Automatically detect, classify, and count electrical, plumbing, and structural symbols in construction blueprints using computer vision — replacing hours of manual takeoff work.
Client Overview
A mid-sized construction and cost estimation firm was performing manual quantity takeoff on complex construction blueprints for commercial and residential projects. Quantity takeoff — the process of counting and cataloguing every symbol, fixture, and component represented in architectural, electrical, plumbing, and mechanical drawings — is a foundational step in construction cost estimation. The firm's estimators were spending between 20 and 40 hours per project manually working through multi-page blueprint sets, counting symbols such as electrical outlets, lighting fixtures, plumbing fixtures, structural support columns, and fire suppression heads, and recording counts in estimation spreadsheets. This process was not only slow but prone to error. Counting hundreds of small, visually similar symbols across dozens of blueprint pages introduced significant human error risk, particularly for large commercial projects where a single page might contain several hundred instances of the same symbol. Miscounts at the takeoff stage propagated directly into cost estimation errors that could result in underbidding — exposing the firm to margin loss on won contracts — or overbidding, with lost opportunities to competitors. The firm was bidding on an increasing number of projects simultaneously, creating pressure on estimators who could not scale their individual throughput. Faster turnaround on preliminary estimates was also becoming a competitive differentiator, as clients increasingly selected estimation firms partly on their ability to deliver preliminary pricing quickly. The manual takeoff bottleneck was limiting both the volume of bids the firm could submit and the speed at which those bids could be delivered.
Our Approach
Zentric Solutions built a computer vision symbol detection and counting system using YOLOv8 trained specifically on construction blueprint symbol libraries. The model was trained on a dataset of annotated blueprints covering electrical, plumbing, mechanical, and structural drawing types, with each symbol class — outlets, switches, lighting fixtures, plumbing fixtures, valves, columns, and more — annotated across diverse blueprint styles, scales, and print quality conditions. Blueprint PDF files were parsed and converted to high-resolution image tiles, with the detection model applied to each tile and results aggregated across pages. The system automatically produced a structured count table for each uploaded blueprint set, listing detected symbol types, counts per page, and totals across the full document set. Detection results were presented in a React dashboard that displayed the original blueprint page with bounding box overlays on every detected symbol, allowing estimators to visually review and verify detections and make manual corrections to edge cases before exporting. Corrections made by estimators were fed back into the training dataset incrementally, continuously improving model accuracy over time on the specific symbol styles encountered in the firm's client blueprints. Export functionality produced count tables in formats compatible with the firm's existing estimation software, eliminating manual re-entry of takeoff data. Blueprint sets that previously required a full day or more of estimator time were processed by the system in under 30 minutes, with estimator review and correction of edge cases adding at most another hour on complex projects. The firm was able to bid on significantly more projects simultaneously and consistently delivered preliminary estimates faster than their competitors, contributing to a measurable improvement in bid win rate.
Tech Stack
Project Tags
Everything you need to know about this project and our approach.
Zentric Solutions delivers cutting-edge digital products that streamline operations, enhance engagement, and drive lasting growth.
