Drive&Act

Drive & Act dataset是最先进的多模式基准，用于在车辆行驶时识别驾驶员的动作。数据集包括3D骨架，以及由帧的6个不同视点和3个模态 (RGB) 帧级分层标记捕获的960万图像。
该数据集具有以下特征
包括12小时的视频数据，共29个长序列；

校准的5个视图的多视图摄像机系统；
多模态视频: RGB、IR和深度；

无标记动作捕捉: 3D身体姿势和头部姿势
83人类注释的分层活动注释。-第1级: 长时间运行的任务 (12)。-第2级: 语义行为 (34) -第3级: 对象交互三合会 (动作，对象，地点) (6 | 17 | 14)。"

The Drive & Act dataset is a state-of-the-art multimodal benchmark for driver action recognition during vehicle travel. It includes 3D skeletons and 9.6 million images captured from 6 distinct viewpoints per frame across three modalities, with frame-level hierarchical annotations.

This dataset features the following characteristics:
- 12 hours of video data, totaling 29 long sequences;
- A calibrated multi-view camera system with 5 viewpoints;
- Multimodal videos: RGB, IR, and depth;
- Markerless motion capture for 3D body poses and head poses;
- 83 human-annotated hierarchical activity annotations, divided into three levels:
- Level 1: Long-running tasks (12 categories);
- Level 2: Semantic behaviors (34 categories);
- Level 3: Object interaction triads (action, object, location) (6 | 17 | 14).

搜集汇总

背景与挑战

背景概述

Drive&Act是一个多模态驾驶员动作识别数据集，包含12小时视频、29个序列，通过5个校准视图采集RGB、IR和深度数据，并提供3D骨架和分层活动注释，适用于细粒度行为分析。