Research On Stereo Vision Technology Of Fruit And Vegetable Harvesting Robot

Fruit and vegetable harvesting operation is a strength, long time-consuming, highly request work, and have certain hazardous. With the aging of the population and the transfer of agricultural labor, harvesting operation urgently needs high efficiency, current, low-cost robot technology to improve productivity and quality of operations, improve the production environment. Along with the development and the application of the technology of precision agriculture, as one of the effective equipments in the field of precision agriculture, the research and application of the robot with intelligent and stereo vision had been paid more attention to and been developed gradually.The principle of stereo vision needs to acquire two or more scene images of different angles at first, and then calculate the disparity between the pixels of these images to obtain the target object's 3D information, this process is similar to the feeling of human beings. Along with the development of stereo vision technology, the trinocular stereo vision system comes forth, through matching among the information of images acquired by three cameras simultaneously, it can solve some problems in binocular stereo vision system effectively. Eliminate the ambiguity and reduce the difficulty of matching.On the base of representative binocular stereo vision technology, this paper introduced some component modules and basic principles of stereo vision system, and narrated some problems and difficulties of vision technology's application research on harvesting machine. According to the introduction of the trinocular stereo vision technology, this paper established a camera model with one rank radial aberration, and used radial alignment constraint method to calibrate the camera. The process of calibration can determine the location, attribute parameters of the camera, and make it be convenient to make sure the relationship between target object and its imaging.Aim at raw images acquired by camera, this paper strengthened the interested target by using image division technology. Then based on classic edge detect technology to acquire edge information of target object, and obtained corners of it by using Harris's method. Corners retained rich characteristic information of target object, they could provide adequate restraint to analysis and processing of the target. Meanwhile, this process reduced the amount of information of image, and also the amount of operation of matching process, and improved the speed of entire system.The most difficult and complex step of stereo vision is namely stereo match. In order to solve this step, this paper advanced trinocular stereo matching method based on corners, and used the depth model of trinocular camera produced by superimpose images to match these corners, which aimed to obtain depth information of the fruit and vegetable. Trincular stereo match method can overcome some natural limitations of binocular stereo vision technology effectively, such as its accuracy depends on the depth of target object and the length of baseline of two cameras. Meanwhile, this paper used corners as matching characteristics in ternary matching, in this situation, matching errors will happen only in occasional circumstance. In this way, this method can ensure the reliability of match result.Finally, this paper made use of stereo vision technology to construct experimental platform including trinocular camera, image acquisition card, image processing computer, and used this system to do a range of processing to three images including the same target, for example, adjustment and edge detect, acquire the depth image of target fruit and vegetable. It can direct autonomy grab of the fruit and vegetable harvesting robot, and achieve the purpose of this paper.