智慧型行動輔助居家照護機器人研發---子計畫三：結合音源定位與姿態辨識之人與機器人強健型感知互動介面研發

Abstract

在眾多的感測技術中，一般均認為影像與聲音是人與機器人最直接的互動介面，雖然 這方面的研究已經進行多年，在實際應用上仍然面臨強韌性不足的問題。互動介面的 強韌性是機器人能否實際走入人類生活的關鍵之一，而其中所引領出的理論問題更需 要深入去探討。本計劃將其於以往的研究成果，直接挑戰強韌性的問題。將預計使用 高維之麥克風陣列與三維影像處理的技術，在一般環境中萃取有效之人的語音與姿態 資訊，以遂行有效的溝通。本計劃預計以三年的時間，研究下列四個主要技術問題並 應用於智慧型行動輔助居家照護機器人。1. 高解析度之全數位式麥克風陣列之研究設 計; 2. 多維麥克風陣列設計與移動式多維麥克風陣列多聲源追蹤研究; 3. 利用單攝影 機之人體姿態辨識、記憶及追.理論之研究與嵌入式平台即時實現; 4. 結合聲源追蹤以 辨識被遮蔽情況下的人體姿態。

Among various sensing technologies, it is commonly recognized that image and voice are still the most straightforward interface when considering the human-robot interaction. Despite many years of research in these two interfaces, the robustness of using image and voice still remains an important issue for real practice. The robustness of any interactive interface for robot is the key to its realization. And the associated theoretical problems need to be studied with a greater depth. This proposal plans to study the robustness problem based on previous research results. Specifically, the proposal will try to establish the robustness results using high-dimensional microphone array and 3-D object recognition technology. The plan covers a 3-year study in the following topics: 1. High-resolution full-digital microphone array design and implementation; 2. Multi-dimensional microphone array and mobile microphone array for sound source tracking; 3. Human posture recognition, memorization and tracking using mono-vision; 4. Combining source tracking for object occlusion recognition.