Research On Viewpoint Selection Algorithm For Three-dimensional Model And Its Application

At present, applications of three-dimensional models are growing rapidly, and the resources of three-dimensional model are more abundant. Users will need to quickly filter out models they need in the three-dimensional model database. Users will distinguish features of three-dimensional models through two-dimensional views of these three-dimensional models. So how to automatically select the viewpoint of three-dimensional model to fit people’s habits of observation and reflect the information of the three-dimensional model on two-dimensional views is becoming an important research issue.The relevant knowledge about how to select the optimal viewpoint of three-dimensional model is introduced in this paper firstly, and then the existing optimal viewpoint selection algorithms are analyzed. But those existing algorithms are not universal for all kinds of models. This paper presents an optimal viewpoint selection algorithm adapted to the features of three-dimensional model and the correction method for the direction of the view. This algorithm is applied in a three-dimensional model retrieval system based on sketch. The main research works include:1. This paper presents an optimal viewpoint selection algorithm adapted to the features of three-dimensional model. Firstly, mixed features of three-dimensional model are extracted. Then in accordance with mixed features, the model is classified by learning classifier. According to the classification result, a viewpoint measure for the optimal viewpoint selection is chosen. Experimental results show that the proposed algorithm can effectively compensate for the limitations of existing single measure algorithms facing with different kinds of models. The view gotten reflects the key features of the three-dimensional model accurately.2. The optimal projection view obtained from existing viewpoint selection algorithm is not sensitive to direction. This paper presents a direction correction algorithm based on feature points matching. Firstly, the classification label of the model is gotten through the classifier. Then the orientation of original view will be corrected in some degrees in order to get the candidate views. Then the direction cost between sample view which is correspond to the classification label and candidate views is calculated. Finally, the view with the smallest direction cost among the candidate views is selected as the corrected view.3. A sketch based three-dimensional model retrieval system is designed base on the above research. In this system, an integrated multi-view matching method is used to improve the stability of retrieval. It can help users to efficiently get the three-dimensional model features through the views which are more compliant with people’s observation habits.