A retail company, whose name is under a non-disclosure agreement, decided to get the most out of modern technology and create a system that, using computer vision, could recognize and classify objects in the image. And it was not about recognizing clothes items from a catalog photo.
We attempted at a large-scale project aiming to analyze the store visitors using video surveillance and computer vision. Looking ahead we must say that the result was not long in coming because soon our customer increased his revenue by 16% using our software.
Do you want to know how?
Our immediate task was to teach the system to recognize and classify details in real-time using machine learning. We needed to create a model that would recognize all available objects in the image for future analyzing. Object detection and segmentation were supposed to define the characteristics of the general traffic of visitors including volume and time of entrance/exit of visitors on the territory of the trading hall.
Sleeves rolled up, developers’ minds sharpened, the following technologies are chosen:
- Deep Learning,
- Computer Vision – Python,
- Machine Learning – Python,
And the path began.
94 days of diligent work resulted in detection with 98% classification accuracy. We created a model that recognizes all available objects from COCO datasets and reached high accuracy in the segmentation of objects and change their background.
Our model recognizes the individual characteristics of visitors. It identifies regular customers and their preferences and client traffic maps through the trading floor area.
As a result of our collaboration, our customer gets a better understanding of marketing and boosted its revenues by 16%. He also plans to enhance this solution for the whole network of shops.