Research On Target Recognition Method In 3D Point Cloud Data

With the rapid development of science and technology, the Internet and intelligence will bring more extensive and profound influence to us. With smart factories, intelligent production, and intelligent logistics as the theme, the fourth industrial revolution is on stage. The robot will become an important symbol of the fourth industrial. The intelligent robots are largely dependent on the perception of the environment. As one of the basic functions of intelligent robots, object recognition has attracted more and more attention. With the development of 3D scanning technology, object recognition based on 3D data is becoming a hot research topic which gathered scholars’ attention.The thesis studied the algorithm of object recognition based on 3D point clouds for intelligent robots. This thesis focuses on the preprocessing of 3D point clouds, the feature extraction from 3D point clouds, and object recognition based on 3D point clouds. The main research contents and achievements are the following aspects:1. Based on the relevant technical accumulation of our team, the research has been accomplished the acquisition of the 3D data of the object, and preprocessed the 3D data. Full investigated the technology of “Noise reducing” and “Simplify”. Based on this, used a new algorithm of “Noise reducing” based on the K-D tree. The validity and performance of the proposed method are theoretically analyzed and tested with sufficient data.2. Full investigated the algorithm of feature extraction based on 3D point clouds. Analyzed the relationship between the geometric discontinuity points and the geometric properties of 3D point cloud data. Based on this, we proposed a new algorithm of feature extraction based on normal vectors and curvatures. Achieved the object identified fast by using the features extracted from the point clouds and the typical characteristics of the object. The experiment shows that the algorithm of object recognition based on local features and which is under specific environment has a higher recognition performance.3. In order to solve the problem of the object recognition based on local features in some complex scenes, we propose a novel method that creates a global model description based on oriented point pair features and matches that model locally using a fast voting scheme. We analyzed the main steps of the algorithm, built a point cloud model database for the interested object and completed the recognition in the scene point cloud data.