Processed Data for ConfLab: A Data Collection Concept, Dataset, and Benchmark for Machine Analysis of Free-Standing Social Interactions in the Wild

This folder contains processed data (video and wearable sensors) for 16 mins<br> of interaction that were annotated for the Conflab dataset.<br> <br> <strong>./cameras</strong><br> ./video_segments contains the overhead video recordings for 5 cameras<br> in MP4 files. These are split into 2 min segments, down-scaled to 960x540px,<br> and denoised. The resulting files are the ones that were used to annotate poses<br> and actions.<br> <br> For applications that require higher resolutions, the original video files in the<br> "data_raw" folder are at 1920x1080 resolution and a script is provided to extract<br> the same 2min segments.<br> <br> ./camera-calibration contains the camera instrinsic files obtained from:<br> https://github.com/idiap/multicamera-calibration. Camera extrinsic parameters<br> can be calculated using the existing intrinsic parameters and the instructions<br> in the multicamera-calibration repo. The coordinates in the image are provided<br> by the crosses marked on the floor, which are visible in the video recordings.<br> The crosses are 1m apart (=100cm). <br> <br> <strong>./wearables</strong><br> contains one single dataframe with aggregated sensor information<br> (accelerometer, gyroscope, magnetometer, rotation, proximity) per each person.<br> The data has been interpolated and imputed at 50Hz (fixed). The code used to obtain<br> these files from the raw data can be found at:<br> https://github.com/TUDelft-SPC-Lab/conflab/blob/master/preprocessing/midge/preprocess.ipynb<br> <br> This code can be used to preprocess larger segments of the werable data<br> from the "data_raw" folder.<br> <br> Audio files are not part of this folder, but are provided in the "data_raw" file.

本文件夹存储了为Conflab数据集标注的16分钟交互场景的处理后数据，涵盖视频与可穿戴传感器数据两类。

**./cameras**
./video_segments 目录内存储了5路顶置摄像头的MP4格式录像片段。这些录像被切分为2分钟的片段，分辨率下调至960×540像素，并已完成降噪处理，生成的文件用于姿态与动作标注。

若应用场景需要更高分辨率的视频，可使用"data_raw"文件夹内的原始视频文件（分辨率为1920×1080），同时提供了用于提取相同2分钟片段的脚本。

./camera-calibration 目录内包含相机内参文件，内参获取自 https://github.com/idiap/multicamera-calibration。可利用现有内参结合该多相机标定仓库中的说明计算相机外参。图像坐标由地面上的十字标记提供，该标记在录像中可见，相邻十字间距为1米（即100厘米）。

**./wearables**
./wearables 目录内包含一个聚合了每位受试者传感器信息的单一数据帧，传感器类型涵盖加速度计（accelerometer）、陀螺仪（gyroscope）、磁力计（magnetometer）、旋转矢量传感器（rotation）与接近传感器（proximity）。该数据已完成插值与补全，采样率固定为50Hz。从原始数据生成此类文件的代码可参见：https://github.com/TUDelft-SPC-Lab/conflab/blob/master/preprocessing/midge/preprocess.ipynb。

该代码可用于预处理"data_raw"文件夹内更多片段的可穿戴传感器数据。

音频文件未包含在本文件夹中，可在"data_raw"文件夹中获取。