Masakatsu Kourogi

Real-time image mosaicing from a video sequence

The paper describes a fast and robust image registration method that can be used to create a panoramic image/video from video sequences. To estimate alignment parameters for image registration, the method computes pseudo motion vectors that are rough estimates of optical flows at each selected pixel. Using the proposed method, we implemented a software system that can, with a low-cost PC, create and display panoramic images/videos in real time.

A panorama-based technique for annotation overlay and its real-time implementation

The panorama-based annotation method described uses a panoramic image as the source of information about the positions of the annotations. It finds image alignment parameters between an input frame and the panoramic image and then maps the positions of annotations from the panoramic image to the input frame and displays the input frame overlaid with those annotations. Camera movement from place to place is made possible by preparing a set of panoramic images in advance. The panoramic image that gives the least mean squares error of the image alignment is selected automatically and is appropriately switched as the camera moves around. The position of the camera can be tracked by monitoring the switching of selected panoramic images. Experimental results show that this method can find image alignment parameters, display input frames overlaid with the annotations, and switch the panoramic image appropriately in real-time.

Improvement of panorama-based annotation overlay using omnidirectional vision and inertial sensors

Annotation overlay on live video frames is an essential feature of augmented reality (AR), and is a well-suited application for wearable computers. A novel method of annotation overlay and its real-time implementation is presented. This method uses a set of panoramic images captured by omnidirectional vision at various points of environment and annotations attached on the images. The method overlays the annotations according to the image alignment between the input frames and the panoramic images. It uses inertial sensors not only to produce robust results of image registration but also to improve processing throughput and delay.

A Method Applied for Soccer's Behaviors Using Proper Feedback and Feedforward Control

As a practical way of achieving “Athletic Intelligence(AI)”, we have tried to build a robot capable of playing soccer. In this paper, we made a mobile robot that can shoot and dribble a ball. In order to shoot the rolling ball, the robot must predict the future position of the ball so that it can move ahead of the ball. That is, the robot should be controlled by feedforward control rather than feedback control because feedback control does not allow enough time to catch the ball. Therefore, we think that it is important that the robot has internal model with which it can predict the target position. As long as the ball is within the field of view, the robot is under feedforward control. At the final stage of shooting, control is switched to feedback to minimum errors. When dribbling the ball through the flags, the robot must move without touching the flags and also keep the ball in front under feedback control. Since the robot has an internal model, it should follow the target using feedback control. We have checked that proper use of both control improves the two athletic movements so the robot must have an internal model that can predict the future state of the target object.