[3F4-GS-10k] AI応用：行動分析

[3F4-GS-10k-01] Behavior Analysis System for In-Store Displays Using Multiple Deep Learning Models

〇Yuri Nishikawa^1,2, Jun Ozawa^1,2 (1. National Institute of Advanced Industrial Science and Technology, 2. Panasonic Corporation)

Keywords:Behavior analysis system, marketing analysis, action localization

It is important in marketing analysis to understand customer behavior in front of in-store displays and their subsequent actions. On the other hand, there are ongoing studies on action detection techniques using deep learning, but what can be obtained by such techniques are primitive action labels such as "standing" and "walking," and further innovations are needed to extract detailed store behavior. In this study, we propose a system that analyzes the status of the display at the entrance of a restaurant and subsequent user behavior by combining a deep learning model for action detection (SlowFast), and a multi-person tracking model (FairMOT). The effectiveness of the system is discussed through evaluation using images of people entering a restaurant.