There are two main reasons for researching group activity recognition.
In many sports, good results are gained by using tactics with data. Taking baseball as an example, in the major league in 2017 the total number of home runs increased by 6105. It is a surprising result because the past home run record is 5693 in 2000. Why did such a record come out? The reason is in data analysis. In the Major League it is now possible to measure the hitting speed and hitting angle by applying a measuring instrument that measures the trajectory of a missile called trackman to baseball. Statistically analyzing the data showed that if the hitting speed is over 158 kilometers and the angle is around 30 degrees, it will be hit with a probability of 80%, and most of it will be a home run (It is called Barrel Zone ). Based on that data, this record was born as major leagers trained in accordance with that "Barrel zone". In this way, data analysis has greatly changed the tactics of sports, and there are cases where good results are obtained.
By applying such data analysis to team sports, it is a goal to find out what kind of tactics to take from the huge game data leads to scoring and lead to final victory (Big Tactical analysis of team sports by data analysis). As a result, sports technology is expected to develop further.
Currently, various technologies are adopted in many sports broadcasts. As a remarkable example, "Eye vision" of Mr. Kanade at Carnegie Mellon University is famous Super Bowl in 2001. This is an image effect technology that synchronizes and controls the pan, tilt, and zoom of 30 TV cameras installed so as to surround the ground on the stadium, so you can see the players from various angles of 360 degrees It came to be like. In this way, various methods are adopted for sports broadcasting so that viewers enjoy it.
My goal is to make sports broadcasting better by superimposing the tactics information on the video in sports where tactics and rules are complex. Taking an example of volleyball as an example, suppose that a spike was decided at a certain position. By analyzing the game data so far, by superimposing and displaying the success rate of the position where the spike was decided this time on the video, it is understood that the viewer understands whether the play was good by numerical value. This allows you to understand the goodness of play without specialized knowledge, so viewers are expected to be more interested in that sport.
In order to achieve these two goals by tactical analysis of big data analysis, we are studying group activity recognition which is the fundamental technology.