Development of simple jaw-tracking system that applies pattern recognition technology

我们试图建立一种神经网络算法，从口腔周围的运动中推测切牙运动，以开发一种新的颌骨跟踪系统，而不需要复杂的设备围绕受试者嘴巴。在下颌切牙上安装一个直径为10 mm的圆形下颌运动检测标记器，用于检测真实的下颌运动，并在受试者的口腔周围安装8个相同的标记，以检测口腔周围的运动，以神经网络算法估计下颌运动。在受试者面前用数码摄像机记录下神经网络学习1次的侧缘运动和最大张开次数，其中4次和用于验证的手势动作。在运动学习中，神经网络软件(Neurosim/L，富士通)学习切牙运动和口周运动之间的关系。在用于验证的运动中，利用学习的神经网络从口周运动中推定切牙运动。将假定的切牙坐标与测量的切牙坐标进行比较，并进行简单回归分析。利用高精度的神经网络从口周运动中推算出切牙运动的y值。在横向运动中，它们的x值也被推定得很高。然而，在最大张口和咀嚼口香糖时，x值并不精确。这可以通过修改神经网络的学习方法来改善。提示神经网络可以从口腔周围组织的运动中推测出切牙的运动。

We are attempted to establish a neural-network algorithm to presume incisal movement from peri-oral movements in order to develop a new jaw-tracking system without complicated devices around subject's mouth. A marker to detect jaw movements for control, with a circle of 10 mm diameter, was attached to the mandibular incisors to detect real jaw movements and eight same markers were attached around the subject's mouth to detect peri-oral movements to estimate jaw movements by neural network algorithm. Lateral border movement and maximum opening one time for neural network learning and four times of them and masiticatory movements for verification were recorded by a digital camcorder in front of the subject. In the movement for learning, neural network software (Neurosim/L, Fujitsu) learned the relation between the incisal movements and peri-oral movements. In the movements for verification, incisal movements were presumed from the peri-oral movements by using the learned neural network. Presumed incisal coordinates and measured incisal coordinates were compared and simple regression analyses were carried out. The y values of the incisal movements are presumed from peri-oral movements by utilizing neural network with high precision. In lateral movements, the x values of them are also presumed with high precision. However, in maxmum opening and gum chewing, the x values were not precise. This may be improved by modification of the learning methods of the neural network. It was suggeted that the incisal movement can be presumed from the movement of the peri oral tissue by the neural network.