Researches On Scene-based Plant 3D Reconstruction Technology

Automatically farm-produces-harvesting robot based on computer vision technical have good application foreground. In aboard, the researches have been done many years, and have had some results; but in my country, the researches about text are getting behind and there are have long distance to practice application. The study of this paper object is plant scenes, because the plant components appearance in a scene has certain stable, we can reconstruction 3D model of a plant scene with plant component models. This 3D model can provide farm-produces-harvesting robot getting the location of the plant components and automatically walking or harvestingAccording to the above-mentioned purpose, this text main work as follows:1. We analyzed the appearance feature of each plant component, and withdraw the color feature of the plant component, the plant component connectivity feature, the venation feature in a plant field scene. That set foundation for the mathematics of the plant scenes.2. We Analyze the algorithm of image processing in 3D reconstruction and find out the suitable numerical image processing algorithm of this research. These algorithms include the edge detection algorithm, portrait algorithm of the color covariance and the image dividing algorithm etc.3. According to the biology characteristics of the plant components, we create the leaf component model , stem component model, fruit component model, flower component model and appendix model. Then the plant component lib is created.4. We create the algorithm, which combination the plant model lib with images which are get from the left camera and the right camera, to reconstruct the 3D model of plant scenes. This algorithm has better efficiency than getting 3D model from images directly. At the same times, we use the non-key-frame images to certificate the 3D plant scene model, which obtained good affection.At the end of the text, we tallied up the main research work and discussed a future research direction. This research can avail the farm-produces-harvesting-robot to self-determinate walking, recognition the fruits and self-harvesting in the future.