CN106327467A - Method for quickly tracking and indentifying target fruits picked by apple picking robot - Google Patents

Abstract

The invention discloses a method for quickly tracking and indentifying target fruits picked by an apple picking robot. The method comprises the steps of: image segmentation; target fruit determination; identification area extraction; fast template extraction and identification; and target fruit positioning. Through the adoption of the method, the picking speed of the apple picking robot is improved, the different between the picking speed of the apple picking robot and the manual picking speed is reduced, and the practicability is high.

Description

A kind of apple picking robot objective fruit quick Tracking Recognition method

Technical field

The invention belongs to field of image recognition, particularly to apple picking robot objective fruit quick Tracking Recognition method.

Background technology

From the sixties in 20th century, after American Schertz and Brown proposes to use picking robot fruit, various fruits Vegetables picking robot technology is widely studied, but fruit and vegetable picking speed all ratios of model machine are relatively low in early days, wherein plucks a Fructus Mali pumilae Time be tens of second.2008, the apple picking robot AFPM that Baeten etc. develops, to diameter 6～11cm The average plucking time of Fructus Mali pumilae is 9 seconds.The research of domestic fruit and vegetable picking robot is started late, and Some Universities and scientific research institutions are to various Fruit and vegetable picking robot has carried out research in succession, and achieves initial achievements and develop some model machines, wherein China in 2009 When the apple picking robot of research institute of mechanization science in agriculture and Jiangsu University's joint research and development single fruit in laboratory conditions is plucked Between be 15 seconds, substantially reflect the technical merit of current domestic apple picking robot picking rate, but compared to manually adopting Pluck speed and there is also bigger gap, also need to reduce picking process further and process the time.The apple picking robot of joint research and development During objective fruit barycenter Step wise approximation picture centre, need multi collect image trace identification.

Summary of the invention

The purpose of the present invention is to propose to a kind of apple picking robot objective fruit quick Tracking Recognition method, the method can improve Herba Marsileae Quadrifoliae The picking rate of fruit picking machine device people, reduces and the gap of artificial picking rate, practical.

The technical scheme realizing the object of the invention is:

A kind of apple picking robot objective fruit quick Tracking Recognition method, concrete steps include:

1) image segmentation

It is chosen at one group of picture of apple orchard shooting under natural environment, selects Apple and the branch of background, greenery and dead zone, sky Territory, carries out statistical analysis to the value of its R, G, B color factor, with aberration R-G and 2R-G-B method by Apple from In background separated；

2) objective fruit determines

Use 8 neighbourhood signatures's methods that the above-mentioned fruit handled well segmentation image is marked, and two dimension is asked in labelling fruit region Center-of-mass coordinate, formula is

In formula: the horizontal stroke of i, j fruit image pixel, vertical coordinate

The total pixel number of n fruit image

Ω belongs to the collection of pixels of same fruit image

Calculate its length of side simultaneously, finally determine objective fruit with the nearest principle in range image center；

3) extracted region is identified

Utilize the objective fruit information that prior image frame identifies to reduce the recognition time of present image, analogize frame by frame with this and successively decrease, Utilize the center-of-mass coordinate of objective fruit in prior image frame and self size and picture centre coordinate to determine the treatment region of rear two field picture Territory；

4) fast Template is extracted and is identified

In target figure, search for the subimage matched according to known template figure, use and quickly go mean normalization product correlation al gorithm The objective fruit of two field picture after match cognization；

5) objective fruit location

To going mean normalization product correlation al gorithm to be accelerated Optimal improvements, carry out template in the rear two field picture region constantly reduced Join, position objective fruit.

During image segmentation, owing to having bigger color distinction between Apple and its background, select based on color characteristic for this Image partition method.

Objective fruit is undoubtedly the main related information between the gathered image of apple picking robot.When objective fruit determines, Single Mechanical hands picking robot when carrying out picking fruit, can only the most single harvesting, therefore when image has multiple fruit, Must determine out the objective fruit that will carry out plucking.

When identifying extracted region, it is Step wise approximation picture centre owing to gathering the center-of-mass coordinate of objective fruit in image, so phase Gathering image for first width, the processing region of subsequent acquisition gained image can be substantially reduced, such that it is able to greatly reduce image procossing Time, and then shorten the overall plucking time of picking robot, strengthen it and pluck rapidity.

Fast Template is extracted when identifying, rear two field picture is closed by the center-of-mass coordinate and self size utilizing prior image frame objective fruit Manage downscaled images processing region thus reach to reduce the purpose of image recognition time.For the target fruit of two field picture after reducing further Real recognition time, after utilizing prior image frame related information to reduce while two field picture processing region, extracts the target of first two field picture Fruit as the template of successive image identification, uses and quickly goes mean normalization product correlation al gorithm to carry out the mesh of two field picture after match cognization Mark fruit.Algorithms of template matching recognition is exactly the process searching for the subimage that matches according to known template figure in target figure.Go all Value normalization product correlation al gorithm is insensitive to the change of brightness of image and grade, for other correlation matching algorithms, and robust Property strong, precision is high.

One apple picking robot objective fruit of the present invention quick Tracking Recognition method, can improve the harvesting of apple picking robot Speed, reduces and the gap of artificial picking rate, practical.

Detailed description of the invention

Below present invention is further elaborated, but is not limitation of the invention.

A kind of apple picking robot objective fruit quick Tracking Recognition method, concrete steps include:

1) image segmentation

It is chosen at one group of picture of apple orchard shooting under natural environment, selects Apple and the branch of background, greenery and sky areas, The value of its R, G, B color factor is carried out statistical analysis, by aberration R-G and 2R-G-B method by Apple from background In separated；Owing to having bigger color distinction between Apple and its background, select figure based on color characteristic for this As dividing method.

2) objective fruit determines

Use 8 neighbourhood signatures's methods that the above-mentioned fruit handled well segmentation image is marked, and two dimension is asked in labelling fruit region Center-of-mass coordinate, formula is

In formula: the horizontal stroke of i, j fruit image pixel, vertical coordinate

The total pixel number of n fruit image

Ω belongs to the collection of pixels of same fruit image

Calculate its length of side simultaneously, finally determine objective fruit with the nearest principle in range image center.

3) extracted region is identified

The objective fruit information of prior image frame can be that rear two field picture target recognition is offered reference, and i.e. utilizes objective fruit in prior image frame Center-of-mass coordinate and self size and picture centre coordinate determine the processing region of rear two field picture.Owing to gathering target fruit in image Real center-of-mass coordinate is Step wise approximation picture centre, so gathering image, the process of subsequent acquisition gained image relative to first width Region can be substantially reduced, such that it is able to greatly reduce image processing time, and then the overall plucking time of shortening picking robot, Strengthen it and pluck rapidity.

4) fast Template is extracted and is identified

After in above-mentioned steps, two field picture carrys out reasonable downscaled images by the center-of-mass coordinate and self size utilizing prior image frame objective fruit Processing region thus reach to reduce the purpose of image recognition time, and its image-recognizing method is identical.In order to reduce further The objective fruit recognition time of rear two field picture, after utilizing prior image frame related information to reduce while two field picture processing region, carries Take the objective fruit template as successive image identification of first two field picture, use and quickly go mean normalization product correlation al gorithm to mate The objective fruit of two field picture after identification.Algorithms of template matching recognition searches for, according to known template figure, the son that matches exactly in target figure The process of image.Go mean normalization product correlation al gorithm insensitive to the change of brightness of image and grade, relative to other relevant For joining algorithm, strong robustness, precision is high.

5) objective fruit location

To going mean normalization product correlation al gorithm to be accelerated Optimal improvements, carry out template in the rear two field picture region constantly reduced Join, position objective fruit.

Claims (1)

1. an apple picking robot objective fruit quick Tracking Recognition method, is characterized in that concrete steps include:

1) image segmentation

It is chosen at one group of picture of apple orchard shooting under natural environment, selects Apple and the branch of background, greenery and dead zone, sky Territory, carries out statistical analysis to the value of its R, G, B color factor, with aberration R-G and 2R-G-B method by Apple from In background separated；

2) objective fruit determines

Use 8 neighbourhood signatures's methods that the above-mentioned fruit handled well segmentation image is marked, and two dimension is asked in labelling fruit region Center-of-mass coordinate, formula is $\left\{\begin{array}{c}x=\underset{i,j\∈\mathrm{\Ω}}{\Σ}\frac{i}{n}\\ y=\underset{i,j\∈\mathrm{\Ω}}{\Σ}\frac{j}{n}\end{array}\right.$

In formula: the horizontal stroke of i, j fruit image pixel, vertical coordinate

The total pixel number of n fruit image

Ω belongs to the collection of pixels of same fruit image

Calculate its length of side simultaneously, finally determine objective fruit with the nearest principle in range image center；

3) extracted region is identified

Utilize the objective fruit information that prior image frame identifies to reduce the recognition time of present image, analogize frame by frame with this and successively decrease, Utilize the center-of-mass coordinate of objective fruit in prior image frame and self size and picture centre coordinate to determine the treatment region of rear two field picture Territory；

4) fast Template is extracted and is identified

In target figure, search for the subimage matched according to known template figure, use and quickly go mean normalization product correlation al gorithm The objective fruit of two field picture after match cognization；

5) objective fruit location

To going mean normalization product correlation al gorithm to be accelerated Optimal improvements, carry out template in the rear two field picture region constantly reduced Join, position objective fruit.