Article Preview
Top1. Introduction
With the development of Internet technology and the popularity of video acquisition equipment, video has become the main carrier of information. At present, the amount of video data is growing explosively, hence, how to analyze and understand the content of video becomes more and more important. As one of the important tasks of video understanding, human action recognition has become the research hotpot of computer vision. Action recognition is to learn the appearance and motion information contained in the video by modeling the spatial-temporal information of the pre segmented time-domain sequence, so as to establish the mapping relationship between the video content and the action category, so that the computer can effectively be competent for the task of video understanding. Action recognition has broad application prospects, such as action analysis, intelligent monitoring, human-computer interaction, video information retrieval and so on. However, most of the current action recognition methods are designed for human daily activities, and the action recognition for basketball is still a challenging problem. Therefore, it is of great practical significance to introduce video based action recognition methods into the field of basketball.
Basketball originated in the United States in 1891 and officially became an Olympic sport in 1936. It is a competition between the players of the two teams on the playing field, and the final result is the score obtained according to the basket. Basketball is a sport that takes into account individual ability and team cooperation. Individual technical level and team tactical level are very important for the game, which requires coaches to formulate detailed training plans in the process of player training. Up to now, in the field of sports, coaches mainly rely on observing athletes' performance on the spot to develop appropriate training plans for athletes, which has high requirements for coaches' professional quality. Unfortunately, with the increasing enthusiasm for sports in recent years, the number of professional coaches has become difficult to meet the needs of sports enthusiasts. Therefore, it is more and more urgent to study the artificial intelligence algorithms that can replace coaches’ work.
The basic actions of basketball include dribbling, shooting, etc. Dribbling is the most basic action in basketball, and shooting is a necessary skill for scoring, and the accuracy of basic actions has a great impact on the score of the game. The result of professional basketball players' shooting score is related to the angle and strength of shooting action, so the practice of shooting action can improve the technical level of players. At present, the training of players in basketball is mainly aimed at basic actions, and the traditional training manner is that the coach observes the shooting action of the players, judges the standardization of the action according to his own experience, and then guides the players. However, because this manner relies on the coach's intuitive sense of judgment and lacks corresponding evaluation, it cannot give players a judgment standard. And during the training, players will analyze the hand feeling of their own shooting action, resulting in some judgment errors, which are inconsistent with the requirements of the standard action. The long-term training of non-standard actions will not only have a certain impact on the shooting results, but also have a certain sports injury to the players themselves (Zhu 2017). Therefore, the research on the assessment algorithm of action specification based on video can help players find the gap with the standard action, and improve the training according to the shortcomings of their own actions, so as to improve the intuitiveness of training and the rapidity of feedback. At the same time, the standardization of training actions can also protect the sports health of basketball players.