Depth Info Acquisition For Tree Images Based On Stereo Vision

Presently the forestry plant diseases and insect pests restrict development of forestation, virescence and entironment, however the chemical treatment, as the traditional solution, is of very low efficiency. So, developing precision pesticide application methods based on machine vision technique is necessary. Because the tree of precision pesticide application technique is the 3D object, the method of adjustment to spraying ranges according to individual depth (distance) info of different position of tree crown was suggested in this paper.Firstly, the author developed a binocular stereo vision system with parallel optical axes based on tree images according to precision pesticide application technique, binocular stereo vision basic principle and parallel optical axes vision model. This system included two steps: offline camera calibration and online 3D reconstruction.Secondly, camera model, camera calibration and many calibration methods were researched, and we calibrated cameras by Tsai two steps method and then got inner and external parameters of the cameras. These parameters can be used to rectificate the distorted tree images.Finally, LoG filtering operator was used to preprocess the simulating tree images, gained ideal stereo disparity image make use of area-correlation match algorithm based on epipolar line restriction, feature restriction and consistency restriction, using VC++ as the development platform through OpenGL complete tree image 3D reconstruction.The research achievement in the paper can provide valuable experiences both theoretically and practically for implement of dimensional precision pesticide application.