SALSA

SALSA：协同社交场景分析数据集包含 60 分钟内涉及 18 名受试者的室内社交活动的不间断记录。它是行为分析和社会信号处理社区的丰富而广泛的存储库。除了原始的多模态数据外，SALSA 还包含整个事件持续时间的位置、姿势和 F 形注释，用于评估目的，以及有关参与者性格特征的信息。 场景和角色。 SALSA 被记录在一个常规的室内空间中，捕获的社交活动涉及 18 名参与者，由两个持续时间大致相等的部分组成。第一部分包括海报展示环节，研究生展示了四项研究。第五个人主持了海报会议。在下半场，所有参与者都可以在鸡尾酒会上自由地就食物和饮料进行互动。 传感器。数据由摄像头网络和目标佩戴的可穿戴徽章捕获。摄像头网络包括四个同步的静态 RGB 摄像头（1024×768 分辨率），以每秒 15 帧 (fps) 的速度运行。每个参与者在录音过程中都佩戴了社会计量徽章，这是一个 9×6×0.5 厘米的盒子，配备了四个传感器，即麦克风、红外 (IR) 光束和探测器、蓝牙探测器和加速度计。 注释。使用专用的多视图场景注释工具，每 45 帧（3 秒）对每个目标的位置、头部和身体方向进行注释。带注释的位置和头部/身体方向用于推断 F 形。在收集数据之前，所有参与者都填写了大五人格问卷。大五问卷的名字来源于它认为构成人格的五个特征：外向性；宜人性；认真;情绪稳定;创造力。SALSA: The Collaborative Social Scene Analysis Dataset consists of uninterrupted recordings of indoor social activities involving 18 subjects over a 60-minute period. It is a rich and extensive repository for the communities of behavior analysis and social signal processing. In addition to the raw multimodal data, SALSA also includes location, posture, and F-form annotations for the entire duration of the event for evaluation purposes, as well as information about participants' personality traits. Scenarios and Roles. SALSA was recorded in a conventional indoor space. The captured social activities involved 18 participants and consisted of two roughly equal-duration segments. The first segment included a poster session, where graduate students presented four studies. A fifth individual moderated the poster session. In the second half, all participants were free to interact over food and drinks at a cocktail party. Sensors. The data was captured by a camera network and target-worn wearable badges. The camera network includes four synchronized static RGB cameras (1024×768 resolution) operating at 15 frames per second (fps). Each participant wore a sociometric badge during the recording session: a 9×6×0.5 cm box equipped with four sensors: a microphone, infrared (IR) beam and detector, Bluetooth detector, and accelerometer. Annotations. Using a dedicated multi-view scene annotation tool, the location, head, and body orientation of each target were annotated every 45 frames (3 seconds). The annotated location and head/body orientation are used to infer F-form. Prior to data collection, all participants completed the Big Five Personality Questionnaire. The Big Five Questionnaire derives its name from the five traits it identifies as constituting personality: extraversion; agreeableness; conscientiousness; emotional stability; and creativity.

搜集汇总

背景与挑战

背景概述

SALSA 是一个多模态数据集，记录了18名参与者在室内社交场景中的60分钟活动，用于行为分析和社会信号处理研究。它通过摄像头网络和可穿戴设备采集数据，并提供位置、姿势、F形注释及参与者人格特征信息。