Abstract
OpenPose, which is developed by Carnegie Mellon University (CMU) presented in CVPR 2017, takes in real-time motion images via a simple web camera and is capable of recognizing skeletons of multiple persons in these images. It also generates recognized skeleton point coordinates to files. OpenPose is featured by CMU’s original top-down method for real-time recognition and it is open online especially for research purposes. Thus we aimed to build a posture analysis model using OpenPose skeletal recognition data and verifying the practicality of OpenPose by verifying the accuracy of the model. As a posture analysis model, we adopted a logistic regression model that predicts the shooting probability of the basketball free throw with skeleton posture data as explanatory variables and the fact whether the ball enters the basket or not as a binary target variable. As the result, sufficiently significant prediction accuracy was obtained. Therefore, posture analysis using OpenPose has been verified to be practical with our model. We consider that with many skeleton data which are easily provided by a simple web camera, OpenPose makes statistical diagnostic approach possible. We also consider it could lower costs (in both financial and time-wise) of such an analysis which has previously required more equipments and more time for preparation regarding motion capture analysis systems.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
OpenPose realizes three-dimensional acquisition by stereo (compound eye) camera in March 2017, but in this research, OpenPose of 2D position recognition version using monocular Web camera is used because of easy operation and sufficient use frequency.
- 2.
Our experimental machine is CPU: AMD Ryzen 7 1800X, MEMORY: 16 GB, GPU: NVidia GeForce GTX 1080ti, OS: Ubuntu 14.04 LTS, CUDA version: 8.0, cuDNN version: 5.1 for CUDA8.0.
- 3.
Since many same records are generated according to this table, we added a small perturbation of \(\mathcal {N}(0,0.01)\) to level value to avoid rank deficient by same records. For example 7.0026 at level 7.
References
CMU-Perceptual-Computing-Lab. https://github.com/CMU-Perceptual-Computing-Lab/openpose
Cao, Z., Simon, T., Wei, S., Sheikh, Y.: Realtime multi-person 2D pose estimation using part affinity fields. In: CVPR 2017 (2017)
MSCOCO keypoint evaluation metric. http://mscoco.org/dataset/#keypoints-eval
Hulka, K., Cuberek, R., Svoboda, Z.: Time-motion analysis of basketball players: a reliability assessment of Video Manual Motion Tracker 1.0 software. J. Sports Sci. 32(1), 53–59 (2013)
Wang, X.: An optimization method of basketball teaching and training system design based on motion capture technology (2017)
Dutta, T.: Evaluation of the Kinect sensor for 3-D kinematic measurement in the workplace. Appl. Ergon. 43(4), 645–649 (2012)
Shanahan, J.K.: A model for predicting the probability of a win in basketball. University of Iowa (1984)
Inamura, T., Nakamura, Y., Toshima, I.: Embodied symbol emergence based on mimesis theory. Int. J. Robot. Res. 23(4), 363–377 (2004)
Bishop, C.M.: Pattern Recognition and Machine Learning. Springer, New York (2006). 4.3.2 Logistic regression
Platt, J.C.: Fast training of support vector machines using sequential minimal optimization (1999)
Chen, T., Guestrin, C.: XGBoost: a scalable tree boosting system. arXiv:2016.02754 (2016)
Kudo, Y., Ogaki, K., Matusi, Y., Odagiri, Y.: Unsupervised adversarial learning of 3D human pose from 2D joint locations. arXiv:1803.08244 (2018)
ESPN Western Conference Player Free-Throw Shooting Statistics - 2018–19. http://www.espn.com/nba/statistics/player/_/stat/free-throws/sort/freeThrowPct/league/west/qualified/false/order/false
Acknowledgment
We would like to thank Basketball Club Team of Tokyo Metropolitan College of Technology and the Exchange Students Basketball Community at Tokyo International Exchange Center by JASSO (Japan Student Services Organization) for their cooperation in our taking movies of their basketball free throw. We would like to special thank to Dr. Atushi Shibata of AIIT for provision of experimental computation environment.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
A Appendix: Discussion on the AR Value
A Appendix: Discussion on the AR Value
In order to consider obtained our AR value in this experiment, we compared NBA (National Basketball Association) Free Throw data [13]. Figure 13 shows the histogram of free throw success rate of NBA’s 238 players who threw more than 5 times last year. As shown in Table 2, we generated simulation data according to the number of NBA histogram. These data are consisted of the level and the binary flag in each record. The level is set according to the success rate, but the success rate less than 0.5 was compiled to level 4 because of very few people. The binary flag is set randomly according to the success rate. But we generated 2380 records by multiplying the number by 10 to avoid bias of the random.
We made logistic regression using the binary flag as a target variable and the levelFootnote 3 as a explanatory variable to obtain the AR value. As the result we obtain AR = 35%. In this simulation, even if the level as a explanatory variable has a strong correlation with the success rate explicitly, the AR was only 35%. We thought that the low AR value is due to the relatively small number of people at high and low levels. Because we obtained AR = 60% in the case of same number at each level in our simulation. Assuming expert or beginner players were somewhat few in our experiment, our experiment AR = 41% can be considered as sufficiently significant accuracy.
Histogram of NBA Free Throw success rate
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this paper
Cite this paper
Nakai, M., Tsunoda, Y., Hayashi, H., Murakoshi, H. (2019). Prediction of Basketball Free Throw Shooting by OpenPose. In: Kojima, K., Sakamoto, M., Mineshima, K., Satoh, K. (eds) New Frontiers in Artificial Intelligence. JSAI-isAI 2018. Lecture Notes in Computer Science(), vol 11717. Springer, Cham. https://doi.org/10.1007/978-3-030-31605-1_31
Download citation
DOI: https://doi.org/10.1007/978-3-030-31605-1_31
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-31604-4
Online ISBN: 978-3-030-31605-1
eBook Packages: Computer ScienceComputer Science (R0)