An Erhu Bowing Self-Learning System Using YOLO Deep Learning Techniques
| Autor: | PENG, CHANG-JAN, 彭昶然 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 Generally, when the beginners of erhu are practicing, it is difficult for them to know whether his posture is correct. Unless the teacher gives guidance in the face-to-face, the wrong posture will not be corrected immediately, and the wrong habits will be developed. The “Level, Straight and Stable” of erhu is the basic threshold that the erhu learners must learn. How to solve this problem for the learners of erhu is the goal of this thesis. We define the rule of erhu bowing, “Level, Straight and Stable”, and design an auxiliary system that uses “Level” and “Straight” as the research focus to detect the trajectory of erhu bowing and the slope of the bow. This system can help erhu learners record the status and results of each exercise. When the erhu teacher can't give guidance on the spot, the assistant system will effectively assist the erhu learners to self-learning and record the results. In order to achieve this goal, we have developed a set of erhu bowing self-learning system. Through the object detection technology based on convolutional neural network, we can capture the erhu bowing trajectory and slope of the bow when the learners are practicing, and through the auxiliary system to record the practice process. Therefore, the erhu teacher can refer to the practice record to develop the next learning plan for learners. In the experimental stage, we will lighten and optimize the model in order to implement our object detection system on smart phones. We have prepared three different proportions of training set pictures to train the model separately. The results show that the training set contains two different scales of images, and the lightweight model it trains works best on smart phones. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|