Research On Object Pose Estimation Algorithm Based On Machine Vision

Industrial robots need to use a computer vision system to detect and locate objects and automatically pick up disordered objects.With the development of intelligent industrial production,the estimation of the six-dimensional pose of three-dimensional objects has become one of the problems that need to be solved urgently.For the three-dimensional objects in industrial production,the materials are mostly metal or plastic products,the surface texture is scarce,it is difficult to extract stable feature descriptors,the pose change range is large,the image and pose search space is large,and the search process takes a lot of time.At the same time,it is greatly affected by the external environment,and it is prone to false detection and inaccurate positioning.In order to improve the intelligence of industrial robots and complete the task of picking up disordered three-dimensional objects,this thesis deeply researches the target pose estimation algorithm in the field of machine vision.(1)The research summarizes and evaluates several types of common pose estimation algorithms,from simpler pure object detectors to two-stage 6D pose estimators.Comprehensive analysis of some commonly used performance indicators,comparison of algorithms,analysis of advantages and disadvantages,which is helpful to choose suitable methods for different application scenarios;(2)The current internationally leading pose estimation method is based on a deep convolutional neural network method,which can even detect and locate objects without prior special training for the objects.The accuracy rate of up to 90% is achieved on pre-trained objects,but it is still difficult to apply in practice due to various reasons.This article elaborates on the challenges faced by current pose estimation algorithms;(3)A 6D pose estimation method based on 3D CAD model for bin-picking application is proposed,the algorithm process is introduced in detail,and simulation experiments are carried out to prove the effectiveness of the algorithm;(4)A cube pose estimation algorithm based on feature corner points is proposed.Based on the Raspberry Pi platform,a feature point extraction module and a pose solving module are written using Python to realize a demonstration experiment of a robot grabbing a cube.With the widespread application of artificial intelligence technology,machine vision systems as powerful visual perception "organs" have penetrated into all walks of life.This article is researching this advanced technology,and proposes a 6D pose estimation algorithm based on 3D CAD model and a cube pose estimation algorithm based on feature corners,which realizes the robot picking operation of 3D targets and reduces external constraints.Rules have improved the flexibility and intelligence of automated production lines.