多混合特徵法於自主投籃姿勢辨識

10.29428/9789860544169.201801.0085

夏至賢；陳昭和；郭景明；李永祥

Kinect感應器 ； 身體骨架動態節點資訊 ； 投籃分析 ； Kinect sensors ； Human skeleton node information ； Field goal shooting analyses

NCS 2017 全國計算機會議

2017（2018 / 01 / 01）

443 - 448

繁體中文

本文提出一個基於定點射籃動作並配合數值分析方法作籃球投籃姿勢以辨識進球與否。利用Kinect 感應器擷取球員身體骨架動態節點資訊並捕捉全身動態動作進行研究分析。整體投籃辨識系統的架構，主要是從球員在罰球線投籃，由投籃預備期到籃球投射出去的軌跡過程，本文記錄身體連續動作骨架動態資訊分析。從籃球員身體的關節三維動態資訊中，擷取出完整的投籃動作中，作為分析與辨識。由實驗得知，本文所提出的籃球員自主投籃姿勢辨識系統進球辨識率可達75%， 非進球辨識率達66.7%以及辨識時間為15.625ms。

This work proposes an analysis method base on field goal shooting postures that can predict good field goals. This work analyses the basketball player skeleton information which obtains from the Kinect camera. Based on the movement of the data captured from Kinect, we divide the field goal shooting posture into three steps, before-, mid-, and finish-action. The proposed analysis method abstracts complete field goal shooting movement from a series of continuous frames containing (x,y,z)-coordinate information of the joint positions. As results, the recognition rate of good field goals in all field goal shooting posture is up to 75%, while the recognition rate of missed field goal shots is 66.7%. The proposed technique has process ability in real-time up to15.625ms.