Human Part Semantic Segmentation Using CDGNET Architecture for Human Activity Recognition

Abstract

The segmentation of human body parts is a task that entails assigning labels to pixels in an image to identify the corresponding body part classes. To enhance accuracy, a technique known as sample class distribution was developed, considering the hierarchical structure of the human body and the unique positioning of each part. This technique involves gathering and applying primary human parsing labels in both vertical and horizontal dimensions to exploit the distribution of classes. By combining these guided features, a spatial guidance map is generated and incorporated into the backbone network. These semantic-guided features contribute to the effective recognition of human activity through semantic segmentation-enabled human pose. To assess the effectiveness of this approach, extensive experiments were performed on a large dataset called CIHP, using metrics such as mean IOU, pixel accuracy, and mean accuracy.