| There are two important subjects in competitive sports training, namely the athlete (student athlete)'s individual mechanics training and the team's tactics training.;For the student athlete's individual mechanics training, many researchers seek to capture and visualize the student's motions in 3D virtual environments. Despite most virtual training systems are able to visualize students' motions from various angles, those systems leave the students themselves to figure out how to revise their motions to improve performance. We present a training system that is able to compare the coach's and the student's motions and quantify their distances. In addition, based on the motion comparison, we are able to automatically generate training advice to tell students where and how to improve.;Besides individual mechanics training, tactics training is an important training aspect for team sports. Tactics training studies the current state of the game from existing video footage, to determine a good tactic move such as where to attack or which player to defend. Among all types of video footage, broadcast matching replays are more widely studied because they are easily accessible and unlikely to have some tactical reservations. Because broadcast sports videos consist of both tactic relevant shots and irrelevant shots, many efforts have been made to automatically segment videos to separate these types of shots. Some methods use domain knowledge of the target sports activity, and are not able to be applied to other sports activities. Other systems use supervised learning approach to improve frame classification and shot boundary detection accuracy, but without focus on maintaining the integrity of the tactic relevant segments and the video structure. We introduce a novel method, named Segmentation based on distance dependent Chinese Restaurant Process (S-CRP), to segment broadcast sports videos into high quality semantic shots without the use of domain knowledge and more sophisticated classifiers. In addition, we also introduced a new performance metric, namely Levenshtein distance Ratio, to provide a more accurate measure of how well the segmentation result maintains the structure of the original video.;After sports videos segmentation, researchers have made progresses in further tactic analysis by automatically detecting and tracking players' positions or ball movements in order to capture interesting player behaviors or important events in the game. However, current tactic analysis systems provide only low level assistances in tactics training. They are able to capture certain events in the game and help the team with statistical analysis, but they provide little help towards finding a better tactic. We introduce a novel tactics training system that is able to consider players' attributes together with their positions to accurately estimate each player's offense threat or defense ability. Our system formulates an optimization problem to find the optimal defense tactic that minimizes the offense team's threat. |
