Research On Recognizing And Locating Tomatoes Base On Binocular Stereo Vision Technology

Automatic identification and location of fruit was very important for the fruit picking robot, in order to prepare for the picking, the identification and location with high accuracy were essential. Automatic identification and location of fruit involved in a wide range of related technologies, including image processing, pattern recognition, stereo vision, etc. The main research topics consist of some parts, the calibration of binocular stereo vision system was the first step, target recognition was completed using pattern recognition theory in the processed image, received disparity map after match the stereo images, and then acquired the three-dimensional information of target.In chapter 1, the main significances of the research were presented.then domestic and international visual recognition technology research and development trends and the process of visual recognition were described also, and analyzed three levels of visual recognition:the underlying processing, intermediate processing, advanced processing. Afterward the identification method and spatial orientation methods were summarized. Finally, the main elements of the research, key technologies and technical route were discussed.Chapter 2 illustrated the basic components of binocular stereo vision system, based on the consideration of non-linear radial distortion and other factors, the pinhole camera imaging model was established. According to project requirements, considering the existed experimental conditions, the parameters of binocular camera were obtained using planar pattern calibration. In addition, the binocular stereo vision principle was the basic preconditions of the follow-up 3D reconstruction and positioning.Chapter 3 indicated that image processing was an important part of target recognition. Combined with the practical requirements of the subject, brief description of basic procedure of the image processing was completed, for instance:morphological processing, object silhouette edge detection, image segmentation and other aspects. These were the basis for subsequent target recognition. Color, contour, texture and other identifying features were extracted from the image, besides as a support vector machine (SVM) classifier input. Before SVM classifier application, a set of data was used for training, the use of cross-validation and grid search techniques to optimize the SVM RBF kernel parameters. Afterward tomato picture was took for the SVM classifier to recognition experiments. Finally a multi-class recognition direction was presented.Chapter 4 described image correction method by perspective transformation, to obtain binocular stereo vision with the standard epipolar geometry. Error energy minimization region based stereo matching algorithms were developed for extraction depth information from two color stereo image pair., and generated disparity map, introduced the 3D reconstruction-related technologies, ultimately calculated 3D coordinates of the target information of tomato by the disparity map.Chapter 5 summarized the main conclusions of project, and future trends of research direction were discussed.