Article ID: 680
This study presents a customer behavior analysis system that integrates multisensor-based event recognition with continuous two-dimensional motion tracking using light detection and ranging (LiDAR) in a retail-like experimental environment. Conventional in-store behavior research has predominantly focused on purchase outcomes or discrete shelf-dwell events, thereby limiting the ability to capture exploratory movements and behavioral transitions that occur before explicit decision-making. To address the aforementioned limitation, this study extends an existing sensor-fusion architecture comprising ultrasonic proximity sensing, load-cell-based pickand-return detection, and gesture sensing by incorporating a LiDAR-based trajectory-tracking module. All sensor streams were synchronized via MQTT-based communication, enabling the reconstruction of a unified shopping context along a temporally aligned fused timeline. Controlled experiments were conducted in a mock store layout, in which six participants navigated the space under clockwise, counter-clockwise, and center-crossing movement conditions, yielding approximately 90,000 LiDAR trajectory samples. Based on these data, multiple mobility indicators, including the total movement distance, dwell ratio, and convex-hull-based exploration area, were quantitatively derived. Furthermore, kernel density estimation was applied to the spatial trajectory data, revealing localized hotspots corresponding to prolonged dwell behavior in front of product shelves. The results indicated that reductions in walking speed, increased dwell density, and the emergence of microlooping movement patterns consistently preceded picking actions. Overall, this study demonstrates that integrating event-based sensing with continuous spatial behavior modeling provides a robust framework for capturing behavioral processes preceding purchase actions in retail environments. The proposed framework contributes not only to retail behavior research but also to the broader field of sensor-driven behavioral analytics by offering a scalable approach for the empirical observation of behavioral patterns derived from movement dynamics.
