移動軌跡解析による動物体の有生性推定

抄録

The purpose of this work is to achieve computer discrimination between living matters that can create the subjective impression that it is alive and nonliving matters by locomotion analysis. First of all, moving object is detected using frame difference method and extracted as a silhouette image. Then the center of mass of the silhouette is computed to represent the object region. From a video sequence, the trajectory of the center of the moving area is obtained. We use the change of speed and the directional change of the center as the moving object's features to discriminate living matters from nonliving matters. AdaBoost was employed to learn the features. An experiment was conducted to evaluate the discrimination method using 77 moving images. The accuracy of discrimination was 85.7%.