Determination of a pedestrian who does a prespecified body motion in a structured environment

The authors propose an algorithm to extract a pedestrian among other passers-by who does a specific body motion. They consider the situation where a person gives instructions to a robot, by his body motions. The situation is simplified by restricting the scene to an in-house situation, and it is also assumed that the flow rate is not very crowded. Under these assumptions the algorithm is straightforward. The scene is divided into blocks, and the difference of a scene from the background is computed. If the difference is large, it is judged that the block has a temporal object (or not background). The difference is filtered through a low pass function, which becomes the criterion function to identify the stop state of a moving object. Three benchmark tests were done. At least within the tested conditions the algorithm works correctly. In the most unfavorable situation, in which a noise covers an entire object, the algorithm shows stable detectability, although it shows a temporal disturbance