Bodily Emotion Recognition Dataset

Abstract: The Bodily Emotion Recognition Dataset (BERD) was developed as part of the research study titled "Human-Computer Interaction Methodologies for Bodily Emotion Recognition: Optimizing Data Collection Strategies" This dataset comprises motion capture data collected from participants performing emotional body movements under various experimental conditions. It is designed to facilitate the development and evaluation of automatic emotion recognition (AER) systems using bodily movement data. The dataset offers insights into the effects of participant acting expertise, motion capture device types, and emotional stimuli on bodily emotion recognition accuracy. Key Features: 1. (Expertise) Participant Data: Actors: Professional actors with at least three years of acting experience. Non-actors: General participants with no formal acting training. Test: General participants with no formal acting training for test 2. (Devices) Motion Capture Devices: Marker-based motion capture (Optitrack system with 18 infrared cameras). Pose estimation using RGB videos. Kinect motion capture. Mobile phone motion capture (iPhone 12 with ARKit). 3. (Stimulus) Emotional Stimulus Type: Word instructions (e.g., "happy," "sad"). Picture stimuli (Karolinska Directed Emotional Faces dataset). Video stimuli (validated emotional film clips). 4. Emotions Recorded: Seven categories: happy, sad, surprised, angry, disgusted, fearful, and neutral. 5. Data Format: Skeletal data represented as 3D joint coordinates. Sampling rate: 30 frames per second. File format: CSV. Potential Applications: Developing deep learning models for bodily emotion recognition. Studying the impact of data collection conditions on emotion recognition accuracy. Citation: If you use this dataset in your research, please cite it as follows: Cho, Y., Jung, M., Bae, J., & Kim, K. (2024). Behavioral Emotion Recognition Dataset (BERD). Zenodo. DOI: [https://zenodo.org/records/14576033]