Tokuo Tsuji

High-speed 3D image acquisition using coded structured light projection

In various application fields, high-speed cameras are used to analyze high-speed phenomena. Coded structured light projection methods have been proposed for acquiring three-dimensional images. Most of them are not suitable for measuring high-speed phenomena because the errors are caused when the measured objects move due to light projection of multiple patterns. In this paper, we propose a new coded structured light projection method that can select not only a temporal encoding but also a spatial encoding adaptively for obtaining three-dimensional images at a high-speed frame rate. We also develop an evaluation system that uses a DLP projector and an off-line high-speed video camera, and verify the effectiveness of the proposed method by showing the obtained three-dimensional shapes for moving objects.

A Real-time Finger-tapping Interface Using High-speed Vision System

In this paper, we propose a finger-tapping interface called FINGERTAP, which can estimate the contact states and forces on human fingertips in real time by using only the fingertip positions measured by high-speed vision cameras. In particular, we focus on the high-frequency component that develops at the moment a fingertip contacts something. We introduce a real-time estimation algorithm for the fingertip contact state and force by using frequency filtering to detect the high-frequency movement component of the fingertip positions. We also introduce a clickable pointing interface and a virtual musical instrument as a demonstration of an application using the FINGERTAP system.

Real-time Sound Source Localization Based on Audiovisual Frequency Integration

We propose a pixelwise sound source localization algorithm based on audiovisual frequency integration. The localization is realized by detecting the common vibration dynamics of sound sources in the audio and the brightness signal. In order to detect the common vibration dynamics, temporal correlation values between the two signals are calculated in the algorithm. Several experimental results are shown for vibrated objects, and the pixelwise sound source localization images are obtained

2P1-G05 Fast Planning for Safe Motion Path by using Bounding Volumes

経路計画では,初期位置姿勢と目標位置姿勢が与えられた際 に,障害物を回避してその間をつなぐ動作経路を生成する.これ は,ロボットの状態の各自由度を軸とするコンフィグレーション 空間の経路探索問題となる.探索により生成された経路は,各コ ンフィグレーションをノードとし,そのノード間を線分でつない だグラフとして表現できる.これまで,PRMや RRTなどのラ ンダム探索に基づく経路探索手法 [2]が提案されている.これら の手法で生成された経路 (Fig.1(a))は,いくつもの折り返しから なる冗長な経路であることが多く,探索後に経路の短縮を行う手 法も提案されている.例えば,経路上に 2つノードをランダムに 生成し,そのノード間をショートカットできるか確認し,ノード 間に障害物がなければ,ショートカットして経路を短縮する.こ の処理を単純に繰り返すと,より線路長の短い経路 (Fig.1(b))を 生成することができる.一方で,障害物をかすめるような経路が 生成され,実機で実行すると衝突の危険性の高い経路となる. したがって,ショートカット生成の際に障害物とロボットの間 でクリアランスを確保する手法が必要である.その手法の一つと して,障害物との距離が一定以下になるショートカットは採用し ない手法が提案されている (Fig.1(c)).この手法では,狭隘な経 路を通る場合やランダム探索で生成されたノードが環境と接近し た場合など,初期経路が設定した距離以下の部分はショートカッ トされずに冗長な経路が維持されることが,問題点としてあげら れる.さらに,クリアランスの確認のための計算コストも問題と なる.経路のクリアランスは,ノード間の線分上でサンプリング された複数点で確認する必要がある.この確認を密に行えば安全 (a) Result of PRM (b) Greedy shortcut method

1A1-W04 Indoor Human Behavior Estimation by combining Hierarchical Hidden Markov Model and Laser Sensing System

This paper presents a system for estimating indoor human behavior using laser range finders on the floor. The proposed method uses hierarchical hidden Markov model(H-HMM) composed of an action estimate layer and a behavior estimate layer. The former is constructed by two kinds of HMMs: one is the HMM for estimating each action, and the other is the HMM for deciding the human action considering the action continuity. In the latter layer, one HMM learns each behavior by using as the features the relative relationship among the actions and the furniture. Our behavior estimation using such features enable to recognize the behaviors robustly even thought the indoor environment is changed.

A Motion-based System to Evaluate Infant Movements Using Real-time Video Analysis

This paper proposes a marker-less motion measurement and analysis system for quantitative evaluation of movements of infants. In this system, movements of infants are measured using a single video camera, and then changes of body position and motion in infants are calculated from binarized images, which are extracted using background subtraction or inter-frame difference. Furthermore, eight indices are derived for quantitative evaluation of movements of infants, such as body activity level and amount of body motion. Base on these data, doctors can therefore get an intuitive understanding of the movements of infants without long-period observation. This is considered helpful for supporting diagnosis and detecting of infants’ disability or function diseases in the early stages. In this paper, the evaluation indices and features of movement between 14 full-term infants (FTIs) and 11 low-birthweight infants (LBWIs) are compared using the prototype developed. The experimental results show that, with some LBWIs, the upper body moves more than the lower body compared with FTIs, demonstrating that the proposed system can quantitatively evaluate the difference between the movements of FTIs and LBWIs.