| Apples are rich in nutrition and loved by people.However,most of the current apple picking operations are mainly manual picking,with a low degree of mechanization.Therefore,it is of great significance to study the apple picking robot and realize the automatic picking of apples.Apple will shake under the influence of environmental wind or picking operations.The position obtained by the traditional apple recognition algorithm is only a momentary position,which cannot accurately guide the apple picking robot to pick.It needs to track the position of the apple in real time.In order to realize the accurate and rapid picking of the oscillating fruit under the motion state of the apple picking robot,a method for oscillating apple tracking and recognition of the apple picking robot is proposed.The main research content and results of this article are as follows:(1)An experimental platform was built in the laboratory environment to simulate the orchard environment.In this paper,the median filtering of the image is first used to protect the edge of the apple as much as possible while reducing noise.After comparing a variety of color features,it is proposed to use 2R-G-B color features to process the original image to improve the distinction between apple and background,and at the same time convert the three-channel RGB image into a single-channel grayscale image.Then use dynamic threshold method(OTSU)segmentation to get apple binary image.(2)An Apple rapid recognition algorithm is proposed.In this paper,the hole filling algorithm is used to fill the holes formed by the calyx at the tail of the apple fruit.Then use a circular template with a radius of 5 pixels to open the image to remove small noise.After removing the non-interesting connected regions,for the defects of different shooting angles,different apple shapes and overlapped occlusions,and the problem that the global Hough circle transform is more time-consuming,the local parameter adaptive Hough transform is used Method to identify apples.Experiments show that the recognition method in this paper can identify apples quickly and accurately.(3)An Apple fast tracking algorithm is proposed.This paper first extracts the rectangular area where the target fruit is located as a template image in the 2R-G-B image,and extracts Shi-Tomasi corner points in the grayscale image.Due to the information correlation between the front and back images,this paper uses this information correlation to propose the use of pyramidal LK optical flow method to track Shi-Tomasi corner points,and then use the forward and backward optical flow error method to remove tracking unstable corner points,and then based on The Random Sampling Consensus Algorithm(RANSAC)calculates the affine transformation matrix to predict the position of the apple and correct the template scale,and finally uses the de-averaged normalized cross-correlation template matching algorithm to search for the best position of the apple within 1.2 times the prediction area to achieve the target apple Tracking.The experiment shows that the running time is 25 ms without correction template,and the tracking error is greater than 8%;the running time is 130 ms without predicting the position of the apple,and some apple tracking errors occur;the running time of the algorithm in this paper is 25 ms,and the tracking error is less than 4 %,The speed and efficiency of tracking and identification have been greatly improved.The above results show that the Apple Picking Robot Oscillating Apple Tracking and Recognition method proposed in this paper can achieve better tracking and identification of oscillating apples in motion,which is of great significance to automatic apple picking.reference. |
PDF Full Text Request