Core Concepts
OpenPack is a large-scale multimodal dataset for recognizing complex packaging operations in logistics environments, containing sensor data, IoT device readings, and rich metadata to enable research on advanced activity recognition methods.
Abstract
The OpenPack dataset is the largest publicly available dataset for recognizing complex packaging work activities in industrial logistics environments. It contains 53.8 hours of multimodal sensor data, including acceleration, gyroscope, depth images, LiDAR point clouds, and readings from IoT devices like barcode scanners, collected from 16 subjects with varying levels of packaging experience.
The dataset provides 10 classes of packaging operations and 32 action classes, with rich metadata on the subjects and order details. This enables research on advanced activity recognition methods that can leverage contextual information beyond just sensor data.
The benchmark results show that existing state-of-the-art models struggle to achieve high accuracy, especially in challenging scenarios with variations in working speed, item characteristics, and occlusions. This highlights the need for developing speed-invariant, metadata-aided, and multi-modal fusion techniques to enable robust recognition of complex industrial work activities.
OpenPack presents opportunities for various research directions, including transfer learning, skill assessment, mistake detection, and fatigue estimation, in addition to the core task of activity recognition. The dataset is expected to contribute significantly to advancing sensor-based work activity recognition in industrial domains.
Stats
The total duration of the dataset is 53.8 hours, consisting of 104 data collection sessions.
The dataset contains 20,161 instances of packaging operations and 53,286 instances of actions.
The number of subjects is 16, with varying levels of packaging work experience.
Quotes
"OpenPack is the largest multimodal work activity dataset in the industrial domain, including sensory data from body-worn inertial measurement units (IMUs), depth images, and LiDAR point clouds."
"OpenPack is the first large-scale dataset for complex packaging work recognition that contains readings from IoT-enabled devices."