Detection of Workers’ Behaviour in the Manufacturing Plant using Deep Learning

Abstract

In the modern manufacturing landscape, optimizing productivity is a paramount challenge, particularly in dynamic, non-concentrative environments where human activities are diverse and complex. Accurately monitoring and analyzing worker behavior is crucial for enhancing manufacturing processes, but traditional methods fall short in these settings due to their reliance on simplistic global image features and manual classification. Addressing this gap, this paper introduces a groundbreaking vision-based capture technology, integrated into a manufacturing monitoring system. This technology significantly advances productivity by providing a nuanced assessment of worker behavior. It departs from conventional approaches by employing gait recognition techniques, which effectively match input sequences with predefined models. This method adeptly navigates the hurdles of data scarcity, diverse human behaviors, and visual variations typical in manufacturing environments. Utilizing machine learning algorithms, our system learns and detects intricate activities from worker behavior sequences, offering a sophisticated analysis of worker efficiency. The primary aim is to quantify human behavior based on learning rates, thereby facilitating improved production control. Our findings are promising, demonstrating an impressive 99% accuracy in behavior detection. This high level of precision underscores the potential of our technology to transform manufacturing productivity and worker monitoring practices.