| Summary: | 碩士 === 國立臺灣海洋大學 === 通訊與導航工程學系 === 103 === The application of MEMs－based smart phone device has become more and more popular in our daily life. However, due to the limits of product specifications or cost considerations, there exists a great amount of acceleration errors while executing the trajectory recognition experiment. These errors will gradually stack up and failed to rebuild the device position information, which may raise the difficulty of trajectory recognition.
In this thesis, a novel trajectory recognition algorithm based on model identification is proposed. In the proposed algorithm, the acceleration data from accelerometer-based device will be analyzed for elimination measurement bias, then built the auto-regressive (AR) model for corresponding type of trajectory. We then propose the method of dynamic time wrapping (DTW) to reach the purpose of recognition by solving the best correlation between unknown testing data and existing AR models. Based on the current simulation results, the algorithm proposed in this thesis offers better performance of operation time than the existing trajectory recognition method, which usually used neural network －that need additional cost on training and learning.
|
|---|
