Table 2 Comparison between the proposed HA4M dataset and existing vision-based datasets on assembly actions.

From: The HA4M dataset: Multi-Modal Monitoring of an assembly task for Human Action recognition in Manufacturing

Dataset

Visual Sensors

Environment

Data Modalities

Task

Assembly10141

Eight RGB Cameras mounted on a scaffold around a table and four monochrome cameras mounted on an headset

Laboratory

RGB frames, 3D hand poses

Assembly and Disassembly of toy vehicles

Meccano42

One Intel RealSense SR300 camera mounted on an headset

Laboratory

RGB videos

Assembly of a toy motorbike

IKEA-ASM43

Three Microsoft Kinect v2

Offices, Labs and Family Homes

RGB videos, Depth videos, 3D Skeleton Joints

Furniture Assembly

HA4M

Microsoft Azure Kinect

Laboratory

RGB frames, Depth maps, IR frames, RGB-Depth-Aligned frames, Point Clouds, Skeleton Data

Assembly of an EGT

  1. For each dataset, information about the cameras used for data acquisition, the type of environment where acquisitions were made, the type of provided data and the assembly task are given.
Back to article page