Study on Integrated Recognition Systems

在该项目中，我们讨论了一种综合考虑语音、图像和视觉信息的识别系统，以及一种基于对话的主动识别系统，用于询问不可理解的含义或寻求更好的视角来识别对象。继上一财年之后，今年也主要讨论了以下三点。当机器人在给定的视点下不能识别物体时，主动识别机器人应该如何或在哪里移动才能更清楚地识别对象？2.在计算机视觉领域，运动或行为是由三维几何数据的变化来表示的。另一方面，推理或规划等人工智能方法不是基于几何数据的表达，而是基于逻辑符号数据的表达。人工智能方法与计算机视觉领域在数据表达方面存在差距。提出了一种基于框架文法将三维几何数据转换为谓词逻辑表达式的方法，并对该方法的有效性进行了检验。我们已经制作了一个主动识别机器人，以改变更好的视角，更清楚地识别对象，并检验机器人动作的有效性。

In the project, we discusses an integrated recognition system considering the fusion of speech information, image and visual information, and an active recognition system based on a dialogue to ask non understandable meanings or seeking a better point of view to recognize an object. In succession from the last fiscal year, also this year mainly discuss the following three points.1. How or where an active recognition robot should move to recognize objects more clearly, in the case when a robot could not recognize under the given point of view ?2. In the field of computer vision, a movement or behavior is represented by changes of 3-dimensional geometrical data. On the other hand, AI approaches such as inference or planning is not based on the expression of the geometrical data, but the expression of logical symbolic data. There has been a gap of data expression between AI approach and the field of computer vision. We propose a method of converting the 3-dimensional geometrical data into an expression of predicate logic based on the frame grammar, and examine the effectiveness of our approach.3. We have made an active recognition robot to change a better point of view to recognize the object more clearly, and examine the effectiveness of robot actions.