Construction Site Monitoring with Computer Vision

ML-based construction project tracking system analyzing photos and drawings

Overview

During my internship at Magikkraft (IIM Ahmedabad Incubatee) from July to August 2025, I worked on developing a comprehensive computer vision and machine learning-based construction project tracking system that analyzes site photos and engineering drawings to monitor project progress.

Problem Statement

Construction project monitoring typically requires manual inspection and comparison of site progress against engineering drawings. This process is time-consuming, error-prone, and doesn’t scale well for large projects. The challenge was to develop an automated system that could:

  • Detect and count construction elements from site photos
  • Compare real-world progress with 2D engineering drawings
  • Provide accurate tracking for Oil & Gas plant construction projects
  • Handle various object detection tasks with high accuracy

Methodology

The project involved several key components:

  • Object Detection for Steel Pipes: Developed a CV model for detecting steel pipes on Oil & Gas Plant Construction sites using EfficientDet and YOLO architectures
  • Drawing-to-Photo Comparison: Prepared a Vision Model for comparing 2D engineering drawings with actual site photographs
  • Architecture Exploration: Explored and evaluated different architectures for object detection and counting tasks
  • CAD File Interpretation: Developed models for interpreting and extracting information from CAD files

Technologies

  • Deep Learning Frameworks: PyTorch, TensorFlow
  • Object Detection: EfficientDet, YOLO (YOLOv5, YOLOv8)
  • Computer Vision: OpenCV, PIL
  • CAD Processing: Custom parsers for engineering drawing formats
  • Data Processing: pandas, NumPy
  • Model Training: Weights & Biases for experiment tracking

Outcomes

The project successfully delivered:

  • A robust steel pipe detection model with high accuracy for Oil & Gas construction sites
  • An innovative vision system that bridges the gap between engineering drawings and physical construction progress
  • Scalable object detection and counting capabilities for various construction elements
  • Foundation for automated construction project tracking and progress monitoring

This work has practical applications in reducing manual inspection time, improving construction quality assurance, and enabling data-driven project management decisions.