Shihua Tian and Bin Feng. Research on Basketball Training Attitude Motion Capture Based on Mathematical Similarity Matching Statistical Analysis. Dynamic Systems and Applications 29 (2020) No. 3, 812-822
https://doi.org/10.46719/dsa202029333
ABSTRACT.
Basketball is a world-famous ball sport and a core event of the Olympic Games. It is a hand-centered body-oriented sports. In the field of basketball training, the existing training plan is mainly based on the manual observation and personal experience of the coach, which is inevitably subjective. Nowadays, video image processing technology and image statistical analysis technology are applied to the daily training of athletes. Judging the athlete’s sports posture can improve the discriminating level of the coaches. At the same time, it also aims at the athletes’ irregular movements and obvious sports characteristics. Obviously calibrated. The human movements embodied in basketball are more complicated, and the movement is very large. The changes and physical confrontation are very frequent. The analysis and recognition of basketball training posture movement has a pioneering effect on the practice of basketball competition and training, as well as the help of coaches and athletes. In this paper, the mathematical similarity matching statistical analysis is applied to the gesture recognition in basketball. The main researches are as follows: (1) By extracting the feature points in the line target, a matching model based on spatial position and topological relationship is established. Similarity matching is performed on feature points. (2) Based on the non-training infrared image target recognition method, this paper deeply studies the feature extraction method of adaptive regression kernel and applies it to the extraction of local structural features of images; (3) Based on the above research, the recognition method of basketball posture is proposed. This paper analyzes the behaviors of the legs and arms in slow walking, running jumps, dribble shooting, passing and catching basketball, as well as the corresponding signal waveform characteristics. A two-stage data partitioning method for basketball is proposed. The unit motion data is analyzed to realize feature extraction. In order to select the classifier that is most suitable for basketball gesture recognition, the constructed feature vector is trained by four different classifiers to construct different classifiers to realize the action division. Experiments show that the experimental method designed for the common gestures in basketball can obtain the basketball action information of the testee in real time, realize the accurate extraction of single action data, and complete the recognition of basketball posture. The average accuracy is up to 98.85% has certain practical value in basketball gesture recognition.
Keywords: mathematical similarity, statistical analysis, basketball training posture