【作者】 孙建刚；

【导师】 李铁军；

【作者基本信息】 河北工业大学 ， 机械电子工程， 2006， 硕士

【摘要】 多模式信息融合技术是一门新兴的实践应用技术,有着广泛的应用领域,它为各领域的信息处理以及决策支持提供了可靠的手段,具有很高的研究价值以及应用潜力。本文模拟人类多模式信息融合的机制,进行了视触信息融合的研究,通过注意力机制提高计算机信息处理的效率。本文做了如下几方面的研究:首先,分析总结了神经生理学、心理学在人类视触融合机制的最新研究成果。比较了单模式和多模式在物体识别中的优缺点。通过比较发现,多模式对物体识别具有识别效率高,抗干扰性强的优点。其次,基于神经生理学的最新研究成果,本文建立了机器人视触融合驱动视觉注意的理论模型。根据物体的视觉和触觉信息显著度的不同,选择显著度较大的物体驱动视觉注意,使该物体位于图像分辨率最高的区域,即摄像机光轴与物体形心重合。通过该方法,机器人可以获得更加精确的物体视觉信息,从而提高物体识别的效率。机器人能够通过注意力机制实现相关参数的实时自动更新,提高系统的自适应能力。基于该理论模型的机器人通过学习能够补偿由于环境变化和自身机械结构变化而引起的误差,矫正CCD摄像机的图像畸变。使运动精度不断提高,从而达到规定的要求。最后,本文根据视觉注意的焦点,建立了基于视觉注意的物体识别的理论模型。在注意前期获得图像显著特征的过程中,根据触觉模式提供的物体位置信息,得到物体在图像中的相应位置,确定图像处理的区域,提高了图像处理的效率。基于注意焦点的物体识别可以获得更加精确的图像信息,提高物体识别的准确度。更多还原

【Abstract】 Multimodal cues fusion is a developing and practical technology in the field of robot investigation and application. It is a valuable method for information processing and decision-making in many fields, also with huge potential application. By introducing the model of visual attention mechanisms based on visual-tactile cues fusion, the robot perception system can improve the efficiency of information processing to match the need of real-time demand. The main contents of this paper are as forllows:Firstly, the latest research achievements in neurophysiology and psychics in human visuo-tactile fusion is analysed and summarized. The superiority and defects between the single model perception and multimodal perception are compared. The result is that the multimodal object recognition is more superior than single modal object recognition, not only in the efficiency and anti-jamming characteristic.Secondly, this paper is based on the latest research achievements in neurophysiology. The visuo-tactile coordination theoretical model is set up. According to the different salience of objects in visual information and tactile information, the model will choose the object which is more conspicuous to drive visual attention and to make the axis of vidicon and the center of the object to superposition. Consequently, the robot can get some precise visual information in interested region and obtain more detail in object recognition. In the process of information, the relevant parameter is need not to be provided beforehand, the robot can not only ceaseless update those parameters to adapt the change of the environment but also automatically compensate the error which is produced by circumstance and mechanism itself. At the same time, it can rectify some image aberration of CCD camera in learning process.Finally, the paper set up an object recognition theoretical model that is based on human visual attention mechanism. By acquiring the cues of the orientation of the object through tactile perception, the robot can reduce some preparing work of image processing so the speed of image processing is increased, the processing time is decreased greatly . Based on the focus of visual attention the model can improve the efficiency of image processing and get more accurate image information to implement accurate object recognition.更多还原