The Design And Implementation On Registration System Of Depth Image Of Plants In 3D Construction

Three dimensional visualization analysis of plant morphology is a technical problem in the field of agricultural research. At present, the measurements of morphological characteristics information of plants mainly rely on artificial measurement, such as eye-measurement. As a result, it leads to varieties of disadvantages, such as low speed, high strength, strong subjectivity and large errors. It can effectively avoid these problems by using computer vision technology to study the 3D shape of plants. The critical problems are to get the depth image of plants and registration on varieties of depth images obtained from different perspectives. To solve this problem, this study takes neighborhood geometric feature constraint point cloud registration method as the object of study. We analyzed its principle, realization process, main advantages and practicability. A software system for registration on depth image data of plants is designed and developed based on the OpenGL graphics library by using this algorithm. The main contents are as follows:(1)Collection and preprocessing of experimental samples. According to the existing research results of the research group and the needs of the research project,we collected several depth images of a variety of plants as the sample data using the 3D scanning system developed by research group. In order to ensure the accuracy and efficiency of registration,denoising processing is necessary for the collected depth images. For this reason, the collected depth images of a variety of plants was denoised by using unidirectional multilevel median filtering algorithm. As a result, the noise points in the depth images were removed and the sharper and higher quality depth images were selected as the experimental samples.(2)Development of the registration system for depth image of plants. A registration system for depth images of plants was designed and developed based on Visual Studio 2010 combined with OpenGL graphics library with reading and display function, registration function and interactive control function of the depth image of plants implemented. Firstly, the feasibility of the system was analyzed from three aspects of theory, technology and economy. Next, the basic situation of the system was analyzed by using the structured analysis method to confirm the basic function of the system. Then, according to the function that the system needs to be implemented, system class and data processing were designed to confirm the system function module and technical route using software design process and technology. Finally, according to the algorithm flow of local geometric feature constraint point cloud registration algorithm, coding and implementation of each function module were completed with the help of OpenGL graphics library.(3)System stability and universal applicability testing. The stability and robustness of the system were tested by using automated test software AutoRunner, which proves that there is no abnormal phenomenon such as the collapse of the system during the operation. We compared the system function and the requirements of users, which proves that the expected basic function of the system has been implemented. The universal applicability of the system was tested by using the classical Bunny point cloud model, which proves that the system has good universal applicability. The analysis of the results of the two groups of comparative experiments confirmed that the system has a good registration effect for different kinds of plants with different forms and has a certain practical value. The experimental data of varities of point clouds verifies that the registration accuracy of the algorithm doesn’t significantly decrease with the increase of the number of point cloud data when the amount of point cloud data reaches about 6000.