To create a DEM using satellite images, correspondence detection in a stereo pair is operated. For a stereo pair, two or more in-track images are used for most cases. To use images between different orbits for a stereo pair, there are some subject about parallax, time difference, and environment. In this paper, we propose a selection method of stereo pairs for creating a precise DEM. Our method uses parallax, time difference and similarity between satellite images. And our method selects stereo pairs for every points for calculating height. We evaluate our method using ALOS/PRISM images, and the result shows the DEM provided by our method using different track images has almost same precision in comparison with the computed one using in-track images.
We proposed a novel edge extraction method from 3D point cloud. In our proposal method, we calculate inner products of a normal vector from a point and its around points, and take a solution of the average of the vicinities as the characteristics of them. When the normal vector at a center of plane is vertical to the adjoining continuous plane, or is not vertical, the inner product comes different from the others. Therefore, the averages of inner product around the edges and the others on planes, are speci_cally different and we can divide them into two classes clearly. Our experimental results showed that the proposal method is useful to the 3D edge extraction method.
Multi-view video makes users possible to watch live concerts or sports events more freely from various viewpoints. However, users should select a camera view which is showing captures their own preferred view at each event. In this paper, we explore an assistant method for viewing multi-view video of soccer game. We assume that users watch soccer game with following targets, which are a ball or a player who keeps a ball. Our proposed method learns user's interest for watching multi-view video contents by using Support Vector Machine and geometric features of the targets and then estimates which camera view agrees with the user's interest. Through the evaluation experiment, the effectiveness of our proposed method was confirmed.
This work focuses on a similarity between clothes on blurred images shoot by the security camera and clothes of the accused on clear images and proposes a method. The proposed method provides a similarity based on the marginal distribution and probability of occurrence of edge of each images. Experimental results show that the proposed method can authenticate efficiently.
As video content applications, video search techniques are often used. The techniques of video identification methods are required for the various applications. They are also used as the guarantee methods of correctness content from the aspect of copyright protection. Then, we focus on a video identification method as video retrieval techniques. From the viewpoint of the accuracy and the processing time for moving information, we compare traditional method with proposed schemes, and consider them. Moreover we propose a video delivery system to provide secure content services. Especially, we also propose a generating method of identification code using video indexing, and explain the concept. Then, the performance for identification is considered by experimental simulation.
We propose a method which quantitatively evaluates landscapes using images of landscape taken by a fixed camera. While evaluation of landscapes has been studied, it is difficult to evaluate landscape images using existing methods, because these methods only provide qualitative evaluation and landscapes change by various factors such as time, weather and season. In addition, evaluation elements of landscape are not always one and are different by each person. In our proposed method, we quantitatively evaluate landscape images taken by a fixed camera. In particular, our system enables for users to select evaluation items, so that landscapes can be quantitatively evaluated under sense of each user.
We propose a shading method for computer generated cel animation. Conventional cel shading generates a single color shade like cel animation depending on 3DCG model shape and light sources. However, the shape of its shade often becomes too complicated for cel animation. To solve this problem, we use 3D polygon models with smoothed normal vectors. First, we generate a smoothed shape model from the original complex shape model. Then, we copy the normal vectors of the smoothed shape model onto the original complex shape model. By rendering this model, we obtain a shading result suitable for cel animation.