A Novel Algorithm For Simplifying Descriptors In 3D Point Cloud

3D feature descriptors play an important role in the field of computer vision because it is a prerequisite for many 3D vision applications.Existing algorithms for 3D object recognition can broadly be classified into two categories: global feature based and local feature based algorithms.The local feature of the detected target can be used to achieve the purpose of identifying the target.Especially in the case of complex image content,large noise interference,local occlusion and clutter,pose changes,etc.,therefore,it is so effective to use local features for target recognition,that local feature descriptors have been widely studied in recent years.Although many such descriptors exist so far,most of them exist in the form of floating-point data,resulting in computational complexity in the feature matching process.In this paper,the simplified development of local feature descriptors in complex scenes is studied and used for target recognition.The key techniques in the algorithm process,such as model library construction,point cloud segmentation,feature extraction and description,feature matching,etc.,are analyzed and conducted a series of comparative analysis and verification.The work of the thesis mainly includes the following parts:1.Construct a point cloud model library for the working objects of the modular robotic arm sorting platform.Firstly,the noise point and outliers in the point cloud are removed by filtering.A region growing segmentation algorithm is designed to extract the object cluster.The ISS algorithm is used to detect the key points of the cluster and calculate the corresponding local feature descriptors to establish k-d tree.The tree sequence table implements the construction of the object model library.2.In order to improve the matching efficiency,this paper proposes a new simplified method.The method contains two parameters,simplifying the unit and the number of coded bits .For different original descriptors,the method proposed in this paper is adopted.The parameters can produce different simplified descriptors.Then use the Chebyshev inequality mathematical model to convert floating-point data into binary strings to obtain a simpler and more efficient feature descriptor CI-SHOT(Chebyshev's Inequality Signature of Histogram of Orientations).3.Then,we compare CI-SHOT and SHOT with another binary simplified descriptor B-SHOT.The experimental results on the dataset show that CI-SHOT has obvious advantages in key point detection and matching performance.Finally,the modular robot arm intelligent sensing and automatic working hardware and software system is constructed,and the new descriptor is used on the experimental platform,and the obvious effect is obtained.