Retailers often lack reliable data about who is actually visiting their stores. Traditional methods — surveys, manual counting, or specialized camera systems — are expensive, inaccurate, and difficult to scale across locations.
The goal of this feasibility study was to determine whether existing security camera footage, even from older or low-quality cameras, could be used to generate high-level demographic insights for marketing and operational planning — without facial recognition, cloud processing, or collecting personal data.
We evaluated a local video analysis pipeline that runs entirely on standard on-site hardware. Using modern object detection and vision-language models, the system detects and tracks people entering the store and classifies aggregate demographics such as adults vs. children and broad gender categories — based on full-body visual cues rather than faces.
All processing is performed locally, with no video uploads and no dependency on cloud services. This approach keeps infrastructure costs low while working with existing camera installations.
The study demonstrated that useful, non-identifying demographic data can be extracted reliably from ordinary security footage. Retailers can understand visitor volume, gender distribution, and family shopping patterns over time.
These insights can support decisions around store layout, product placement, staffing, and marketing strategy, while avoiding expensive hardware upgrades or privacy-sensitive technologies.