Picking Robot Target Identification And Positioning Research

In the natural environment, Oranges target’s background is very complex, there is very common phenomenon with the branches and leaves covering or superposition between the fruits, the complexity of this environment makes the identification of machine vision system difficult, leading to picking robot cannot identify the target effectively and accurately. In order to solve this problem, this paper has studied the orange target image in the complex environment. The main work is as follows:This paper has studied the method of identification of the outline of the orange target in complex environment, firstly introduces the traditional edge detection algorithm, and tests the orange image, see from the test result, can’t extract the target contour from complex environment. This paper will apply K-means clustering algorithm and Canny algorithm method fusion to complete contour detection, using the K-means cluster algorithm to divide the target area from the image, then in combination with the Canny detection algorithm, detects the outline of the target area, thus accomplish the target recognition, the orange image test results verify the validity of the method.This paper has studied the separation of the overlapping orange target outline, by comparison of the characteristics and principle of the Corrosion Stripping and Watershed segmentation method divide overlapping (adjacency) target contour algorithm, studied the separation the outline of the overlapping target based on the K-means clustering algorithm.using the algorithm to test the target image of double adjacency and overlapping, the separation of overlapping target contour is also relatively completed in the experiment, show that the algorithm has stronger stability.The target contour matching is studied, in order to describe the outline of the target, import the geometric invariance as the description of the outline, the results of the differential method are used as the contour matching measure of two images, show that the geometrical invariant moment parameter has a good effect on the characteristic description of the orange target, has good match ability. Then the gradient based Hough transform circle was introduced to reconstruct the shape of the circular orange target, the fruit can be effectively positioned.This paper calculates the target depth based on the monocular vision, camera collects two images from the same scene, extract matching feature points in two images, and combine the camera imaging model, calculate the depth information of the space object from the camera.Finally, introduces the experimental hardware system, then think of orange and jujube as the experimental subject, the identification and positioning of target profile in complex environment is completed, verified the effectiveness of the algorithm in this paper.