Study On Vision System Based On The Four Degrees Of Freedom Tomato Picking Robot

Tomato planting area in China almost covered the whole country, and is forefront of the world. Because artificial picking tomato is high cost, long time, and difficult to guarantee the quality, it is very difficult to large-scale and industrialized cultivation. Using visual system to identify and locate the fruit in natural scenes, it is an important step in robot automation picking. Study on the visual system is great significance to reality the automated picking. Based on the four degrees of freedom tomato picking robot of the project team research and development, the vision system of the robot is researched and developed in paper.Main contents including:The existing tomato cultivation mode, growth properties, and the relationship between color change in the process of the tomato mature and the maturity of the fruit is studied. On this basis, hardware and software platforms of tomato harvesting robot vision system, and the flow chart of software system are established.Through analysis and study on existing image acquisition methods, the image acquisition method of the visual system is determined, and image acquisition software system of the robot visual is designed. The paper selected camera directly connects with mainframe to real-time image acquisition. Based on VFW, image acquisition system software is developed with Visual C++ 6.0 editing software. Test results obtain the acquisition system better meet actual needs, and improve the image acquisition flexibility, reduce system costs.This paper is the first to identify picking tomato of different maturity. Through the distribution statistical analysis of picking tomatoes and background color components, access to effective color indicators of segment picking tomato and background. Bring forward equal scale gray about color image, the color images segmentation will be turned into a gray image segmentation problem. Test result got to that the gray image can be better retain differences information of color images indicators, thus simplified the difficulty of segmentation and deal with image, and increased processing speed. The light and acquisition equipment brought about image noise is solved by using light compensation method and median filtering method. Threshold segmentation method used image segmentation in paper. The best combinations of the segmentation methods and color indicators are determined. The original Otsu method is improved. The image under different circumstances can be better segmented through the improve Otsu method.Based on the color and shape feature, the tomato is identified and the center of the tomato is got. Determined the shape features of the picking object, and calculated the features. The trip marking algorithm is used to the goal marker, the search direction rotation algorithm is used to goal outline extraction. Through morphological processing, object extraction, and filling empty, gradually removed the image interference and accessed to better integrated tomato. In the process of the tomato center acquired, the paper will research to three mutual locations, and the single fruit in the image can be better extracted.Base on the mission requirements and laboratory conditions, the paper adopted the plane template calibration method between the traditional calibration method and self-calibration method. Through self-made plane template, camera is calibrated, and the camera's internal and external parameters are accessed. To have finished the nonlinear optimization of the camera distortion compensates and to obtain optimal parameters with Levenberg-Marquardt iterative optimization algorithm.The binocular stereo camera calibration is finished.Two core issues of the Binocular Stereo Vision are three-dimensional reconstruction and feature match. In the paper, working space of the four degrees of freedom tomato picking robot is analyzed, the vision scope of the visual identification and positioning system and the baseline value of the cameras in the binocular vision system are established. Based on the center characteristics of the tomato and epipolar constraint, the only constraint, parallax gradient constraint, goal matching is realized. In three-dimensional reconstruction, space coordinates of the picking tomatoes is achieved with the parallel binocular stereo model. Through studied conversion relationship between camera coordinates and robot coordinates, the tomato coordinate in the robot space coordinates is final achieved.It has developed to the vision system of four degrees of freedom tomato picking robot. In the laboratory environment, tomato picking robot vision system is tested, and the error of the identification and location is analyzed. The testing results indicate that: The identification of the mature tomato can be up to 98%; the identification of green ripeing tomato can be up to 89 %; positioning error is in 15 mm. The effect of recognition and positioning can meet requirements of the four degrees of freedom tomato picking robot.