E-commerce companies are constantly looking for on-boarding new customers. One of the techniques they use is marketing through network of existing customers. An existing customer say - Noah has shared their social network co-ordinates in their profile, it is possible to extract images of people with whom Noah is mostly seen. Specific promotions can be sent to Noah's friends and improve the likelihood of on-boarding them as well.
Use-case:
Expected solution should be able to Utilise recorded video feeds from security cameras typically 3MP to 5MP and night vision enabled.
Leverage images on social media through web-scraping
From this feed identify a registered users or set of visitors (Customers, Store staff, Brand representatives, House-keeping staff)
Record timestamps when each individual was identified and the camera which identified them.
Success criteria:
Accurate - Registered users should be identified in at least 95% of cases
Unbiased - Ability to prove that it works equally well for various demographics (Age, Gender, Race)
Creative - Generate metrics that are insightful for store operations and security team (e.g. anomalies in time of entry or exit)
Scalable - Solution should be able to scale to a store with >10K sq. feet floor space
Extendable - Provide interface for adding more users to the recognition model