2.2 6).,.,.,. Yang, 7).,,.,,. 2.3 SIFT SIFT (Scale-Invariant Feature Transform) 8).,. SIFT,,. SIFT, Mean-Shift 9)., SIFT,., SIFT,. 3.,.,,,,,.,,,., 1,

Size: px

Start display at page:

Download "2.2 6).,.,.,. Yang, 7).,,.,,. 2.3 SIFT SIFT (Scale-Invariant Feature Transform) 8).,. SIFT,,. SIFT, Mean-Shift 9)., SIFT,., SIFT,. 3.,.,,,,,.,,,., 1,"

ちかこうばら
5 years ago
Views:

1 1 1 2,,.,.,,, SIFT.,,. Pitching Motion Analysis Using Image Processing Shinya Kasahara, 1 Issei Fujishiro 1 and Yoshio Ohno 2 At present, analysis of pitching motion from baseball videos is timeconsuming since it is often performed manually by experts with professional knowledge and expertise. In this article, we propose a method to detect the habit of a pitcher effectively from a strategical perspective. We focus on the direction of a glove that is most likely to be detected as a habit. We track the glove using the particle filter and analyze the gradient directions of the glove using the SIFT. We applied the proposed method to a certain pitcher who is known for his habit and proved the effectiveness of our method. 1 Keio University 2 Tohoku University of Community Service and Science 1.,.,,, 1) 5).,,.,.,,,,.,.,.,, SIFT,.,, 3. 4, 5., 6, 7,. 2.,,, SIFT. 2.1,.,. 1),,, 2).,,,.,,. 1 c 2011 Information Processing Society of Japan

2.2 6).,.,.,. Yang, 7).,,.,,. 2.3 SIFT SIFT (Scale-Invariant Feature Transform) 8).,. SIFT,,. SIFT, Mean-Shift 9)., SIFT,., SIFT,. 3.,.,,,,,.,,,., 1, 2. 1,, 2,., 1. 1.. Fig.

The pitcher has a habit to be close contact with the glove when throwing a fork ball.,. SIFT,. SIFT,. 4.1 SIFT,. DoG (Difference-of-Gaussian),.

2 2.2 6).,.,.,. Yang, 7).,,.,,. 2.3 SIFT SIFT (Scale-Invariant Feature Transform) 8).,. SIFT,,. SIFT, Mean-Shift 9)., SIFT,., SIFT,. 3.,.,,,,,.,,,., 1, 2. 1,, 2,., Fig. 1 Straight (left) and breaking ball (right). The pitcher has a habit in terms of the directions of the glove and arm Fig. 2 Straight (left) and fork ball (right). The pitcher has a habit to be close contact with the glove when throwing a fork ball.,. SIFT,. SIFT,. 4.1 SIFT,. DoG (Difference-of-Gaussian),. DoG D(x, y, σ), D(x, y, σ) = L(x, y, kσ) L(x, y, σ) L(x, y, σ) = G(x, y, σ) I(x, y) ( ) G(x, y, σ) = 1 2πσ exp x2 + y 2 2 2σ 2. L(x, y, σ), I(x, y), G(x, y, σ), k,. DoG,. DoG, 26..,, DoG. 2 c 2011 Information Processing Society of Japan

3 H = [ D xx D xy D xy D yy ], 1 α, 2 β, α β β α,., DoG,. x = (x, y, σ) T DoG D(x), D(x) = D + DT x x xt 2 D x 2 x. x 0, D x + 2 D x ˆx = D D ˆx = x2 x., ˆx, DoG. DoG D(ˆx),. 4.2,. L(u, v),,., m(u, v) θ(u, v),. m(u, v) = (L(u + 1, v) L(u 1, v)) 2 + (L(u, v + 1) L(u, v 1)) ( ) 2 L(u, v + 1) L(u, v 1) θ(u, v) = tan 1 L(u + 1, v) L(u 1, v),. 36,,.,,. 80%,., 80%, ,. 3. SIFT, , ,. 5.,, SIFT,. 5.1 RGB.,,., b(t), f(t), α, b(t) = αf(t) + (1 α)b(t 1). 4.,.,,, 10)., x t, 2 u t, v t, x t = f(x t 1, u t ) + v t., 2,. 3 c 2011 Information Processing Society of Japan

3 The gradient direction histograms for a straight (left) and a breaking ball (right).

4 4.,,. Fig. 4 Background image created by the weighted mean (left) and a result of background differenced image (right). The background differencing with the weighted mean made the background removed while leaving the background , Fig. 3 The gradient direction histograms for a straight (left) and a breaking ball (right). The gradient directions are concentrated around 60 degrees and 240 degrees when he throws a straight, while they are concentrated around 30 degrees and 210 degrees when throwing a breaking ball. 5.2 SIFT SIFT. SIFT,,, ), SIFT., 2 I, M, H, H = 35 i=0 min(i i, M i )., while undersimulation do for all particle do calculatelikelihood end for dosift calculateopticalflow for all particle do doresampling dopredicting end for end while calculatesimilarity 4 c 2011 Information Processing Society of Japan

5 5,,.,. Fig. 5 Histograms between the same stuff (left) and the different stuff (center, right). The same stuff has a high degree of similarity whereas the different stuff a low degree of similarity., SIFT.,., SIFT,.,. 6. CPU Intel Core2 3.00GHz, C++, API OpenCV.,, 2, 2. 5.,, A B, A B, 4 A A, A B, B A, B B.. 7. SIFT,.,,.,,.,,.,,. 5 c 2011 Information Processing Society of Japan

6 1) :, D-II, Vol. J88-D-II, No. 8, pp , ) :, No. 3, pp , ) Hsuan-sheng Chen et al.: Pitch-By-Pitch Extraction from Single View Baseball Video Sequences, in Proc. ICME 2007, pp , ) Hua-Tsung Chen et al.: A Trajectory-Based Ball Tracking Framework with Visual Enrichment for Broadcast Baseball Videos, Journal of Information Science and Engineering, Vol. 24 No. 1, pp , ) Wei-Ta Chu and Ja-Ling Wu: Development of Realistic Applications Based on Explicit Event Detection in Broadcasting Baseball Videos, in Proc. 12th International Conference on Multi Media Modeling (MMM 2006), pp , ) Carlo Tomasi and Takeo Kanade: Detection and Tracking of Point Features, Technical Report, CMUCS , ) Changjiang Yang et al.: Fast Multiple Object Tracking via a Hierarchical Particle Filter, in Proc. ICCV 2005, pp , ) David G. Lowe: Distinctive image features from scale-invariant keypoints, International Journal of Computer Vision, Vol. 60, No. 2, pp , ) : SIFT Mean-Shift, CVIM 157, pp , ) :,, Vol. 23, pp , ) Michael J. Swain and Dana H. Ballard: Color Indexing, International Journal of Computer Vision, Vol. 7, No. 1, pp , c 2011 Information Processing Society of Japan

1 Fig. 1 Extraction of motion,.,,, 4,,, 3., 1, 2. 2.,. CHLAC,. 2.1,. (256 ).,., CHLAC. CHLAC, HLAC. 2.3 (HLAC ) r,.,. HLAC. N. 2 HLAC Fig. 2

1 Fig. 1 Extraction of motion,.,,, 4,,, 3., 1, 2. 2.,. CHLAC,. 2.1,. (256 ).,., CHLAC. CHLAC, HLAC. 2.3 (HLAC ) r,.,. HLAC. N. 2 HLAC Fig. 2 CHLAC 1 2 3 3,. (CHLAC), 1).,.,, CHLAC,.,. Suspicious Behavior Detection based on CHLAC Method Hideaki Imanishi, 1 Toyohiro Hayashi, 2 Shuichi Enokida 3 and Toshiaki Ejima 3 We have proposed a method for