VARSew: A Literature Review and Benchmark Dataset for Visual Recognition in Garment Sewing

Human action recognition (HAR)is the method of analyzing human behavior with the aid of computer and machine vision technology. Different application domains have seen substantial recent advancements in this field of study. However, the absence of suitable, sizeable trained datasets still limits HAR in several areas of domains. Many factors impact the availability of suitable HAR data in specific domains. One of those interesting industrial domains is garment sewing. Many world reports were published to illustrate the progress of the worldwide garment sewing industry. To our knowledge, HAR data for the garment sewing industry is limited. Thus, this paper reviews recent works in HAR approaches and proposes a new HAR dataset, called VARSew, specifically designed for recognizing human sewing actions during production time. VARSew consists of 3,121 high-resolution videos of around 49,936 frames. The dataset captures the actions of 34 human Sewing Machine Operators and Maintenance Machine Operators, offering research opportunities for both binary and multi-class classification problems.