A Sports Video Behavior Recognition Using Local Spatiotemporal Patterns

Abstract

These days, many researchers are interested in sports video behavior recognition (SVBR), which has emerged as a core research domain for visual understanding and analysis of athlete performance. Despite significant results in simple scenarios, SVBR remains a difficult task due to numerous challenges such as independence, obstruction, and interclass visual appearance in real situations. Recognizing the sports video behavior is one of the crucial challenges in computer vision for sports video behavior data, with implementation in monitoring, mental illness diagnostics, video information extraction, and multitarget tracking. This paper provides a local pattern activity discrimination model for the detection and localization of active individuals in a video to address issues of multitarget tracking in video behavior recognition caused by mutual occlusion of targets and the complexity of ambient background. This work solves the problem of low recognition accuracy due to incomplete extraction of trajectories or overly complex backgrounds using a trajectory-based approach. Our model is applied to the video segments of 38 matches associated with goal events in Euro 2012, and an average accuracy of 91.3% is obtained. The experimental results verify the high accuracy and applicability of our method for the recognition of target object behavior in the videos.