Handball action dataset - UNIRI-HBD

The dataset consists of 751 videos, each containing the performance one of the handball actions out of 7 categories (passing, shooting, jump-shot, dribbling, running, crossing, defence). The videos were manually extracted from longer videos recorded in handball practice sessions. The scenes were shot with stationary GoPro cameras that were mounted on the left or right side of the playground, from different angles. The videos were recorded in at least full HD (1920 × 1080) resolution at 30 or more frames per second. On average 12 players appear in each scene, but mostly one or two players perform the action of interest. The performances of the actions of interest are extracted in separate video clips showing only that action in isolation. The isolated action clips are obtained first by automatic detection of players using Mask RCNN detector and tracking with Deep SORT, then by manual selection of appropriate tracked player.

该数据集包含751个视频，每个视频均展示了7种手球动作类别（传球、射门、跳投、运球、奔跑、横传、防守）之一。视频系人工从更长时段的手球训练录像中提取而来。拍摄场景采用固定式GoPro摄像机，摄像机安装于训练场地的左侧或右侧，并从不同角度进行拍摄。视频的录制分辨率至少为全高清（1920 × 1080），帧率至少为每秒30帧。平均每个场景中约有12名球员出现，但通常只有一名或两名球员执行感兴趣的动作。感兴趣动作的表现被提取为独立的视频片段，这些片段仅展示该动作的孤立表现。首先，通过使用Mask RCNN检测器自动检测球员并利用Deep SORT进行跟踪，然后通过人工选择适当的追踪球员来获得这些孤立动作片段。