智慧羽球拍應用於球種分類

Abstract

一場羽球比賽中，選手的球路選擇是實戰輸贏的重要資訊，所謂的球路即為一個羽球選手的對戰習慣或策略。本研究中，我們基於穿戴式設備與行動計算技術開發一套羽球球種記錄系統，透過智慧球拍即時記錄選手的擊球資料。該系統分成兩大部分，分別為擊球偵測與球種分類，在擊球偵測中，我們設計一個聲紋偵測演算法，分析不同球拍的聲紋特徵並根據聲紋判斷擊球事件與擊球時間點，該演算法的準確度可達100%，且在擊球點的時間精度上，誤差平均值為12ms，相較於過去文獻少了26ms；而在球種分類中，我們依據聲紋偵測演算法釋出的擊球時間點擷取慣性感測資料，並分析Random Forest、SMO（Polynomial Kernel, RBF Kernel）和Naïve Bayes四種分類器對於球種分類的準確度，其中SMO（Polynomial Kernel）最適合應用於球種分類，其平均準確度高達96.83%。本研究的球種記錄系統相較於目前市面上的產品，具有相當大的進步，未來可與雲端服務結合，可用於長期收集個人的揮拍紀錄，用以找出選手自身的弱點，提升對戰能力。

In a badminton competition, each player's style of play is winning tactic, style of play means each badminton player's favorite shots, weaknesses, like and dislike. In this study, we develop a badminton stroke recorder system which is based on wearable devices and mobile computing, and log stroke types with a smart racket in real time. The system divides into two parts, stroke detection and stroke classification, respectively. In stroke detection, we design a voiceprint-based detection algorithm, it can analyze the voiceprint features of different rackets and detect stroke event and stroke time with voiceprints. The accuracy of detection is 100% in our experiment, and the precision of stroke time is 12ms in average. Voiceprint-based detection algorithm reduces 26ms compared with the existing study[6]; In stroke classification, we get motion data based on the stroke time extracted from voiceprint-based detection algorithm, and evaluate accuracy with four classifiers, Random Forest, SMO（Polynomial Kernel, RBF Kernel）and Naïve Bayes, respectively. SMO（Polynomial Kernel）has the best accuracy, the average accuracy of classification is 96.83%. Our system has significant improvement compared with the commercial products. In the future, the system can combine with cloud services, collect the stroke data and find the weaknesses of each player to enhance his/her abilities of badminton.