2018 年 6 巻 1 号 p. 1-12
Recently, we have reached to the era using many videos by various image display devices. In particular, since an opportunity of viewing a content in a large size display or a mobile device has been increased, it has become one of the important problems to prevent visually induced motion sickness (VIMS). Although various methods to cope this issue had been studied, most of them were not practical. For example, to employ biological signals as a signal analysis approach to detect VIMS is difficult to suppress a cost and to reduce weight in a general home TV or a mobile device with a biological signal measurement device. In employing global motion vectors, it is also difficult to shorten processing time of estimating motion information, because they are employed as an image processing approach. Therefore in this study, in order to decrease an adverse effect on a human body due to VIMS, the author focuses on screen shake (SS) as VIMS, and proposes a determination method of SS in consideration of high-speed processing. The histogram is calculated by using motion vectors which are obtained by a simple block matching method. After that, histograms of two types are employed as a motion analysis (MA). First type is a histogram of motion direction, and second type is a histogram of motion magnitude. In addition, a frequency analysis (FA) is performed in horizontal directions, i.e., right and left direction. Thereby, it is possible to extract a change point of motion information caused by SS. The proposed method (PM) finally uses the combination processing of MA of each histogram and FA to enhance accuracy. Generating pseudo swing images according to five kinds of shake type (stype), the simulation experiments are carried out to evaluate PM. As the results, the results of accuracy ratio are 1.000 in stype 1 and 2, it is greater than 0.714 in stype 3, and it is greater than 0.573 except for s2 sequence in stype 4. However, the results of 0.202–0.366 obtained in stype 5 are lower than in the other stypes, and consequently there is a problem in a slow pseudo swing. By the experimental results, it was revealed that the pseudo motions affected the values of the degree of similarity and PM could simply determine the state of SS. Next, the result of 5.53 × 10 -4 sec. per frame in PM was obtained as the evaluation result in the processing time. Hence, it was found that PM can realize a real-time processing. Moreover, it was revealed that PM is more practical than the conventional method, in which newly obtained motion vectors in spatial domain are used, since PM can use coded motion vectors. In the further study, an evaluation by using practical images, and an evaluation of the swing in vertical direction or 2D swing direction will be required.
