標題: | 空間與時間最佳化的籃球影片相機校正系統 Spatially and temporally optimized camera calibration for broadcast basketball videos |
作者: | 鄭為之 王昱舜 Cheng, Wei-Chih 多媒體工程研究所 |
關鍵字: | 籃球影片;相機校正;時間與空間;Basketball videos;Camera calibration;Spatial and temporal |
公開日期: | 2016 |
摘要: | 我們提供了一個籃球影片的相機校正方法,將原始的court view shot籃球影片中球員的位置轉換到標準的籃球戰術板,相較於前人獨立使用空間或時間的概念來做球場位置的校正而產生的ㄧ些問題,我們的方法考量影片的時間和空間特徵,最佳化的方式結合了兩者的優點而能夠有效解決這些問題。
在籃球影片中常常有亮度變化、球員遮蔽特徵點的問題發生,在這些情況之下通常會發生偵測的偏差和不穩定等等問題,我們的系統考慮了特徵點的移動軌跡,以移動軌跡預測的特徵點位置取代偏差較大的原偵測位置,彌補以上提到的問題。另外若只以時間概念校正球場會造成校正偏差的累積,最終會導致球場之中線條或圖案的對應失敗。但若以空間概念輔助時間概念,我們可以減少或避免計算偏差的持續累積而提高對應的準確。利用球員追蹤搭配我們的相機校正功能,我們可以準確的將某球員的位置投影至戰術板上,日後可以應用的領域包括戰術分析、取得特定球員軌跡、球員資訊的統計甚至是擴增實境的等等。 We introduced a camera calibration technique for broadcast basketball videos. Our method correctly transform basketball players’ position from broadcast video into standard basketball court model. Compare with previous works which only used spatial or temporal features to calibrate video camera independently, which exist some artifacts,our method automatically generate an optimal result using the spatial features and temporal features in the video.Due to illumination changes and player occlusions,it is hard to detect stable and reliable court features, our method however, overcome the problem by taking temporal coherence into consideration; besides,the accumulation errors caused by similarity transformation would decrease thanks to spatial features detection.Using our camera calibration technique, we can transform the positions of basketball players into standard court model coordinate, we believe our method would be beneficial to tactic analysts, users who want to retrieve players’ trajectories, and augmented reality. |
URI: | http://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT070256650 http://hdl.handle.net/11536/143212 |
Appears in Collections: | Thesis |