基于比赛环境特征的多目标运动员跟踪方法

Multi-object player tracking method based on game environment feature

鉴于运动员跟踪与移动的复杂程度和环境等因素高度相关,提出一种基于当前运动轨迹和赛场环境特征的跟踪检测方法。从噪声检测中提取赛场环境特征集,对选手空间分布情况等当前赛场环境特征进行描述;基于当前轨迹和比赛环境特征训练随机决策森林,建立运动员各时间点的亲密度模型,该环境条件性运动模型包含目标关联性,可保持可跟踪性。对分别包括10名和22名选手的篮球和足球比赛进行实验,各选取持续几分钟的序列作为样本,实验结果表明,该模型能够显著提高运动员跟踪性能。

Concerning that player tracking is highly correlated with the complexity of movements and game environment,a tracking model based on the game environment feature was proposed.A set of game environment features was extracted from noisy detection to describe the current state of the match,such that how the players are spatially distributed.A random decision forest was trained based on current trajectory and game environment features to build an appropriate simplified affinity model for each player and time instant.The environment-conditioned motion models implicitly incorporate complex inter-object correlations while remaining tractable.Significant performance improvements are showed over existing multi-target tracking algorithms on basketball and football sequences in which contain 10 and 22 players respectively in the duration of several minutes.