Precise Human Activity Recognition for The OpenPack Challenge 2022
Shubham M Wagh · 2023
This technical report provides an overview of our method for the OpenPack Challenge 2022, which is part of the Behavior Analysis and Recognition for Knowledge Discovery Workshop at PerCom 2023. The challenge involves proposing an activity recognition approach by predicting activity classes for each 1-second-long time slot using sensor data from packaging works in the logistics industry. To accomplish this, we employed various data augmentation techniques and trained a deep neural network that incorporates spatio-temporal features from multiple sensor time-series data and a self-attention mechanism for selecting and learning essential time points. Our model achieved an average macro F1-score of 91.12% and generalised well on the submission set. This experiment is a part of the challenge conducted by our team “Shubham Wagh” and secured 5thplace in the competition.