Predictive Human Trajectory Classification in Industrial Settings With Transfer Learning and Data Augmentation

Read full text Summarise with VIRAT
Authors: Even Falkenberg Langås, Atle Aalerud, Svein Olav Nyberg, Muhammad Hamza Zafar, Daniel Hagen, Filippo Sanfilippo
Cited by: 0
References: 24

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The successful realisation of Industry 5.0 relies on seamless collaboration between humans and robots in shared workspaces. This paper investigates predictive human trajectory classification (HTC) in a factory setting to enhance safety, productivity, and adaptability in human-machine environments. Using a state-of-the-art light detection and ranging (LiDAR) system with reflective mirrors to extend the field of view, motion data are collected across a 26 m × 20 m indoor industrial floor. Human trajectories are classified to predict workstation transitions using a stacked long-short term memory (LSTM) network. To address the challenges of limited real-world data, transfer learning is applied by first training the model on synthetically generated trajectories. The paper compares a traditional stochastic method, specifically the Ornstein-Uhlenbeck (OU) process, with a modern learning based method, represented by a timeseries generative adversarial network (TimeGAN). Further on, the algorithm is fine-tuned on the real-world dataset. The comparison shows that in data-scarce scenarios, OU outperforms TimeGAN in most metrics. The LSTM model trained on OU-based data and real-world data shows an overall accuracy gain of 4.8 percentage points compared to training on real-world data only. It is able to classify the goal workstation three seconds in advance with an accuracy of 95.6%, demonstrating the potential for predictive models to improve workplace operations by reducing stall time, optimising workstation readiness, and reducing the need for traditional safety cages.

Fields of Study & Subjects

Robot Manipulation and Learning Autonomous Vehicle Technology and Safety Data Management and Algorithms

Publication details

Journal / source
IEEE Access
Volume
14
Pages
47071–47084
Publication date
1 January 2026
Language
EN