JPH04167007A - Line detection device - Google Patents

Abstract

PURPOSE:To prevent a line along which this vehicle travels from being missed by estimating a line position to be detected from a past line position and selecting a line which is close to the estimated position. CONSTITUTION:An image of each time obtained by a camera 10 is processed to detect the line and at least the last line position and line immediately before the last line are stored in a line position storage means 30. Then a CPU 26 estimate the position of the line to be detected this time from the variation state of the line positions which are stored and selects the line which is the closest to the predicted position. Consequently, even when there are plural lines in the image obtained by the camera 10, the line along which this vehicle travels can accurately be selected.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a line detection device for detecting guidelines provided along a road for automatic driving of vehicles and the like.

[Prior Art] Conventionally, various driving control devices have been installed in vehicles in order to reduce the burden on drivers. Among these, research is being conducted into autonomous driving, which also automatically handles steering wheel operations, and consideration is being given to equipping regular cars with mechanisms that would enable autonomous driving.

In order to achieve this automatic driving, it is necessary to constantly know the position of the vehicle with respect to the road (road) and control the vehicle so that it travels along the road.

As a method for detecting this running route, it has been proposed to photograph a white line formed along the running route using a television camera, and detecting the position of the vehicle with respect to the running route from the position of this white line.

Roads usually have center lines, lane dividing lines, etc., and it is thought that if position detection for autonomous driving is performed by white line detection, the equipment on the road side can be made very simple. .

For example, Japanese Unexamined Patent Publication No. 63-40913 discloses a method of detecting a white line in front of the vehicle using a CCD camera mounted on a vehicle. That is, in this example, C
A portion where the brightness of the road surface obtained by the CD camera is equal to or higher than a predetermined value is determined to be a white line. On ordinary roads, white lines are formed on black parts such as asphalt, and the brightness of the white line parts is sufficiently higher than other parts, so that the white lines can be detected.

[Problems to be Solved by the Invention] However, when the vehicle is running, the attitude of the vehicle may change significantly. In such cases, with conventional devices, the white line may move out of the field of view and be lost.

In order to solve this problem, it is conceivable to widen the field of view of the CCD camera. However, if the viewing angle is widened, multiple white lines will come into view on a road with multiple lanes. In other words, if the width of the negative lane is approximately 3.7m, two white lines will be visible if the field of view is 4m wide, and the width of the 8m lane will be approximately 3.7m.
If you look at the width of the road surface, you will see three white lines.

Therefore, it is necessary to always determine which white line the vehicle should follow from among a plurality of white lines.

The present invention has been made in view of the above problems, and it is an object of the present invention to provide a line detection device that can effectively determine the white line that the vehicle should follow.

[Means for Solving the Problems] A line detection device according to the present invention includes a camera that repeatedly outputs images of a road in a predetermined range, and a line that processes each image output from this camera to detect a guideline. a detection means, a line position storage means whose contents are sequentially updated according to the line position detected by the line detection means and which stores at least the line position from the previous time, and a plurality of lines stored in the line position storage means. A line position T-side means for predicting the currently detected line position from the state of position change, and a line selection means for selecting the detection line closest to the predicted position obtained by this line position predicting means. Features.

[Function] In the line detection device according to the present invention, the line is detected by processing each image obtained by the camera, and the line position storage means stores at least the previous line position and the line position before the previous time. I'll keep it.

Then, the position of the currently detected line is predicted from the stored change state of the line position, and the line closest to this predicted position is selected.

Therefore, even if there are multiple lines in the image obtained by the camera, it is possible to accurately select the line that the vehicle should follow.

[Example] Hereinafter, a line detection device according to the present invention will be described based on the drawings.

FIG. 1 is a block diagram showing the configuration of a line detection device according to the present invention, which includes a CCD camera 10. As shown in FIG. This C
The CD camera 10 includes an optical system including a lens 12 and an aperture 14, and a CCD linear sensor 16 having a photoelectric conversion element arranged in the negative dimension. The CCD camera 10 is mounted at a position where it can obtain an image of the front of the vehicle, and outputs information regarding the brightness of each photoelectric element output from the CCD linear sensor 16 as a serial signal.

This CCD camera 10 includes a white line detection ECU 2.
0 is connected. This white line detection ECU 20 is a CCD
An amplifier 22 that amplifies the signal manually input from the camera 10 and blunts the waveform, an AD converter 24, a CPU 26,
ROM2 that stores programs related to CPU operation
8, a RAM 30 used as a storage area during processing by the CPU 26, and a 110 section 32 for input/output operations. Information about the white line detected by the white line detection ECU 20 is then supplied to the vehicle control ECLT 40, which controls the steering etc. of the vehicle so that the vehicle runs along the detected white line. .

Next, the white line detection mechanism in the CCD camera 10 will be explained based on FIG. 2.

The CCD camera 10 is installed in the vehicle so as to capture an image at a distance L1 in front of the lens 12. The focal length of the lens 12 is f, and an image is formed on the CCD linear sensor 16 located at a distance L2 behind the lens 12. Therefore, CCD linear sensor 1
The width of the front image that can be detected at the width W of 6 is W.

Therefore, if the front Ll of the lens 12 is located at a position shifted by X from the white line or its center line, the CCD linear sensor 1
In step 6, the position where this white line is imaged is uniquely determined. The width W of this CCD linear sensor 16 is
If the number of elements in the CCD linear sensor 16 is N, and the number of elements from the center of the CCD linear sensor 16 is I, then the distance X from the center of the white line is expressed as follows.

From the convex lens formula, 1/Ll +1/L2 -1/f And from that magnification, W-w-L1/L2 It is.

Therefore, W−w・(LL +f)/f, and x −(W/N) conflict (N/2−′I).

Therefore, by detecting which element in the CCD linear sensor 16 the white line is detected, the position X of the white line can be detected.

Next, we will develop a white line detection ECU 2 using the principle described above.
The signal processing (white line detection) from the CCD camera 10 in 0 will be explained.

FIG. 3 is a flowchart illustrating the overall operation of the white line detection process. First, exposure control is performed so that the amount of exposure in the CCD camera 10 falls within a predetermined range (Sl). Then, since the luminance signal for each element from the CCD camera 10, which has undergone predetermined exposure control, is output as serial data, this is converted into digital data by the AD converter 24 (S2). Here, since the data output from the CCD camera 10 is determined by the clock ratio supplied to the horizontal output shift register of the CCDC linear sensor 16, it is assumed that the clock frequency in the AD converter 24 is also synchronized therewith.

Therefore, the respective luminance signals obtained from each of the N elements in the CCD linear sensor 16 are sequentially output from the AD converter 24 as N pieces of digital data. This digital signal is stored at a predetermined location in RAM 30.

Then, maximum value detection and average value calculation of the luminance signal in the CCD camera 10 obtained in this way are performed, and a threshold value is calculated based on this maximum value and average value (S3
).

Next, each of the N pieces of digital data obtained by AD conversion is binarized using the obtained threshold value, and the left and right edges of the white line are detected from the change state of this binarized data. (S4). Then, edge correction is performed based on the number of detected left and right edges, taking into account the case where a white line appears at the edge of the image (S5).

In this way, multiple white line edges are detected, and the white line that the own vehicle should follow is detected from among these, or at least the current white line position is predicted from the change state of the white line position from the previous time or the time before the previous time, and the predicted position is Select the closest one.

Then, the coordinates of the selected white line are output (S6).

As explained above, the coordinates of this white line correspond to the distance from the center of the white line, and by supplying this X to the vehicle control ECU 40, the vehicle control ECU 40 performs steering control based on this X, and the lane ( Vehicle running along the white line) is achieved.

In this way, according to the present embodiment, since the white line that the own vehicle should follow is predicted and selected from a plurality of detected white line positions, accurate line white line detection can be performed even when the vehicle attitude changes significantly. I can do it.

Threshold Calculation The operation of threshold calculation in this embodiment will be explained based on FIG.

As an initial setting, the average value avb of three variables,
Maximum value maxd, AD conversion data counter adcount
The three variables of t are reset to O, and the start address (eoooh) in RAM 30 of the luminance data obtained by AD conversion is set to variable A (S10
1-5103). Note that the values of each variable set here are stored in the RAM 30.

Next, it is determined whether adcount has reached 400h (1024) (S104). This is a CCD
This corresponds to the number of effective elements in the linear sensor 16 (the number of elements within the range of width W), that is, the number of data output from the n1 AD converter 24 when photographing at -degrees. Therefore, ad
When count reaches 1024, it can be determined that processing of all data has been completed.

If adcount is less than 1024, the data processing has not finished yet, so the variables bc, ax
, upload the following data manually (S105).

c-A Bz-adcount Ip-wbc+ax Here, the variable up is sequentially designated as an address in the RAM 30 where 1024 luminance signals obtained by the CCD linear sensor 16 are stored. Then, the data at the address specified by this variable up is read and inputted as the value of the variable DATA (S105).

Next, to detect the maximum value, DATA is set to the maximum value max
It is determined whether or not it is smaller than d (S106). and,
If DATA is greater than maxd, this should be the maximum value, input DATA into the variable maxd,
This value is updated (5107). On the other hand, if DATA is not the maximum, there is no need to update the maximum value maxd, so the process skips through 5107 and moves to the next process.

Next, in order to calculate the average value avb, the values obtained by dividing the DATA value by the number of data 1024 are sequentially added (3108)
. Then, add 1 to adcount (5109)
, returns to 5104.

Therefore, calculation of maximum value and average value (S 107°510
8) will be repeated until the number of data reaches 1024, and when all (1024) data have been processed, the answer will be YES in 5102, and the threshold value t will be reached.
hj! 0 is calculated using the following formula (S110).

thJjO-(avb+maxd)/2 In this way, in this embodiment, the intermediate value between the average value and the maximum value of data for all elements obtained by the CCD linear sensor 16 becomes the threshold value th1o. .

Binarization and Edge Detection Process Next, the binarization process and edge detection process will be explained based on FIG.

First, set initial values for each variable as follows.

(5201).

White line section continuous count permission flag cokl=0°Road surface section continuous count permission flag cok2-0. White line continuation counter fcl-0. Road surface continuation counter fc2-0. White line left edge candidate coordinates wn01-0° White line cloth edge candidate coordinates wn0
2=o, white line detection flag fwl-0. White line right edge counter wcl-O1 road surface detection flag fw2-1. White line left edge counter wc2 = 0 Also, as in the case of threshold calculation described above, input the start address (eoooh) of the luminance data to variable A.
02) Set the AD conversion data counter adcount, which represents 1024 data numbers, to 0 (S
2 Q 3).

First, it is determined whether adcount has reached 1024 (400h) and processing of all data has been completed (S204).

Then, if the processing is not completed, variables bc, a
The values of A, adcount, and bc+ax are assigned to x and up, and the value of the luminance data at the address determined by up is input to variable DATA (S205).

Next, binarization processing is performed, but first the variable DATA is compared with the threshold value ttB (S206).

Here, this ttB is the t calculated in 8110 above.
h, different from JO. That is, this threshold value trB is not the threshold value calculated in the current process, but the threshold value calculated in the previous process.

If DATA≦ttB, it is a road surface area, so variables fc1-0. Variable wn01-0. that indicates the detected coordinates of the left edge of the white line. A variable f w 2-1 indicating road surface detection is set (S 207).

Next, it is determined whether it is fwl-1 (S208). f
w1 is initially set to O in 5201, and when a white line is detected, that is, DATA>t
It is set to 1 in the case of hf. Therefore,
At 8206, the data is ≦thJ! and 5208
The fact that fwl is 1 means that the right edge of the white line has been detected. Therefore, if it is fwl-1 in 5208, the road surface is detected, so it is set to fwl-0, and the variable wn02-adcount indicating the detection position of the right edge and the road surface continuation count permission flag cok2-1 are set ( S209).

Here, if an edge is detected, its adc.

Is it possible to store the value of unt as it is as the edge position? In this example, after detecting the right edge, the position is determined to be the right edge only if the next data is also not a white line, to prevent the occurrence of misjudgment. suppress.

For this purpose, the road surface continuous count permission flag cok2-1
It is said that

Next, it is determined whether the road surface continuation permission flag cok2 is 1 (S210). If cok2 is 1, add 1 to the road surface continuation counter fC2 (5211), c
If ok2 is not 1, this addition is not performed and it is then determined whether fC2 is 2 (S212).

If this fC2 passes through 5211 twice in a row, it will be -
In other words, if two consecutive pieces of data are equal to or less than the threshold th1, 2
It becomes. Therefore, if it is fc2 or 2, the value of the white line edge candidate coordinate wn02 set in S'209 above is set to the array variable Wn2[WC2] indicating the white line edge coordinate specified by the right edge counter WC2. Remember. Also, the variable cok2. At the same time as resetting fc2 to 0, 1 is added to the right edge counter WC2. Therefore, in the case of the next right edge detection, wc2 is increased by one. Therefore, the position of the next right edge can be stored in the array variable wn2. fc at 5212
In the case of 2 x 2 and when the process of 8213 is completed, the process for the right edge has been completed, so the adcount
Add 1 to t (5214) and return to 5204.

On the other hand, in 5206, it is determined that DATA>t J,
When the white line portion is detected, the left edge is detected as follows.

First, the road surface continuation counter fc2. Left edge candidate coordinates wn
02 to 0, white line detection flag f w 1 to 1
(S215). That is, since fc2 is a variable indicating the number of consecutive road surface data, it is reset to 0 when data regarding a white line is detected. Also,
w n O2 is a variable regarding the candidate coordinates of the right edge, but it is reset here because the loop has entered the left edge detection loop. Further, fwl is a flag indicating that a white line portion has been detected, and is designated as fw-1 here.

Next, it is determined whether fw2-1 or not (S216).

Since this fw2 is set to 1 when the road surface is detected, by being f w 2-1 here, it is recognized that the vehicle has moved from the road surface to the white line, that is, a left edge. Then, a variable f indicating that the road surface has been detected
Reset w2 to O and set a to the left edge candidate coordinate wn01.
Enter the value of dcount t and set the left edge detection counter c.
okl is set to 1 (S217).

Then, using the white line continuation count permission flag C0k1 and the white line continuation counter fcl, it is determined whether or not the right edge of two consecutive white line parts has been detected.

For this purpose, first determine whether cokl is 1 (5
218), and if cokl is 1, it means that a left edge has been detected, and 1 is added to the left edge counter fcl, which indicates the number of white lines detected at the time of left edge detection (5219).

Next, determine whether fcl is 2 or not, and if it is 2,
Candidate coordinates wn entered in 5217, which means that there are two consecutive white line data after moving from the road surface to a white line.
The value of 01 is set to the array variable Wn12[wn
O1]. As a result, the position of the detected white line cloth edge is stored. Also, since this input was made, the flag cokl=1. At the same time as resetting fc1-0, add 1 to the variable wcl indicating the number of right edges (
S221). Then, add 1 to adcount and return to S 214° 5204).

This operation can be repeated up to adcountm1024 to store as many right edge and left edge positions as detected. In addition, the detected number is WCI, w
It is stored in c2. and in 5204 adco
If it is determined that unt has reached 10247:
Since white line edge detection has been completed for all data, the binarization and white line edge detection operations are completed, and detected edge correction processing is performed.

Detection Edge Correction Next, detection edge correction will be explained based on FIG.

As described above, the positions of the left and right edges are set in the array variable wnl.
[wcl] - wn2 When input to [wc2], subtract 1 from wcl and wc2.

This is done during the edge detection process described above.

Since 1 is added to wc2 (S213.5221), wcl is actually used as an array variable.

This is to return the value to wc2 (S401).

Then, wcl and wc2 are compared (5402).

If wcl<wc2, it means that the number of left edges is smaller than the number of right edges, and it means that a white line is over the left edge of the camera image. For this reason,
By inserting a left edge at this left edge, it is possible to recognize the white line that spans the left edge of the screen.

Therefore, the value of wcl is assigned to the variable j (8403), and this value of j is compared with 1 (5404).

Then, in the case of ``>1, the value of j is assigned to the variable aX, and the value of the array variable wnl[ax specified by this aX is assigned to the variable up (S405). Next, variable a
Substitute j11 for X, and set the variable u obtained in 8405 above to the array variable wnl[ax] specified by this aX.
The value of p is manually calculated (S406). This allows wnl[
The value of j is converted to the value of wnl[j+1]. Then, the value of j is subtracted by 1 (5407) and the process returns to 5404.

In this way, this process is repeated until j becomes 1, so the values input to the array variable wn1 [wcl] are converted to w and n 1 [wc 1 + 11, respectively. In other words, if there are n values as the value of wnl and these are stored as values of wn 1 [11 to wn 1 [n], this process will result in Wnl [2] to
wnl is converted to the value of [n+1].

Next, the value O is assigned to wnlrll (S407).

As a result, one value is forcibly inserted as the value of the array variable wnl, and even if the left edge cannot be detected, this one is also inserted.

On the other hand, in 5402, if wcl≧WC2, either the right edge is missing or the number of both edges is the same. Therefore, next, wc2 and wcl are compared (5409). Here, if WC2≧Wc1, it means that they are equal, and no correction processing is necessary.

On the other hand, if wc2 is smaller than wcl, it is necessary to correct the right edge insertion.

Therefore, 1 is added to wc2, and 1023 is forcibly inserted into the array variable wn2[wc2] specified by this wc2. As a result, the value at the right end of the field of view is inserted as the right edge.

In this way, if the left edge is missing, input the left edge of the camera's field of view as the value of the left edge, and if the line reaches the right edge and the right edge is missing, the value is 1023, which is the right edge of the camera's field of view. is input as the right edge position.

White Line Coordinate Output/Preliminary M1 Value Calculation Next, the output of white line coordinates and the next white line element #j will be explained based on FIG. 7.

If the edge can be detected as described above, first
Determine whether both cl or and WC2 are 0 (S3
01), wcl and WC2 represent the number of white line position detections, and if both are 0, it indicates that no white line has been detected. Therefore, in this case, the error code 256 (80h) is output serially (S
OUT) (S302), and the process ends. Therefore, in this case, the threshold value ttB is not updated, and binarization is performed using the currently used threshold value in the next processing loop as well.

Next, wcl and WC2 are compared (S303).

If the two are not equal, the number of right edges and the number of left edges are different, and the white line was not detected accurately, and the process advances to step 5302.

On the other hand, if it is wcl-wc2, the white line detection has been performed normally, and the white line position coordinates at this time are output as follows. First, a variable nn-1 indicating the number of white lines is set (5304), and this nn is compared with the number Wc1 of detected white lines (S305).

If it is less than nn or wc1, the center position W of the white line is calculated from the left and right edge coordinates of the white line using the following formula (S 306
).

w-(wnl [nnl tenwn2 [nnl)/
Second, it is determined whether the calculated white line position W is within a predetermined range. That is, WJ! has a certain degree of width with respect to the predicted white line position wJ calculated and stored in the previous processing loop! ! Determine whether it is within the range of ±α. (S307). This α is preferably set to 1/2 or less of the normal lane spacing.

If this judgment indicates that it is outside the range, n n l:: 1 is added (S30g), the process returns to 5305, and nn is determined as WC.
Repeat this until it is equal to l.

On the other hand, if W is within the predetermined range (±α), this W is the desired white line position, so W is output (S309)
, and wJ! 0. wJ! 1. wj! , the values of the previous time, last time, and current time are exchanged (S 310).
.

wj! O-w, jl wJ71-wJ W1■W In this way, when data replacement is completed, a new predicted value WJ1 is calculated based on the replaced data.
1 is calculated and stored for the next prediction (S311)
.

wJlJl-3 (wJ-wj!1) +wj! 0 Here, this formula is for FIG. 8(A).

The explanation will be based on (B).

The time required for one processing routine in the white line detection ECU 20 is Δt, the horizontal axis is time, and the vertical axis is CCD camera 1.
The white line position detected by 0 is taken. And Figure 8 (
Consider the case where the vehicle curves to the left as shown in A).

Current location wj! , Previous position W! ! 1. Position W10 before last
, and if the difference between the position before the previous time and the previous time is Δ1, the difference between the previous low and current position is Δ2, and the predicted difference between this time and the next time is Δ3,
It is considered best to estimate that the amount of change from Δ2→Δ3 is the same as the amount of change from Δ1−Δ2. That is, the processing cycle Δt is minute, and before and after the time Δt,
It is considered that the amount of steering is almost the same. and,
This is because if the steering amount is the same, the amount of change in the white line detection position is the same. Therefore, as shown in FIG. 8(B), if time is plotted on the horizontal axis and differences Δ1 to Δ3 are plotted on the vertical axis, then the straight line Δ1
It is estimated that Δ3 is located on the straight line connecting Δ2 and Δ3.

Therefore, Δ3 is expressed as follows.

Δ3-((Δ2-Δ1)/Δt) ・Δ is ten Δ2 ■ Δ2
-Δ1+Δ2 Therefore, wJlJl is expressed as follows, and Wle is calculated using this formula (S311).

wJlJl−wjj+Δ3=wJl+2・Δ2−Δ1−wJ! +2 (w!!-w, el) -(w j l -w, l O) -3 (wj!-wl! 1) +wj! 0 In this way, the predicted white line position W used in the next processing loop is
1 can be calculated and can be used to select the white line, so accurate white line position detection can be performed at all times.

Furthermore, when the line detection device starts operating, it is necessary to select a white line for the vehicle to follow from among a plurality of white lines. Therefore, wj! is the initial value when the line detection device starts operating. -wJ! 1-wj! 0 - Set as the center coordinate of the CCD linear sensor 16, and when receiving an instruction from the driver to start operation (when receiving an instruction for automatic driving by white line detection), the white line reflected near the center of the CCD linear sensor 16 should be set by the military. Learn as a white line to follow. Normally, the command to start automatic driving is often when driving on a straight road, so the white line that your troops should follow is the CCD.
This is because it is thought that the image appears in the center of the linear sensor. Note that the state in which the vehicle is moving straight may be overridden by a signal from the steering angle sensor, and the white line position may be learned in that state.

Here, the state when such a white line detection position is predicted will be explained based on FIGS. 9(A) and 9(B). Here, on an actual road, the vehicle travels within a lane between two white lines, and it is preferable to detect the two white lines and control the vehicle to drive within them. Therefore, in the example of FIG. 9, two white lines are recognized.

In addition, in the above-mentioned flowchart, only the predictive detection of one white line was described, but the same method can be easily extended to the detection of two white lines.

FIG. 9(B) shows the results of a simulation using the device of the present invention. Three lanes (four white lines) are shown on the CCD linear sensor 16, and the vehicle moves in a sinusoidal manner. This is a plot of the estimated position of the own lane (the center position of the two white lines) and the actual lane position (the center position of the two white lines) found from the estimated position.

First, in FIG. 9(A), two white lines located near the center of the white lines in front of the own vehicle are recognized as the own lane (lock-up). And after that, the next white line position WJIj from this time, last time, and the time before last! is calculated as described above.

As shown in FIG. 9(B), the estimated value is very close to the actual lane position, and it is understood that the line detection device can select the correct white line (lane).

Note that depending on the curve, only one white line forming the vehicle's own lane can be detected. However, since the lane width is constant, it is understood that the lane position (white line position) can be estimated with fairly good accuracy even if the lane is estimated from the white line on one side.

If Δ3 is predicted by taking into account the history before the previous time, it will no longer be a straight line as shown in Figure 8 CB), and although the estimation method will be complicated, it will further increase the accuracy of white line detection. I can do it.

And threshold thj! The threshold value t calculated this time is
hJ! Enter 0 to update the value (S312).

By updating the threshold value thJl in this manner, the time delay of the threshold value can be suppressed to a minimum.

Then, abnormal threshold values calculated when there is no white line can be eliminated. Therefore, the threshold value can be changed depending on the condition of the road surface, and it is also possible to accurately determine whether there is a white line.

[Effects of the Invention] As explained above, according to the line detection device according to the present invention, the line position that will be detected in the future is estimated from the past line position, and a line close to the estimated position is selected. Therefore, it is possible to prevent the vehicle from losing sight of the line it should follow. Therefore, highly accurate automatic operation can be achieved.

[Brief explanation of drawings]

Fig. 1 is a configuration block diagram showing an embodiment of a line detection device according to the present invention, Fig. 2 is an explanatory diagram for explaining position detection in a COD camera, and Fig. 3 is an explanation of processing operations in a white line detection ECU. FIG. 4 is a flowchart showing the operation of threshold calculation. FIG. 5 is a flowchart showing the operation of edge detection and binarization processing. FIG. 6 is a flowchart showing the operation of edge correction. 8 is an explanatory diagram showing a white line position prediction method. FIG. 9 is an explanatory diagram showing an example of white line position prediction. 10... CCD camera 20... White line detection ECU 24... AD converter 26... CPU

Claims (1)

[Claims] A line detection device mounted on a vehicle that detects a guideline provided along a road, comprising a camera that repeatedly outputs images of a road in a predetermined range, and one output from this camera. a line detection means for processing each image and detecting a guideline; and a line position storage means for sequentially updating the contents based on the line position detected by the line detection means, and storing at least the previous line position and the line position before the previous time. , line position prediction means for predicting the currently detected line position from the state of change of a plurality of line positions stored in the line position storage means, and a detection line closest to the predicted position obtained by this line position prediction means. A line detection device comprising: a line selection means for selecting;