Tileylioğlu E, Yilmaz A
Annu Int Conf IEEE Eng Med Biol Soc. 2015;2015:4820-3. doi: 10.1109/EMBC.2015.7319472.
In this study, two gait phase estimation methods which utilize a rule based quantization and an artificial neural network model respectively are developed and applied for the microcontroller based semi-active knee prosthesis in order to respond user demands and adapt environmental conditions. In this context, an experimental environment in which gait data collected synchronously from both inertial and image based measurement systems has been set up. The inertial measurement system that incorporates MEM accelerometers and gyroscopes is used to perform direct motion measurement through the microcontroller, while the image based measurement system is employed for producing the verification data and assessing the success of the prosthesis. Embedded algorithms dynamically normalize the input data prior to gait phase estimation. The real time analyses of two methods revealed that embedded ANN based approach performs slightly better in comparison with the rule based algorithm and has advantage of being easily-scalable, thus able to accommodate additional input parameters considering the microcontroller constraints.
在本研究中，开发了两种分别利用基于规则的量化方法和人工神经网络模型的步态阶段估计方法，并将其应用于基于微控制器的半主动膝关节假体，以响应用户需求并适应环境条件。在此背景下，建立了一个实验环境，在该环境中从惯性和基于图像的测量系统同步收集步态数据。结合MEM加速度计和陀螺仪的惯性测量系统用于通过微控制器进行直接运动测量，而基于图像的测量系统则用于生成验证数据并评估假体的性能。嵌入式算法在步态阶段估计之前动态地对输入数据进行归一化处理。两种方法的实时分析表明，基于嵌入式人工神经网络的方法与基于规则的算法相比表现稍好，并且具有易于扩展的优势，因此考虑到微控制器的限制能够容纳额外的输入参数。
