square value of the lateral velocity according to the equation (3). ^* Is calculated, and the increment Δv2 ^* is added to the mean square value v2 ^* (previous value) of the lateral speed to obtain the current value of the mean square value v2 ^* of the lateral speed as shown in Equation (5). Also, using the lateral acceleration square average value a2 ^* (previous value), the current input lateral acceleration a, and the meander frequency observation time constant T, calculate the square acceleration average Δa2 ^* of the lateral acceleration according to equation (4). Then, as shown in the equation (6), the increment Δa2 ^* is added to the mean square value a2 ^* (previous value) of the lateral acceleration to obtain the present value of the mean square value a2 ^* of the lateral acceleration. Then, the meander frequency ω is calculated by the equation (7) using the mean square value v2 ^* (current value) of the lateral velocity and the mean square value a2 ^* (current value) of the lateral acceleration.

Further, the ECU 20 performs phase adjustment in order to adjust the time difference between the detection position (front position) of the curvature ρ _{cam from} the camera image and the detection position (own vehicle position) of the steering angle δ. Specifically, using the curvature ρ _fit (previous value) adjusted in phase, the curvature ρ _cam based on the camera image input this time, the vehicle speed V, and the distance between the camera and the curvature measurement point, the curvature increment Δρ _filter is calculated by the equation (8). calculate. Then, as shown in Expression (9), a value obtained by adding the increment Δρ _filter to the increment Δρ _cam of the curvature ρ _cam by the camera image is set as the current value of the phase-adjusted curvature ρ _filter . The phase-adjusted curvature ρ _fit (current value) is a value that is phase-adjusted so that the curvature ρ _cam at the forward position based on the camera image detected this time becomes the position of the _host vehicle.

Further, the ECU 20 calculates the lateral acceleration gain G with respect to the rudder angle by the equation (10) using the vehicle speed V, the stability factor A, the wheel base l, and the steering gear ratio N that are input this time.

The ECU 20 determines the entrance / exit of the curve and sets the transient flag tr. In the case of the curve entrance / exit (when tr = 1), the meandering damping ratio ζ and the meandering frequency ω are set. Specifically, as shown by the conditional expression of Expression (11), it is determined whether or not the absolute value of the curvature ρ _cam at the front position of the camera detection is larger than the threshold ρo _cam . If the curvature [rho absolute value is larger than the threshold .rho.o _cam of _cam a curvature showing a circular white line in the forward position, showing the white line is a straight line in a forward position when the absolute value of the curvature [rho _cam following threshold .rho.o _cam Curvature. Furthermore, as shown by the conditional expression of Expression (12), it is determined whether or not the absolute value of the curvature ρ at the vehicle position is larger than a threshold value ρo. When the absolute value of the curvature ρ is larger than the threshold ρo, the white line is a curvature showing a circle at the own vehicle position, and when the absolute value of the curvature ρ is less than the threshold ρo, the white line is a curvature showing a straight line at the own vehicle position. is there.

Therefore, when the absolute value of the case and the curvature [rho absolute value of the curvature [rho _cam is greater than the threshold value .rho.o _cam is greater than the threshold .rho.o, as shown in (13), a circular from the circle (in cornering), Transient Flag tr Is set to 0. If the absolute value of the curvature [rho _cam absolute value if and curvature [rho threshold .rho.o _cam greater is less than the threshold value .rho.o, as shown in (14), a straight line from the circle (in the curve exit travel), as Transient Flag tr 1, 1 is set as the meandering damping ratio ζ, the crsonicoid damping coefficient ζn, and 1 is set as the meandering frequency ω. If the absolute value of the absolute value of the threshold .rho.o _cam following cases and the curvature [rho curvature [rho _cam is greater than the threshold value .rho.o, as shown in (15), a circular from a straight line (in curve entrance traveling), a Transient Flag tr 1, 1 is set as the meandering damping ratio ζ, the crosonoid divergence coefficient ζp is set, and 1 is set as the meandering frequency ω. When the absolute value of the curvature ρ _cam is equal to or smaller than the threshold ρo _cam and the absolute value of the curvature ρ is equal to or smaller than the threshold ρo, the straight line is straight (during straight line travel) as shown in (16), and the transient flag tr is 0. And 1 is set as the straight line state flag straight.

In the ECU 20, each set value, the curvature lost flag flag_lost input from each sensor 10, 11, 12, 13, the lateral speed v, the lateral position y, the steering angle δ, the lateral acceleration a, the vehicle speed V, the calculated meander frequency ω, and the rudder The Kalman gain of the Kalman filter is calculated by using the lateral acceleration gain G with respect to the angle, the phase-adjusted curvature ρ _fit , the set transient flag tr, the straight state flag straight, and the meandering damping ratio ζ, and the Kalman gain is calculated based on the Kalman gain. Driven to estimate the curvature ρ at the vehicle position. Below, the Kalman filter used by ECU20 is demonstrated concretely.

  The Kalman filter applies the model expression of the above equation (1) when the transient flag tr = 0 (when it is not the curve entrance / exit), and the above equation (when the transient flag tr = 1 (when the curve entrance / exit)) The model formula of 2) is configured to be applied. Further, the Kalman filter is configured such that the filter does not act when the straight state flag straight = 0 (in the case of a straight road). Further, the Kalman filter is configured such that when the curvature lost flag flag_lost = 1 (when no curvature is detected by camera detection), the Kalman filter estimation function is emphasized.

The Kalman filter is configured to satisfy the above conditions. The state equation of the Kalman filter is Expression (17). The state quantity x in the state equation (17) is each value shown by the equation (18). The input amount u in the state equation (17) is the steering angle δ as shown in the equation (19). A in the state equation (17) is a 4 × 4 matrix represented by the equation (20). B in the state equation (17) is a four-dimensional vector represented by the equation (21).

Moreover, the process noise P is Formula (22). System noise Q is expressed by equation (23). Depending on the balance between the process noise P and the system noise Q, the degree of emphasis between the correction function and the estimation function in the Kalman filter changes. Note that σ _δ ² is the variance of the steering angle δ.

The output equation of the Kalman filter is Expression (24). The observation amount z in the output equation (24) is each value indicated by the equation (25). In particular, the steering angle δ is an output. C in the output equation (25) is a 4 × 4 matrix represented by the equation (26). Here, r is the yaw angular velocity of the vehicle with respect to the white line, and θ is the yaw angle of the vehicle with respect to the white line.

  When the transient flag tr = 1, − (1−tr) ω = 0 in the matrix of Expression (20), and the model expression of Expression (2) is applied. When the transient flag tr = 0, − (1-tr) ω = −ω in the matrix of the equation (20) is obtained, and the model equation of the above equation (1) is applied.

In the case where the curvature lost flag flag_lost = 1 (when the curvature cannot be detected from the camera image), P in the equation (22) is almost zero. When the curvature lost flag flag_lost = 0 (when the curvature can be detected from the camera image), P = W_str ^straight in Expression (22). As a result, the ratio between the process noise P and the system noise Q changes, so the degree of emphasis between the Kalman filter estimation function and the correction function changes, especially when the curvature cannot be detected from the camera image. Will be.

  When the curvature lost flag flag_lost = 0 and the straight line state flag straight = 1 (in the case of a straight road), P = W_str in equation (22). When the curvature lost flag flag_lost = 0 and the straight line state flag straight = 0 (in the case of a curved road), P = 1 in the equation (22). As a result, the Kalman filter does not work on straight roads, and the Kalman filter works on curved roads.

  Then, when A is obtained by equation (20), B is obtained by equation (21), C is obtained by equation (26), P is obtained by equation (22), and Q is obtained by equation (24), the Kalman gain is A, B, C, P , Q is uniquely determined. Then, the Kalman filter is driven using the obtained A, B, C, P, and Q. As a result, the observed values shown in Expression (25) are obtained, and the curvature ρ is output.

  With reference to FIG.1 and FIG.2, operation | movement of the curvature estimation apparatus 1 is demonstrated. In particular, the processing in the ECU 20 will be described along the flowchart of FIG. FIG. 3 is a flowchart showing the flow of curvature estimation processing in the ECU of FIG. The curvature estimation apparatus 1 starts when the ignition switch is turned on, stops when the ignition switch is turned off, and repeatedly performs the following operation at predetermined time intervals while the ignition switch is turned on.

In the camera sensor 10, the front of the host vehicle is imaged by the camera, the curvature ρ _cam , the lateral position y, the lateral velocity v, and the lateral acceleration α are calculated based on the captured image by the image processing device, and the curvature lost flag flag_lost is set. , Curvature ρ _cam , lateral position y, lateral velocity v, lateral acceleration α and curvature lost flag flag_lost are output to the ECU 20.

  The steering angle sensor 11 detects the steering angle δ and outputs data on the steering angle δ to the ECU 20. Lateral acceleration sensor 12 detects lateral acceleration a and outputs data of lateral acceleration a to ECU 20. The vehicle speed sensor 13 detects the vehicle speed V and outputs data on the vehicle speed V to the ECU 20.

  At start-up, the ECU 20 reads the above set values (S1). Further, the ECU 20 sets each initial value described above (S2).

  Each time the data is output from the camera sensor 10, the steering angle sensor 11, the lateral acceleration sensor 12, and the vehicle speed sensor 13, the ECU 20 reads each data (S3).

The ECU 20 uses the lateral velocity v read this time to calculate an increase Δv2 ^{* of the} mean square value of the lateral velocity by the equation (3), and the increment Δv2 ^* and the mean square value v2 ^{* of the} lateral velocity (previous value). Is used to calculate the mean square value v2 ^* (current value) of the lateral velocity according to the equation (5) (S4). Further, the ECU 20 calculates the increment Δa2 ^{* of the} mean square value of the lateral acceleration by the equation (4) using the lateral acceleration a read this time, and calculates the increment Δa2 ^* and the mean square value of the lateral acceleration a2 ^* (previous time). Value) is used to calculate the mean square value a2 ^* (current value) of the lateral acceleration by equation (6) (S4). Then, the ECU 20 calculates the meander frequency ω by the equation (7) using the mean square value v2 ^* (current value) of the lateral velocity and the mean square value a2 ^* (current value) of the lateral acceleration (S4). .

Further, the ECU 20 calculates the curvature increment Δρ _fit adjusted by the equation (8) using the curvature ρ _cam and the vehicle speed V based on the camera image read this time. Then, the ECU 20 calculates a curvature ρ _filt (current value) obtained by adjusting the phase of the curvature ρ _cam of the camera image by using the increment Δρ _{filter according} to the equation (9) (S4).

  Further, the ECU 20 calculates the lateral acceleration gain G with respect to the rudder angle by the equation (10) using the vehicle speed V read this time (S4).

The ECU 20 determines whether or not the absolute value of the curvature ρ _cam at the front position read this time is larger than the curvature threshold value ρo _cam (conditional expression (11)) (S5). When it is determined that the absolute value of the curvature ρ _cam at the front position is larger than the curvature threshold value ρo _cam , the ECU 20 determines whether the absolute value of the curvature ρ at the vehicle position is larger than the threshold value ρo (conditional expression (12)). Is determined (S5). When it is determined that the absolute value of the curvature ρ _cam at the front position is larger than the threshold value ρo _cam and the absolute value of the curvature ρ at the host vehicle position is larger than the threshold value ρo, the ECU 20 sets 0 as the transient flag tr (S5). ). When it is determined that the absolute value of the curvature ρ _cam at the front position is greater than the threshold value ρo _cam and the absolute value of the curvature ρ at the host vehicle position is equal to or less than the threshold value ρo, 1 is set as the transient flag tr, and the meandering damping ratio ζ is set. The clothonoid attenuation coefficient ζn is set, and 1 is set as the meander frequency ω (S5).

When the absolute value of the curvature ρ _cam at the front position is determined to be equal to or less than the curvature threshold value ρo _cam , the ECU 20 determines whether the absolute value of the curvature ρ at the vehicle position is larger than the threshold value ρo (conditional expression (12) ) Is determined (S5). When it is determined that the absolute value of the curvature ρ _cam at the front position is equal to or less than the threshold value ρo _cam and the absolute value of the curvature ρ at the host vehicle position is equal to or less than the threshold value ρo, the ECU 20 sets 0 as the transient flag tr and sets the straight line state flag 1 is set as straight (S5). When it is determined that the absolute value of the curvature ρ _cam at the front position is less than or equal to the threshold value ρo _cam and the absolute value of the curvature ρ at the host vehicle position is greater than the threshold value ρo, 1 is set as the transient flag tr, and the meandering damping ratio ζ is set. The clothonoid divergence coefficient ζp is set, and 1 is set as the meander frequency ω (S5).

Then, the ECU 20 obtains A by equation (20) using the time constant τ of the lateral acceleration with respect to the steering angle, the meandering frequency ω, the meandering damping ratio ζ, and the transient flag tr (S6). Further, the ECU 20 obtains B by the equation (21) using the time constant τ of the lateral acceleration with respect to the steering angle and the lateral acceleration gain G with respect to the steering angle (S6). Further, the ECU 20 obtains C by the equation (26) using the vehicle speed V (S6). Further, the ECU ²⁰ obtains the process noise P by the equation (22) using the straight line state flag straight, the curvature lost flag flag_lost, the bottom weighting coefficient base W_lost, the straight road weighting coefficient W_str, and the steering angle variance σ _δ ² (S6). ). Further, the ECU ²⁰ obtains the system noise Q by the equation (23) using the curvature dispersion σ _ρ ² , the lateral acceleration dispersion σ _a ² , the lateral velocity dispersion σ _v ² , and the lateral position dispersion σ _y ² (S6). Then, the ECU 20 calculates a Kalman gain based on the A, B, C, P, and Q (S6). Further, the ECU 20 sets the state quantity x by the equation (18) using the lateral acceleration a, the vehicle speed V, the lateral speed v, the lateral position y, and the phase-adjusted curvature ρ _filter that are read this time, and the steering angle that is read this time. The input amount u is set according to the equation (19) using δ (S6). Then, the ECU 20 uses the A, B, C, P, Q (Kalman gain), the state quantity x, and the input quantity u to perform a Kalman filter having the equation (17) as a state equation and the equation (24) as an output equation. Driven to calculate the observation amount z shown in the equation (25) (S6). Then, the ECU 20 outputs the curvature ρ at the own vehicle position included in the observation amount z to a driving support device or the like (S7).

  The ECU 20 determines whether or not the ignition switch is turned off (S8). When it is determined that the ignition switch is kept on, the ECU 20 returns to S3 after a predetermined time and repeats the above processing. When it is determined that the ignition switch is OFF, the ECU 20 ends the process.

According to this curvature estimation apparatus 1, a Kalman filter incorporating a curvature model in which the curvature of the road is associated with meandering by the driver's steering is configured to estimate the curvature ρ, so that it is included in the curvature ρ _cam detected from the camera image. Noise can be suppressed, and the curvature of the road (white line constituting the lane) can be estimated with high accuracy. In addition, since the configuration is simple with the Kalman filter, the calculation load is small.

  Moreover, according to the curvature estimation apparatus 1, the meandering frequency ω corresponding to the steering characteristic for each driver is obtained, and the meandering frequency ω is applied to the curvature model, whereby the curvature model according to the steering characteristic of the driver is obtained. Can be used, and the curvature of the road can be estimated with higher accuracy.

  Further, according to the curvature estimation device 1, the curve entrance / exit is determined, and the curve entrance / exit is configured with a Kalman filter incorporating a curvature model that is a function of monotonic increase or decrease, thereby estimating the curvature ρ. It can be estimated with higher accuracy.

Moreover, according to the curvature estimation apparatus 1, it is determined whether or not the road is a straight road, and the Kalman filter is not applied to the curvature ρ _cam detected from the camera image on the straight road. Can be suppressed.

  As mentioned above, although embodiment which concerns on this invention was described, this invention is implemented in various forms, without being limited to the said embodiment.

  For example, in the present embodiment, the present invention is applied to the curvature estimation device, but it may be configured to be incorporated as a curve estimation function in various devices using curve information such as a lane keeping device.

  In the present embodiment, an example of the observation amount is shown, but other observation amounts may be used. However, it is essential that the observed quantity includes the curvature ρ. In the present embodiment, the meandering frequency is obtained for each driver. However, a general meandering frequency of the driver obtained by experiments or the like may be set.

  In this embodiment, an example of the method for determining the curve entrance / exit is shown, but the curve entrance / exit may be determined by another method. In the case of the curve entrance / exit, a curvature model formula different from the curvature model formula of the curve is used, but the same model formula may be applied.

  DESCRIPTION OF SYMBOLS 1 ... Curvature estimation apparatus, 10 ... Camera sensor, 11 ... Rudder angle sensor, 12 ... Lateral acceleration sensor, 13 ... Vehicle speed sensor, 20 ... ECU.

Claims (4)

A curvature estimation device that estimates the curvature of a road based on an image captured by an imaging means,
Based on the meandering by the steering of the driver of the vehicle, a steering wheel angle is set as an input, a road model is output with the road curvature as an output, and actual vehicle driving data is applied to the Kalman filter incorporating the road model, A curvature estimation apparatus that estimates a curvature based on a result of driving the Kalman filter with respect to a curvature of a road obtained from an image captured by the imaging unit.   The curvature estimation apparatus according to claim 1, wherein a meander frequency by steering of a vehicle driver is calculated, and the meander frequency is applied to the road model.   3. The curvature estimation apparatus according to claim 1, wherein, when the road is an entrance / exit of a curve, the curvature is estimated using a monotonically increasing function or a monotonically decreasing function.   The curvature estimation apparatus according to any one of claims 1 to 3, wherein when the road is a straight line, the Kalman filter is not driven.

Images