Real-time environmental analysis for industrial vehicles based on synthetic sensor data and deep learning

Abstract

Industrial vehicles often operate in the same environment as employees do. Monitoring their environment in real-time is essential to prevent accidents and injuries to employees. Machine learning methods are well-suited for people detection, but they require large amounts of training data. This article describes a real-time environmental analysis system based on 3D time-of-flight sensors and Deep Learning methods. Automatically labeled synthetic data as well as publically available real-world data were used for training. An in-situ evaluation was carried out on a real forklift truck in a warehouse. The results show that inexpensive and quickly generated synthetic data increases the robustness of environmental analysis.