Research On Tree Branches Information Acquisition For Apple Harvesting Robot

In China, such a large agricultural country, it is necessary to study the harvest mechanization. Today, fruit harvesting robot research focuses on the identification of target fruit, positioning and picking, however, there are few researches on how to avoid the collision between the manipulator and fruit tree branches. The working environment of the manipulator is complex, In order to avoid damaging to the robot caused by colliding between branches and the manipulator, it is necessary to obtain apple tree branches spacial information.The objective of this research was to study the vision system of obstacle avoidance for apple harvesting robot using computer vision and image processing techniques. The ripe tall apple trees are the research object and recognition and location of the tree branches is the aims. Several main points in this research are as follows:1.The color information of the several regions of the branches image was analyzed to find a suitable segmentation algorithm. In this research, apple branches image was divided into three regions:branches region, ripe apples region and the green area. The distribution of the three regions in the commonly used color space including RGB, HSV, YIQ, YCbCr, HSI, XYZ, Lab, I1I2I3and YUV color space was analyzed with matlab programming. The feasible segmentation method was summarized and a layer stripped segmentation algorithm was proposed, that is, to remove the green area and then the ripe apple region. Experiment results show that the proposed segmentation method can effectively identify the apple branches region.2. The image of apple branches obtained after segmentation with the proposed method was processed. The segmented image was processed by binary, and then the noise was removed by closing operation with the third order matrix structural elements. This process can fill some small voids of the branches region. Each connected region was marked and the region area was calculated. The small objects were deleted whose areas were below the threshold. Cracks, holes and the interrupted are filled to strengthen the branches connected to get a more real image of the apple tree branches.3. Apple tree branches characteristics were obtained. The principle and method to obtain the skeleton of the apple tree branches were analyzed based on mathematical morphology refinement algorithm. The structural elements to refine the apple tree branches were chosen. On the premise of maintaining branches connected, the skeleton was obtained by corrosion with mathematical morphology operator. That is the axis of the apple tree branches. Using the hit or miss transform was proposed to get the endpoint and the bifurcation point of the apple tree branches. Furthermore, combining with distance transform method, the diameters of the apple tree branches were calculated.4. According to the requirements for access to the depth information of the branches, camera calibration and binocular match were processed. The current camera calibration methods and the advantages and disadvantages of several methods were summarized. The camera parameters were obtained by Zhang Zhengyou calibration method. Combining with binocular camera imaging principle, the common match methods-SIFT match, SURF match and Harris corner detector match were summed up and match experiment was done.