Visual Behavior Analysis in Retail Scenario

Abstract

The retail world is today highly competitive and has seen its logics completely revolutionized by the introduction of e-commerce that have prompted a reaction from the retail market, requiring greater attention to the consumer. We therefore moved from the world of traditional marketing (generic flyer) to that of 1to1 marketing (specific attention to the customer, profiling and personalization of the assortment offer). In this context the need arises to introduce innovative tools that can allow the physical sales spaces to be kept competitive, interacting more with the customer in order to create a more relevant commercial proposal. As a consequence, the computer vision represented one of the possible means to carry out the behavioral analysis of the consumer useful for dynamically adapting the assortment proposal. DOOH (Digital Out Of Home) in its most widespread form of interactive point-of sale kiosks is one of the best tools to get in touch with the customer, create a synergy with him, listen to his needs in order to improve the offer, the level of service and therefore customer satisfaction. Next to DOOH, it is necessary to introduce further and time-continuous monitoring tools, which map the entire customer's shopping experience into the point of sale. For this purpose the egocentric vision is introduced through the use of cam narratives on board the trolleys, which allow a timely story of the consumer, called Visual Market Basket Analysis (evolution of Market Basket Analysis), which generates process functional alerts to the improvement of the service offered. The story of these approaches is provided in this PhD thesis, which tells the three-year course carried out, its experiments and possible future developments. This study has been conducted thanks to the support of Centro Studi S.r.l., a sister company of a privately owned consumer goods distribution company called Orizzonti Holding Group, located in southern Italy. The study has been implemented through an industrial application approach, in a real context (Futura Supermarkets). Consequently, the PhD thesis has considered the typical difficulties of a challenging environment, starting from the creation and acquisition of a dataset to the integration of the approach in the current business processes.