The client is a Washington, DC-based owns a chain of shopping malls that draws customers attention with its world-class services and various retails shops franchises sell impressive products.
It is imperative for shopping mall owners to obtain an enhanced understanding of the types of data that are available and draw insights from them. Our client asked for an experienced team of software engineers who can analyze the data collected from CCTV footage to understand the customer’s behavioral pattern.
The client’s main motive was to understand its customers and their shopping patterns, which can help the retailers of the mall in growing their businesses.
CCTV records the videos which are excessive in length. Therefore to find out details about a particular activity of the customers or understand their behavioral patterns becomes difficult.
With the recorded videos on CCTV analyzing the actions and studying them isn’t feasible. Therefore we came up with a solution that by using machine learning technology, we can collect data of customers’ movements and gestures that showcase their interests.
At Sigma, we came up with software, EagleEyeViewer that can find out at which place the customers are spending more time and what interests them more. This software can also figure out how much time the customers are spending at the mall, which shops grab their attention, discounts attract them more or do any other marketing strategy grabs their attention. EagleEyeViewer gives alerts about all these customer behavioral patterns to the client.
EagleEyeViewer also gives alerts about any outlandish behavior of the customers, which can say that were there any cases where someone tried to steal something from any of the shops.
We created datasets from the collected CCTV videos and to analyze the customers’ activities. With the help of EagleEyeViewer, we were able to catch the customers’ actions very quickly from a large number of video collections that the CCTV captured.
Through EagleEyeViewer, RNN models were implemented, which proved to be useful in creating and naming the datasets. The accuracy level of fetching the activities using EagleEyeViewer is 91%
- The implementation of LSTM models for creating and naming datasets was a huge and time-consuming task.
- High-END GPU configurations were required to process videos & images, which is a complex task, and besides this, it costs a lot.
- Various CNN and LSTM models like Inception V3.