Towards Multimodal Prediction of Time-continuous Emotion using Pose Feature Engineering and a Transformer Encoder.