Drishti

Inspiration

Construction is one of the largest industries in the country, yet it remains one of the least monitored in real time. Anyone who has spent time on a construction site knows how much happens during a normal workday that never gets documented. A supervisor can't be everywhere at once, so details like who is actively working, which crew is waiting for materials, or whether someone ignored a safety protocol are often known only to the people who witnessed them. Over time, this has simply become the accepted way of working.

As we explored this problem, one thing stood out. Almost every construction site already has CCTV cameras running throughout the day, but the footage is rarely used unless an incident has already happened. That made us wonder: what if those same cameras could become an active source of insights instead of just a recording system? That question became the starting point for Drishti.

What it does

Drishti transforms footage from existing construction site cameras into information that contractors can use immediately.

It monitors worker activity across different zones and identifies where time is being spent productively and where crews are waiting without progress. Instead of relying on assumptions, contractors receive a clear picture of on-site activity.

At the same time, Drishti continuously watches for safety violations such as missing helmets, missing safety vests, or unauthorized entry into restricted areas. Rather than discovering these issues days later during a manual review, they can be addressed while work is still in progress.

The feature we're most excited about focuses on material management. Drishti observes the stockyard, estimates available material using computer vision, and combines that information with worker idle time. If construction materials like cement or rebar begin running low while waiting time starts increasing, the system recommends reordering before work comes to a halt. Every recommendation is backed by data rather than assumptions.

How we built it

One of our earliest decisions was to keep the scope focused. It would have been tempting to promise a long list of features, but we wanted every feature we included to work well.

We focused on three core capabilities that together tell a complete story: activity and idle time tracking, safety monitoring, and material depletion analysis.

To accelerate development, we built on existing pretrained object detection and tracking models instead of training everything from scratch. At this stage, our priority is to demonstrate a complete working solution on real construction footage rather than develop entirely new computer vision models.

Our current prototype is being developed in stages. We started with safety detection because it provides the most visible demonstration, and we're now integrating idle time tracking followed by the material correlation engine.

Challenges we ran into

The biggest challenge wasn't building any individual computer vision model. The real difficulty was combining two completely different signals, worker idle time and visual material stock levels, into recommendations that contractors could genuinely trust.

Another important consideration was privacy and user acceptance. We never wanted Drishti to feel like a surveillance tool designed to monitor individuals. From the beginning, we chose to focus on zone-level insights and aggregated productivity metrics instead of identifying or ranking workers.

Construction sites also present difficult visual conditions. Lighting changes throughout the day, cameras are positioned for security rather than AI analysis, and workers or equipment frequently block the view. Designing a system that performs reliably under these conditions required much more effort than we initially expected.

Accomplishments that we're proud of

One achievement we're particularly proud of is the way the idea evolved. We didn't settle for the first concept we came up with. Instead, we repeatedly refined the scope until every feature contributed to a single, well-connected product.

We're especially excited about the material depletion module. From our research, we found very few solutions that combine visual inventory estimation with workforce idle time to predict supply shortages. Even better, this works using the cameras that construction sites already have, without requiring additional hardware.

What we learned

Throughout this project, we discovered that construction technology is far more competitive than it initially appears. Several established companies already provide computer vision solutions for progress monitoring, which pushed us to find a genuinely unique direction instead of building another version of an existing product.

We also learned that saying no to features can be just as valuable as adding new ones. Every feature we removed made the remaining product easier to build, easier to explain, and ultimately stronger.

Perhaps the most important lesson was that technology like this must always consider the people using it. Building trust and respecting worker privacy cannot be treated as an afterthought; it has to be part of the design from the very beginning.

What's next for Drishti

Our next goal is to move beyond detecting activity and begin recognizing specific construction tasks. Instead of simply knowing that work is happening, Drishti will understand what type of work is being performed.

After that, we plan to introduce workforce pairing recommendations and intelligent scheduling that adapts automatically whenever delays begin to develop.

Looking further ahead, we want to deploy Drishti on a live construction site alongside an actual contractor. Comparing its insights with traditional manual supervision will allow us to validate whether our productivity analysis and delay predictions perform as effectively in the real world as they do in our demonstrations.