Research On 3D Object Recognition Method Based On Point Cloud

Machine vision is a subject that combines the theory and technology of multiple disciplines.Object recognition is an important research topic in the field of machine vision.It focuses on image acquisition,processing analysis,output and reality,that is,transforming the surface information of the measured object into image data,and then extracting various features of the image.With the rapid development of three-dimensional(3D)scanning technology,computer technology and target recognition technology,target recognition technology based on 3D point cloud data has become a research hotspot in the field of machine vision and has been widely appilied in civil and military fields.The main work of this paper is divided into three parts:Firstly,a bilateral filtering point cloud comprehensive denoising algorithm based on fuzzy C-means clustering and normal correction is designed.The filtering method combines the advantages of normal vector correction,fuzzy C-means clustering and bilateral filtering.The fuzzy C-means clustering method is used to remove the outlier noise,and then the normal vector of the point cloud surface is modified.On this basis,bilateral filtering is performed.It can remove large-scale noise and smooth small-scale noise,and has the characteristics of universality,robustness and high efficiency.It can maintain the key characteristics of the model,remove a large range of noise,and has high running time efficiency.Then,a object recognition algorithm is designed based on Keypoints-based surface representation(KSR).The key points are detected by SIFT algorithm,then the KSR descriptor is calculated,and the local reference frame(LRF)is established at the same time.The calculation of KSR reduces the complexity of the algorithm and improves the calculation speed of the algorithm.Hypothesis is generated by KSR matching and verified by ICP algorithm,then recognition and segmentation of object are completed.The performance of the recognition algorithm is tested based on the data set.Finally,a point cloud recognition algorithm based on covariance feature combination optimization is designed.This method can randomly combine the color,location,depth,geometry and other feature information of 3D point cloud and select the optimal combination.It has the advantages of compactness and flexibility.The genetic algorithm is used to optimize the parameters of support vector machine(SVM)classifier,and SVM is used to classify features.The experiment is divided into three parts: entity classification,category classification and scene object recognition,and compares the recognition effects of various combination features.The simulation environment of this paper adopts VC++ environment and third-party tool point cloud library to verify and analyze the feasibility and effectiveness of the algorithm proposed in each chapter in the selt-built data sets,RGB-D data sets and the corresponding complex scenarios.The experimental results show that all the methods proposed in this paper can achieve the desired results and have certain practical engineering application value.