# Manually classified dataset of leaning and standing personnel images for construction site monitoring and neural network training

## TL;DR

This paper introduces a labeled dataset of construction site images showing workers either standing or leaning, to improve AI models for monitoring safety and activities.

## Contribution

The novel contribution is a manually labeled dataset with bounding boxes for training and validating AI models in construction site monitoring.

## Key findings

- The dataset includes 1,214 images with two classes: standing and leaning personnel.
- Bounding box coordinates are provided for each image to support precise neural network training.
- The dataset is publicly available for academic and industrial use in computer vision and safety monitoring.

## Abstract

This data paper presents a manually labeled dataset of 1,214 images of personnel captured from a construction site using four static cameras. There are two classes, standing and people leaning. The classification is stored in accompanying text files and bounding box coordinates for every image. The compilation was done to support the developing and validation computer vision and AI models for construction site monitoring. This dataset addresses the challenges of finding personnel in different poses within complex construction environments. The resource will enhance construction site safety monitoring and personnel activity analysis by allowing more precise neural network training. The dataset is stored in a public repository, making it openly accessible for academic and industrial purposes regarding computer vision, civil engineering, and workplace safety.