A Computer Vision Solution for a Retail Company

Infopulse cognitive computing specialists worked closely with the client’s team for a seamless implementation of the solution according to the set objectives. Our experts provided a full suite of services from the initial environment assessment to system configuration and data science models training.

Operations

Infopulse team enabled the smooth performance of the system with an end-to-end approach to its development and implementation:

• Advised on the hardware choice, including servers and cameras, according to the data processing requirements;
• Located, configured, and calibrated 50+ cameras on-premises to best meet the project objectives;
• Enabled on-premises computing and data processing without any external service connected;
• Continuously researched for state-of-the-art data science models to utilize in the project for the best result;
• Customized the data science models by collecting, labeling, and feeding the models with real-life data from the project site to enable their training;
• Gathered extended data on customer behavior, including the number of people and movement tracking;
• Provided data visualization dashboards, such as people tracking graph and store heatmap;
• Ensured secure on-premises data storage with role-based access;
• Utilized AWS cloud platform for secure backups and a testing environment.

Data Science

Infopulse developed a custom computer vision solution for intelligent customer behavior analytics with features including people detection, recognition, and movement tracking.

People counting:

• Set up a high-precision face camera at the mall entrance with a scanning frequency of 10 frames per second;

• Enabled real-time people tracking and counting once entering the store;

• Facilitated further re-identification with image segmentation and detecting unique features for each visitor.

Flowline analysis:

• Introduced a full-body image re-identification model, which is not sensitive to wearing masks, hoods, scarves, glasses, and allows for precise customer flow analysis even during the pandemic and obligatory wearing of personal protective equipment;

• Utilized a person re-identification model to track customers moving around the store. Collecting the data from different cameras, the solution can track customers’ paths through the sections of the mall;

• Implemented a lazy tracking model to meet the data processing capabilities.

Heatmap:

• Enabled real-time people counting across different areas of the mall;

• Configured scanning frequency with 1 frame per 4 seconds for cost- and resource-efficiency;

• Implemented a computer vision model to detect crowd density within a camera range and count the number of people per frame;

• Enabled heatmap dashboard update with a 10-second interval for real-time data visualization.

Age & Gender Recognition:

• Applied a tailored approach to the computer vision solution to eliminate the side-effects of pandemic protective measures, which include face covers;

• Implemented a custom data set, based on production data, to enable age and gender recognition with an image of the face and upper torso.

Infopulse client leverages their retail strategy with insightful data and real-time analytics provided by an advanced computer vision solution:

• Quick time-to-market: first working version deployed in 3 months after the start;
• Continuous video analysis with 50+ cameras: real-time people counting and movement tracking;
• Lowered energy costs by optimizing customer flow and dynamically adjusting power consumption thanks to the customer flow data analysis;
• Data-driven marketing and promotion, better customer experience and optimized store layout based on customer flow and demographic analysis;
• Enhanced store security by implementing accurate facial recognition of habitual criminals and sending immediate alerts to the store staff;
• High precision of computer vision models:
  • 95% for people counting;
  • 72% for movement tracking and re-identification;
  • 89% for age & gender detection;
  • 90% for blacklists;
  • 91% for heatmap.

Satisfied with the quality of the Infopulse project delivery and our dedication to the project, the client plans to continue our cooperation and further develop the AI-powered surveillance solution.