JP2949736B2 - Fruit detection method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a method for detecting fruits and the like.

(Problems to be Solved by the Related Art and the Invention) For example, when taking out an apple fruit from an image of an apple tree, the background of the apple fruit, the leaves of the tree, the sky, etc.
The relationship between the density (brightness), noise, and the objects, etc., is a determining factor for deriving an image processing procedure.

An object of the present invention is to improve the detection accuracy when an object such as a fruit is separated from a background.

(Means for Solving the Problems) According to the present invention, in a red signal density histogram of a red object fruit or the like captured by a camera, an average signal value of the red signal density histogram is used in binarizing an image. Calculate and calculate the signal value of the valley portion between the zero signal value of the red signal value and the average signal value, and compare the signal value of the valley portion with the average signal value to obtain a smaller signal. A configuration of a method for detecting a fruit or the like characterized by performing a binarization process using a value.

(Operation and Effect of the Invention) An image of a red fruit or the like in front is captured by a camera, and an image density histogram of a red signal is created. The average signal value of the red signal density histogram is calculated using the histogram as a threshold value for binarization processing, and the signal value of the valley between the zero signal value of the red signal value and the average signal value is calculated. The signal value of the valley portion is compared with the average signal value, and the smaller signal value is used to perform the binarization processing.

Since the valley portion of the histogram is a blurred portion of the outline of a red fruit or the like, the detection accuracy of the contour position of the red fruit or the like can be improved by the above-described method, and the background and the fruit or the like can be accurately detected. Can be separated, and the area, width, and the like of the fruit can be calculated.

(Embodiment) In the illustrated example, an image of an object such as a fruit in front is captured by a camera 1 of an image sensor, and red R, green G, and blue B signals of each pixel are input to an image frame memory via an input interface 2. 3 is stored. For example, in an example of tomato fruit detection, this image frame memory 3
And only the R signal is taken into the CPU, and the CPU calculates the image density histogram. In this histogram, the difference between the highest density value (Hmax) and the lowest density value (Hmin) is obtained, and the coefficient α for changing and adjusting this difference is 1 except for 0.
The binarized reference density value (y) = α (Hmax−
Hmin) to binarize the image and detect the fruit. These results are output to the image monitor 5 via the output interface.

As described above, by providing the coefficient α that enables the binarization reference density value to be changed in the image binarization process, the image portion of the fruit is more clearly distinguished from the portion that is not, and only the image portion of the fruit is extracted. Therefore, the accuracy of fruit detection can be improved.

In FIGS. 4 and 5, as a threshold value for separating the background and the fruit, a binary value is used by using the smaller one of the average signal value and the signal value of the valley portion of the signal histogram. Configuration.

For example, when detecting tomato fruits, the camera 1 stores red, green, and blue signals in the image frame memory 3 via the input interface 2. Only the red signal is taken into the CPU from the image frame memory 3 to calculate an average signal value a. Also, a histogram of the red signal of each pixel is created. A valley of the histogram is found in the range from the red signal value 0 to the average signal value a. The red signal is binarized using the red signal value b corresponding to the valley as a threshold. When there is no valley in the histogram, the red signal is binarized using the average signal value a as a threshold.

Since the valley portion of this histogram is a blurred portion of the outline portion of the tomato, the detection accuracy of the outline position of the fruit can be improved by the above method, and the area, width, etc. of the fruit can be accurately calculated. Can be.

6 and 7, the configuration is such that the signal value of each pixel after the background and the fruit are separated is less than or equal to or less than a specified value.

For example, in the detection example of tomato, red,
Input green and blue signals and create a histogram of red signals in CPU. In the histogram of each pixel signal value, a peak value within a 1/8 range of the signal is obtained. By removing signals below the peak value on the lower side of the histogram, it is possible to remove the influence of a very dark portion in the camera image, for example, a shadow. In addition, by removing signals that are higher than the peak value on the higher side, it is possible to eliminate the effects of very bright parts in the camera image, for example, strong light in the background. Can be prevented. Therefore, the accuracy of fruit detection can be improved.

8 and 9, an area for each density of image density is obtained, this area is compared with the previously input area data for each fruit, and a fruit is detected based on the result. I have.

For example, in detecting a tomato fruit, the camera 1 captures an image of the forward tomato fruit, and stores the red, green, and blue signals of each pixel in the image frame memory 3 via the input interface 2. This image frame memory 3
From the CPU, the difference between this density value and the density value of the background is taken out, the contour is detected, and the area is obtained. The area data for each fruit stored in the memory in advance is designated by a selection switch and compared to detect a target tomato fruit.

In this way, the image captured by the camera 1 is divided into the image portion of the fruit and the portion other than the image portion, and only the image portion of the fruit is extracted, so that the detection accuracy of the fruit can be improved.

[Brief description of the drawings]

1 shows an embodiment of the present invention. FIG. 1 is a control block diagram, FIG. 2 is a flowchart, FIG. 3 is an explanatory diagram, FIG. 4 is a flowchart of another detection method, and FIG. FIG. 6 is a flowchart of another detection method,
FIG. 7 is an explanatory diagram, FIG. 8 is a flowchart of another invention, and FIG. 9 is an explanatory diagram thereof. In the figure, reference numeral 1 denotes a camera, 2 denotes an input interface, 3 denotes an image frame memory, 4 denotes an output interface, 5
Indicates an image monitor.

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁶ Identification code FI H04N 9/64 G06F 15/62 380

Claims (1)

(57) [Claims] In a red signal density histogram of fruits of a red object captured by a camera, an average signal value of the red signal density histogram is calculated in binarizing an image, and a red signal value of zero is calculated. A signal value of a valley portion between the signal value and the average signal value is calculated, and the signal value of the valley portion is compared with the average signal value, and a binarization process is performed using a smaller signal value. A method for detecting fruits and the like, wherein