Determine Next Best View Using Occlusion Information And Counter Information In The Depth Image

With the development of computer vision and the popularity of image collecting equipment Kinect, image processing is becoming one hot topic in the computer vision field. Based on the visual object of depth image, Next Best View is one of research subjects in many fields, such as 3D reconstruction, object recognition, robot navigation etc.In order to obtain the 3D information of visual object fast and accurately, the method to calculating Next Best View based on the visual object of depth information is proposed in this paper.Firstly, a method based on the visual object of depth image is proposed by using occlusion information and counter information to determine Next Best View. The proposed method determining Next Best View take the following steps. First, detect the obtained depth image in the current view to acquire occlusion information. Secondly,according to the occlusion detection result and counter information of visual object, the unknown region is constructed and then the unknown region is modeled by using the triangulation. Thirdly, the objective function is constructed utilizing the midpoint, normal vector, and area corresponding to the triangles after modeling. Lastly, Next Best View is calculated by optimizing the objective function.Secondly, the method based on unsupervised learning is proposed according to the research of depth image and the characters of unsupervised cluster. The proposed method uses the machine learning theory into Next Best View. The related feature is extracted by each normal vector and midpoint of the triangle and then the Next Best View is calculated by the improved K-means algorithm.At last, through comparing with the experiment of existing related literatures, the feasibility and effectiveness of proposed method are proved, and the experiment result is analyzed and concluded.