VideoSift: Analysis of Video for Retail

Client Company

Shopping Malls in Russia and Algeria

Challenge

One of our clients, a shopping mall in Algeria, was looking for visual statistics to find the best spots to place tea and coffee terminals, ATMs, and OOH billboards.

Another shopping mall we worked with later had a more challenging task: they wanted to track users not only visually but also process their mobile data and purchase history to offer personalized mobile ads when clients entered the mall.

Solution Description

A solution that analyzes IP-camera videos from shopping malls in real-time to recognize and count every visitor in view; collect statistics on the visitors for further analysis and reporting, and build visits heat map.

VideoSift addresses the clients’ daily needs by providing:

Visualization of visitors’ movements across the mall
Statistics on visitors entry hours, identification of peak hours and conversions into buyers at any moment
Statistics on all clients for any timeframe for making the comparative study to achieve the best traffic monetization

Technology background

The system is based on Neural Networks and Machine Learning, the main functions include

Video Processing
based on an open-cv library (a classical pipeline for video processing)
Object Detection
a fast open-source solution built on http://dlib.net/
Person Recognition
use of machine-learning models based on TensorFlow framework
Unique Counting
proprietary hash function algorithm for further advanced identification
Gender and Age Identification
based on neural networks

Results and Current Status

The application has proven its reliability for two retail clients, and Inventale is currently looking for business opportunities for it in the following fields: