JPH10111929A - Weed recognizer for lawn mower - Google Patents

Abstract

PROBLEM TO BE SOLVED: To identify the weeds among the lawn grass through the binarization of an image that is inputted by a visual sensor. SOLUTION: When a lawn image is inputted by a visual sensor 1 of a lawn mower running on the lawn, the number of pixels and the image pickup height are set so as to secure the lawn grass width equivalent to 1/2 to 1 pixel, for example, in regard to the relation between the number of pickup pixels and the pickup height set by the sensor 1. In such an input image, every attentional pixel is decided in all pixels. Then the value (a) obtained by subtracting the minimum density value from the maximum one among 9 pixels including an attentional pixel and its 8 nearby pixels is compared with the value (b) obtained by dividing the mean density value of those 9 pixels by a prescribed coefficient. If the value (a) is equal to or larger than the value (b), the attentional pixel is set at 0. If the value (a) is smaller than the value (b), the attentional pixel is set at 1 and the entire image is successively binarized.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a weed recognizing device for a turf weeding machine, and belongs to a technical field such as a device for identifying a weed in a turf and removing only the weed.

[0002]

2. Description of the Related Art When trying to remove weeds in turf, it is necessary to first identify only weeds. As a means for identifying such weeds, the weeds are identified by the spectral reflection characteristics of leaves. It is possible to do,
Turf and weeds are almost indistinguishable because the spectral reflection characteristics of the leaves are approximately the same, and it is difficult to discern rice weeds by general structural analysis. Was considered extremely difficult.

Accordingly, the present invention makes it possible to identify a weed by binarizing an image with a specific mathematical expression based on the difference in the width of the turf and the leaf of the weed.

[0004]

SUMMARY OF THE INVENTION The present invention relates to a turf weeding machine having a visual sensor 1 capable of recognizing turf and weeds on turf. The image is binarized based on a comparison result between a value obtained by subtracting the minimum value from the maximum value of each density value of the pixel of interest and its eight neighboring pixels, and a value obtained by dividing the average value of the same density value by a constant coefficient. The weed recognition device is characterized by identifying weeds.

[0005]

According to the above arrangement, when a turf image is input by the visual sensor 1 of the turf weeder running on the turf,
In the relationship between the number of image pixels and the image height by the visual sensor 1, for example, all the images input by setting the number of pixels or the image height so that the width of the turf leaf is 1/2 to 1 pixel are set. A pixel is a pixel of interest, one pixel at a time, and nine pixels including the pixel of interest and eight neighboring pixels are used. The a value obtained by subtracting the minimum value from the maximum value of each density value in the nine pixels, and Is compared with the b value obtained by dividing the average value of the respective density values by a preset coefficient. If the a value is equal to or greater than the b value, the target pixel is set to 0, and if the a value is smaller than the b value. If the pixel of interest is 1, the entire image is sequentially binarized.

[0006]

As described above, when weeds on a turf are identified, all the pixels of the input image are binarized by a specific equation, so that the turf has a leaf width of about 1 pixel. Since the input is a small point like noise, weeds can be extracted in a state where the number of pixels is larger than that and the outline can be recognized, so that weeds can be easily identified by shape. In addition, it is possible to identify the weeds of the same rice family as the lawn by the difference in leaf width.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to a turf weeding machine for discriminating and removing weeds in turf. FIG.
Indicates a lawn weeder. A pair of left and right front wheels 3 and a rear wheel 4 are respectively supported on the left and right sides of the front and rear positions of the traveling frame 2 framed in a rectangular shape, and a wheel drive motor 5 is separately connected to the front wheels 3 in an interlocking manner. In contrast, a steering angle sensor 6 for detecting a steering angle for steering control so as to be steerable and turnable is provided at a detectable position. The left and right front frames 7 and the left and right rear frames 8 are respectively set up on the left and right sides of the four corners of the running frame 2, and the upper ends of the left and right front frames 7 and the rear frame 8 are connected in a rectangular shape by the upper frame 9. The traveling device 10 is formed by forming the above.

A battery 12 as a power source is mounted on a gantry 11 protruding from the left and right front frames 8 to the upper side of the front wheels 3, and a lawn is provided at a left and right position on an upper front side of the front frame 8. A travel control box 14 containing a travel controller 14a for controlling travel of the ground weeding machine,
Image processing controller 15a for controlling identification and removal of weeds on turf, and controller 1 for weed removal
5b, and a weed control box 15 having the interior 5b.

The four corners of the upper frame 9 are aligned with X in a plane.
An upper auxiliary frame 16 that cross-connects in the shape of a letter is provided, and a cylindrical holding cover 17 is fixed to the center of the crossing of the upper auxiliary frame 16 as shown in FIG. The guide rail 18 provided at an appropriate position of
An inverted L-shaped camera base 19 that slides up and down and moves up and down is provided. A CCD camera 20 that captures a lawn and inputs an image is attached to the lower surface of the camera base 19, and is mounted on the upper surface of the camera base 19. The visual sensor 1 is constructed by extending a slide screw 21 upward and screwing the slide screw 21 to adjust the vertical height of the CCD camera 20. Lights 23 for imaging illumination of the CCD camera 20 are disposed at appropriate positions at four corners of the upper frame 9.

On the left and right frames 2 of the traveling frame 2, right and left slide receiving members 24a slidable in the front-rear direction by meshing with a rack and a pinion are supported, and a slide frame 24 is provided between the left and right sliding receiving members 24a. A chemical liquid ejector 25 having an injection nozzle or the like downwardly slidable left and right by rotation of a screw shaft or the like is supported on the slide frame 24, and the slide frame 24 and the chemical liquid are supported. An XY table type manipulator M is formed by the sliding action in the front-rear and left-right directions with the ejector 25.

A chemical solution tank 26 for filling a herbicide is mounted on the traveling frame 2 from the position in contact with the rear surface of the left and right rear frames 8 toward the rear side.
A hose or the like capable of flowing a chemical solution is provided between and the chemical solution injector 25. For the protection of the above-described devices and the safety of workers, the entire grass lawn weeder is covered with an exterior cover 27, and at an appropriate position on the upper surface of the exterior cover 27,
A remote control receiving antenna 28 connected to each of the controllers 14a, 15a, 15b is mounted.

When weeds in a turf are to be removed by a turf weeder, a signal is received from a receiving antenna 28 by remote control as shown in a flowchart (all work steps) of FIG. The wheel drive motor 5 is driven by the controller 14a and moved by an arbitrary distance. At this time, when steering in the traveling direction is required, the steering angle of the front wheel 3 is detected by the steering angle sensor 6 to determine the steering direction.

When the movement of an arbitrary distance is completed, the movement is temporarily stopped, and in the image processed by the image processing controller 15a, the width of the turf leaf is set to 1/2 to 1 pixel so that the visual sensor 1 is used. The number of pixels or the imaging height of the CCD camera 20 is set by the adjustment dial 22 and the light 2
An image of a lawn is input by lighting of the number 3. In addition, the traveling of the lawn weeder is not performed by remote control, but by the traveling controller 14.
a, the height of the CCD camera 20 may be adjusted by remote control or autonomous control by the image processing controller 15a.

By scanning on the original image (see FIGS. 11A and 11B) having a density value of 256 gradations of the input image, as shown in the flowchart (image processing routine) of FIG.
When each pixel in the filter frame (pixel area) is detected as a density value of 9 pixels by the pixel of interest in the block 1 area and the block 2 area, and 8 neighboring pixels as shown in FIG. 5, for example, As shown in the calculation formula of FIG. 6, each detected density value is 100 (a) in the block 1 area.
(Value) ≧ 47.7 (b value), the target pixel is set to 0, and in the block 2 area, 30 (a value) <52
When (b value) is reached, the pixel of interest is set to 1, and the binarization of the pixel of interest in the entire image is sequentially terminated. (See FIGS. 11B and 11F.) The coefficients used in this calculation are set based on experimental values and the like, and are slightly variable depending on the amount of light.

As described above, the maximum value, the minimum value, and the average value of the detected density values are obtained, and in the case of maximum value−minimum value ≧ average value / coefficient, the pixel of interest is set to 0, and the maximum value−minimum value <average value / In the case of a coefficient, it is checked whether or not the binarization processing of all the pixels has been completed by using a mathematical expression in which the pixel of interest is 1, and if NO, return to the beginning; if YES, each of noise clear 1 and noise clear 2 Continue the process.

In the noise clear 1, in a binary image, weeds are a set of pixels that are hardened to some extent, whereas grass is small dots or lines, so that these small points and lines are removed by a simple operation. As shown in the flowchart of FIG. 7 (noise clear 1), when scanning is performed pixel by pixel in the X direction, as shown in FIG.
Check if they are arranged in the order of 1, 0, Y
In the case of ES, the value of the pixel which is sandwiched between 0 and independently becomes 1 is changed to 0.

It is checked whether this process has been completed for all pixels in the X direction. If YES, scanning is performed one pixel at a time in the Y direction, and pixels are arranged in the Y direction in the order of 0, 1, 0. Check if it is, and if yes,
The value of a pixel which is sandwiched between 0 and independently becomes 1 is changed to 0. As described above, in the noise removal by the noise clear 1, the image is input such that the width of the turf leaf is about one pixel. Therefore, in the binary image, the leaf is shown in a state where one pixel is arranged, and is extremely small. It is possible to remove only the part which becomes noise and is obviously grass. It is not necessary to take a long time to remove noise as in the case of the filter processing, or the pixel indicating the edge portion of the weed is not removed and the original shape is not changed. Therefore, small noise can be processed in a short time. At the same time, the shape of the edge of the group of pixels indicating the weeds can be left in substantially the same state as that of the input image, and the weeds can be easily recognized. (See FIGS. 11 c and 11 g) In the noise clear 2, the shape of the edge of the weed is left as it is by using an improved filter on the image obtained by the process of the noise clear 1, and is removed by the noise clear 1 process. This is a method for removing noise that could not be performed.
As shown in the flow chart (Noise Clear 2), all the pixels of the image after the noise clear 1 processing are set as a target pixel one by one, and nine pixels of the target pixel and eight neighboring pixels are used as shown in FIG. The number of 1 pixels in the block 1 area and the block 2 area in the 3 × 3 filter frame is counted, and if this count is equal to or less than a set value set by an experimental value or the like, one pixel is counted. , And if it exceeds this set value, it remains at 1. In the block 1 area, marking is performed when the count number 4 <the set value 5, and in the block 2 area, marking is performed when the count number 5 = the set value 5. It is checked whether or not this marking has been completed for all the pixels, and if YES, it is checked whether or not there is one pixel around the marking pixel.

By this check, it can be seen that in the block 1 area, it is connected to the unmarked 1 in the upper row of the marking. Remain without being shrunk. In the block 2 area, all the surroundings are 0.
Since it is a pixel, it is determined as an isolated small figure that is not connected and removed. Thus, the processing of all the marking pixels is completed.

As described above, in the noise removal by the noise clear 2, only the aggregate of pixels having a small area can be removed, so that the weed portion can be left as it is, and the hole ( The shape change is small because the portion (0 in the massive figure) is not increased. In addition, unlike a conventional noise elimination filter, processing is performed only by the number of 1s in the filter frame without considering connection to the outside of the frame. Therefore, the shape of the edge of the set of pixels representing the weeds can be left in substantially the same state as the input image, and the weeds can be easily recognized. (FIG. 11
(See d and h.) When the above image processing is completed, it is checked whether or not weeds are present in the range as shown in the flowchart of FIG. 3, and if YES, the weed removal controller 1 is checked.
The manipulator M is actuated by 5b. By the operation of the manipulator M, the slide frame 24 is slid back and forth, and the chemical liquid ejector 25 is slid left and right to move the chemical liquid ejector 25 to the weed position, and the chemical liquid is sprayed from the nozzle of the injector 25 onto the weed. When the predetermined course or the entire range of weeds is removed by spraying the chemical, the operation is terminated.

In addition, instead of the chemical solution weeding, it is also possible to perform powdered weeding, heated weeding, mechanical weeding, or the like. Also, when inputting an image on the turf as described above,
The CCD camera 2 so that the width of the turf leaf is about 1 pixel.
It is necessary to set the total number of pixels or the height position of 0, but when making this setting, the turf currently used in golf courses and the like is roughly classified into Koryo turf and bent turf, and the former is Since the leaves are wider than the latter, it is necessary to change the height of the camera 20 so that each pixel has an appropriate pixel state.

Therefore, when an image of Koryo turf is input while the height of the camera 20 is set so as to adapt to the bent turf, the width of the Koryo turf leaf is indicated by 2 to 3 pixels, and is recognized as a weed. At the same time, the amount of noise during identification (binarization) increases. Conversely, when images of bent turf are input with the height of the camera 20 set to adapt to Koryo turf, the width of the bent turf leaves is too small to capture the characteristics of bent turf, and thus weeds are not captured. It is easy to be recognized as.

For this reason, as shown in FIG.
When the vertical height of the camera 20 is adjusted, the detection switches 30 respectively set at the vertical position are turned ON / OFF by the projections 29 provided at appropriate positions of the camera 20 so that the respective set heights of the bent grass and the Koryo grass are adjusted. Can be accurately and easily set to each appropriate height position in either manual adjustment or automatic adjustment. The operation range of the manipulator M is changed and set according to the height position of the camera 20.

By these actions, when the image is binarized, the number of pixels in the turf portion is reduced to 1 (not removed), the noise in the image is reduced and reduced, and each of the turf and weeds is reduced. , And it becomes easy to identify even rice weeds similar to turf by the width of leaves.

[Brief description of the drawings]

FIG. 1 is a perspective view showing the overall configuration of a lawn weeding machine.

FIG. 2 is a side view showing the structure of the visual sensor.

FIG. 3 is a flowchart showing an operation procedure for removing weeds on a turf.

FIG. 4 is a flowchart illustrating an image processing procedure for identifying weeds on a turf.

FIG. 5 is a pixel diagram showing a detected density value of each pixel in a 3 × 3 filter frame in image processing.

FIG. 6 is a formula for binarizing each pixel in a 3 × 3 filter frame in image processing.

FIG. 7 is a flowchart showing a procedure of noise clear 1 in image processing.

FIG. 8 is a pixel diagram showing an array of pixels in XY directions in image processing.

FIG. 9 is a flowchart showing a procedure of noise clear 2 in image processing.

FIG. 10 is a pixel diagram showing a marking state in a 3 × 3 filter frame in image processing.

FIG. 11 is an image diagram showing a transition state of processing contents in image processing.

[Explanation of symbols]

 1. Vision sensor

Claims (1)

[Claims] 1. A lawn weeding machine having a visual sensor 1 capable of recognizing turf and weeds on a turf, wherein an image of a turf is input by the visual sensor 1, and a pixel of interest of this image and its eight neighboring pixels Binarizing an image based on a comparison result between a value obtained by subtracting a minimum value from a maximum value of each density value and a value obtained by dividing an average value of the same density values by a constant coefficient, and identifying weeds. Weed recognition device.