Research, Artificial Three-dimensional Object Recognition Based On Geometric Properties

Three-dimensional object recognition is one of the most important research topics in computer vision.Generally,the existing three-dimensional object recognition technology is a process which acquired features from the scene image,and then compared them with models in data base.Three are three important issues in this process:the extraction and expression of scene features,the structure of three dimensional object models,the matching method of scene features and object models. The most major issue is the matching method.Geometric features are free from illumination,color and texture,and they widely exist in the artificial objects.For this,the paper presents a new three-dimensional artificial object recognition method based on geometric features.We detected line segments based on freeman chain,and recognized objects through the matching result of line segments between scenes and images.Compared to the recognition base on feature points,this method uses fewer features and it is robust enough to have a lower error probability.The shape features correspondence computing algorithm can commonly be divided into several categories:interpretation tree,pose clustering and hypothesize and test and so on.Interpretation trees are not generally able to work with single images.Pose clustering is noise sensitive and high-dimensional.The traditional hypothesize and test also has a very high complexity when the scene is complex.We use hypothesize and test method.First we pre-match the feature points using geometric invariants,then assume and solve the plane homography matrix between scenes to model.After that we match the line segments on the homography plane,and by this we verify the assumption.Compared to the traditional hypothesize and test, our method scattered the global coordinate transform solving over the various planes homography transform solving.The complexity was greatly reduced,so it can recognize object fast and accurately.The paper proposed a line segment based three-dimensional object recognition method with a comparatively high automation level.To validate our method,a large number of experiments were carried out.In addition,according to the experimental results,we improved the matching method.