博士論文概要

全文

(1)早稲田大学大学院情報生産システム研究科. 博士論文概要. 論. 文. 題. 目. Automatic Data Acquisition Based on Abrupt Motion Feature and Spatial Importance for 3D Volleyball Analysis. 申請者 Xina CHENG 情報生産システム工学専攻画像情報システム研究. ２０１８年４月.

(2) Sports analysis technologies have attracted increasing attention with the hosting of big sports events in Japan, such as the 2020 Olympics. In p a r t i c u l a r, v i s i o n s e n s o r b a s e d s p o r t s d a t a a n a l y s i s t e c h n o l o g y i s e x p e c t e d to provide copious game data for the realization of professional strategy analysis systems and sports broadcasting systems. The TV broadcasting system for swimming and the soccer strategy analysis system have been d e v e l o p e d a n d p u t i n t o p r a c t i c a l u s e . H o w e v e r, t h e s e t e c h n o l o g i e s w o r k f o r the data acquisition of 1D and 2D movement. For expanding them to a wider range of sports such as volleyball, pingpong and tennis, the 3D s p o r t s a n a l y s i s i s e x p e c t e d . P a r t i c u l a r l y, t h e v o l l e y b a l l r e q u i r e s 3 D d a t a f o r 3 D m o v e m e n t s o f b o t h t h e b a l l a n d t h e p l a y e r. In order to realize the 3D sports analysis system for volleyball, the technologies that acquire the game data (the position and velocity of the ball and players, the motion of players and the data representing the play e f f i c a c y, e t c . ) a n d t h e i r r e a l - t i m e i m p l e m e n t a t i o n a r e r e q u i r e d . S i n c e t h e m o v e m e n t o f t h e b a l l i s d e e p l y i n v o l v e d i n e v e r y s i t u a t i o n d u r i n g t h e p l a y, the ball data holds the key of further data acquisition and analysis. The ball related data include the physical, event and strategy data. First, the physical data represent the 3D position and velocity of the ball. Second, the event data represent what kind of play has been done, such as receive, t o s s a n d s p i k e . F i n a l l y, t h e s t r a t e g y d a t a a r e v a l u e s d e c i d i n g t h e p l a y e f f i c a c y, s u c h a s t h e n u m b e r o f a v a i l a b l e a t t a c k e r s a n d b l o c k e r s , t h e attack tempo, the attack type, and the set zone. The physical data can be o b t a i n e d b y t h e 3 D t r a c k i n g t e c h n o l o g y, w h i c h s u f f e r s f r o m t h e a b r u p t b a l l motion, the ball moving out of the frame and the occlusion. The event data c a n b e o b t a i n e d f r o m t h e b a l l t r a j e c t o r y. H o w e v e r, n o i s e i n t h e t r a j e c t o r y affects the detection accuracy and the event change is difficult to predict bec aus e of t he unk now n t im e o f t he p l ayer ’s act i o n t hat c hanges t he e ve nt . To o b t a i n t h e s t r a t e g y d a t a , t h e r o l e o f e a c h p l a y e r ( a t t a c k e r, b l o c k e r, setter) is required, however it cannot be automatically obtained only from the physical data and event data without manually labeling. To s o l v e a b o v e p r o b l e m s a n d r e a l i z e a 3 D s p o r t s a n a l y s i s s y s t e m f o r volleyball that enables automatic acquisition of ball related data (physical data, event data, and strategy data), the abrupt motion feature and spatial importance are proposed in this dissertation. The abrupt motion feature is defined by the velocity change in the abrupt situation and the g e n e r a l s i t u a t i o n o f t h e b a l l a n d p l a y e r. B y c o m b i n i n g t h e a b r u p t m o t i o n feature in different situations or using it as the likelihood, it is possible 2.

(3) t o a c q u i r e s e v e r a l k i n d s o f d a t a a c c u r a t e l y. I n a d d i t i o n , t h e s p a t i a l importance indicates the probability distribution of the ball candidates, the event change positions and the locations of each player role. Based on these, target state of tracking and detection can be estimated accurately so that a higher performance of data acquisition can be achieved. This dissertation is organized as follows. Chapter 1 explains the significance of 3D sports analysis technology and the necessity of automatic data acquisition based on vision sensors. Then, the points of focus and the objectives of this dissertation are described. In chapter 2, to acquire the physical data of the ball such as 3D position and velocity from the multi-view game data, the abrupt motion adaptive system model and the ball candidate distribution based recovery are p r o p o s e d . I n D e g u c h i ' s w o r k [ M VA , 2 0 0 2 ] , t h e s y s t e m m o d e l i s p r o p o s e d a s s u m i n g t h a t t h e b a l l m o t i o n i s f r e e f a l l a n d b o u n c i n g . H o w e v e r, i t i s unsuitable for volleyball, in which the ball direction changes due to external factors by players. The proposed system model combines two different system noises that cover the motion of the ball both in general situation and abrupt situation. Combination ratio of these two noises and n u m b e r o f p a r t i c l e s a r e a d a p t i v e t o t h e e s t i m a t e d m o t i o n p r o b a b i l i t y. Then, the ball candidate distribution is used to recover the tracking lost caused by the ball moving out of the frame. The anti-occlusion observation model solves the occlusion problem by eliminating influence of the camera with less confidence. In this dissertation (including all chapters), the experimental HDTV video sequences are the 2014 Inter High School Men's Vo l l e y b a l l G a m e s h e l d i n J a p a n , w h i c h a r e c a p t u r e d b y f o u r c a m e r a s located at each corner of the court. Here, the experimental results show that the success rate achieved by the proposals of 3D ball tracking is 99.42% and the average position precision error is 5.32(cm). In chapter 3, to acquire the event data based on the ball trajectory and game videos, the multiple event change feature based observation and the spatial. event. change. probability. distribution. based. prediction. are. proposed. As the conventional work of the ball event detection, Almaljai [ I C I P, 2 0 1 0 ] d e t e c t s t h e e v e n t c h a n g e b y u s i n g e n t i r e 3 D b a l l t r a j e c t o r y. H o w e v e r, n o i s e i n t h e t r a j e c t o r y i n f l u e n c e s t h e d e t e c t i o n p e r f o r m a n c e . Besides using the abruptness of trajectory in which noise exists, the proposed observation also uses the area of players' skin as likelihood to evaluate the event change. When players' hands are close to the ball, the ball is more likely to be hit. So the area of players' skin in the region 3.

(4) around the ball is calculated as the likelihood of event change. The probability distribution of event change is approximated according to the fact that the probabilities of event change in different spatial locations are different. For example, the probability becomes large in the spike and block areas. Thus event changes are accurately predicted. Experimental results show the detection rate of ball event achieves 92.43%. In chapter 4, to automatically acquire the strategy data showing the attack efficacy from the game videos, the relative motion abruptness and relative court zone division based player role detection is proposed. For the. strategy. data. analysis,. the. Data. Vo l l e y. system. is. widely. used.. H o w e v e r, i t s i n p u t i s p e r f o r m e d m a n u a l l y b y v i s u a l o b s e r v a t i o n , w h i c h i s inaccurate. for. human. mistake. and. requires. much. manually labeling.. Therefore, there is a demand for a technology that automatically and accurately acquires the strategy data. All the strategy data are defined based on player role, which cannot be obtained directly from the physical and event data because of consisting strategy information. In the proposed method for player role detection, the relative motion abruptness, which is calculated from the ball motion and player motion, represents the effort of e a c h p l a y e r t o t h e p l a y.. This feature is used to filter out the candidate. r o l e f o r e a c h p l a y e r. B y d i v i d i n g t h e c o u r t z o n e i n t o s e t t e r, a t t a c k e r a n d b l o c k e r z o n e s b a s e d o n t h e d i s t a n c e b e t w e e n t h e b a l l a n d t h e p l a y e r, a court zone based filter is modeled to further detect the player role. Based on the detected role of each player and their physical data and event data, the. strategy. data. indicating. the. attack. efficacy. variables. can. be. calculated. The detection rate of the set zone, the number of available attackers, the attack tempo and the number of blockers are 100%, 100%, 97.8%, and 100%. Compared with the detection rate of manual acquisition, it achieves 8.3% average improvement. In chapter 5, the overall dissertation is summarized and the future works. are. described.. To. realize. a. 3D. volleyball. analysis. system,. algorithms for the automatic acquisition of the physical data, event data and. strategy. data. are. proposed.. The. motion features. in. abrupt. and. general situation are combined in system model and are used as likelihood in observation. The spatial probability distribution indicates the ball and players' exiting area. They are used to achieve an accuracy of 92~99% for the game data acquisition, which greatly contributes to the academic. These results are highly evaluated in various practical applications, such as the game strategy system and the TV sports broadcasting system. 4.

(5)

博 士 論 文 概 要

博士論文概要