Presenter(s)

Anurag Mallik

Comments

3:00-4:15, Kennedy Union Ballroom

Download

Download Project (1.0 MB)

Description

Construction sites pose significant risks to workers due to the presence of various hazardous objects. These items can be sharp or blunt, or even cause electrocution. With the common denominator being that all can lead to serious accidents. During construction, all personnel on site must take proper precautions when handling hazardous objects. While larger items are easily visible, smaller tools like nails, hammers, and drill machines often go unnoticed. If these objects are left unattended in active areas, they can lead to life-threatening incidents in construction zones. In this research work, a computer vision algorithm has been proposed to identify construction tools in a construction zone as part of a larger pipeline for construction hazard detection. Specifically, this work focuses on images of construction zones which have been captured from various angles. The Segment Anything Model (SAM) is used to segment these images allowing them to analyze regions of interest. Regions selected for further processing are done by calculating bounding boxes which are based on the segmented areas. Through experimentation, an optimal size of the bounding boxes to reduce the number of boxes to processes showed bounding boxes that capture an area of 3% to 8% of the total image size are typically relevant to detect a hazardous object in a construction scenario. These resulting bounding box areas in the image are fed to a feature extractor, DINOv2 that returns the object feature matrix to feed into a fully connected three-layer neural network classifier. The classifier identifies an object feature set belonging to one of the twenty predefined hazardous objects if the score of the highest probability output node of the classifier is beyond a predefined threshold value.

Publication Date

4-23-2025

Project Designation

Graduate Research

Primary Advisor

Vijayan K. Asari, Theus H. Aspiras, Tam Nguyen

Primary Advisor's Department

Electrical and Computer Engineering

Keywords

Stander Symposium, School of Engineering

Institutional Learning Goals

Practical Wisdom; Community; Diversity

Autonomous Vision System for Hazardous Object Detection in Construction Sites

Share

COinS