There was an Indian client who was looking for a solution to reliably count people, chairs, and animals from images and videos as manual counting is very time-consuming, tiresome, and inaccurate. The request of the client was to create a tool that would detect the objects and display their number at the current moment without manual counting.
The client wanted a working demonstration that finds and counts people, chairs and animals in photos and videos, shows a label and confidence level for each one, and reports totals for every frame. We built the whole solution, from idea to a tested system.
Our AI engineer trained a YOLO vision model using Python, PyTorch and OpenCV, starting from a ready-made model and teaching it the client’s own scenes. Our data engineer prepared and checked the labelled pictures, our backend developer built the video processing and reports, and our tester compared the counts with manual counts. Our project lead kept communication clear.
The main challenges were hidden chairs in busy rooms, flickering counts, far-away subjects and look-alikes such as dogs and cats. We solved them by training on cluttered scenes, smoothing counts over a short time window and tuning the model for speed. The client now saves much manual effort, gets time-stamped reports and has a solid base that can detect more objects.
Python, ReactJS, PostgreSQL, PyTorch, Docker
YOLO AI/ML Fire, Smoke & Occupancy Monitoring Dashboard
View DetailsYOLO Vision AI Fire & Smoke Detection Solution
View DetailsChat bot developed in AI/ML for Fast Track Sourcing
View DetailsNBA Game prediction model using ML & Blockchain
View Details