An Edge AI Study for Human Action Recognition on Industry 4.0 Using Weightless Neural Networks

The environment degradation and global warming are some of the greatest global challenges currently faced. Information and Communication Technologies (ICTs) perform a vital role in solving environmental problems caused by nature's degradation. To this end, EdgeAI is emerging as a promising solution. Edge devices endowed with intelligence capacities can perform several tasks with lower latency in comparison with the cloud. A target domain for Edge Intelligence is the Industrial loT. Edge devices can perform real-time human action recognition on the factory floor for quality, health, safety and environmental (QHSE) compliance. An important application of EdgeAI for QHSE compliance is monitoring and analyzing human actions in manufacturing and other industrial environments. In this context, we present our exploratory and experimental work, where the main premise is the investigation of the application of Weightless Neural Networks in Video Applications in the context of Industry 4.0 with Edge devices. The work aims to explore the potential of implementing Weightless Neural Networks in edge devices (Edge) without the use of accelerators such as VPUs, GPUs and TPUs. We propose a WNN application for video detection on Edge devices. We aim to have a similar accuracy to a CNN reducing resources from the device in terms of processing and memory. We performed experiments to show the feasibility of our proposal.